diff options
| author | Allan Sandfeld Jensen <allan.jensen@qt.io> | 2020-10-29 10:46:47 +0100 |
|---|---|---|
| committer | Allan Sandfeld Jensen <allan.jensen@qt.io> | 2020-11-02 12:02:10 +0000 |
| commit | 99677208ff3b216fdfec551fbe548da5520cd6fb (patch) | |
| tree | 476a4865c10320249360e859d8fdd3e01833b03a /chromium/third_party/dawn | |
| parent | c30a6232df03e1efbd9f3b226777b07e087a1122 (diff) | |
| download | qtwebengine-chromium-99677208ff3b216fdfec551fbe548da5520cd6fb.tar.gz | |
BASELINE: Update Chromium to 86.0.4240.124
Change-Id: Ide0ff151e94cd665ae6521a446995d34a9d1d644
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Diffstat (limited to 'chromium/third_party/dawn')
270 files changed, 8680 insertions, 4295 deletions
diff --git a/chromium/third_party/dawn/DEPS b/chromium/third_party/dawn/DEPS index 13c8d7911a7..e9550496188 100644 --- a/chromium/third_party/dawn/DEPS +++ b/chromium/third_party/dawn/DEPS @@ -1,5 +1,9 @@ use_relative_paths = True -use_relative_hooks = True + +gclient_gn_args_file = 'build/config/gclient_args.gni' +gclient_gn_args = [ + 'mac_xcode_version', +] vars = { 'chromium_git': 'https://chromium.googlesource.com', @@ -8,24 +12,25 @@ vars = { 'swiftshader_git': 'https://swiftshader.googlesource.com', 'dawn_standalone': True, + + # This can be overridden, e.g. with custom_vars, to download a nonstandard + # Xcode version in build/mac_toolchain.py instead of downloading the + # prebuilt pinned revision. + 'mac_xcode_version': 'default', } deps = { # Dependencies required to use GN/Clang in standalone 'build': { - 'url': '{chromium_git}/chromium/src/build@896323eeda1bd1b01156b70625d5e14de225ebc3', + 'url': '{chromium_git}/chromium/src/build@b8f14c09b76ae3bd6edabe45105527a97e1e16bd', 'condition': 'dawn_standalone', }, 'buildtools': { - 'url': '{chromium_git}/chromium/src/buildtools@2c41dfb19abe40908834803b6fed797b0f341fe1', + 'url': '{chromium_git}/chromium/src/buildtools@eb3987ec709b39469423100c1e77f0446890e059', 'condition': 'dawn_standalone', }, 'tools/clang': { - 'url': '{chromium_git}/chromium/src/tools/clang@698732d5db36040c07d5cc5f9137fcc943494c11', - 'condition': 'dawn_standalone', - }, - 'third_party/binutils': { - 'url': '{chromium_git}/chromium/src/third_party/binutils@f9ce777698a819dff4d6a033b31122d91a49b62e', + 'url': '{chromium_git}/chromium/src/tools/clang@d027d75e8dd91140115a4cc9c7c3598c44bbf634', 'condition': 'dawn_standalone', }, 'tools/clang/dsymutil': { @@ -61,31 +66,31 @@ deps = { # SPIRV-Cross 'third_party/spirv-cross': { - 'url': '{chromium_git}/external/github.com/KhronosGroup/SPIRV-Cross@2e7a5625835380857e785716efd8b2720bf2c84a', + 'url': '{chromium_git}/external/github.com/KhronosGroup/SPIRV-Cross@4c7944bb4260ab0466817c932a9673b6cf59438e', 'condition': 'dawn_standalone', }, # SPIRV compiler dependencies: SPIRV-Tools, SPIRV-headers, glslang and shaderc 'third_party/SPIRV-Tools': { - 'url': '{chromium_git}/external/github.com/KhronosGroup/SPIRV-Tools@52a5f074e9bb6712487653cf360771e98a1ebe97', + 'url': '{chromium_git}/external/github.com/KhronosGroup/SPIRV-Tools@1023dd7a04be15064188d0e511e1708ef1c5af4a', 'condition': 'dawn_standalone', }, 'third_party/spirv-headers': { - 'url': '{chromium_git}/external/github.com/KhronosGroup/SPIRV-Headers@11d7637e7a43cd88cfd4e42c99581dcb682936aa', + 'url': '{chromium_git}/external/github.com/KhronosGroup/SPIRV-Headers@3fdabd0da2932c276b25b9b4a988ba134eba1aa6', 'condition': 'dawn_standalone', }, 'third_party/glslang': { - 'url': '{chromium_git}/external/github.com/KhronosGroup/glslang@839704450200e407490c538418f4d1a493b789ab', + 'url': '{chromium_git}/external/github.com/KhronosGroup/glslang@f257e0ea6b9aeab2dc7af3207ac6d29d2bbc01d0', 'condition': 'dawn_standalone', }, 'third_party/shaderc': { - 'url': '{chromium_git}/external/github.com/google/shaderc@caa519ca532a6a3a0279509fce2ceb791c4f4651', + 'url': '{chromium_git}/external/github.com/google/shaderc@21b36f7368092216ecfaa017e95c383c2ed9db70', 'condition': 'dawn_standalone', }, # WGSL support 'third_party/tint': { - 'url': '{dawn_git}/tint@0c647a8896d127ba2526f18ed034f48bb35c878a', + 'url': '{dawn_git}/tint@b08e25388629f5a8f467b726ace18df529bba3ef', 'condition': 'dawn_standalone', }, @@ -101,19 +106,13 @@ deps = { 'condition': 'dawn_standalone', }, - # Our own pre-compiled Linux clang-format 7.0 for presubmit - 'third_party/clang-format': { - 'url': '{dawn_git}/clang-format@2451c56cd368676cdb230fd5ad11731ab859f1a3', - 'condition': 'dawn_standalone and checkout_linux', - }, - # Khronos Vulkan headers, validation layers and loader. 'third_party/vulkan-headers': { - 'url': '{chromium_git}/external/github.com/KhronosGroup/Vulkan-Headers@09531f27933bf04bffde9074acb302e026e8f181', + 'url': '{chromium_git}/external/github.com/KhronosGroup/Vulkan-Headers@4c079bf40c2587220dbf157d825d3185c9adc896', 'condition': 'dawn_standalone', }, 'third_party/vulkan-validation-layers': { - 'url': '{chromium_git}/external/github.com/KhronosGroup/Vulkan-ValidationLayers@31bdb2d4376919f5964dd8314dd2192810f08b36', + 'url': '{chromium_git}/external/github.com/KhronosGroup/Vulkan-ValidationLayers@e8b96e86fe2edfaee274b98fbbe1bd65579b0904', 'condition': 'dawn_standalone', }, 'third_party/vulkan-loader': { @@ -122,51 +121,13 @@ deps = { }, 'third_party/swiftshader': { - 'url': '{swiftshader_git}/SwiftShader@d25ce872522427bd7a957acf7126dc5aae9d26fc', + 'url': '{swiftshader_git}/SwiftShader@e8dd233c7a85f3c689caf06c226a7f8405a480d3', 'condition': 'dawn_standalone', }, } hooks = [ - # Pull clang-format binaries using checked-in hashes. - { - 'name': 'clang_format_win', - 'pattern': '.', - 'condition': 'host_os == "win" and dawn_standalone', - 'action': [ 'download_from_google_storage', - '--no_resume', - '--platform=win32', - '--no_auth', - '--bucket', 'chromium-clang-format', - '-s', 'buildtools/win/clang-format.exe.sha1', - ], - }, - { - 'name': 'clang_format_mac', - 'pattern': '.', - 'condition': 'host_os == "mac" and dawn_standalone', - 'action': [ 'download_from_google_storage', - '--no_resume', - '--platform=darwin', - '--no_auth', - '--bucket', 'chromium-clang-format', - '-s', 'buildtools/mac/clang-format.sha1', - ], - }, - { - 'name': 'clang_format_linux', - 'pattern': '.', - 'condition': 'host_os == "linux" and dawn_standalone', - 'action': [ 'download_from_google_storage', - '--no_resume', - '--platform=linux*', - '--no_auth', - '--bucket', 'chromium-clang-format', - '-s', 'buildtools/linux64/clang-format.sha1', - ], - }, - # Pull the compilers and system libraries for hermetic builds { 'name': 'sysroot_x86', @@ -208,20 +169,11 @@ hooks = [ '-s', 'build/toolchain/win/rc/win/rc.exe.sha1', ], }, - # Pull binutils for linux hermetic builds - { - 'name': 'binutils', - 'pattern': 'src/third_party/binutils', - 'condition': 'host_os == "linux" and dawn_standalone', - 'action': [ - 'python', - 'third_party/binutils/download.py', - ], - }, # Update build/util/LASTCHANGE. { 'name': 'lastchange', 'pattern': '.', + 'condition': 'dawn_standalone', 'action': ['python', 'build/util/lastchange.py', '-o', 'build/util/LASTCHANGE'], }, diff --git a/chromium/third_party/dawn/PRESUBMIT.py b/chromium/third_party/dawn/PRESUBMIT.py index ae10063e7d5..4ad052cb994 100644 --- a/chromium/third_party/dawn/PRESUBMIT.py +++ b/chromium/third_party/dawn/PRESUBMIT.py @@ -16,54 +16,21 @@ import os import platform import subprocess -def _DoClangFormat(input_api, output_api): - # Our binary clang-format is a linux binary compiled for x64 - if platform.system() != 'Linux' or platform.architecture()[0] != '64bit': - return [output_api.PresubmitNotifyResult('Skipping clang-format')] - - # We need to know which commit to diff against. It doesn't seem to be exposed anywhere - # except in that private member of presubmit_support.Change. This is likely to break - # but hopefully we have an updated clang-format in CPID/GS before it does. - upstream_commit = input_api.change._upstream - if upstream_commit == None: - return [] - - lint_cmd = [ - 'scripts/lint_clang_format.sh', - 'third_party/clang-format/clang-format', - upstream_commit - ] - - # Make clang-format use our linux x64 sysroot because it is compiled with a version of - # stdlibc++ that's incompatible with the old libraries present on the bots. - env = { - 'LD_LIBRARY_PATH': os.path.join( - os.getcwd(), - 'build', - 'linux', - 'debian_sid_amd64-sysroot', - 'usr', - 'lib', - 'x86_64-linux-gnu' - ) - } - - # Call the linting script and forward the output as a notification or as an error - try: - output = subprocess.check_output(lint_cmd, env=env); - return [output_api.PresubmitNotifyResult(output)] - except subprocess.CalledProcessError as e: - return [output_api.PresubmitError(e.output)] def _DoCommonChecks(input_api, output_api): results = [] - results.extend(input_api.canned_checks.CheckChangedLUCIConfigs(input_api, output_api)) - results.extend(input_api.canned_checks.CheckGNFormatted(input_api, output_api)) - results.extend(_DoClangFormat(input_api, output_api)) + results.extend( + input_api.canned_checks.CheckChangedLUCIConfigs(input_api, output_api)) + results.extend( + input_api.canned_checks.CheckPatchFormatted(input_api, + output_api, + check_python=True)) return results + def CheckChangeOnUpload(input_api, output_api): return _DoCommonChecks(input_api, output_api) + def CheckChangeOnCommit(input_api, output_api): return _DoCommonChecks(input_api, output_api) diff --git a/chromium/third_party/dawn/dawn.json b/chromium/third_party/dawn/dawn.json index f313dcf851e..791763c92d8 100644 --- a/chromium/third_party/dawn/dawn.json +++ b/chromium/third_party/dawn/dawn.json @@ -92,11 +92,11 @@ {"name": "visibility", "type": "shader stage"}, {"name": "type", "type": "binding type"}, {"name": "has dynamic offset", "type": "bool", "default": "false"}, + {"name": "min buffer binding size", "type": "uint64_t", "default": "0"}, {"name": "multisampled", "type": "bool", "default": "false"}, {"name": "view dimension", "type": "texture view dimension", "default": "undefined"}, {"name": "texture component type", "type": "texture component type", "default": "float"}, - {"name": "storage texture format", "type": "texture format", "default": "undefined"}, - {"name": "min buffer binding size", "type": "uint64_t", "default": "0"} + {"name": "storage texture format", "type": "texture format", "default": "undefined"} ] }, "bind group layout descriptor": { @@ -117,9 +117,9 @@ {"value": 3, "name": "sampler"}, {"value": 4, "name": "comparison sampler"}, {"value": 5, "name": "sampled texture"}, - {"value": 6, "name": "storage texture"}, - {"value": 7, "name": "readonly storage texture"}, - {"value": 8, "name": "writeonly storage texture"} + {"value": 6, "name": "readonly storage texture"}, + {"value": 7, "name": "writeonly storage texture"}, + {"value": 8, "name": "storage texture"} ] }, "blend descriptor": { @@ -176,14 +176,6 @@ "category": "object", "methods": [ { - "name": "set sub data", - "args": [ - {"name": "start", "type": "uint64_t"}, - {"name": "count", "type": "uint64_t"}, - {"name": "data", "type": "void", "annotation": "const*", "length": "count"} - ] - }, - { "name": "map read async", "args": [ {"name": "callback", "type": "buffer map read callback"}, @@ -198,6 +190,32 @@ ] }, { + "name": "map async", + "args": [ + {"name": "mode", "type": "map mode"}, + {"name": "offset", "type": "size_t"}, + {"name": "size", "type": "size_t"}, + {"name": "callback", "type": "buffer map callback"}, + {"name": "userdata", "type": "void", "annotation": "*"} + ] + }, + { + "name": "get mapped range", + "returns": "void *", + "args": [ + {"name": "offset", "type": "size_t", "default": 0}, + {"name": "size", "type": "size_t", "default": 0} + ] + }, + { + "name": "get const mapped range", + "returns": "void const *", + "args": [ + {"name": "offset", "type": "size_t", "default": 0}, + {"name": "size", "type": "size_t", "default": 0} + ] + }, + { "name": "unmap" }, { @@ -209,10 +227,8 @@ "category": "structure", "extensible": true, "members": [ - {"name": "buffer", "type": "buffer"}, - {"name": "offset", "type": "uint64_t", "default": 0}, - {"name": "bytes per row", "type": "uint32_t"}, - {"name": "rows per image", "type": "uint32_t", "default": 0} + {"name": "layout", "type": "texture data layout"}, + {"name": "buffer", "type": "buffer"} ] }, "buffer descriptor": { @@ -221,7 +237,15 @@ "members": [ {"name": "label", "type": "char", "annotation": "const*", "length": "strlen", "optional": true}, {"name": "usage", "type": "buffer usage"}, - {"name": "size", "type": "uint64_t"} + {"name": "size", "type": "uint64_t"}, + {"name": "mapped at creation", "type": "bool", "default": "false"} + ] + }, + "buffer map callback": { + "category": "callback", + "args": [ + {"name": "status", "type": "buffer map async status"}, + {"name": "userdata", "type": "void", "annotation": "*"} ] }, "buffer map read callback": { @@ -263,7 +287,8 @@ {"value": 32, "name": "vertex"}, {"value": 64, "name": "uniform"}, {"value": 128, "name": "storage"}, - {"value": 256, "name": "indirect"} + {"value": 256, "name": "indirect"}, + {"value": 512, "name": "query resolve"} ] }, "char": { @@ -371,7 +396,7 @@ { "name": "insert debug marker", "args": [ - {"name": "group label", "type": "char", "annotation": "const*", "length": "strlen"} + {"name": "marker label", "type": "char", "annotation": "const*", "length": "strlen"} ] }, { @@ -383,6 +408,23 @@ "args": [ {"name": "group label", "type": "char", "annotation": "const*", "length": "strlen"} ] + }, + { + "name": "resolve query set", + "args": [ + {"name": "query set", "type": "query set"}, + {"name": "first query", "type": "uint32_t"}, + {"name": "query count", "type": "uint32_t"}, + {"name": "destination", "type": "buffer"}, + {"name": "destination offset", "type": "uint64_t"} + ] + }, + { + "name": "write timestamp", + "args": [ + {"name": "query set", "type": "query set"}, + {"name": "query index", "type": "uint32_t"} + ] } ] }, @@ -396,7 +438,7 @@ "compare function": { "category": "enum", "values": [ - {"value": 0, "name": "undefined"}, + {"value": 0, "name": "undefined", "jsrepr": "undefined"}, {"value": 1, "name": "never"}, {"value": 2, "name": "less"}, {"value": 3, "name": "less equal"}, @@ -420,7 +462,7 @@ { "name": "insert debug marker", "args": [ - {"name": "group label", "type": "char", "annotation": "const*", "length": "strlen"} + {"name": "marker label", "type": "char", "annotation": "const*", "length": "strlen"} ] }, { @@ -449,6 +491,13 @@ ] }, { + "name": "write timestamp", + "args": [ + {"name": "query set", "type": "query set"}, + {"name": "query index", "type": "uint32_t"} + ] + }, + { "name": "dispatch", "args": [ {"name": "x", "type": "uint32_t"}, @@ -548,17 +597,17 @@ ] }, { - "name": "create render pipeline", - "returns": "render pipeline", + "name": "create pipeline layout", + "returns": "pipeline layout", "args": [ - {"name": "descriptor", "type": "render pipeline descriptor", "annotation": "const*"} + {"name": "descriptor", "type": "pipeline layout descriptor", "annotation": "const*"} ] }, { - "name": "create pipeline layout", - "returns": "pipeline layout", + "name": "create query set", + "returns": "query set", "args": [ - {"name": "descriptor", "type": "pipeline layout descriptor", "annotation": "const*"} + {"name": "descriptor", "type": "query set descriptor", "annotation": "const*"} ] }, { @@ -569,6 +618,13 @@ ] }, { + "name": "create render pipeline", + "returns": "render pipeline", + "args": [ + {"name": "descriptor", "type": "render pipeline descriptor", "annotation": "const*"} + ] + }, + { "name": "create sampler", "returns": "sampler", "args": [ @@ -642,13 +698,6 @@ {"name": "callback", "type": "error callback"}, {"name": "userdata", "type": "void", "annotation": "*"} ] - }, - { - "name": "create query set", - "returns": "query set", - "args": [ - {"name": "descriptor", "type": "query set descriptor", "annotation": "const*"} - ] } ] }, @@ -840,6 +889,14 @@ {"value": 1, "name": "load"} ] }, + "map mode": { + "category": "bitmask", + "values": [ + {"value": 0, "name": "none"}, + {"value": 1, "name": "read"}, + {"value": 2, "name": "write"} + ] + }, "store op": { "category": "enum", "values": [ @@ -867,7 +924,7 @@ {"name": "bind group layouts", "type": "bind group layout", "annotation": "const*", "length": "bind group layout count"} ] }, - "pipeline statistics name": { + "pipeline statistic name": { "category": "enum", "values": [ {"value": 0, "name": "vertex shader invocations"}, @@ -918,8 +975,8 @@ {"name": "label", "type": "char", "annotation": "const*", "length": "strlen", "optional": true}, {"name": "type", "type": "query type"}, {"name": "count", "type": "uint32_t"}, - {"name": "pipeline statistics count", "type": "uint32_t", "default": "0"}, - {"name": "pipeline statistics", "type": "pipeline statistics name", "annotation": "const*", "length": "pipeline statistics count"} + {"name": "pipeline statistics", "type": "pipeline statistic name", "annotation": "const*", "length": "pipeline statistics count"}, + {"name": "pipeline statistics count", "type": "uint32_t", "default": "0"} ] }, "query type": { @@ -962,6 +1019,16 @@ {"name": "data", "type": "void", "annotation": "const*", "length": "size"}, {"name": "size", "type": "size_t"} ] + }, + { + "name": "write texture", + "args": [ + {"name": "destination", "type": "texture copy view", "annotation": "const*"}, + {"name": "data", "type": "void", "annotation": "const*", "length": "data size"}, + {"name": "data size", "type": "size_t"}, + {"name": "data layout", "type": "texture data layout", "annotation": "const*"}, + {"name": "write size", "type": "extent 3D", "annotation": "const*"} + ] } ] }, @@ -1036,7 +1103,7 @@ { "name": "insert debug marker", "args": [ - {"name": "group label", "type": "char", "annotation": "const*", "length": "strlen"} + {"name": "marker label", "type": "char", "annotation": "const*", "length": "strlen"} ] }, { @@ -1114,9 +1181,11 @@ {"name": "depth load op", "type": "load op"}, {"name": "depth store op", "type": "store op"}, {"name": "clear depth", "type": "float"}, + {"name": "depth read only", "type": "bool", "default": "false"}, {"name": "stencil load op", "type": "load op"}, {"name": "stencil store op", "type": "store op"}, - {"name": "clear stencil", "type": "uint32_t", "default": "0"} + {"name": "clear stencil", "type": "uint32_t", "default": "0"}, + {"name": "stencil read only", "type": "bool", "default": "false"} ] }, @@ -1127,7 +1196,8 @@ {"name": "label", "type": "char", "annotation": "const*", "length": "strlen", "optional": true}, {"name": "color attachment count", "type": "uint32_t"}, {"name": "color attachments", "type": "render pass color attachment descriptor", "annotation": "const*", "length": "color attachment count"}, - {"name": "depth stencil attachment", "type": "render pass depth stencil attachment descriptor", "annotation": "const*", "optional": true} + {"name": "depth stencil attachment", "type": "render pass depth stencil attachment descriptor", "annotation": "const*", "optional": true}, + {"name": "occlusion query set", "type": "query set", "optional": true} ] }, "render pass encoder": { @@ -1191,7 +1261,7 @@ { "name": "insert debug marker", "args": [ - {"name": "group label", "type": "char", "annotation": "const*", "length": "strlen"} + {"name": "marker label", "type": "char", "annotation": "const*", "length": "strlen"} ] }, { @@ -1254,6 +1324,13 @@ ] }, { + "name": "write timestamp", + "args": [ + {"name": "query set", "type": "query set"}, + {"name": "query index", "type": "uint32_t"} + ] + }, + { "name": "end pass" } ] @@ -1389,11 +1466,11 @@ {"name": "label", "type": "char", "annotation": "const*", "length": "strlen", "optional": true} ] }, - "surface descriptor from HTML canvas id": { + "surface descriptor from canvas HTML selector": { "category": "structure", "chained": true, "members": [ - {"name": "id", "type": "char", "annotation": "const*", "length": "strlen"} + {"name": "selector", "type": "char", "annotation": "const*", "length": "strlen"} ] }, "surface descriptor from metal layer": { @@ -1459,7 +1536,7 @@ {"value": 1, "name": "surface descriptor from metal layer"}, {"value": 2, "name": "surface descriptor from windows HWND"}, {"value": 3, "name": "surface descriptor from xlib"}, - {"value": 4, "name": "surface descriptor from HTML canvas id"}, + {"value": 4, "name": "surface descriptor from canvas HTML selector"}, {"value": 5, "name": "shader module SPIRV descriptor"}, {"value": 6, "name": "shader module WGSL descriptor"}, {"value": 7, "name": "sampler descriptor dummy anisotropic filtering"}, @@ -1503,8 +1580,17 @@ "members": [ {"name": "texture", "type": "texture"}, {"name": "mip level", "type": "uint32_t", "default": "0"}, - {"name": "array layer", "type": "uint32_t", "default": "0"}, - {"name": "origin", "type": "origin 3D"} + {"name": "origin", "type": "origin 3D"}, + {"name": "aspect", "type": "texture aspect", "default": "all"} + ] + }, + "texture data layout": { + "category": "structure", + "extensible": true, + "members": [ + {"name": "offset", "type": "uint64_t", "default": 0}, + {"name": "bytes per row", "type": "uint32_t"}, + {"name": "rows per image", "type": "uint32_t", "default": 0} ] }, "texture descriptor": { @@ -1515,7 +1601,6 @@ {"name": "usage", "type": "texture usage"}, {"name": "dimension", "type": "texture dimension", "default": "2D"}, {"name": "size", "type": "extent 3D"}, - {"name": "array layer count", "type": "uint32_t", "default": "1"}, {"name": "format", "type": "texture format"}, {"name": "mip level count", "type": "uint32_t", "default": 1}, {"name": "sample count", "type": "uint32_t", "default": 1} @@ -1560,37 +1645,38 @@ {"value": 23, "name": "BGRA8 unorm"}, {"value": 24, "name": "BGRA8 unorm srgb"}, {"value": 25, "name": "RGB10 A2 unorm"}, - {"value": 26, "name": "RG11 B10 float"}, + {"value": 26, "name": "RG11 B10 ufloat"}, + {"value": 27, "name": "RGB9 E5 ufloat"}, - {"value": 27, "name": "RG32 float"}, - {"value": 28, "name": "RG32 uint"}, - {"value": 29, "name": "RG32 sint"}, - {"value": 30, "name": "RGBA16 uint"}, - {"value": 31, "name": "RGBA16 sint"}, - {"value": 32, "name": "RGBA16 float"}, + {"value": 28, "name": "RG32 float"}, + {"value": 29, "name": "RG32 uint"}, + {"value": 30, "name": "RG32 sint"}, + {"value": 31, "name": "RGBA16 uint"}, + {"value": 32, "name": "RGBA16 sint"}, + {"value": 33, "name": "RGBA16 float"}, - {"value": 33, "name": "RGBA32 float"}, - {"value": 34, "name": "RGBA32 uint"}, - {"value": 35, "name": "RGBA32 sint"}, + {"value": 34, "name": "RGBA32 float"}, + {"value": 35, "name": "RGBA32 uint"}, + {"value": 36, "name": "RGBA32 sint"}, - {"value": 36, "name": "depth32 float"}, - {"value": 37, "name": "depth24 plus"}, - {"value": 38, "name": "depth24 plus stencil8"}, + {"value": 37, "name": "depth32 float"}, + {"value": 38, "name": "depth24 plus"}, + {"value": 39, "name": "depth24 plus stencil8"}, - {"value": 39, "name": "BC1 RGBA unorm"}, - {"value": 40, "name": "BC1 RGBA unorm srgb"}, - {"value": 41, "name": "BC2 RGBA unorm"}, - {"value": 42, "name": "BC2 RGBA unorm srgb"}, - {"value": 43, "name": "BC3 RGBA unorm"}, - {"value": 44, "name": "BC3 RGBA unorm srgb"}, - {"value": 45, "name": "BC4 R unorm"}, - {"value": 46, "name": "BC4 R snorm"}, - {"value": 47, "name": "BC5 RG unorm"}, - {"value": 48, "name": "BC5 RG snorm"}, - {"value": 49, "name": "BC6H RGB ufloat"}, - {"value": 50, "name": "BC6H RGB sfloat"}, - {"value": 51, "name": "BC7 RGBA unorm"}, - {"value": 52, "name": "BC7 RGBA unorm srgb"} + {"value": 40, "name": "BC1 RGBA unorm"}, + {"value": 41, "name": "BC1 RGBA unorm srgb"}, + {"value": 42, "name": "BC2 RGBA unorm"}, + {"value": 43, "name": "BC2 RGBA unorm srgb"}, + {"value": 44, "name": "BC3 RGBA unorm"}, + {"value": 45, "name": "BC3 RGBA unorm srgb"}, + {"value": 46, "name": "BC4 R unorm"}, + {"value": 47, "name": "BC4 R snorm"}, + {"value": 48, "name": "BC5 RG unorm"}, + {"value": 49, "name": "BC5 RG snorm"}, + {"value": 50, "name": "BC6H RGB ufloat"}, + {"value": 51, "name": "BC6H RGB sfloat"}, + {"value": 52, "name": "BC7 RGBA unorm"}, + {"value": 53, "name": "BC7 RGBA unorm srgb"} ] }, "texture usage": { @@ -1690,6 +1776,12 @@ "void": { "category": "native" }, + "void *": { + "category": "native" + }, + "void const *": { + "category": "native" + }, "uint32_t": { "category": "native" }, diff --git a/chromium/third_party/dawn/dawn_wire.json b/chromium/third_party/dawn/dawn_wire.json index 5291c3e228c..b2256a5a52a 100644 --- a/chromium/third_party/dawn/dawn_wire.json +++ b/chromium/third_party/dawn/dawn_wire.json @@ -14,26 +14,25 @@ "See the License for the specific language governing permissions and", "limitations under the License." ], + "_todos": [ + "Remove usage of size_t because it is not network transparent" + ], "commands": { "buffer map async": [ { "name": "buffer id", "type": "ObjectId" }, { "name": "request serial", "type": "uint32_t" }, - { "name": "is write", "type": "bool" }, + { "name": "mode", "type": "map mode" }, + { "name": "offset", "type": "size_t"}, + { "name": "size", "type": "size_t"}, { "name": "handle create info length", "type": "uint64_t" }, { "name": "handle create info", "type": "uint8_t", "annotation": "const*", "length": "handle create info length", "skip_serialize": true} ], - "buffer set sub data internal": [ - {"name": "buffer id", "type": "ObjectId" }, - {"name": "start", "type": "uint64_t"}, - {"name": "count", "type": "uint64_t"}, - {"name": "data", "type": "uint8_t", "annotation": "const*", "length": "count"} - ], "buffer update mapped data": [ { "name": "buffer id", "type": "ObjectId" }, { "name": "write flush info length", "type": "uint64_t" }, { "name": "write flush info", "type": "uint8_t", "annotation": "const*", "length": "write flush info length", "skip_serialize": true} ], - "device create buffer mapped": [ + "device create buffer": [ { "name": "device", "type": "device" }, { "name": "descriptor", "type": "buffer descriptor", "annotation": "const*" }, { "name": "result", "type": "ObjectHandle", "handle_type": "buffer" }, @@ -54,20 +53,23 @@ {"name": "buffer offset", "type": "uint64_t"}, {"name": "data", "type": "uint8_t", "annotation": "const*", "length": "size"}, {"name": "size", "type": "size_t"} + ], + "queue write texture internal": [ + {"name": "queue id", "type": "ObjectId" }, + {"name": "destination", "type": "texture copy view", "annotation": "const*"}, + {"name": "data", "type": "uint8_t", "annotation": "const*", "length": "data size"}, + {"name": "data size", "type": "size_t"}, + {"name": "data layout", "type": "texture data layout", "annotation": "const*"}, + {"name": "writeSize", "type": "extent 3D", "annotation": "const*"} ] }, "return commands": { - "buffer map read async callback": [ + "buffer map async callback": [ { "name": "buffer", "type": "ObjectHandle", "handle_type": "buffer" }, { "name": "request serial", "type": "uint32_t" }, { "name": "status", "type": "uint32_t" }, - { "name": "initial data info length", "type": "uint64_t" }, - { "name": "initial data info", "type": "uint8_t", "annotation": "const*", "length": "initial data info length", "skip_serialize": true } - ], - "buffer map write async callback": [ - { "name": "buffer", "type": "ObjectHandle", "handle_type": "buffer" }, - { "name": "request serial", "type": "uint32_t" }, - { "name": "status", "type": "uint32_t" } + { "name": "read initial data info length", "type": "uint64_t" }, + { "name": "read initial data info", "type": "uint8_t", "annotation": "const*", "length": "read initial data info length", "skip_serialize": true } ], "device uncaptured error callback": [ { "name": "type", "type": "error type"}, @@ -94,22 +96,27 @@ "SurfaceDescriptorFromXlib" ], "client_side_commands": [ + "BufferMapAsync", "BufferMapReadAsync", "BufferMapWriteAsync", - "BufferSetSubData", + "BufferGetConstMappedRange", + "BufferGetMappedRange", + "DeviceCreateBuffer", + "DeviceCreateBufferMapped", "DevicePopErrorScope", "DeviceSetDeviceLostCallback", "DeviceSetUncapturedErrorCallback", "FenceGetCompletedValue", "FenceOnCompletion", - "QueueWriteBuffer" + "QueueWriteBuffer", + "QueueWriteTexture" ], "client_handwritten_commands": [ "BufferDestroy", "BufferUnmap", - "DeviceCreateBuffer", - "DeviceCreateBufferMapped", + "DeviceCreateErrorBuffer", "DeviceGetDefaultQueue", + "DeviceInjectError", "DevicePushErrorScope", "QueueCreateFence", "QueueSignal" @@ -117,7 +124,8 @@ "client_special_objects": [ "Buffer", "Device", - "Fence" + "Fence", + "Queue" ], "server_custom_pre_handler_commands": [ "BufferDestroy", diff --git a/chromium/third_party/dawn/examples/Animometer.cpp b/chromium/third_party/dawn/examples/Animometer.cpp index dfd041cabe0..5c197c4cf55 100644 --- a/chromium/third_party/dawn/examples/Animometer.cpp +++ b/chromium/third_party/dawn/examples/Animometer.cpp @@ -18,8 +18,8 @@ #include "utils/SystemUtils.h" #include "utils/WGPUHelpers.h" -#include <cstdlib> #include <cstdio> +#include <cstdlib> #include <vector> wgpu::Device device; @@ -138,8 +138,7 @@ void init() { bufferDesc.usage = wgpu::BufferUsage::CopyDst | wgpu::BufferUsage::Uniform; ubo = device.CreateBuffer(&bufferDesc); - bindGroup = - utils::MakeBindGroup(device, bgl, {{0, ubo, 0, sizeof(ShaderData)}}); + bindGroup = utils::MakeBindGroup(device, bgl, {{0, ubo, 0, sizeof(ShaderData)}}); } void frame() { diff --git a/chromium/third_party/dawn/examples/BUILD.gn b/chromium/third_party/dawn/examples/BUILD.gn index 39ef9c521f7..96ae72fbe9c 100644 --- a/chromium/third_party/dawn/examples/BUILD.gn +++ b/chromium/third_party/dawn/examples/BUILD.gn @@ -49,9 +49,7 @@ static_library("dawn_sample_utils") { # Template for samples to avoid listing dawn_sample_utils as a dep every time template("dawn_sample") { executable(target_name) { - deps = [ - ":dawn_sample_utils", - ] + deps = [ ":dawn_sample_utils" ] forward_variables_from(invoker, "*", [ "deps" ]) if (defined(invoker.deps)) { @@ -61,43 +59,27 @@ template("dawn_sample") { } dawn_sample("CppHelloTriangle") { - sources = [ - "CppHelloTriangle.cpp", - ] + sources = [ "CppHelloTriangle.cpp" ] } dawn_sample("CHelloTriangle") { - sources = [ - "CHelloTriangle.cpp", - ] + sources = [ "CHelloTriangle.cpp" ] } dawn_sample("ComputeBoids") { - sources = [ - "ComputeBoids.cpp", - ] - deps = [ - "${dawn_root}/third_party/gn/glm", - ] + sources = [ "ComputeBoids.cpp" ] + deps = [ "${dawn_root}/third_party/gn/glm" ] } dawn_sample("Animometer") { - sources = [ - "Animometer.cpp", - ] + sources = [ "Animometer.cpp" ] } dawn_sample("CubeReflection") { - sources = [ - "CubeReflection.cpp", - ] - deps = [ - "${dawn_root}/third_party/gn/glm", - ] + sources = [ "CubeReflection.cpp" ] + deps = [ "${dawn_root}/third_party/gn/glm" ] } dawn_sample("ManualSwapChainTest") { - sources = [ - "ManualSwapChainTest.cpp", - ] + sources = [ "ManualSwapChainTest.cpp" ] } diff --git a/chromium/third_party/dawn/examples/CHelloTriangle.cpp b/chromium/third_party/dawn/examples/CHelloTriangle.cpp index 048fba1987d..22d820295c4 100644 --- a/chromium/third_party/dawn/examples/CHelloTriangle.cpp +++ b/chromium/third_party/dawn/examples/CHelloTriangle.cpp @@ -120,7 +120,7 @@ void frame() { { colorAttachment.attachment = backbufferView; colorAttachment.resolveTarget = nullptr; - colorAttachment.clearColor = { 0.0f, 0.0f, 0.0f, 0.0f }; + colorAttachment.clearColor = {0.0f, 0.0f, 0.0f, 0.0f}; colorAttachment.loadOp = WGPULoadOp_Clear; colorAttachment.storeOp = WGPUStoreOp_Store; renderpassInfo.colorAttachmentCount = 1; diff --git a/chromium/third_party/dawn/examples/ComputeBoids.cpp b/chromium/third_party/dawn/examples/ComputeBoids.cpp index 59538dfa651..f958ab4002d 100644 --- a/chromium/third_party/dawn/examples/ComputeBoids.cpp +++ b/chromium/third_party/dawn/examples/ComputeBoids.cpp @@ -67,7 +67,7 @@ void initBuffers() { modelBuffer = utils::CreateBufferFromData(device, model, sizeof(model), wgpu::BufferUsage::Vertex); - SimParams params = { 0.04f, 0.1f, 0.025f, 0.025f, 0.02f, 0.05f, 0.005f, kNumParticles }; + SimParams params = {0.04f, 0.1f, 0.025f, 0.025f, 0.02f, 0.05f, 0.005f, kNumParticles}; updateParams = utils::CreateBufferFromData(device, ¶ms, sizeof(params), wgpu::BufferUsage::Uniform); @@ -75,8 +75,7 @@ void initBuffers() { { std::mt19937 generator; std::uniform_real_distribution<float> dist(-1.0f, 1.0f); - for (auto& p : initialParticles) - { + for (auto& p : initialParticles) { p.pos = glm::vec2(dist(generator), dist(generator)); p.vel = glm::vec2(dist(generator), dist(generator)) * 0.1f; } @@ -253,11 +252,13 @@ void initSim() { updatePipeline = device.CreateComputePipeline(&csDesc); for (uint32_t i = 0; i < 2; ++i) { - updateBGs[i] = utils::MakeBindGroup(device, bgl, { - {0, updateParams, 0, sizeof(SimParams)}, - {1, particleBuffers[i], 0, kNumParticles * sizeof(Particle)}, - {2, particleBuffers[(i + 1) % 2], 0, kNumParticles * sizeof(Particle)}, - }); + updateBGs[i] = utils::MakeBindGroup( + device, bgl, + { + {0, updateParams, 0, sizeof(SimParams)}, + {1, particleBuffers[i], 0, kNumParticles * sizeof(Particle)}, + {2, particleBuffers[(i + 1) % 2], 0, kNumParticles * sizeof(Particle)}, + }); } } diff --git a/chromium/third_party/dawn/examples/CppHelloTriangle.cpp b/chromium/third_party/dawn/examples/CppHelloTriangle.cpp index fbc598daa0e..378afa8db5e 100644 --- a/chromium/third_party/dawn/examples/CppHelloTriangle.cpp +++ b/chromium/third_party/dawn/examples/CppHelloTriangle.cpp @@ -36,15 +36,15 @@ wgpu::BindGroup bindGroup; void initBuffers() { static const uint32_t indexData[3] = { - 0, 1, 2, + 0, + 1, + 2, }; indexBuffer = utils::CreateBufferFromData(device, indexData, sizeof(indexData), wgpu::BufferUsage::Index); static const float vertexData[12] = { - 0.0f, 0.5f, 0.0f, 1.0f, - -0.5f, -0.5f, 0.0f, 1.0f, - 0.5f, -0.5f, 0.0f, 1.0f, + 0.0f, 0.5f, 0.0f, 1.0f, -0.5f, -0.5f, 0.0f, 1.0f, 0.5f, -0.5f, 0.0f, 1.0f, }; vertexBuffer = utils::CreateBufferFromData(device, vertexData, sizeof(vertexData), wgpu::BufferUsage::Vertex); @@ -141,17 +141,19 @@ void init() { wgpu::TextureView view = texture.CreateView(); - bindGroup = utils::MakeBindGroup(device, bgl, { - {0, sampler}, - {1, view} - }); + bindGroup = utils::MakeBindGroup(device, bgl, {{0, sampler}, {1, view}}); } -struct {uint32_t a; float b;} s; +struct { + uint32_t a; + float b; +} s; void frame() { s.a = (s.a + 1) % 256; s.b += 0.02f; - if (s.b >= 1.0f) {s.b = 0.0f;} + if (s.b >= 1.0f) { + s.b = 0.0f; + } wgpu::TextureView backbufferView = swapchain.GetCurrentTextureView(); utils::ComboRenderPassDescriptor renderPass({backbufferView}, depthStencilView); diff --git a/chromium/third_party/dawn/examples/CubeReflection.cpp b/chromium/third_party/dawn/examples/CubeReflection.cpp index becec8728b4..4ff18e003c3 100644 --- a/chromium/third_party/dawn/examples/CubeReflection.cpp +++ b/chromium/third_party/dawn/examples/CubeReflection.cpp @@ -18,10 +18,10 @@ #include "utils/SystemUtils.h" #include "utils/WGPUHelpers.h" -#include <vector> #include <glm/glm.hpp> #include <glm/gtc/matrix_transform.hpp> #include <glm/gtc/type_ptr.hpp> +#include <vector> wgpu::Device device; @@ -43,67 +43,44 @@ wgpu::RenderPipeline planePipeline; wgpu::RenderPipeline reflectionPipeline; void initBuffers() { - static const uint32_t indexData[6*6] = { - 0, 1, 2, - 0, 2, 3, + static const uint32_t indexData[6 * 6] = {0, 1, 2, 0, 2, 3, - 4, 5, 6, - 4, 6, 7, + 4, 5, 6, 4, 6, 7, - 8, 9, 10, - 8, 10, 11, + 8, 9, 10, 8, 10, 11, - 12, 13, 14, - 12, 14, 15, + 12, 13, 14, 12, 14, 15, - 16, 17, 18, - 16, 18, 19, + 16, 17, 18, 16, 18, 19, - 20, 21, 22, - 20, 22, 23 - }; + 20, 21, 22, 20, 22, 23}; indexBuffer = utils::CreateBufferFromData(device, indexData, sizeof(indexData), wgpu::BufferUsage::Index); static const float vertexData[6 * 4 * 6] = { - -1.0, -1.0, 1.0, 1.0, 0.0, 0.0, - 1.0, -1.0, 1.0, 1.0, 0.0, 0.0, - 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, - -1.0, 1.0, 1.0, 1.0, 0.0, 0.0, - - -1.0, -1.0, -1.0, 1.0, 1.0, 0.0, - -1.0, 1.0, -1.0, 1.0, 1.0, 0.0, - 1.0, 1.0, -1.0, 1.0, 1.0, 0.0, - 1.0, -1.0, -1.0, 1.0, 1.0, 0.0, - - -1.0, 1.0, -1.0, 1.0, 0.0, 1.0, - -1.0, 1.0, 1.0, 1.0, 0.0, 1.0, - 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, - 1.0, 1.0, -1.0, 1.0, 0.0, 1.0, - - -1.0, -1.0, -1.0, 0.0, 1.0, 0.0, - 1.0, -1.0, -1.0, 0.0, 1.0, 0.0, - 1.0, -1.0, 1.0, 0.0, 1.0, 0.0, - -1.0, -1.0, 1.0, 0.0, 1.0, 0.0, - - 1.0, -1.0, -1.0, 0.0, 1.0, 1.0, - 1.0, 1.0, -1.0, 0.0, 1.0, 1.0, - 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, - 1.0, -1.0, 1.0, 0.0, 1.0, 1.0, - - -1.0, -1.0, -1.0, 1.0, 1.0, 1.0, - -1.0, -1.0, 1.0, 1.0, 1.0, 1.0, - -1.0, 1.0, 1.0, 1.0, 1.0, 1.0, - -1.0, 1.0, -1.0, 1.0, 1.0, 1.0 - }; + -1.0, -1.0, 1.0, 1.0, 0.0, 0.0, 1.0, -1.0, 1.0, 1.0, 0.0, 0.0, + 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, -1.0, 1.0, 1.0, 1.0, 0.0, 0.0, + + -1.0, -1.0, -1.0, 1.0, 1.0, 0.0, -1.0, 1.0, -1.0, 1.0, 1.0, 0.0, + 1.0, 1.0, -1.0, 1.0, 1.0, 0.0, 1.0, -1.0, -1.0, 1.0, 1.0, 0.0, + + -1.0, 1.0, -1.0, 1.0, 0.0, 1.0, -1.0, 1.0, 1.0, 1.0, 0.0, 1.0, + 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, -1.0, 1.0, 0.0, 1.0, + + -1.0, -1.0, -1.0, 0.0, 1.0, 0.0, 1.0, -1.0, -1.0, 0.0, 1.0, 0.0, + 1.0, -1.0, 1.0, 0.0, 1.0, 0.0, -1.0, -1.0, 1.0, 0.0, 1.0, 0.0, + + 1.0, -1.0, -1.0, 0.0, 1.0, 1.0, 1.0, 1.0, -1.0, 0.0, 1.0, 1.0, + 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, -1.0, 1.0, 0.0, 1.0, 1.0, + + -1.0, -1.0, -1.0, 1.0, 1.0, 1.0, -1.0, -1.0, 1.0, 1.0, 1.0, 1.0, + -1.0, 1.0, 1.0, 1.0, 1.0, 1.0, -1.0, 1.0, -1.0, 1.0, 1.0, 1.0}; vertexBuffer = utils::CreateBufferFromData(device, vertexData, sizeof(vertexData), wgpu::BufferUsage::Vertex); static const float planeData[6 * 4] = { - -2.0, -1.0, -2.0, 0.5, 0.5, 0.5, - 2.0, -1.0, -2.0, 0.5, 0.5, 0.5, - 2.0, -1.0, 2.0, 0.5, 0.5, 0.5, - -2.0, -1.0, 2.0, 0.5, 0.5, 0.5, + -2.0, -1.0, -2.0, 0.5, 0.5, 0.5, 2.0, -1.0, -2.0, 0.5, 0.5, 0.5, + 2.0, -1.0, 2.0, 0.5, 0.5, 0.5, -2.0, -1.0, 2.0, 0.5, 0.5, 0.5, }; planeBuffer = utils::CreateBufferFromData(device, planeData, sizeof(planeData), wgpu::BufferUsage::Vertex); @@ -191,15 +168,13 @@ void init() { transformBuffer[1] = utils::CreateBufferFromData(device, &transform, sizeof(glm::mat4), wgpu::BufferUsage::Uniform); - bindGroup[0] = utils::MakeBindGroup(device, bgl, { - {0, cameraBuffer, 0, sizeof(CameraData)}, - {1, transformBuffer[0], 0, sizeof(glm::mat4)} - }); + bindGroup[0] = utils::MakeBindGroup( + device, bgl, + {{0, cameraBuffer, 0, sizeof(CameraData)}, {1, transformBuffer[0], 0, sizeof(glm::mat4)}}); - bindGroup[1] = utils::MakeBindGroup(device, bgl, { - {0, cameraBuffer, 0, sizeof(CameraData)}, - {1, transformBuffer[1], 0, sizeof(glm::mat4)} - }); + bindGroup[1] = utils::MakeBindGroup( + device, bgl, + {{0, cameraBuffer, 0, sizeof(CameraData)}, {1, transformBuffer[1], 0, sizeof(glm::mat4)}}); depthStencilView = CreateDefaultDepthStencilView(device); @@ -250,17 +225,20 @@ void init() { cameraData.proj = glm::perspective(glm::radians(45.0f), 1.f, 1.0f, 100.0f); } -struct {uint32_t a; float b;} s; +struct { + uint32_t a; + float b; +} s; void frame() { s.a = (s.a + 1) % 256; s.b += 0.01f; - if (s.b >= 1.0f) {s.b = 0.0f;} + if (s.b >= 1.0f) { + s.b = 0.0f; + } - cameraData.view = glm::lookAt( - glm::vec3(8.f * std::sin(glm::radians(s.b * 360.f)), 2.f, 8.f * std::cos(glm::radians(s.b * 360.f))), - glm::vec3(0.0f, 0.0f, 0.0f), - glm::vec3(0.0f, 1.0f, 0.0f) - ); + cameraData.view = glm::lookAt(glm::vec3(8.f * std::sin(glm::radians(s.b * 360.f)), 2.f, + 8.f * std::cos(glm::radians(s.b * 360.f))), + glm::vec3(0.0f, 0.0f, 0.0f), glm::vec3(0.0f, 1.0f, 0.0f)); queue.WriteBuffer(cameraBuffer, 0, &cameraData, sizeof(CameraData)); diff --git a/chromium/third_party/dawn/examples/ManualSwapChainTest.cpp b/chromium/third_party/dawn/examples/ManualSwapChainTest.cpp index ee924d3dba8..9e3990d820b 100644 --- a/chromium/third_party/dawn/examples/ManualSwapChainTest.cpp +++ b/chromium/third_party/dawn/examples/ManualSwapChainTest.cpp @@ -150,7 +150,8 @@ void DoRender(WindowData* data) { utils::ComboRenderPassDescriptor desc({view}); desc.cColorAttachments[0].loadOp = wgpu::LoadOp::Clear; - desc.cColorAttachments[0].clearColor = {data->clearCycle, 1.0f - data->clearCycle, 0.0f, 1.0f}; + desc.cColorAttachments[0].clearColor = {data->clearCycle, 1.0f - data->clearCycle, 0.0f, + 1.0f}; wgpu::RenderPassEncoder pass = encoder.BeginRenderPass(&desc); pass.EndPass(); diff --git a/chromium/third_party/dawn/examples/SampleUtils.cpp b/chromium/third_party/dawn/examples/SampleUtils.cpp index 17b7c9e7c57..ee7e0061563 100644 --- a/chromium/third_party/dawn/examples/SampleUtils.cpp +++ b/chromium/third_party/dawn/examples/SampleUtils.cpp @@ -60,7 +60,7 @@ void PrintGLFWError(int code, const char* message) { enum class CmdBufType { None, Terrible, - //TODO(cwallez@chromium.org) double terrible cmdbuf + // TODO(cwallez@chromium.org): double terrible cmdbuf }; // Default to D3D12, Metal, Vulkan, OpenGL in that order as D3D12 and Metal are the preferred on @@ -74,7 +74,7 @@ static wgpu::BackendType backendType = wgpu::BackendType::Vulkan; #elif defined(DAWN_ENABLE_BACKEND_OPENGL) static wgpu::BackendType backendType = wgpu::BackendType::OpenGL; #else - #error +# error #endif static CmdBufType cmdBufType = CmdBufType::Terrible; @@ -136,31 +136,29 @@ wgpu::Device CreateCppDawnDevice() { cDevice = backendDevice; break; - case CmdBufType::Terrible: - { - c2sBuf = new utils::TerribleCommandBuffer(); - s2cBuf = new utils::TerribleCommandBuffer(); + case CmdBufType::Terrible: { + c2sBuf = new utils::TerribleCommandBuffer(); + s2cBuf = new utils::TerribleCommandBuffer(); - dawn_wire::WireServerDescriptor serverDesc = {}; - serverDesc.device = backendDevice; - serverDesc.procs = &backendProcs; - serverDesc.serializer = s2cBuf; + dawn_wire::WireServerDescriptor serverDesc = {}; + serverDesc.device = backendDevice; + serverDesc.procs = &backendProcs; + serverDesc.serializer = s2cBuf; - wireServer = new dawn_wire::WireServer(serverDesc); - c2sBuf->SetHandler(wireServer); + wireServer = new dawn_wire::WireServer(serverDesc); + c2sBuf->SetHandler(wireServer); - dawn_wire::WireClientDescriptor clientDesc = {}; - clientDesc.serializer = c2sBuf; + dawn_wire::WireClientDescriptor clientDesc = {}; + clientDesc.serializer = c2sBuf; - wireClient = new dawn_wire::WireClient(clientDesc); - WGPUDevice clientDevice = wireClient->GetDevice(); - DawnProcTable clientProcs = dawn_wire::WireClient::GetProcs(); - s2cBuf->SetHandler(wireClient); + wireClient = new dawn_wire::WireClient(clientDesc); + WGPUDevice clientDevice = wireClient->GetDevice(); + DawnProcTable clientProcs = dawn_wire::WireClient::GetProcs(); + s2cBuf->SetHandler(wireClient); - procs = clientProcs; - cDevice = clientDevice; - } - break; + procs = clientProcs; + cDevice = clientDevice; + } break; } dawnProcSetProcs(&procs); @@ -221,7 +219,8 @@ bool InitSample(int argc, const char** argv) { backendType = wgpu::BackendType::Vulkan; continue; } - fprintf(stderr, "--backend expects a backend name (opengl, metal, d3d12, null, vulkan)\n"); + fprintf(stderr, + "--backend expects a backend name (opengl, metal, d3d12, null, vulkan)\n"); return false; } if (std::string("-c") == argv[i] || std::string("--command-buffer") == argv[i]) { diff --git a/chromium/third_party/dawn/generator/BUILD.gn b/chromium/third_party/dawn/generator/BUILD.gn index 2970e2277c7..acf48736b3e 100644 --- a/chromium/third_party/dawn/generator/BUILD.gn +++ b/chromium/third_party/dawn/generator/BUILD.gn @@ -20,14 +20,15 @@ import("dawn_generator.gni") # files but we can't just put dawn_gen_root because there are more than # autogenerated sources there. _stale_dirs = [ - "dawn", - "dawn_native", - "dawn_wire", - "mock", - "src" + "dawn", + "dawn_native", + "dawn_wire", + "mock", + "src", ] -_allowed_output_dirs_file = "${dawn_gen_root}/removed_stale_autogen_files.allowed_output_dirs" +_allowed_output_dirs_file = + "${dawn_gen_root}/removed_stale_autogen_files.allowed_output_dirs" write_file(_allowed_output_dirs_file, dawn_allowed_gen_output_dirs) _stale_dirs_file = "${dawn_gen_root}/removed_stale_autogen_files.stale_dirs" @@ -52,8 +53,11 @@ action("remove_stale_autogen_files") { # Have the "list of file" inputs as a dependency so that the action reruns # as soon as they change. - inputs = [_allowed_output_dirs_file, _stale_dirs_file] + inputs = [ + _allowed_output_dirs_file, + _stale_dirs_file, + ] # Output a stamp file so we don't re-run this action on every build. - outputs = [_stamp_file] + outputs = [ _stamp_file ] } diff --git a/chromium/third_party/dawn/generator/dawn_json_generator.py b/chromium/third_party/dawn/generator/dawn_json_generator.py index 6037aa8fab5..b7d9003e9d2 100644 --- a/chromium/third_party/dawn/generator/dawn_json_generator.py +++ b/chromium/third_party/dawn/generator/dawn_json_generator.py @@ -22,6 +22,7 @@ from generator_lib import Generator, run_generator, FileRender # OBJECT MODEL ############################################################ + class Name: def __init__(self, name, native=False): self.native = native @@ -44,7 +45,8 @@ class Name: return ''.join(self.chunks) def camelCase(self): - return self.chunks[0] + ''.join([self.CamelChunk(chunk) for chunk in self.chunks[1:]]) + return self.chunks[0] + ''.join( + [self.CamelChunk(chunk) for chunk in self.chunks[1:]]) def CamelCase(self): return ''.join([self.CamelChunk(chunk) for chunk in self.chunks]) @@ -63,9 +65,11 @@ class Name: result += chunk.lower() return result + def concat_names(*names): return ' '.join([name.canonical_case() for name in names]) + class Type: def __init__(self, name, json_data, native=False): self.json_data = json_data @@ -74,7 +78,10 @@ class Type: self.category = json_data['category'] self.javascript = self.json_data.get('javascript', True) + EnumValue = namedtuple('EnumValue', ['name', 'value', 'valid', 'jsrepr']) + + class EnumType(Type): def __init__(self, name, json_data): Type.__init__(self, name, json_data) @@ -87,42 +94,55 @@ class EnumType(Type): if value != lastValue + 1: self.contiguousFromZero = False lastValue = value - self.values.append(EnumValue( - Name(m['name']), - value, - m.get('valid', True), - m.get('jsrepr', None))) + self.values.append( + EnumValue(Name(m['name']), value, m.get('valid', True), + m.get('jsrepr', None))) # Assert that all values are unique in enums all_values = set() for value in self.values: if value.value in all_values: - raise Exception("Duplicate value {} in enum {}".format(value.value, name)) + raise Exception("Duplicate value {} in enum {}".format( + value.value, name)) all_values.add(value.value) + BitmaskValue = namedtuple('BitmaskValue', ['name', 'value']) + + class BitmaskType(Type): def __init__(self, name, json_data): Type.__init__(self, name, json_data) - self.values = [BitmaskValue(Name(m['name']), m['value']) for m in self.json_data['values']] + self.values = [ + BitmaskValue(Name(m['name']), m['value']) + for m in self.json_data['values'] + ] self.full_mask = 0 for value in self.values: self.full_mask = self.full_mask | value.value + class CallbackType(Type): def __init__(self, name, json_data): Type.__init__(self, name, json_data) self.arguments = [] + class NativeType(Type): def __init__(self, name, json_data): Type.__init__(self, name, json_data, native=True) + # Methods and structures are both "records", so record members correspond to # method arguments or structure members. class RecordMember: - def __init__(self, name, typ, annotation, optional=False, - is_return_value=False, default_value=None, + def __init__(self, + name, + typ, + annotation, + optional=False, + is_return_value=False, + default_value=None, skip_serialize=False): self.name = name self.type = typ @@ -138,13 +158,17 @@ class RecordMember: assert self.type.dict_name == "ObjectHandle" self.handle_type = handle_type + Method = namedtuple('Method', ['name', 'return_type', 'arguments']) + + class ObjectType(Type): def __init__(self, name, json_data): Type.__init__(self, name, json_data) self.methods = [] self.built_type = None + class Record: def __init__(self, name): self.name = Name(name) @@ -160,12 +184,15 @@ class Record: else: return False - self.may_have_dawn_object = any(may_have_dawn_object(member) for member in self.members) + self.may_have_dawn_object = any( + may_have_dawn_object(member) for member in self.members) - # set may_have_dawn_object to true if the type is chained or extensible. Chained structs - # may contain a Dawn object. + # Set may_have_dawn_object to true if the type is chained or + # extensible. Chained structs may contain a Dawn object. if isinstance(self, StructureType): - self.may_have_dawn_object = self.may_have_dawn_object or self.chained or self.extensible + self.may_have_dawn_object = (self.may_have_dawn_object + or self.chained or self.extensible) + class StructureType(Record, Type): def __init__(self, name, json_data): @@ -173,9 +200,11 @@ class StructureType(Record, Type): Type.__init__(self, name, json_data) self.chained = json_data.get("chained", False) self.extensible = json_data.get("extensible", False) - # Chained structs inherit from wgpu::ChainedStruct which has nextInChain so setting - # both extensible and chained would result in two nextInChain members. - assert(not (self.extensible and self.chained)) + # Chained structs inherit from wgpu::ChainedStruct, which has + # nextInChain, so setting both extensible and chained would result in + # two nextInChain members. + assert not (self.extensible and self.chained) + class Command(Record): def __init__(self, name, members=None): @@ -184,11 +213,13 @@ class Command(Record): self.derived_object = None self.derived_method = None + def linked_record_members(json_data, types): members = [] members_by_name = {} for m in json_data: - member = RecordMember(Name(m['name']), types[m['type']], + member = RecordMember(Name(m['name']), + types[m['type']], m.get('annotation', 'value'), optional=m.get('optional', False), is_return_value=m.get('is_return_value', False), @@ -207,7 +238,7 @@ def linked_record_members(json_data, types): member.length = "constant" member.constant_length = 1 else: - assert(False) + assert False elif m['length'] == 'strlen': member.length = 'strlen' else: @@ -215,31 +246,42 @@ def linked_record_members(json_data, types): return members + ############################################################ # PARSE ############################################################ + def link_object(obj, types): def make_method(json_data): arguments = linked_record_members(json_data.get('args', []), types) - return Method(Name(json_data['name']), types[json_data.get('returns', 'void')], arguments) + return Method(Name(json_data['name']), + types[json_data.get('returns', 'void')], arguments) obj.methods = [make_method(m) for m in obj.json_data.get('methods', [])] obj.methods.sort(key=lambda method: method.name.canonical_case()) + def link_structure(struct, types): struct.members = linked_record_members(struct.json_data['members'], types) + def link_callback(callback, types): - callback.arguments = linked_record_members(callback.json_data['args'], types) - -# Sort structures so that if struct A has struct B as a member, then B is listed before A -# This is a form of topological sort where we try to keep the order reasonably similar to the -# original order (though th sort isn't technically stable). -# It works by computing for each struct type what is the depth of its DAG of dependents, then -# resorting based on that depth using Python's stable sort. This makes a toposort because if -# A depends on B then its depth will be bigger than B's. It is also nice because all nodes -# with the same depth are kept in the input order. + callback.arguments = linked_record_members(callback.json_data['args'], + types) + + +# Sort structures so that if struct A has struct B as a member, then B is +# listed before A. +# +# This is a form of topological sort where we try to keep the order reasonably +# similar to the original order (though the sort isn't technically stable). +# +# It works by computing for each struct type what is the depth of its DAG of +# dependents, then resorting based on that depth using Python's stable sort. +# This makes a toposort because if A depends on B then its depth will be bigger +# than B's. It is also nice because all nodes with the same depth are kept in +# the input order. def topo_sort_structure(structs): for struct in structs: struct.visited = False @@ -252,7 +294,8 @@ def topo_sort_structure(structs): max_dependent_depth = 0 for member in struct.members: if member.type.category == 'structure': - max_dependent_depth = max(max_dependent_depth, compute_depth(member.type) + 1) + max_dependent_depth = max(max_dependent_depth, + compute_depth(member.type) + 1) struct.subdag_depth = max_dependent_depth struct.visited = True @@ -269,6 +312,7 @@ def topo_sort_structure(structs): return result + def parse_json(json): category_to_parser = { 'bitmask': BitmaskType, @@ -303,22 +347,22 @@ def parse_json(json): link_callback(callback, types) for category in by_category.keys(): - by_category[category] = sorted(by_category[category], key=lambda typ: typ.name.canonical_case()) + by_category[category] = sorted( + by_category[category], key=lambda typ: typ.name.canonical_case()) by_category['structure'] = topo_sort_structure(by_category['structure']) for struct in by_category['structure']: struct.update_metadata() - return { - 'types': types, - 'by_category': by_category - } + return {'types': types, 'by_category': by_category} + ############################################################ # WIRE STUFF ############################################################ + # Create wire commands from api methods def compute_wire_params(api_params, wire_json): wire_params = api_params.copy() @@ -327,7 +371,8 @@ def compute_wire_params(api_params, wire_json): commands = [] return_commands = [] - wire_json['special items']['client_handwritten_commands'] += wire_json['special items']['client_side_commands'] + wire_json['special items']['client_handwritten_commands'] += wire_json[ + 'special items']['client_side_commands'] # Generate commands from object methods for api_object in wire_params['by_category']['object']: @@ -335,21 +380,33 @@ def compute_wire_params(api_params, wire_json): command_name = concat_names(api_object.name, method.name) command_suffix = Name(command_name).CamelCase() - # Only object return values or void are supported. Other methods must be handwritten. - if method.return_type.category != 'object' and method.return_type.name.canonical_case() != 'void': - assert(command_suffix in wire_json['special items']['client_handwritten_commands']) + # Only object return values or void are supported. + # Other methods must be handwritten. + is_object = method.return_type.category == 'object' + is_void = method.return_type.name.canonical_case() == 'void' + if not (is_object or is_void): + assert command_suffix in ( + wire_json['special items']['client_handwritten_commands']) continue - if command_suffix in wire_json['special items']['client_side_commands']: + if command_suffix in ( + wire_json['special items']['client_side_commands']): continue # Create object method commands by prepending "self" - members = [RecordMember(Name('self'), types[api_object.dict_name], 'value')] + members = [ + RecordMember(Name('self'), types[api_object.dict_name], + 'value') + ] members += method.arguments - # Client->Server commands that return an object return the result object handle + # Client->Server commands that return an object return the + # result object handle if method.return_type.category == 'object': - result = RecordMember(Name('result'), types['ObjectHandle'], 'value', is_return_value=True) + result = RecordMember(Name('result'), + types['ObjectHandle'], + 'value', + is_return_value=True) result.set_handle_type(method.return_type) members.append(result) @@ -362,7 +419,8 @@ def compute_wire_params(api_params, wire_json): commands.append(Command(name, linked_record_members(json_data, types))) for (name, json_data) in wire_json['return commands'].items(): - return_commands.append(Command(name, linked_record_members(json_data, types))) + return_commands.append( + Command(name, linked_record_members(json_data, types))) wire_params['cmd_records'] = { 'command': commands, @@ -378,12 +436,16 @@ def compute_wire_params(api_params, wire_json): return wire_params + ############################################################# # Generator ############################################################# + def as_varName(*names): - return names[0].camelCase() + ''.join([name.CamelCase() for name in names[1:]]) + return names[0].camelCase() + ''.join( + [name.CamelCase() for name in names[1:]]) + def as_cType(name): if name.native: @@ -391,27 +453,32 @@ def as_cType(name): else: return 'WGPU' + name.CamelCase() + def as_cTypeDawn(name): if name.native: return name.concatcase() else: return 'Dawn' + name.CamelCase() + def as_cTypeEnumSpecialCase(typ): if typ.category == 'bitmask': return as_cType(typ.name) + 'Flags' return as_cType(typ.name) + def as_cppType(name): if name.native: return name.concatcase() else: return name.CamelCase() + def as_jsEnumValue(value): if value.jsrepr: return value.jsrepr return "'" + value.name.js_enum_case() + "'" + def convert_cType_to_cppType(typ, annotation, arg, indent=0): if typ.category == 'native': return arg @@ -422,18 +489,20 @@ def convert_cType_to_cppType(typ, annotation, arg, indent=0): converted_members = [ convert_cType_to_cppType( member.type, member.annotation, - '{}.{}'.format(arg, as_varName(member.name)), - indent + 1) - for member in typ.members] + '{}.{}'.format(arg, as_varName(member.name)), indent + 1) + for member in typ.members + ] - converted_members = [(' ' * 4) + m for m in converted_members ] + converted_members = [(' ' * 4) + m for m in converted_members] converted_members = ',\n'.join(converted_members) return as_cppType(typ.name) + ' {\n' + converted_members + '\n}' else: return 'static_cast<{}>({})'.format(as_cppType(typ.name), arg) else: - return 'reinterpret_cast<{} {}>({})'.format(as_cppType(typ.name), annotation, arg) + return 'reinterpret_cast<{} {}>({})'.format(as_cppType(typ.name), + annotation, arg) + def decorate(name, typ, arg): if arg.annotation == 'value': @@ -445,46 +514,57 @@ def decorate(name, typ, arg): elif arg.annotation == 'const*const*': return 'const ' + typ + '* const * ' + name else: - assert(False) + assert False + def annotated(typ, arg): name = as_varName(arg.name) return decorate(name, typ, arg) + def as_cEnum(type_name, value_name): - assert(not type_name.native and not value_name.native) + assert not type_name.native and not value_name.native return 'WGPU' + type_name.CamelCase() + '_' + value_name.CamelCase() + def as_cEnumDawn(type_name, value_name): - assert(not type_name.native and not value_name.native) - return 'DAWN' + '_' + type_name.SNAKE_CASE() + '_' + value_name.SNAKE_CASE() + assert not type_name.native and not value_name.native + return ('DAWN' + '_' + type_name.SNAKE_CASE() + '_' + + value_name.SNAKE_CASE()) + def as_cppEnum(value_name): - assert(not value_name.native) + assert not value_name.native if value_name.concatcase()[0].isdigit(): return "e" + value_name.CamelCase() return value_name.CamelCase() + def as_cMethod(type_name, method_name): - assert(not type_name.native and not method_name.native) + assert not type_name.native and not method_name.native return 'wgpu' + type_name.CamelCase() + method_name.CamelCase() + def as_cMethodDawn(type_name, method_name): - assert(not type_name.native and not method_name.native) + assert not type_name.native and not method_name.native return 'dawn' + type_name.CamelCase() + method_name.CamelCase() + def as_MethodSuffix(type_name, method_name): - assert(not type_name.native and not method_name.native) + assert not type_name.native and not method_name.native return type_name.CamelCase() + method_name.CamelCase() + def as_cProc(type_name, method_name): - assert(not type_name.native and not method_name.native) + assert not type_name.native and not method_name.native return 'WGPU' + 'Proc' + type_name.CamelCase() + method_name.CamelCase() + def as_cProcDawn(type_name, method_name): - assert(not type_name.native and not method_name.native) + assert not type_name.native and not method_name.native return 'Dawn' + 'Proc' + type_name.CamelCase() + method_name.CamelCase() + def as_frontendType(typ): if typ.category == 'object': return typ.name.CamelCase() + 'Base*' @@ -495,6 +575,7 @@ def as_frontendType(typ): else: return as_cType(typ.name) + def as_wireType(typ): if typ.category == 'object': return typ.name.CamelCase() + '*' @@ -503,31 +584,50 @@ def as_wireType(typ): else: return as_cppType(typ.name) + def c_methods(types, typ): return typ.methods + [ Method(Name('reference'), types['void'], []), Method(Name('release'), types['void'], []), ] + def get_c_methods_sorted_by_name(api_params): unsorted = [(as_MethodSuffix(typ.name, method.name), typ, method) \ for typ in api_params['by_category']['object'] \ for method in c_methods(api_params['types'], typ) ] return [(typ, method) for (_, typ, method) in sorted(unsorted)] + def has_callback_arguments(method): return any(arg.type.category == 'callback' for arg in method.arguments) + class MultiGeneratorFromDawnJSON(Generator): def get_description(self): return 'Generates code for various target from Dawn.json.' def add_commandline_arguments(self, parser): - allowed_targets = ['dawn_headers', 'dawncpp_headers', 'dawncpp', 'dawn_proc', 'mock_webgpu', 'dawn_wire', "dawn_native_utils"] - - parser.add_argument('--dawn-json', required=True, type=str, help ='The DAWN JSON definition to use.') - parser.add_argument('--wire-json', default=None, type=str, help='The DAWN WIRE JSON definition to use.') - parser.add_argument('--targets', required=True, type=str, help='Comma-separated subset of targets to output. Available targets: ' + ', '.join(allowed_targets)) + allowed_targets = [ + 'dawn_headers', 'dawncpp_headers', 'dawncpp', 'dawn_proc', + 'mock_webgpu', 'dawn_wire', "dawn_native_utils" + ] + + parser.add_argument('--dawn-json', + required=True, + type=str, + help='The DAWN JSON definition to use.') + parser.add_argument('--wire-json', + default=None, + type=str, + help='The DAWN WIRE JSON definition to use.') + parser.add_argument( + '--targets', + required=True, + type=str, + help= + 'Comma-separated subset of targets to output. Available targets: ' + + ', '.join(allowed_targets)) def get_file_renders(self, args): with open(args.dawn_json) as f: @@ -543,9 +643,10 @@ class MultiGeneratorFromDawnJSON(Generator): base_params = { 'Name': lambda name: Name(name), - - 'as_annotated_cType': lambda arg: annotated(as_cTypeEnumSpecialCase(arg.type), arg), - 'as_annotated_cppType': lambda arg: annotated(as_cppType(arg.type.name), arg), + 'as_annotated_cType': \ + lambda arg: annotated(as_cTypeEnumSpecialCase(arg.type), arg), + 'as_annotated_cppType': \ + lambda arg: annotated(as_cppType(arg.type.name), arg), 'as_cEnum': as_cEnum, 'as_cEnumDawn': as_cEnumDawn, 'as_cppEnum': as_cppEnum, @@ -562,79 +663,145 @@ class MultiGeneratorFromDawnJSON(Generator): 'as_varName': as_varName, 'decorate': decorate, 'c_methods': lambda typ: c_methods(api_params['types'], typ), - 'c_methods_sorted_by_name': get_c_methods_sorted_by_name(api_params), + 'c_methods_sorted_by_name': \ + get_c_methods_sorted_by_name(api_params), } renders = [] if 'dawn_headers' in targets: - renders.append(FileRender('webgpu.h', 'src/include/dawn/webgpu.h', [base_params, api_params])) - renders.append(FileRender('dawn_proc_table.h', 'src/include/dawn/dawn_proc_table.h', [base_params, api_params])) + renders.append( + FileRender('webgpu.h', 'src/include/dawn/webgpu.h', + [base_params, api_params])) + renders.append( + FileRender('dawn_proc_table.h', + 'src/include/dawn/dawn_proc_table.h', + [base_params, api_params])) if 'dawncpp_headers' in targets: - renders.append(FileRender('webgpu_cpp.h', 'src/include/dawn/webgpu_cpp.h', [base_params, api_params])) + renders.append( + FileRender('webgpu_cpp.h', 'src/include/dawn/webgpu_cpp.h', + [base_params, api_params])) if 'dawn_proc' in targets: - renders.append(FileRender('dawn_proc.c', 'src/dawn/dawn_proc.c', [base_params, api_params])) + renders.append( + FileRender('dawn_proc.c', 'src/dawn/dawn_proc.c', + [base_params, api_params])) if 'dawncpp' in targets: - renders.append(FileRender('webgpu_cpp.cpp', 'src/dawn/webgpu_cpp.cpp', [base_params, api_params])) + renders.append( + FileRender('webgpu_cpp.cpp', 'src/dawn/webgpu_cpp.cpp', + [base_params, api_params])) if 'emscripten_bits' in targets: - renders.append(FileRender('webgpu_struct_info.json', 'src/dawn/webgpu_struct_info.json', [base_params, api_params])) - renders.append(FileRender('library_webgpu_enum_tables.js', 'src/dawn/library_webgpu_enum_tables.js', [base_params, api_params])) + renders.append( + FileRender('webgpu_struct_info.json', + 'src/dawn/webgpu_struct_info.json', + [base_params, api_params])) + renders.append( + FileRender('library_webgpu_enum_tables.js', + 'src/dawn/library_webgpu_enum_tables.js', + [base_params, api_params])) if 'mock_webgpu' in targets: mock_params = [ - base_params, - api_params, - { + base_params, api_params, { 'has_callback_arguments': has_callback_arguments } ] - renders.append(FileRender('mock_webgpu.h', 'src/dawn/mock_webgpu.h', mock_params)) - renders.append(FileRender('mock_webgpu.cpp', 'src/dawn/mock_webgpu.cpp', mock_params)) + renders.append( + FileRender('mock_webgpu.h', 'src/dawn/mock_webgpu.h', + mock_params)) + renders.append( + FileRender('mock_webgpu.cpp', 'src/dawn/mock_webgpu.cpp', + mock_params)) if 'dawn_native_utils' in targets: frontend_params = [ base_params, api_params, { - 'as_frontendType': lambda typ: as_frontendType(typ), # TODO as_frontendType and friends take a Type and not a Name :( - 'as_annotated_frontendType': lambda arg: annotated(as_frontendType(arg.type), arg) + # TODO: as_frontendType and co. take a Type, not a Name :( + 'as_frontendType': lambda typ: as_frontendType(typ), + 'as_annotated_frontendType': \ + lambda arg: annotated(as_frontendType(arg.type), arg), } ] - renders.append(FileRender('dawn_native/ValidationUtils.h', 'src/dawn_native/ValidationUtils_autogen.h', frontend_params)) - renders.append(FileRender('dawn_native/ValidationUtils.cpp', 'src/dawn_native/ValidationUtils_autogen.cpp', frontend_params)) - renders.append(FileRender('dawn_native/wgpu_structs.h', 'src/dawn_native/wgpu_structs_autogen.h', frontend_params)) - renders.append(FileRender('dawn_native/wgpu_structs.cpp', 'src/dawn_native/wgpu_structs_autogen.cpp', frontend_params)) - renders.append(FileRender('dawn_native/ProcTable.cpp', 'src/dawn_native/ProcTable.cpp', frontend_params)) + renders.append( + FileRender('dawn_native/ValidationUtils.h', + 'src/dawn_native/ValidationUtils_autogen.h', + frontend_params)) + renders.append( + FileRender('dawn_native/ValidationUtils.cpp', + 'src/dawn_native/ValidationUtils_autogen.cpp', + frontend_params)) + renders.append( + FileRender('dawn_native/wgpu_structs.h', + 'src/dawn_native/wgpu_structs_autogen.h', + frontend_params)) + renders.append( + FileRender('dawn_native/wgpu_structs.cpp', + 'src/dawn_native/wgpu_structs_autogen.cpp', + frontend_params)) + renders.append( + FileRender('dawn_native/ProcTable.cpp', + 'src/dawn_native/ProcTable.cpp', frontend_params)) if 'dawn_wire' in targets: additional_params = compute_wire_params(api_params, wire_json) wire_params = [ - base_params, - api_params, - { + base_params, api_params, { 'as_wireType': as_wireType, - 'as_annotated_wireType': lambda arg: annotated(as_wireType(arg.type), arg), - }, - additional_params + 'as_annotated_wireType': \ + lambda arg: annotated(as_wireType(arg.type), arg), + }, additional_params ] - renders.append(FileRender('dawn_wire/WireCmd.h', 'src/dawn_wire/WireCmd_autogen.h', wire_params)) - renders.append(FileRender('dawn_wire/WireCmd.cpp', 'src/dawn_wire/WireCmd_autogen.cpp', wire_params)) - renders.append(FileRender('dawn_wire/client/ApiObjects.h', 'src/dawn_wire/client/ApiObjects_autogen.h', wire_params)) - renders.append(FileRender('dawn_wire/client/ApiProcs.cpp', 'src/dawn_wire/client/ApiProcs_autogen.cpp', wire_params)) - renders.append(FileRender('dawn_wire/client/ApiProcs.h', 'src/dawn_wire/client/ApiProcs_autogen.h', wire_params)) - renders.append(FileRender('dawn_wire/client/ClientBase.h', 'src/dawn_wire/client/ClientBase_autogen.h', wire_params)) - renders.append(FileRender('dawn_wire/client/ClientHandlers.cpp', 'src/dawn_wire/client/ClientHandlers_autogen.cpp', wire_params)) - renders.append(FileRender('dawn_wire/client/ClientPrototypes.inc', 'src/dawn_wire/client/ClientPrototypes_autogen.inc', wire_params)) - renders.append(FileRender('dawn_wire/server/ServerBase.h', 'src/dawn_wire/server/ServerBase_autogen.h', wire_params)) - renders.append(FileRender('dawn_wire/server/ServerDoers.cpp', 'src/dawn_wire/server/ServerDoers_autogen.cpp', wire_params)) - renders.append(FileRender('dawn_wire/server/ServerHandlers.cpp', 'src/dawn_wire/server/ServerHandlers_autogen.cpp', wire_params)) - renders.append(FileRender('dawn_wire/server/ServerPrototypes.inc', 'src/dawn_wire/server/ServerPrototypes_autogen.inc', wire_params)) + renders.append( + FileRender('dawn_wire/WireCmd.h', + 'src/dawn_wire/WireCmd_autogen.h', wire_params)) + renders.append( + FileRender('dawn_wire/WireCmd.cpp', + 'src/dawn_wire/WireCmd_autogen.cpp', wire_params)) + renders.append( + FileRender('dawn_wire/client/ApiObjects.h', + 'src/dawn_wire/client/ApiObjects_autogen.h', + wire_params)) + renders.append( + FileRender('dawn_wire/client/ApiProcs.cpp', + 'src/dawn_wire/client/ApiProcs_autogen.cpp', + wire_params)) + renders.append( + FileRender('dawn_wire/client/ClientBase.h', + 'src/dawn_wire/client/ClientBase_autogen.h', + wire_params)) + renders.append( + FileRender('dawn_wire/client/ClientHandlers.cpp', + 'src/dawn_wire/client/ClientHandlers_autogen.cpp', + wire_params)) + renders.append( + FileRender( + 'dawn_wire/client/ClientPrototypes.inc', + 'src/dawn_wire/client/ClientPrototypes_autogen.inc', + wire_params)) + renders.append( + FileRender('dawn_wire/server/ServerBase.h', + 'src/dawn_wire/server/ServerBase_autogen.h', + wire_params)) + renders.append( + FileRender('dawn_wire/server/ServerDoers.cpp', + 'src/dawn_wire/server/ServerDoers_autogen.cpp', + wire_params)) + renders.append( + FileRender('dawn_wire/server/ServerHandlers.cpp', + 'src/dawn_wire/server/ServerHandlers_autogen.cpp', + wire_params)) + renders.append( + FileRender( + 'dawn_wire/server/ServerPrototypes.inc', + 'src/dawn_wire/server/ServerPrototypes_autogen.inc', + wire_params)) return renders @@ -644,5 +811,6 @@ class MultiGeneratorFromDawnJSON(Generator): deps += [os.path.abspath(args.wire_json)] return deps + if __name__ == '__main__': sys.exit(run_generator(MultiGeneratorFromDawnJSON())) diff --git a/chromium/third_party/dawn/generator/generator_lib.gni b/chromium/third_party/dawn/generator/generator_lib.gni index 36ac72f2f60..8b9e04cc955 100644 --- a/chromium/third_party/dawn/generator/generator_lib.gni +++ b/chromium/third_party/dawn/generator/generator_lib.gni @@ -135,9 +135,7 @@ template("generator_lib_action") { # outputs are what's expected and write a depfile for Ninja. action(_json_tarball_target) { script = invoker.script - outputs = [ - _json_tarball, - ] + outputs = [ _json_tarball ] depfile = _json_tarball_depfile args = _generator_args if (defined(invoker.deps)) { @@ -153,12 +151,8 @@ template("generator_lib_action") { rebase_path(_gen_dir, root_build_dir), ] - deps = [ - ":${_json_tarball_target}", - ] - inputs = [ - _json_tarball, - ] + deps = [ ":${_json_tarball_target}" ] + inputs = [ _json_tarball ] # The expected output list is relative to the gen_dir but action # target outputs are from the root dir so we need to rebase them. diff --git a/chromium/third_party/dawn/generator/generator_lib.py b/chromium/third_party/dawn/generator/generator_lib.py index 8deec5d1949..5e3734d7833 100644 --- a/chromium/third_party/dawn/generator/generator_lib.py +++ b/chromium/third_party/dawn/generator/generator_lib.py @@ -12,7 +12,6 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. - """Module to create generators that render multiple Jinja2 templates for GN. A helper module that can be used to create generator scripts (clients) @@ -54,6 +53,7 @@ from collections import namedtuple # FileRender = namedtuple('FileRender', ['template', 'output', 'params_dicts']) + # The interface that must be implemented by generators. class Generator: def get_description(self): @@ -72,6 +72,7 @@ class Generator: """Return a list of extra input dependencies.""" return [] + # Allow custom Jinja2 installation path through an additional python # path from the arguments if present. This isn't done through the regular # argparse because PreprocessingLoader uses jinja2 in the global scope before @@ -92,6 +93,7 @@ except ValueError: import jinja2 + # A custom Jinja2 template loader that removes the extra indentation # of the template blocks so that the output is correctly indented class _PreprocessingLoader(jinja2.BaseLoader): @@ -113,18 +115,21 @@ class _PreprocessingLoader(jinja2.BaseLoader): def preprocess(self, source): lines = source.split('\n') - # Compute the current indentation level of the template blocks and remove their indentation + # Compute the current indentation level of the template blocks and + # remove their indentation result = [] indentation_level = 0 - # Filter lines that are pure comments. line_comment_prefix is not enough because it removes - # the comment but doesn't completely remove the line, resulting in more verbose output. + # Filter lines that are pure comments. line_comment_prefix is not + # enough because it removes the comment but doesn't completely remove + # the line, resulting in more verbose output. lines = filter(lambda line: not line.strip().startswith('//*'), lines) # Remove indentation templates have for the Jinja control flow. for line in lines: - # The capture in the regex adds one element per block start or end so we divide by two - # there is also an extra line chunk corresponding to the line end, so we substract it. + # The capture in the regex adds one element per block start or end, + # so we divide by two. There is also an extra line chunk + # corresponding to the line end, so we subtract it. numends = (len(self.blockend.split(line)) - 1) // 2 indentation_level -= numends @@ -142,14 +147,19 @@ class _PreprocessingLoader(jinja2.BaseLoader): elif line.startswith('\t'): line = line[1:] else: - assert(line.strip() == '') + assert line.strip() == '' return line + _FileOutput = namedtuple('FileOutput', ['name', 'content']) + def _do_renders(renders, template_dir): loader = _PreprocessingLoader(template_dir) - env = jinja2.Environment(loader=loader, lstrip_blocks=True, trim_blocks=True, line_comment_prefix='//*') + env = jinja2.Environment(loader=loader, + lstrip_blocks=True, + trim_blocks=True, + line_comment_prefix='//*') def do_assert(expr): assert expr @@ -177,16 +187,17 @@ def _do_renders(renders, template_dir): return outputs + # Compute the list of imported, non-system Python modules. # It assumes that any path outside of the root directory is system. -def _compute_python_dependencies(root_dir = None): +def _compute_python_dependencies(root_dir=None): if not root_dir: # Assume this script is under generator/ by default. root_dir = os.path.join(os.path.dirname(__file__), os.pardir) root_dir = os.path.abspath(root_dir) module_paths = (module.__file__ for module in sys.modules.values() - if module and hasattr(module, '__file__')) + if module and hasattr(module, '__file__')) paths = set() for path in module_paths: @@ -203,37 +214,85 @@ def _compute_python_dependencies(root_dir = None): return paths + def run_generator(generator): parser = argparse.ArgumentParser( - description = generator.get_description(), - formatter_class = argparse.ArgumentDefaultsHelpFormatter, + description=generator.get_description(), + formatter_class=argparse.ArgumentDefaultsHelpFormatter, ) - generator.add_commandline_arguments(parser); - parser.add_argument('--template-dir', default='templates', type=str, help='Directory with template files.') - parser.add_argument(kJinja2Path, default=None, type=str, help='Additional python path to set before loading Jinja2') - parser.add_argument('--output-json-tarball', default=None, type=str, help='Name of the "JSON tarball" to create (tar is too annoying to use in python).') - parser.add_argument('--depfile', default=None, type=str, help='Name of the Ninja depfile to create for the JSON tarball') - parser.add_argument('--expected-outputs-file', default=None, type=str, help="File to compare outputs with and fail if it doesn't match") - parser.add_argument('--root-dir', default=None, type=str, help='Optional source root directory for Python dependency computations') - parser.add_argument('--allowed-output-dirs-file', default=None, type=str, help="File containing a list of allowed directories where files can be output.") - parser.add_argument('--print-cmake-dependencies', default=False, action="store_true", help="Prints a semi-colon separated list of dependencies to stdout and exits.") - parser.add_argument('--print-cmake-outputs', default=False, action="store_true", help="Prints a semi-colon separated list of outputs to stdout and exits.") - parser.add_argument('--output-dir', default=None, type=str, help='Directory where to output generate files.') + generator.add_commandline_arguments(parser) + parser.add_argument('--template-dir', + default='templates', + type=str, + help='Directory with template files.') + parser.add_argument( + kJinja2Path, + default=None, + type=str, + help='Additional python path to set before loading Jinja2') + parser.add_argument( + '--output-json-tarball', + default=None, + type=str, + help=('Name of the "JSON tarball" to create (tar is too annoying ' + 'to use in python).')) + parser.add_argument( + '--depfile', + default=None, + type=str, + help='Name of the Ninja depfile to create for the JSON tarball') + parser.add_argument( + '--expected-outputs-file', + default=None, + type=str, + help="File to compare outputs with and fail if it doesn't match") + parser.add_argument( + '--root-dir', + default=None, + type=str, + help=('Optional source root directory for Python dependency ' + 'computations')) + parser.add_argument( + '--allowed-output-dirs-file', + default=None, + type=str, + help=("File containing a list of allowed directories where files " + "can be output.")) + parser.add_argument( + '--print-cmake-dependencies', + default=False, + action="store_true", + help=("Prints a semi-colon separated list of dependencies to " + "stdout and exits.")) + parser.add_argument( + '--print-cmake-outputs', + default=False, + action="store_true", + help=("Prints a semi-colon separated list of outputs to " + "stdout and exits.")) + parser.add_argument('--output-dir', + default=None, + type=str, + help='Directory where to output generate files.') args = parser.parse_args() - renders = generator.get_file_renders(args); + renders = generator.get_file_renders(args) # Output a list of all dependencies for CMake or the tarball for GN/Ninja. if args.depfile != None or args.print_cmake_dependencies: dependencies = generator.get_dependencies(args) - dependencies += [args.template_dir + os.path.sep + render.template for render in renders] + dependencies += [ + args.template_dir + os.path.sep + render.template + for render in renders + ] dependencies += _compute_python_dependencies(args.root_dir) if args.depfile != None: with open(args.depfile, 'w') as f: - f.write(args.output_json_tarball + ": " + " ".join(dependencies)) + f.write(args.output_json_tarball + ": " + + " ".join(dependencies)) if args.print_cmake_dependencies: sys.stdout.write(";".join(dependencies)) @@ -248,33 +307,42 @@ def run_generator(generator): actual = {render.output for render in renders} if actual != expected: - print("Wrong expected outputs, caller expected:\n " + repr(sorted(expected))) + print("Wrong expected outputs, caller expected:\n " + + repr(sorted(expected))) print("Actual output:\n " + repr(sorted(actual))) return 1 # Print the list of all the outputs for cmake. if args.print_cmake_outputs: - sys.stdout.write(";".join([os.path.join(args.output_dir, render.output) for render in renders])) + sys.stdout.write(";".join([ + os.path.join(args.output_dir, render.output) for render in renders + ])) return 0 outputs = _do_renders(renders, args.template_dir) - # The caller wants to assert that the outputs are only in specific directories. + # The caller wants to assert that the outputs are only in specific + # directories. if args.allowed_output_dirs_file != None: with open(args.allowed_output_dirs_file) as f: allowed_dirs = set([line.strip() for line in f.readlines()]) for directory in allowed_dirs: if not directory.endswith('/'): - print('Allowed directory entry "{}" doesn\'t end with /'.format(directory)) + print('Allowed directory entry "{}" doesn\'t ' + 'end with /'.format(directory)) return 1 def check_in_subdirectory(path, directory): - return path.startswith(directory) and not '/' in path[len(directory):] + return path.startswith( + directory) and not '/' in path[len(directory):] for render in renders: - if not any(check_in_subdirectory(render.output, directory) for directory in allowed_dirs): - print('Output file "{}" is not in the allowed directory list below:'.format(render.output)) + if not any( + check_in_subdirectory(render.output, directory) + for directory in allowed_dirs): + print('Output file "{}" is not in the allowed directory ' + 'list below:'.format(render.output)) for directory in sorted(allowed_dirs): print(' "{}"'.format(directory)) return 1 diff --git a/chromium/third_party/dawn/generator/opengl_loader_generator.py b/chromium/third_party/dawn/generator/opengl_loader_generator.py index 47721e2d4fa..e080c4e800b 100644 --- a/chromium/third_party/dawn/generator/opengl_loader_generator.py +++ b/chromium/third_party/dawn/generator/opengl_loader_generator.py @@ -19,9 +19,10 @@ import xml.etree.ElementTree as etree from generator_lib import Generator, run_generator, FileRender + class ProcName: def __init__(self, gl_name, proc_name=None): - assert(gl_name.startswith('gl')) + assert gl_name.startswith('gl') if proc_name == None: proc_name = gl_name[2:] @@ -40,7 +41,10 @@ class ProcName: def __repr__(self): return 'Proc("{}", "{}")'.format(self.gl_name, self.proc_name) + ProcParam = namedtuple('ProcParam', ['name', 'type']) + + class Proc: def __init__(self, element): # Type declaration for return values and arguments all have the same @@ -72,7 +76,9 @@ class Proc: self.params = [] for param in element.findall('./param'): - self.params.append(ProcParam(param.find('name').text, parse_type_declaration(param))) + self.params.append( + ProcParam( + param.find('name').text, parse_type_declaration(param))) self.gl_name = proto.find('name').text self.alias = None @@ -83,7 +89,7 @@ class Proc: return self.gl_name def ProcName(self): - assert(self.gl_name.startswith('gl')) + assert self.gl_name.startswith('gl') return self.gl_name[2:] def PFNGLPROCNAME(self): @@ -92,11 +98,14 @@ class Proc: def __repr__(self): return 'Proc("{}")'.format(self.gl_name) + EnumDefine = namedtuple('EnumDefine', ['name', 'value']) Version = namedtuple('Version', ['major', 'minor']) VersionBlock = namedtuple('VersionBlock', ['version', 'procs', 'enums']) HeaderBlock = namedtuple('HeaderBlock', ['description', 'procs', 'enums']) -ExtensionBlock = namedtuple('ExtensionBlock', ['extension', 'procs', 'enums', 'supported_specs']) +ExtensionBlock = namedtuple('ExtensionBlock', + ['extension', 'procs', 'enums', 'supported_specs']) + def parse_version(version): return Version(*map(int, version.split('.'))) @@ -107,7 +116,7 @@ def compute_params(root, supported_extensions): all_procs = {} for command in root.findall('''commands[@namespace='GL']/command'''): proc = Proc(command) - assert(proc.gl_name not in all_procs) + assert proc.gl_name not in all_procs all_procs[proc.gl_name] = proc all_enums = {} @@ -117,7 +126,7 @@ def compute_params(root, supported_extensions): if enum_name == 'GL_ACTIVE_PROGRAM_EXT': continue - assert(enum_name not in all_enums) + assert enum_name not in all_enums all_enums[enum_name] = EnumDefine(enum_name, enum.attrib['value']) # Get the list of all Desktop OpenGL function removed by the Core Profile. @@ -126,13 +135,13 @@ def compute_params(root, supported_extensions): core_removed_procs.add(proc.attrib['name']) # Get list of enums and procs per OpenGL ES/Desktop OpenGL version - def parse_version_blocks(api, removed_procs = set()): + def parse_version_blocks(api, removed_procs=set()): blocks = [] for section in root.findall('''feature[@api='{}']'''.format(api)): section_procs = [] for command in section.findall('./require/command'): proc_name = command.attrib['name'] - assert(all_procs[proc_name].alias == None) + assert all_procs[proc_name].alias == None if proc_name not in removed_procs: section_procs.append(all_procs[proc_name]) @@ -140,7 +149,9 @@ def compute_params(root, supported_extensions): for enum in section.findall('./require/enum'): section_enums.append(all_enums[enum.attrib['name']]) - blocks.append(VersionBlock(parse_version(section.attrib['number']), section_procs, section_enums)) + blocks.append( + VersionBlock(parse_version(section.attrib['number']), + section_procs, section_enums)) return blocks @@ -148,12 +159,13 @@ def compute_params(root, supported_extensions): desktop_gl_blocks = parse_version_blocks('gl', core_removed_procs) def parse_extension_block(extension): - section = root.find('''extensions/extension[@name='{}']'''.format(extension)) + section = root.find( + '''extensions/extension[@name='{}']'''.format(extension)) supported_specs = section.attrib['supported'].split('|') section_procs = [] for command in section.findall('./require/command'): proc_name = command.attrib['name'] - assert(all_procs[proc_name].alias == None) + assert all_procs[proc_name].alias == None if proc_name not in removed_procs: section_procs.append(all_procs[proc_name]) @@ -161,10 +173,11 @@ def compute_params(root, supported_extensions): for enum in section.findall('./require/enum'): section_enums.append(all_enums[enum.attrib['name']]) - return ExtensionBlock(extension, section_procs, section_enums, supported_specs) + return ExtensionBlock(extension, section_procs, section_enums, + supported_specs) - extension_desktop_gl_blocks = []; - extension_gles_blocks = []; + extension_desktop_gl_blocks = [] + extension_gles_blocks = [] for extension in supported_extensions: extension_block = parse_extension_block(extension) if 'gl' in extension_block.supported_specs: @@ -176,6 +189,7 @@ def compute_params(root, supported_extensions): already_added_header_procs = set() already_added_header_enums = set() header_blocks = [] + def add_header_block(description, block): block_procs = [] for proc in block.procs: @@ -190,13 +204,18 @@ def compute_params(root, supported_extensions): block_enums.append(enum) if len(block_procs) > 0 or len(block_enums) > 0: - header_blocks.append(HeaderBlock(description, block_procs, block_enums)) + header_blocks.append( + HeaderBlock(description, block_procs, block_enums)) for block in gles_blocks: - add_header_block('OpenGL ES {}.{}'.format(block.version.major, block.version.minor), block) + add_header_block( + 'OpenGL ES {}.{}'.format(block.version.major, block.version.minor), + block) for block in desktop_gl_blocks: - add_header_block('Desktop OpenGL {}.{}'.format(block.version.major, block.version.minor), block) + add_header_block( + 'Desktop OpenGL {}.{}'.format(block.version.major, + block.version.minor), block) for block in extension_desktop_gl_blocks: add_header_block(block.extension, block) @@ -212,30 +231,50 @@ def compute_params(root, supported_extensions): 'header_blocks': header_blocks, } + class OpenGLLoaderGenerator(Generator): def get_description(self): return 'Generates code to load OpenGL function pointers' def add_commandline_arguments(self, parser): - parser.add_argument('--gl-xml', required=True, type=str, help='The Khronos gl.xml to use.') - parser.add_argument('--supported-extensions', required=True, type=str, help ='The JSON file that defines the OpenGL and GLES extensions to use.') + parser.add_argument('--gl-xml', + required=True, + type=str, + help='The Khronos gl.xml to use.') + parser.add_argument( + '--supported-extensions', + required=True, + type=str, + help= + 'The JSON file that defines the OpenGL and GLES extensions to use.' + ) def get_file_renders(self, args): supported_extensions = [] with open(args.supported_extensions) as f: supported_extensions_json = json.loads(f.read()) - supported_extensions = supported_extensions_json['supported_extensions'] + supported_extensions = supported_extensions_json[ + 'supported_extensions'] - params = compute_params(etree.parse(args.gl_xml).getroot(), supported_extensions) + params = compute_params( + etree.parse(args.gl_xml).getroot(), supported_extensions) return [ - FileRender('opengl/OpenGLFunctionsBase.cpp', 'src/dawn_native/opengl/OpenGLFunctionsBase_autogen.cpp', [params]), - FileRender('opengl/OpenGLFunctionsBase.h', 'src/dawn_native/opengl/OpenGLFunctionsBase_autogen.h', [params]), - FileRender('opengl/opengl_platform.h', 'src/dawn_native/opengl/opengl_platform_autogen.h', [params]), + FileRender( + 'opengl/OpenGLFunctionsBase.cpp', + 'src/dawn_native/opengl/OpenGLFunctionsBase_autogen.cpp', + [params]), + FileRender('opengl/OpenGLFunctionsBase.h', + 'src/dawn_native/opengl/OpenGLFunctionsBase_autogen.h', + [params]), + FileRender('opengl/opengl_platform.h', + 'src/dawn_native/opengl/opengl_platform_autogen.h', + [params]), ] def get_dependencies(self, args): return [os.path.abspath(args.gl_xml)] + if __name__ == '__main__': sys.exit(run_generator(OpenGLLoaderGenerator())) diff --git a/chromium/third_party/dawn/generator/remove_files.py b/chromium/third_party/dawn/generator/remove_files.py index 0b8b5216145..6ddf463667d 100644 --- a/chromium/third_party/dawn/generator/remove_files.py +++ b/chromium/third_party/dawn/generator/remove_files.py @@ -15,11 +15,15 @@ import argparse, glob, os, sys + def check_in_subdirectory(path, directory): return path.startswith(directory) and not '/' in path[len(directory):] + def check_is_allowed(path, allowed_dirs): - return any(check_in_subdirectory(path, directory) for directory in allowed_dirs) + return any( + check_in_subdirectory(path, directory) for directory in allowed_dirs) + def get_all_files_in_dir(find_directory): result = [] @@ -27,15 +31,28 @@ def get_all_files_in_dir(find_directory): result += [os.path.join(directory, filename) for filename in files] return result + def run(): # Parse command line arguments parser = argparse.ArgumentParser( - description = "Removes stale autogenerated files from gen/ directories." + description="Removes stale autogenerated files from gen/ directories.") + parser.add_argument( + '--root-dir', + type=str, + help='The root directory, all other paths in files are relative to it.' ) - parser.add_argument('--root-dir', type=str, help='The root directory, all other paths in files are relative to it.') - parser.add_argument('--allowed-output-dirs-file', type=str, help='The file containing a list of allowed directories') - parser.add_argument('--stale-dirs-file', type=str, help='The file containing a list of directories to check for stale files') - parser.add_argument('--stamp', type=str, help='A stamp written once this script completes') + parser.add_argument( + '--allowed-output-dirs-file', + type=str, + help='The file containing a list of allowed directories') + parser.add_argument( + '--stale-dirs-file', + type=str, + help= + 'The file containing a list of directories to check for stale files') + parser.add_argument('--stamp', + type=str, + help='A stamp written once this script completes') args = parser.parse_args() root_dir = args.root_dir @@ -43,11 +60,13 @@ def run(): # Load the list of allowed and stale directories with open(args.allowed_output_dirs_file) as f: - allowed_dirs = set([os.path.join(root_dir, line.strip()) for line in f.readlines()]) + allowed_dirs = set( + [os.path.join(root_dir, line.strip()) for line in f.readlines()]) for directory in allowed_dirs: if not directory.endswith('/'): - print('Allowed directory entry "{}" doesn\'t end with /'.format(directory)) + print('Allowed directory entry "{}" doesn\'t end with /'.format( + directory)) return 1 with open(args.stale_dirs_file) as f: @@ -67,5 +86,6 @@ def run(): return 0 + if __name__ == "__main__": sys.exit(run()) diff --git a/chromium/third_party/dawn/generator/templates/.clang-format b/chromium/third_party/dawn/generator/templates/.clang-format new file mode 100644 index 00000000000..9d159247d51 --- /dev/null +++ b/chromium/third_party/dawn/generator/templates/.clang-format @@ -0,0 +1,2 @@ +DisableFormat: true +SortIncludes: false diff --git a/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiObjects.h b/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiObjects.h index 5c2ae3f613c..288c7004de1 100644 --- a/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiObjects.h +++ b/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiObjects.h @@ -16,10 +16,24 @@ #define DAWNWIRE_CLIENT_APIOBJECTS_AUTOGEN_H_ namespace dawn_wire { namespace client { - {% for type in by_category["object"] if not type.name.CamelCase() in client_special_objects %} - struct {{type.name.CamelCase()}} : ObjectBase { - using ObjectBase::ObjectBase; - }; + + {% for type in by_category["object"] %} + {% set Type = type.name.CamelCase() %} + {% if type.name.CamelCase() in client_special_objects %} + class {{Type}}; + {% else %} + struct {{type.name.CamelCase()}} : ObjectBase { + using ObjectBase::ObjectBase; + }; + {% endif %} + + inline {{Type}}* FromAPI(WGPU{{Type}} obj) { + return reinterpret_cast<{{Type}}*>(obj); + } + inline WGPU{{Type}} ToAPI({{Type}}* obj) { + return reinterpret_cast<WGPU{{Type}}>(obj); + } + {% endfor %} }} // namespace dawn_wire::client diff --git a/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiProcs.cpp b/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiProcs.cpp index a604a7f2a57..3edba1a22b2 100644 --- a/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiProcs.cpp +++ b/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiProcs.cpp @@ -14,7 +14,6 @@ #include "common/Log.h" #include "dawn_wire/client/ApiObjects.h" -#include "dawn_wire/client/ApiProcs_autogen.h" #include "dawn_wire/client/Client.h" #include <algorithm> @@ -157,9 +156,8 @@ namespace dawn_wire { namespace client { } while (false); if (DAWN_UNLIKELY(!sameDevice)) { - ClientDeviceInjectError(reinterpret_cast<WGPUDevice>(device), - WGPUErrorType_Validation, - "All objects must be from the same device."); + device->InjectError(WGPUErrorType_Validation, + "All objects must be from the same device."); {% if method.return_type.category == "object" %} // Allocate an object without registering it on the server. This is backed by a real allocation on // the client so commands can be sent with it. But because it's not allocated on the server, it will @@ -176,8 +174,8 @@ namespace dawn_wire { namespace client { } {% endif %} + auto self = reinterpret_cast<{{as_wireType(type)}}>(cSelf); {% if Suffix not in client_handwritten_commands %} - auto self = reinterpret_cast<{{as_wireType(type)}}>(cSelf); Device* device = self->device; {{Suffix}}Cmd cmd; @@ -202,9 +200,9 @@ namespace dawn_wire { namespace client { return reinterpret_cast<{{as_cType(method.return_type.name)}}>(allocation->object.get()); {% endif %} {% else %} - return ClientHandwritten{{Suffix}}(cSelf + return self->{{method.name.CamelCase()}}( {%- for arg in method.arguments -%} - , {{as_varName(arg.name)}} + {%if not loop.first %}, {% endif %} {{as_varName(arg.name)}} {%- endfor -%}); {% endif %} } @@ -241,6 +239,12 @@ namespace dawn_wire { namespace client { return nullptr; } + void ClientDeviceReference(WGPUDevice) { + } + + void ClientDeviceRelease(WGPUDevice) { + } + struct ProcEntry { WGPUProc proc; const char* name; diff --git a/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiProcs.h b/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiProcs.h deleted file mode 100644 index 2a27c3cc8e3..00000000000 --- a/chromium/third_party/dawn/generator/templates/dawn_wire/client/ApiProcs.h +++ /dev/null @@ -1,41 +0,0 @@ -//* Copyright 2019 The Dawn Authors -//* -//* Licensed under the Apache License, Version 2.0 (the "License"); -//* you may not use this file except in compliance with the License. -//* You may obtain a copy of the License at -//* -//* http://www.apache.org/licenses/LICENSE-2.0 -//* -//* Unless required by applicable law or agreed to in writing, software -//* distributed under the License is distributed on an "AS IS" BASIS, -//* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -//* See the License for the specific language governing permissions and -//* limitations under the License. - -#ifndef DAWNWIRE_CLIENT_APIPROCS_AUTOGEN_H_ -#define DAWNWIRE_CLIENT_APIPROCS_AUTOGEN_H_ - -#include <dawn/webgpu.h> - -namespace dawn_wire { namespace client { - - //* Dawn API - {% for type in by_category["object"] %} - {% set cType = as_cType(type.name) %} - {% for method in c_methods(type) %} - {% set Suffix = as_MethodSuffix(type.name, method.name) %} - {% if Suffix in client_handwritten_commands %} - {% set Suffix = "Handwritten" + Suffix %} - {% endif %} - {{as_cType(method.return_type.name)}} Client{{Suffix}}( - {{-cType}} cSelf - {%- for arg in method.arguments -%} - , {{as_annotated_cType(arg)}} - {%- endfor -%} - ); - {% endfor %} - {% endfor %} - -}} // namespace dawn_wire::client - -#endif // DAWNWIRE_CLIENT_APIPROCS_AUTOGEN_H_ diff --git a/chromium/third_party/dawn/generator/templates/mock_webgpu.cpp b/chromium/third_party/dawn/generator/templates/mock_webgpu.cpp index 6b8c5a6863c..0abfe9a93b4 100644 --- a/chromium/third_party/dawn/generator/templates/mock_webgpu.cpp +++ b/chromium/third_party/dawn/generator/templates/mock_webgpu.cpp @@ -96,6 +96,19 @@ void ProcTableAsClass::BufferMapWriteAsync(WGPUBuffer self, OnBufferMapWriteAsyncCallback(self, callback, userdata); } +void ProcTableAsClass::BufferMapAsync(WGPUBuffer self, + WGPUMapModeFlags mode, + size_t offset, + size_t size, + WGPUBufferMapCallback callback, + void* userdata) { + auto object = reinterpret_cast<ProcTableAsClass::Object*>(self); + object->mapAsyncCallback = callback; + object->userdata = userdata; + + OnBufferMapAsyncCallback(self, callback, userdata); +} + void ProcTableAsClass::FenceOnCompletion(WGPUFence self, uint64_t value, WGPUFenceOnCompletionCallback callback, @@ -135,6 +148,11 @@ void ProcTableAsClass::CallMapWriteCallback(WGPUBuffer buffer, object->mapWriteCallback(status, data, dataLength, object->userdata); } +void ProcTableAsClass::CallMapAsyncCallback(WGPUBuffer buffer, WGPUBufferMapAsyncStatus status) { + auto object = reinterpret_cast<ProcTableAsClass::Object*>(buffer); + object->mapAsyncCallback(status, object->userdata); +} + void ProcTableAsClass::CallFenceOnCompletionCallback(WGPUFence fence, WGPUFenceCompletionStatus status) { auto object = reinterpret_cast<ProcTableAsClass::Object*>(fence); diff --git a/chromium/third_party/dawn/generator/templates/mock_webgpu.h b/chromium/third_party/dawn/generator/templates/mock_webgpu.h index 78b9b78ef30..0a44248e004 100644 --- a/chromium/third_party/dawn/generator/templates/mock_webgpu.h +++ b/chromium/third_party/dawn/generator/templates/mock_webgpu.h @@ -65,6 +65,12 @@ class ProcTableAsClass { void BufferMapWriteAsync(WGPUBuffer self, WGPUBufferMapWriteCallback callback, void* userdata); + void BufferMapAsync(WGPUBuffer self, + WGPUMapModeFlags mode, + size_t offset, + size_t size, + WGPUBufferMapCallback callback, + void* userdata); void FenceOnCompletion(WGPUFence self, uint64_t value, WGPUFenceOnCompletionCallback callback, @@ -86,6 +92,9 @@ class ProcTableAsClass { virtual void OnBufferMapWriteAsyncCallback(WGPUBuffer buffer, WGPUBufferMapWriteCallback callback, void* userdata) = 0; + virtual void OnBufferMapAsyncCallback(WGPUBuffer buffer, + WGPUBufferMapCallback callback, + void* userdata) = 0; virtual void OnFenceOnCompletionCallback(WGPUFence fence, uint64_t value, WGPUFenceOnCompletionCallback callback, @@ -96,6 +105,7 @@ class ProcTableAsClass { void CallDeviceLostCallback(WGPUDevice device, const char* message); void CallMapReadCallback(WGPUBuffer buffer, WGPUBufferMapAsyncStatus status, const void* data, uint64_t dataLength); void CallMapWriteCallback(WGPUBuffer buffer, WGPUBufferMapAsyncStatus status, void* data, uint64_t dataLength); + void CallMapAsyncCallback(WGPUBuffer buffer, WGPUBufferMapAsyncStatus status); void CallFenceOnCompletionCallback(WGPUFence fence, WGPUFenceCompletionStatus status); struct Object { @@ -104,6 +114,7 @@ class ProcTableAsClass { WGPUDeviceLostCallback deviceLostCallback = nullptr; WGPUBufferMapReadCallback mapReadCallback = nullptr; WGPUBufferMapWriteCallback mapWriteCallback = nullptr; + WGPUBufferMapCallback mapAsyncCallback = nullptr; WGPUFenceOnCompletionCallback fenceOnCompletionCallback = nullptr; void* userdata = 0; }; @@ -140,6 +151,10 @@ class MockProcTable : public ProcTableAsClass { MOCK_METHOD(bool, OnDevicePopErrorScopeCallback, (WGPUDevice device, WGPUErrorCallback callback, void* userdata), (override)); MOCK_METHOD(void, OnBufferMapReadAsyncCallback, (WGPUBuffer buffer, WGPUBufferMapReadCallback callback, void* userdata), (override)); MOCK_METHOD(void, OnBufferMapWriteAsyncCallback, (WGPUBuffer buffer, WGPUBufferMapWriteCallback callback, void* userdata), (override)); + MOCK_METHOD(void, + OnBufferMapAsyncCallback, + (WGPUBuffer buffer, WGPUBufferMapCallback callback, void* userdata), + (override)); MOCK_METHOD(void, OnFenceOnCompletionCallback, (WGPUFence fence, uint64_t value, WGPUFenceOnCompletionCallback callback, void* userdata), (override)); }; diff --git a/chromium/third_party/dawn/generator/templates/webgpu.h b/chromium/third_party/dawn/generator/templates/webgpu.h index 4e1c7c8ef34..ec9db20519a 100644 --- a/chromium/third_party/dawn/generator/templates/webgpu.h +++ b/chromium/third_party/dawn/generator/templates/webgpu.h @@ -93,6 +93,8 @@ typedef uint32_t WGPUFlags; {% endif %} {% endfor %} +//* TODO(dawn:22) remove this once the PSA is sent and the deadline passed. +#define WGPUTextureFormat_RG11B10Float WGPUTextureFormat_RG11B10Ufloat typedef struct WGPUChainedStruct { struct WGPUChainedStruct const * next; diff --git a/chromium/third_party/dawn/generator/templates/webgpu_cpp.cpp b/chromium/third_party/dawn/generator/templates/webgpu_cpp.cpp index 7d22f640af4..be17a5d6d82 100644 --- a/chromium/third_party/dawn/generator/templates/webgpu_cpp.cpp +++ b/chromium/third_party/dawn/generator/templates/webgpu_cpp.cpp @@ -11,37 +11,39 @@ //* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. //* See the License for the specific language governing permissions and //* limitations under the License. - #include "dawn/webgpu_cpp.h" namespace wgpu { - {% for type in by_category["enum"] %} {% set CppType = as_cppType(type.name) %} {% set CType = as_cType(type.name) %} + // {{CppType}} + static_assert(sizeof({{CppType}}) == sizeof({{CType}}), "sizeof mismatch for {{CppType}}"); static_assert(alignof({{CppType}}) == alignof({{CType}}), "alignof mismatch for {{CppType}}"); {% for value in type.values %} static_assert(static_cast<uint32_t>({{CppType}}::{{as_cppEnum(value.name)}}) == {{as_cEnum(type.name, value.name)}}, "value mismatch for {{CppType}}::{{as_cppEnum(value.name)}}"); {% endfor %} - - {% endfor %} + {% endfor -%} {% for type in by_category["bitmask"] %} {% set CppType = as_cppType(type.name) %} {% set CType = as_cType(type.name) + "Flags" %} + // {{CppType}} + static_assert(sizeof({{CppType}}) == sizeof({{CType}}), "sizeof mismatch for {{CppType}}"); static_assert(alignof({{CppType}}) == alignof({{CType}}), "alignof mismatch for {{CppType}}"); {% for value in type.values %} static_assert(static_cast<uint32_t>({{CppType}}::{{as_cppEnum(value.name)}}) == {{as_cEnum(type.name, value.name)}}, "value mismatch for {{CppType}}::{{as_cppEnum(value.name)}}"); {% endfor %} - {% endfor %} + // ChainedStruct + static_assert(sizeof(ChainedStruct) == sizeof(WGPUChainedStruct), "sizeof mismatch for ChainedStruct"); static_assert(alignof(ChainedStruct) == alignof(WGPUChainedStruct), @@ -50,11 +52,12 @@ namespace wgpu { "offsetof mismatch for ChainedStruct::nextInChain"); static_assert(offsetof(ChainedStruct, sType) == offsetof(WGPUChainedStruct, sType), "offsetof mismatch for ChainedStruct::sType"); - {% for type in by_category["structure"] %} {% set CppType = as_cppType(type.name) %} {% set CType = as_cType(type.name) %} + // {{CppType}} + static_assert(sizeof({{CppType}}) == sizeof({{CType}}), "sizeof mismatch for {{CppType}}"); static_assert(alignof({{CppType}}) == alignof({{CType}}), "alignof mismatch for {{CppType}}"); @@ -67,17 +70,18 @@ namespace wgpu { static_assert(offsetof({{CppType}}, {{memberName}}) == offsetof({{CType}}, {{memberName}}), "offsetof mismatch for {{CppType}}::{{memberName}}"); {% endfor %} - - {% endfor %} + {% endfor -%} {% for type in by_category["object"] %} {% set CppType = as_cppType(type.name) %} {% set CType = as_cType(type.name) %} + // {{CppType}} + static_assert(sizeof({{CppType}}) == sizeof({{CType}}), "sizeof mismatch for {{CppType}}"); static_assert(alignof({{CppType}}) == alignof({{CType}}), "alignof mismatch for {{CppType}}"); - {% macro render_cpp_method_declaration(type, method) %} + {% macro render_cpp_method_declaration(type, method) -%} {% set CppType = as_cppType(type.name) %} {{as_cppType(method.return_type.name)}} {{CppType}}::{{method.name.CamelCase()}}( {%- for arg in method.arguments -%} @@ -89,9 +93,9 @@ namespace wgpu { {%- endif -%} {%- endfor -%} ) const - {%- endmacro %} + {%- endmacro -%} - {% macro render_cpp_to_c_method_call(type, method) -%} + {%- macro render_cpp_to_c_method_call(type, method) -%} {{as_cMethod(type.name, method.name)}}(Get() {%- for arg in method.arguments -%},{{" "}} {%- if arg.annotation == "value" -%} @@ -109,9 +113,9 @@ namespace wgpu { {%- endif -%} {%- endfor -%} ) - {%- endmacro %} + {%- endmacro -%} - {% for method in type.methods %} + {% for method in type.methods -%} {{render_cpp_method_declaration(type, method)}} { {% if method.return_type.name.concatcase() == "void" %} {{render_cpp_to_c_method_call(type, method)}}; @@ -131,9 +135,10 @@ namespace wgpu { {{as_cMethod(type.name, Name("release"))}}(handle); } } - {% endfor %} + // Instance + Instance CreateInstance(const InstanceDescriptor* descriptor) { const WGPUInstanceDescriptor* cDescriptor = reinterpret_cast<const WGPUInstanceDescriptor*>(descriptor); diff --git a/chromium/third_party/dawn/generator/templates/webgpu_cpp.h b/chromium/third_party/dawn/generator/templates/webgpu_cpp.h index 3dde1e803e3..d02bf986ace 100644 --- a/chromium/third_party/dawn/generator/templates/webgpu_cpp.h +++ b/chromium/third_party/dawn/generator/templates/webgpu_cpp.h @@ -11,7 +11,6 @@ //* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. //* See the License for the specific language governing permissions and //* limitations under the License. - #ifndef WEBGPU_CPP_H_ #define WEBGPU_CPP_H_ @@ -27,6 +26,10 @@ namespace wgpu { {% for value in type.values %} {{as_cppEnum(value.name)}} = 0x{{format(value.value, "08X")}}, {% endfor %} + //* TODO(dawn:22) remove this once the PSA is sent and the deadline passed. + {% if type.name.canonical_case() == "texture format" %} + RG11B10Float = RG11B10Ufloat, + {% endif %} }; {% endfor %} diff --git a/chromium/third_party/dawn/generator/templates/webgpu_struct_info.json b/chromium/third_party/dawn/generator/templates/webgpu_struct_info.json index 5120ba8d5aa..c4b40001ccd 100644 --- a/chromium/third_party/dawn/generator/templates/webgpu_struct_info.json +++ b/chromium/third_party/dawn/generator/templates/webgpu_struct_info.json @@ -23,14 +23,13 @@ "defines": [], "structs": { "WGPUChainedStruct": [ - "nextInChain", + "next", "sType" ], {% for type in by_category["structure"] if type.javascript %} "{{as_cType(type.name)}}": [ {% if type.chained %} - "nextInChain", - "sType" + "chain" {%- elif type.extensible %} "nextInChain" {%- endif %} diff --git a/chromium/third_party/dawn/infra/config/PRESUBMIT.py b/chromium/third_party/dawn/infra/config/PRESUBMIT.py index 8f08f24a495..de9b7eb0236 100644 --- a/chromium/third_party/dawn/infra/config/PRESUBMIT.py +++ b/chromium/third_party/dawn/infra/config/PRESUBMIT.py @@ -14,8 +14,10 @@ def CheckChangeOnUpload(input_api, output_api): - return input_api.canned_checks.CheckChangedLUCIConfigs(input_api, output_api) + return input_api.canned_checks.CheckChangedLUCIConfigs( + input_api, output_api) def CheckChangeOnCommit(input_api, output_api): - return input_api.canned_checks.CheckChangedLUCIConfigs(input_api, output_api) + return input_api.canned_checks.CheckChangedLUCIConfigs( + input_api, output_api) diff --git a/chromium/third_party/dawn/scripts/dawn_features.gni b/chromium/third_party/dawn/scripts/dawn_features.gni index 1cee17450fa..9d3cbfcf250 100644 --- a/chromium/third_party/dawn/scripts/dawn_features.gni +++ b/chromium/third_party/dawn/scripts/dawn_features.gni @@ -69,9 +69,9 @@ declare_args() { # Whether Dawn should enable X11 support. dawn_use_x11 = is_linux && !is_chromeos - # Enable support WGSL for shaders. Currently this is - # experimental, and requires dawn_tint_dir to be set. - dawn_enable_wgsl = false + # Enable support WGSL for shaders. + # Turned off for Skia, because Tint is currently not part of its DEPS. + dawn_enable_wgsl = !defined(is_skia_standalone) } # GN does not allow reading a variable defined in the same declare_args(). diff --git a/chromium/third_party/dawn/scripts/git-clang-format b/chromium/third_party/dawn/scripts/git-clang-format deleted file mode 100755 index 60cd4fb25b6..00000000000 --- a/chromium/third_party/dawn/scripts/git-clang-format +++ /dev/null @@ -1,579 +0,0 @@ -#!/usr/bin/env python -# -#===- git-clang-format - ClangFormat Git Integration ---------*- python -*--===# -# -# The LLVM Compiler Infrastructure -# -# This file is distributed under the University of Illinois Open Source -# License. See LICENSE.TXT for details. -# -#===------------------------------------------------------------------------===# - -r""" -clang-format git integration -============================ - -This file provides a clang-format integration for git. Put it somewhere in your -path and ensure that it is executable. Then, "git clang-format" will invoke -clang-format on the changes in current files or a specific commit. - -For further details, run: -git clang-format -h - -Requires Python 2.7 or Python 3 -""" - -from __future__ import print_function -import argparse -import collections -import contextlib -import errno -import os -import re -import subprocess -import sys - -usage = 'git clang-format [OPTIONS] [<commit>] [<commit>] [--] [<file>...]' - -desc = ''' -If zero or one commits are given, run clang-format on all lines that differ -between the working directory and <commit>, which defaults to HEAD. Changes are -only applied to the working directory. - -If two commits are given (requires --diff), run clang-format on all lines in the -second <commit> that differ from the first <commit>. - -The following git-config settings set the default of the corresponding option: - clangFormat.binary - clangFormat.commit - clangFormat.extension - clangFormat.style -''' - -# Name of the temporary index file in which save the output of clang-format. -# This file is created within the .git directory. -temp_index_basename = 'clang-format-index' - - -Range = collections.namedtuple('Range', 'start, count') - - -def main(): - config = load_git_config() - - # In order to keep '--' yet allow options after positionals, we need to - # check for '--' ourselves. (Setting nargs='*' throws away the '--', while - # nargs=argparse.REMAINDER disallows options after positionals.) - argv = sys.argv[1:] - try: - idx = argv.index('--') - except ValueError: - dash_dash = [] - else: - dash_dash = argv[idx:] - argv = argv[:idx] - - default_extensions = ','.join([ - # From clang/lib/Frontend/FrontendOptions.cpp, all lower case - 'c', 'h', # C - 'm', # ObjC - 'mm', # ObjC++ - 'cc', 'cp', 'cpp', 'c++', 'cxx', 'hpp', # C++ - # Other languages that clang-format supports - 'proto', 'protodevel', # Protocol Buffers - 'java', # Java - 'js', # JavaScript - 'ts', # TypeScript - ]) - - p = argparse.ArgumentParser( - usage=usage, formatter_class=argparse.RawDescriptionHelpFormatter, - description=desc) - p.add_argument('--binary', - default=config.get('clangformat.binary', 'clang-format'), - help='path to clang-format'), - p.add_argument('--commit', - default=config.get('clangformat.commit', 'HEAD'), - help='default commit to use if none is specified'), - p.add_argument('--diff', action='store_true', - help='print a diff instead of applying the changes') - p.add_argument('--extensions', - default=config.get('clangformat.extensions', - default_extensions), - help=('comma-separated list of file extensions to format, ' - 'excluding the period and case-insensitive')), - p.add_argument('-f', '--force', action='store_true', - help='allow changes to unstaged files') - p.add_argument('-p', '--patch', action='store_true', - help='select hunks interactively') - p.add_argument('-q', '--quiet', action='count', default=0, - help='print less information') - p.add_argument('--style', - default=config.get('clangformat.style', None), - help='passed to clang-format'), - p.add_argument('-v', '--verbose', action='count', default=0, - help='print extra information') - # We gather all the remaining positional arguments into 'args' since we need - # to use some heuristics to determine whether or not <commit> was present. - # However, to print pretty messages, we make use of metavar and help. - p.add_argument('args', nargs='*', metavar='<commit>', - help='revision from which to compute the diff') - p.add_argument('ignored', nargs='*', metavar='<file>...', - help='if specified, only consider differences in these files') - opts = p.parse_args(argv) - - opts.verbose -= opts.quiet - del opts.quiet - - commits, files = interpret_args(opts.args, dash_dash, opts.commit) - if len(commits) > 1: - if not opts.diff: - die('--diff is required when two commits are given') - else: - if len(commits) > 2: - die('at most two commits allowed; %d given' % len(commits)) - changed_lines = compute_diff_and_extract_lines(commits, files) - if opts.verbose >= 1: - ignored_files = set(changed_lines) - filter_by_extension(changed_lines, opts.extensions.lower().split(',')) - if opts.verbose >= 1: - ignored_files.difference_update(changed_lines) - if ignored_files: - print('Ignoring changes in the following files (wrong extension):') - for filename in ignored_files: - print(' %s' % filename) - if changed_lines: - print('Running clang-format on the following files:') - for filename in changed_lines: - print(' %s' % filename) - if not changed_lines: - print('no modified files to format') - return - # The computed diff outputs absolute paths, so we must cd before accessing - # those files. - cd_to_toplevel() - if len(commits) > 1: - old_tree = commits[1] - new_tree = run_clang_format_and_save_to_tree(changed_lines, - revision=commits[1], - binary=opts.binary, - style=opts.style) - else: - old_tree = create_tree_from_workdir(changed_lines) - new_tree = run_clang_format_and_save_to_tree(changed_lines, - binary=opts.binary, - style=opts.style) - if opts.verbose >= 1: - print('old tree: %s' % old_tree) - print('new tree: %s' % new_tree) - if old_tree == new_tree: - if opts.verbose >= 0: - print('clang-format did not modify any files') - elif opts.diff: - print_diff(old_tree, new_tree) - else: - changed_files = apply_changes(old_tree, new_tree, force=opts.force, - patch_mode=opts.patch) - if (opts.verbose >= 0 and not opts.patch) or opts.verbose >= 1: - print('changed files:') - for filename in changed_files: - print(' %s' % filename) - - -def load_git_config(non_string_options=None): - """Return the git configuration as a dictionary. - - All options are assumed to be strings unless in `non_string_options`, in which - is a dictionary mapping option name (in lower case) to either "--bool" or - "--int".""" - if non_string_options is None: - non_string_options = {} - out = {} - for entry in run('git', 'config', '--list', '--null').split('\0'): - if entry: - name, value = entry.split('\n', 1) - if name in non_string_options: - value = run('git', 'config', non_string_options[name], name) - out[name] = value - return out - - -def interpret_args(args, dash_dash, default_commit): - """Interpret `args` as "[commits] [--] [files]" and return (commits, files). - - It is assumed that "--" and everything that follows has been removed from - args and placed in `dash_dash`. - - If "--" is present (i.e., `dash_dash` is non-empty), the arguments to its - left (if present) are taken as commits. Otherwise, the arguments are checked - from left to right if they are commits or files. If commits are not given, - a list with `default_commit` is used.""" - if dash_dash: - if len(args) == 0: - commits = [default_commit] - else: - commits = args - for commit in commits: - object_type = get_object_type(commit) - if object_type not in ('commit', 'tag'): - if object_type is None: - die("'%s' is not a commit" % commit) - else: - die("'%s' is a %s, but a commit was expected" % (commit, object_type)) - files = dash_dash[1:] - elif args: - commits = [] - while args: - if not disambiguate_revision(args[0]): - break - commits.append(args.pop(0)) - if not commits: - commits = [default_commit] - files = args - else: - commits = [default_commit] - files = [] - return commits, files - - -def disambiguate_revision(value): - """Returns True if `value` is a revision, False if it is a file, or dies.""" - # If `value` is ambiguous (neither a commit nor a file), the following - # command will die with an appropriate error message. - run('git', 'rev-parse', value, verbose=False) - object_type = get_object_type(value) - if object_type is None: - return False - if object_type in ('commit', 'tag'): - return True - die('`%s` is a %s, but a commit or filename was expected' % - (value, object_type)) - - -def get_object_type(value): - """Returns a string description of an object's type, or None if it is not - a valid git object.""" - cmd = ['git', 'cat-file', '-t', value] - p = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE) - stdout, stderr = p.communicate() - if p.returncode != 0: - return None - return convert_string(stdout.strip()) - - -def compute_diff_and_extract_lines(commits, files): - """Calls compute_diff() followed by extract_lines().""" - diff_process = compute_diff(commits, files) - changed_lines = extract_lines(diff_process.stdout) - diff_process.stdout.close() - diff_process.wait() - if diff_process.returncode != 0: - # Assume error was already printed to stderr. - sys.exit(2) - return changed_lines - - -def compute_diff(commits, files): - """Return a subprocess object producing the diff from `commits`. - - The return value's `stdin` file object will produce a patch with the - differences between the working directory and the first commit if a single - one was specified, or the difference between both specified commits, filtered - on `files` (if non-empty). Zero context lines are used in the patch.""" - git_tool = 'diff-index' - if len(commits) > 1: - git_tool = 'diff-tree' - cmd = ['git', git_tool, '-p', '-U0'] + commits + ['--'] - cmd.extend(files) - p = subprocess.Popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE) - p.stdin.close() - return p - - -def extract_lines(patch_file): - """Extract the changed lines in `patch_file`. - - The return value is a dictionary mapping filename to a list of (start_line, - line_count) pairs. - - The input must have been produced with ``-U0``, meaning unidiff format with - zero lines of context. The return value is a dict mapping filename to a - list of line `Range`s.""" - matches = {} - for line in patch_file: - line = convert_string(line) - match = re.search(r'^\+\+\+\ [^/]+/(.*)', line) - if match: - filename = match.group(1).rstrip('\r\n') - match = re.search(r'^@@ -[0-9,]+ \+(\d+)(,(\d+))?', line) - if match: - start_line = int(match.group(1)) - line_count = 1 - if match.group(3): - line_count = int(match.group(3)) - if line_count > 0: - matches.setdefault(filename, []).append(Range(start_line, line_count)) - return matches - - -def filter_by_extension(dictionary, allowed_extensions): - """Delete every key in `dictionary` that doesn't have an allowed extension. - - `allowed_extensions` must be a collection of lowercase file extensions, - excluding the period.""" - allowed_extensions = frozenset(allowed_extensions) - for filename in list(dictionary.keys()): - base_ext = filename.rsplit('.', 1) - if len(base_ext) == 1 and '' in allowed_extensions: - continue - if len(base_ext) == 1 or base_ext[1].lower() not in allowed_extensions: - del dictionary[filename] - - -def cd_to_toplevel(): - """Change to the top level of the git repository.""" - toplevel = run('git', 'rev-parse', '--show-toplevel') - os.chdir(toplevel) - - -def create_tree_from_workdir(filenames): - """Create a new git tree with the given files from the working directory. - - Returns the object ID (SHA-1) of the created tree.""" - return create_tree(filenames, '--stdin') - - -def run_clang_format_and_save_to_tree(changed_lines, revision=None, - binary='clang-format', style=None): - """Run clang-format on each file and save the result to a git tree. - - Returns the object ID (SHA-1) of the created tree.""" - def iteritems(container): - try: - return container.iteritems() # Python 2 - except AttributeError: - return container.items() # Python 3 - def index_info_generator(): - for filename, line_ranges in iteritems(changed_lines): - if revision: - git_metadata_cmd = ['git', 'ls-tree', - '%s:%s' % (revision, os.path.dirname(filename)), - os.path.basename(filename)] - git_metadata = subprocess.Popen(git_metadata_cmd, stdin=subprocess.PIPE, - stdout=subprocess.PIPE) - stdout = git_metadata.communicate()[0] - mode = oct(int(stdout.split()[0], 8)) - else: - mode = oct(os.stat(filename).st_mode) - # Adjust python3 octal format so that it matches what git expects - if mode.startswith('0o'): - mode = '0' + mode[2:] - blob_id = clang_format_to_blob(filename, line_ranges, - revision=revision, - binary=binary, - style=style) - yield '%s %s\t%s' % (mode, blob_id, filename) - return create_tree(index_info_generator(), '--index-info') - - -def create_tree(input_lines, mode): - """Create a tree object from the given input. - - If mode is '--stdin', it must be a list of filenames. If mode is - '--index-info' is must be a list of values suitable for "git update-index - --index-info", such as "<mode> <SP> <sha1> <TAB> <filename>". Any other mode - is invalid.""" - assert mode in ('--stdin', '--index-info') - cmd = ['git', 'update-index', '--add', '-z', mode] - with temporary_index_file(): - p = subprocess.Popen(cmd, stdin=subprocess.PIPE) - for line in input_lines: - p.stdin.write(to_bytes('%s\0' % line)) - p.stdin.close() - if p.wait() != 0: - die('`%s` failed' % ' '.join(cmd)) - tree_id = run('git', 'write-tree') - return tree_id - - -def clang_format_to_blob(filename, line_ranges, revision=None, - binary='clang-format', style=None): - """Run clang-format on the given file and save the result to a git blob. - - Runs on the file in `revision` if not None, or on the file in the working - directory if `revision` is None. - - Returns the object ID (SHA-1) of the created blob.""" - clang_format_cmd = [binary] - if style: - clang_format_cmd.extend(['-style='+style]) - clang_format_cmd.extend([ - '-lines=%s:%s' % (start_line, start_line+line_count-1) - for start_line, line_count in line_ranges]) - if revision: - clang_format_cmd.extend(['-assume-filename='+filename]) - git_show_cmd = ['git', 'cat-file', 'blob', '%s:%s' % (revision, filename)] - git_show = subprocess.Popen(git_show_cmd, stdin=subprocess.PIPE, - stdout=subprocess.PIPE) - git_show.stdin.close() - clang_format_stdin = git_show.stdout - else: - clang_format_cmd.extend([filename]) - git_show = None - clang_format_stdin = subprocess.PIPE - try: - clang_format = subprocess.Popen(clang_format_cmd, stdin=clang_format_stdin, - stdout=subprocess.PIPE) - if clang_format_stdin == subprocess.PIPE: - clang_format_stdin = clang_format.stdin - except OSError as e: - if e.errno == errno.ENOENT: - die('cannot find executable "%s"' % binary) - else: - raise - clang_format_stdin.close() - hash_object_cmd = ['git', 'hash-object', '-w', '--path='+filename, '--stdin'] - hash_object = subprocess.Popen(hash_object_cmd, stdin=clang_format.stdout, - stdout=subprocess.PIPE) - clang_format.stdout.close() - stdout = hash_object.communicate()[0] - if hash_object.returncode != 0: - die('`%s` failed' % ' '.join(hash_object_cmd)) - if clang_format.wait() != 0: - die('`%s` failed' % ' '.join(clang_format_cmd)) - if git_show and git_show.wait() != 0: - die('`%s` failed' % ' '.join(git_show_cmd)) - return convert_string(stdout).rstrip('\r\n') - - -@contextlib.contextmanager -def temporary_index_file(tree=None): - """Context manager for setting GIT_INDEX_FILE to a temporary file and deleting - the file afterward.""" - index_path = create_temporary_index(tree) - old_index_path = os.environ.get('GIT_INDEX_FILE') - os.environ['GIT_INDEX_FILE'] = index_path - try: - yield - finally: - if old_index_path is None: - del os.environ['GIT_INDEX_FILE'] - else: - os.environ['GIT_INDEX_FILE'] = old_index_path - os.remove(index_path) - - -def create_temporary_index(tree=None): - """Create a temporary index file and return the created file's path. - - If `tree` is not None, use that as the tree to read in. Otherwise, an - empty index is created.""" - gitdir = run('git', 'rev-parse', '--git-dir') - path = os.path.join(gitdir, temp_index_basename) - if tree is None: - tree = '--empty' - run('git', 'read-tree', '--index-output='+path, tree) - return path - - -def print_diff(old_tree, new_tree): - """Print the diff between the two trees to stdout.""" - # We use the porcelain 'diff' and not plumbing 'diff-tree' because the output - # is expected to be viewed by the user, and only the former does nice things - # like color and pagination. - # - # We also only print modified files since `new_tree` only contains the files - # that were modified, so unmodified files would show as deleted without the - # filter. - subprocess.check_call(['git', 'diff', '--diff-filter=M', old_tree, new_tree, - '--']) - - -def apply_changes(old_tree, new_tree, force=False, patch_mode=False): - """Apply the changes in `new_tree` to the working directory. - - Bails if there are local changes in those files and not `force`. If - `patch_mode`, runs `git checkout --patch` to select hunks interactively.""" - changed_files = run('git', 'diff-tree', '--diff-filter=M', '-r', '-z', - '--name-only', old_tree, - new_tree).rstrip('\0').split('\0') - if not force: - unstaged_files = run('git', 'diff-files', '--name-status', *changed_files) - if unstaged_files: - print('The following files would be modified but ' - 'have unstaged changes:', file=sys.stderr) - print(unstaged_files, file=sys.stderr) - print('Please commit, stage, or stash them first.', file=sys.stderr) - sys.exit(2) - if patch_mode: - # In patch mode, we could just as well create an index from the new tree - # and checkout from that, but then the user will be presented with a - # message saying "Discard ... from worktree". Instead, we use the old - # tree as the index and checkout from new_tree, which gives the slightly - # better message, "Apply ... to index and worktree". This is not quite - # right, since it won't be applied to the user's index, but oh well. - with temporary_index_file(old_tree): - subprocess.check_call(['git', 'checkout', '--patch', new_tree]) - index_tree = old_tree - else: - with temporary_index_file(new_tree): - run('git', 'checkout-index', '-a', '-f') - return changed_files - - -def run(*args, **kwargs): - stdin = kwargs.pop('stdin', '') - verbose = kwargs.pop('verbose', True) - strip = kwargs.pop('strip', True) - for name in kwargs: - raise TypeError("run() got an unexpected keyword argument '%s'" % name) - p = subprocess.Popen(args, stdout=subprocess.PIPE, stderr=subprocess.PIPE, - stdin=subprocess.PIPE) - stdout, stderr = p.communicate(input=stdin) - - stdout = convert_string(stdout) - stderr = convert_string(stderr) - - if p.returncode == 0: - if stderr: - if verbose: - print('`%s` printed to stderr:' % ' '.join(args), file=sys.stderr) - print(stderr.rstrip(), file=sys.stderr) - if strip: - stdout = stdout.rstrip('\r\n') - return stdout - if verbose: - print('`%s` returned %s' % (' '.join(args), p.returncode), file=sys.stderr) - if stderr: - print(stderr.rstrip(), file=sys.stderr) - sys.exit(2) - - -def die(message): - print('error:', message, file=sys.stderr) - sys.exit(2) - - -def to_bytes(str_input): - # Encode to UTF-8 to get binary data. - if isinstance(str_input, bytes): - return str_input - return str_input.encode('utf-8') - - -def to_string(bytes_input): - if isinstance(bytes_input, str): - return bytes_input - return bytes_input.encode('utf-8') - - -def convert_string(bytes_input): - try: - return to_string(bytes_input.decode('utf-8')) - except AttributeError: # 'str' object has no attribute 'decode'. - return str(bytes_input) - except UnicodeError: - return str(bytes_input) - -if __name__ == '__main__': - main() diff --git a/chromium/third_party/dawn/scripts/lint_clang_format.sh b/chromium/third_party/dawn/scripts/lint_clang_format.sh deleted file mode 100755 index 82b572c23cd..00000000000 --- a/chromium/third_party/dawn/scripts/lint_clang_format.sh +++ /dev/null @@ -1,44 +0,0 @@ -#!/bin/bash - -# Copyright 2018 The Dawn Authors -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -clang_format=$1 -base_commit=$2 - -echo -skipped_directories="(examples|generator|src/tests/(unittests|end2end)|third_party)" -# Find the files modified that need formatting -files_to_check=$(git diff --diff-filter=ACMR --name-only $base_commit | grep -E "*\.(c|cpp|mm|h)$" | grep -vE "^$skipped_directories/*") -if [ -z "$files_to_check" ]; then - echo "No modified files to format." - exit 0 -fi -echo "Checking formatting diff on these files:" -echo "$files_to_check" -echo -files_to_check=$(echo $files_to_check | tr '\n' ' ') - -# Run git-clang-format, check if it formatted anything -format_output=$(scripts/git-clang-format --binary $clang_format --commit $base_commit --diff --style=file $files_to_check) -if [ "$format_output" == "clang-format did not modify any files" ] || [ "$format_output" == "no modified files to format" ] ; then - exit 0 -fi - -# clang-format made changes, print them and fail Travis -echo "Following formatting changes needed:" -echo -echo "$format_output" -echo -exit 1 diff --git a/chromium/third_party/dawn/scripts/perf_test_runner.py b/chromium/third_party/dawn/scripts/perf_test_runner.py index c4cc55cf818..157d4497a14 100755 --- a/chromium/third_party/dawn/scripts/perf_test_runner.py +++ b/chromium/third_party/dawn/scripts/perf_test_runner.py @@ -22,7 +22,8 @@ import sys import os import re -base_path = os.path.abspath(os.path.join(os.path.dirname(os.path.abspath(__file__)), '..')) +base_path = os.path.abspath( + os.path.join(os.path.dirname(os.path.abspath(__file__)), '..')) # Look for a [Rr]elease build. perftests_paths = glob.glob('out/*elease*') @@ -35,6 +36,7 @@ if sys.platform == 'win32': scores = [] + def mean(data): """Return the sample arithmetic mean of data.""" n = len(data) @@ -102,6 +104,7 @@ if len(sys.argv) >= 2: print('Using test executable: ' + perftests_path) print('Test name: ' + test_name) + def get_results(metric, extra_args=[]): process = subprocess.Popen( [perftests_path, '--gtest_filter=' + test_name] + extra_args, @@ -139,11 +142,14 @@ for experiment in range(max_experiments): if (len(scores) > 1): sys.stdout.write(", mean: %.2f" % mean(scores)) - sys.stdout.write(", variation: %.2f%%" % (coefficient_of_variation(scores) * 100.0)) + sys.stdout.write(", variation: %.2f%%" % + (coefficient_of_variation(scores) * 100.0)) if (len(scores) > 7): truncation_n = len(scores) >> 3 - sys.stdout.write(", truncated mean: %.2f" % truncated_mean(scores, truncation_n)) - sys.stdout.write(", variation: %.2f%%" % (truncated_cov(scores, truncation_n) * 100.0)) + sys.stdout.write(", truncated mean: %.2f" % + truncated_mean(scores, truncation_n)) + sys.stdout.write(", variation: %.2f%%" % + (truncated_cov(scores, truncation_n) * 100.0)) print("") diff --git a/chromium/third_party/dawn/scripts/roll-shader-deps.sh b/chromium/third_party/dawn/scripts/roll-shader-deps.sh index ac6db480ac0..db16800867b 100755 --- a/chromium/third_party/dawn/scripts/roll-shader-deps.sh +++ b/chromium/third_party/dawn/scripts/roll-shader-deps.sh @@ -26,7 +26,7 @@ spirv_cross_dir="third_party/spirv-cross/" spirv_cross_trunk="origin/master" spirv_headers_dir="third_party/spirv-headers/" spirv_headers_trunk="origin/master" -spirv_tools_dir="third_party/spirv-tools/" +spirv_tools_dir="third_party/SPIRV-Tools/" spirv_tools_trunk="origin/master" tint_dir="third_party/tint/" tint_trunk="origin/main" diff --git a/chromium/third_party/dawn/scripts/travis_lint_format.sh b/chromium/third_party/dawn/scripts/travis_lint_format.sh deleted file mode 100755 index 3515d606f64..00000000000 --- a/chromium/third_party/dawn/scripts/travis_lint_format.sh +++ /dev/null @@ -1,27 +0,0 @@ -#!/bin/bash - -# Copyright 2018 The Dawn Authors -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -if [ "$TRAVIS_PULL_REQUEST" == "false" ]; then - echo "Running outside of pull request isn't supported yet" - exit 0 -fi - -# Choose the commit against which to format -base_commit=$(git rev-parse $TRAVIS_BRANCH) -echo "Formatting against $TRAVIS_BRANCH a.k.a. $base_commit..." -echo - -scripts/lint_clang_format.sh $1 $base_commit diff --git a/chromium/third_party/dawn/src/common/BUILD.gn b/chromium/third_party/dawn/src/common/BUILD.gn index 59a23057224..85d63d1ede7 100644 --- a/chromium/third_party/dawn/src/common/BUILD.gn +++ b/chromium/third_party/dawn/src/common/BUILD.gn @@ -84,8 +84,9 @@ config("dawn_internal") { } # Only internal Dawn targets can use this config, this means only targets in - # this BUILD.gn file. - visibility = [ ":*" ] + # this BUILD.gn file and related subdirs. + visibility = [ "../*" ] + cflags = [] # Enable more warnings that were found when using Dawn in other projects @@ -108,6 +109,12 @@ config("dawn_internal") { "-Wtautological-unsigned-zero-compare", ] + # Allow comparison against type limits that might be tautological on 32bit + # or 64bit systems. Without this the following produces an error on 64bit: + # + # if (myUint64 > std::numeric_limits<size_t>::max()) {...} + cflags += [ "-Wno-tautological-type-limit-compare" ] + if (is_win) { cflags += [ # clang-cl doesn't know -pedantic, pass it explicitly to the clang driver @@ -165,13 +172,17 @@ if (is_win || is_linux || is_mac || is_fuchsia || is_android) { "SerialStorage.h", "SlabAllocator.cpp", "SlabAllocator.h", + "StackContainer.h", "SwapChainUtils.h", "SystemUtils.cpp", "SystemUtils.h", "TypedInteger.h", + "UnderlyingType.h", "ityp_array.h", "ityp_bitset.h", "ityp_span.h", + "ityp_stack_vec.h", + "ityp_vector.h", "vulkan_platform.h", "windows_with_undefs.h", "xlib_with_undefs.h", diff --git a/chromium/third_party/dawn/src/common/BitSetIterator.h b/chromium/third_party/dawn/src/common/BitSetIterator.h index d35bc8a2df0..14ba4856ff7 100644 --- a/chromium/third_party/dawn/src/common/BitSetIterator.h +++ b/chromium/third_party/dawn/src/common/BitSetIterator.h @@ -48,7 +48,7 @@ class BitSetIterator final { T operator*() const { using U = UnderlyingType<T>; - ASSERT(mCurrentBit <= std::numeric_limits<U>::max()); + ASSERT(static_cast<U>(mCurrentBit) <= std::numeric_limits<U>::max()); return static_cast<T>(static_cast<U>(mCurrentBit)); } diff --git a/chromium/third_party/dawn/src/common/CMakeLists.txt b/chromium/third_party/dawn/src/common/CMakeLists.txt index 1ab20234b1f..a6f320e33fe 100644 --- a/chromium/third_party/dawn/src/common/CMakeLists.txt +++ b/chromium/third_party/dawn/src/common/CMakeLists.txt @@ -41,6 +41,7 @@ target_sources(dawn_common PRIVATE "SerialStorage.h" "SlabAllocator.cpp" "SlabAllocator.h" + "StackContainer.h" "SwapChainUtils.h" "SystemUtils.cpp" "SystemUtils.h" @@ -49,6 +50,8 @@ target_sources(dawn_common PRIVATE "ityp_array.h" "ityp_bitset.h" "ityp_span.h" + "ityp_stack_vec.h" + "ityp_vector.h" "vulkan_platform.h" "windows_with_undefs.h" "xlib_with_undefs.h" diff --git a/chromium/third_party/dawn/src/common/Constants.h b/chromium/third_party/dawn/src/common/Constants.h index df792cd5d6d..e281646bffa 100644 --- a/chromium/third_party/dawn/src/common/Constants.h +++ b/chromium/third_party/dawn/src/common/Constants.h @@ -18,8 +18,6 @@ #include <cstdint> static constexpr uint32_t kMaxBindGroups = 4u; -// TODO(cwallez@chromium.org): investigate bindgroup limits -static constexpr uint32_t kMaxBindingsPerGroup = 16u; static constexpr uint32_t kMaxVertexAttributes = 16u; // Vulkan has a standalone limit named maxVertexInputAttributeOffset (2047u at least) for vertex // attribute offset. The limit might be meaningless because Vulkan has another limit named @@ -35,13 +33,21 @@ static constexpr uint32_t kMaxColorAttachments = 4u; static constexpr uint32_t kTextureBytesPerRowAlignment = 256u; // Dynamic buffer offsets require offset to be divisible by 256 static constexpr uint64_t kMinDynamicBufferOffsetAlignment = 256u; -// Max numbers of dynamic uniform buffers -static constexpr uint32_t kMaxDynamicUniformBufferCount = 8u; -// Max numbers of dynamic storage buffers -static constexpr uint32_t kMaxDynamicStorageBufferCount = 4u; -// Max numbers of dynamic buffers -static constexpr uint32_t kMaxDynamicBufferCount = - kMaxDynamicUniformBufferCount + kMaxDynamicStorageBufferCount; + +// Per stage limits +static constexpr uint32_t kMaxSampledTexturesPerShaderStage = 16; +static constexpr uint32_t kMaxSamplersPerShaderStage = 16; +static constexpr uint32_t kMaxStorageBuffersPerShaderStage = 6; +static constexpr uint32_t kMaxStorageTexturesPerShaderStage = 4; +static constexpr uint32_t kMaxUniformBuffersPerShaderStage = 12; + +// Per pipeline layout limits +static constexpr uint32_t kMaxDynamicUniformBuffersPerPipelineLayout = 8u; +static constexpr uint32_t kMaxDynamicStorageBuffersPerPipelineLayout = 4u; + +// Max size of uniform buffer binding +static constexpr uint64_t kMaxUniformBufferBindingSize = 16384u; + // Indirect command sizes static constexpr uint64_t kDispatchIndirectSize = 3 * sizeof(uint32_t); static constexpr uint64_t kDrawIndirectSize = 4 * sizeof(uint32_t); @@ -58,4 +64,8 @@ static constexpr uint32_t kMaxTexture2DMipLevels = 14u; static_assert(1 << (kMaxTexture2DMipLevels - 1) == kMaxTextureSize, "kMaxTexture2DMipLevels and kMaxTextureSize size mismatch"); +// Offset alignment for CopyB2B. Strictly speaking this alignment is required only +// on macOS, but we decide to do it on all platforms. +static constexpr uint64_t kCopyBufferToBufferOffsetAlignment = 4u; + #endif // COMMON_CONSTANTS_H_ diff --git a/chromium/third_party/dawn/src/common/GPUInfo.cpp b/chromium/third_party/dawn/src/common/GPUInfo.cpp index 60076d2830b..c3ea9cefb5c 100644 --- a/chromium/third_party/dawn/src/common/GPUInfo.cpp +++ b/chromium/third_party/dawn/src/common/GPUInfo.cpp @@ -36,4 +36,7 @@ namespace gpu_info { bool IsSwiftshader(PCIVendorID vendorId, PCIDeviceID deviceId) { return vendorId == kVendorID_Google && deviceId == kDeviceID_Swiftshader; } + bool IsWARP(PCIVendorID vendorId, PCIDeviceID deviceId) { + return vendorId == kVendorID_Microsoft && deviceId == kDeviceID_WARP; + } } // namespace gpu_info diff --git a/chromium/third_party/dawn/src/common/GPUInfo.h b/chromium/third_party/dawn/src/common/GPUInfo.h index d16c89b9b99..87efbbc9dfb 100644 --- a/chromium/third_party/dawn/src/common/GPUInfo.h +++ b/chromium/third_party/dawn/src/common/GPUInfo.h @@ -29,8 +29,10 @@ namespace gpu_info { static constexpr PCIVendorID kVendorID_Nvidia = 0x10DE; static constexpr PCIVendorID kVendorID_Qualcomm = 0x5143; static constexpr PCIVendorID kVendorID_Google = 0x1AE0; + static constexpr PCIVendorID kVendorID_Microsoft = 0x1414; static constexpr PCIDeviceID kDeviceID_Swiftshader = 0xC0DE; + static constexpr PCIDeviceID kDeviceID_WARP = 0x8c; bool IsAMD(PCIVendorID vendorId); bool IsARM(PCIVendorID vendorId); @@ -39,6 +41,7 @@ namespace gpu_info { bool IsNvidia(PCIVendorID vendorId); bool IsQualcomm(PCIVendorID vendorId); bool IsSwiftshader(PCIVendorID vendorId, PCIDeviceID deviceId); + bool IsWARP(PCIVendorID vendorId, PCIDeviceID deviceId); } // namespace gpu_info #endif // COMMON_GPUINFO_H diff --git a/chromium/third_party/dawn/src/common/HashUtils.h b/chromium/third_party/dawn/src/common/HashUtils.h index 9d10ca713c2..1c33a3f2c1c 100644 --- a/chromium/third_party/dawn/src/common/HashUtils.h +++ b/chromium/third_party/dawn/src/common/HashUtils.h @@ -90,7 +90,7 @@ size_t Hash(const std::bitset<N>& value) { namespace std { template <typename Index, size_t N> - class hash<ityp::bitset<Index, N>> { + struct hash<ityp::bitset<Index, N>> { public: size_t operator()(const ityp::bitset<Index, N>& value) const { return Hash(static_cast<const std::bitset<N>&>(value)); diff --git a/chromium/third_party/dawn/src/common/LinkedList.h b/chromium/third_party/dawn/src/common/LinkedList.h index 69fcf78ab29..47ca1eb1040 100644 --- a/chromium/third_party/dawn/src/common/LinkedList.h +++ b/chromium/third_party/dawn/src/common/LinkedList.h @@ -197,4 +197,4 @@ class LinkedList { private: LinkNode<T> root_; }; -#endif // COMMON_LINKED_LIST_H
\ No newline at end of file +#endif // COMMON_LINKED_LIST_H diff --git a/chromium/third_party/dawn/src/common/Math.cpp b/chromium/third_party/dawn/src/common/Math.cpp index 62807556f55..8e5985f7e73 100644 --- a/chromium/third_party/dawn/src/common/Math.cpp +++ b/chromium/third_party/dawn/src/common/Math.cpp @@ -159,4 +159,4 @@ uint64_t RoundUp(uint64_t n, uint64_t m) { ASSERT(n > 0); ASSERT(m <= std::numeric_limits<uint64_t>::max() - n); return ((n + m - 1) / m) * m; -}
\ No newline at end of file +} diff --git a/chromium/third_party/dawn/src/common/Math.h b/chromium/third_party/dawn/src/common/Math.h index db941f279a9..c673785e260 100644 --- a/chromium/third_party/dawn/src/common/Math.h +++ b/chromium/third_party/dawn/src/common/Math.h @@ -31,6 +31,22 @@ uint32_t Log2(uint64_t value); bool IsPowerOfTwo(uint64_t n); uint64_t RoundUp(uint64_t n, uint64_t m); +constexpr uint32_t ConstexprLog2(uint64_t v) { + return v <= 1 ? 0 : 1 + ConstexprLog2(v / 2); +} + +constexpr uint32_t ConstexprLog2Ceil(uint64_t v) { + return v <= 1 ? 0 : ConstexprLog2(v - 1) + 1; +} + +inline uint32_t Log2Ceil(uint32_t v) { + return v <= 1 ? 0 : Log2(v - 1) + 1; +} + +inline uint32_t Log2Ceil(uint64_t v) { + return v <= 1 ? 0 : Log2(v - 1) + 1; +} + uint64_t NextPowerOfTwo(uint64_t n); bool IsPtrAligned(const void* ptr, size_t alignment); void* AlignVoidPtr(void* ptr, size_t alignment); diff --git a/chromium/third_party/dawn/src/common/Platform.h b/chromium/third_party/dawn/src/common/Platform.h index bc64db26d8f..af7b1751518 100644 --- a/chromium/third_party/dawn/src/common/Platform.h +++ b/chromium/third_party/dawn/src/common/Platform.h @@ -41,6 +41,10 @@ # define DAWN_PLATFORM_FUCHSIA 1 # define DAWN_PLATFORM_POSIX 1 +#elif defined(__EMSCRIPTEN__) +# define DAWN_PLATFORM_EMSCRIPTEN 1 +# define DAWN_PLATFORM_POSIX 1 + #else # error "Unsupported platform." #endif @@ -60,7 +64,7 @@ # define DAWN_PLATFORM_64_BIT 1 static_assert(sizeof(sizeof(char)) == 8, "Expect sizeof(size_t) == 8"); #elif defined(_WIN32) || defined(__arm__) || defined(__i386__) || defined(__mips32__) || \ - defined(__s390__) + defined(__s390__) || defined(__EMSCRIPTEN__) # define DAWN_PLATFORM_32_BIT 1 static_assert(sizeof(sizeof(char)) == 4, "Expect sizeof(size_t) == 4"); #else diff --git a/chromium/third_party/dawn/src/common/RefCounted.cpp b/chromium/third_party/dawn/src/common/RefCounted.cpp index bb0f76d8842..af38fc668e2 100644 --- a/chromium/third_party/dawn/src/common/RefCounted.cpp +++ b/chromium/third_party/dawn/src/common/RefCounted.cpp @@ -75,4 +75,4 @@ void RefCounted::Release() { void RefCounted::DeleteThis() { delete this; -}
\ No newline at end of file +} diff --git a/chromium/third_party/dawn/src/common/RefCounted.h b/chromium/third_party/dawn/src/common/RefCounted.h index b055d7b71e6..90144a6043b 100644 --- a/chromium/third_party/dawn/src/common/RefCounted.h +++ b/chromium/third_party/dawn/src/common/RefCounted.h @@ -122,6 +122,14 @@ class Ref { mPointee = nullptr; } + bool operator==(const T* other) const { + return mPointee == other; + } + + bool operator!=(const T* other) const { + return mPointee != other; + } + operator bool() { return mPointee != nullptr; } diff --git a/chromium/third_party/dawn/src/common/SerialStorage.h b/chromium/third_party/dawn/src/common/SerialStorage.h index 6f382134a72..d71c7dd07d5 100644 --- a/chromium/third_party/dawn/src/common/SerialStorage.h +++ b/chromium/third_party/dawn/src/common/SerialStorage.h @@ -280,8 +280,8 @@ SerialStorage<Derived>::ConstIterator::ConstIterator( } template <typename Derived> -typename SerialStorage<Derived>::ConstIterator& SerialStorage<Derived>::ConstIterator:: -operator++() { +typename SerialStorage<Derived>::ConstIterator& +SerialStorage<Derived>::ConstIterator::operator++() { const Value* vectorData = mStorageIterator->second.data(); if (mSerialIterator == nullptr) { diff --git a/chromium/third_party/dawn/src/common/StackContainer.h b/chromium/third_party/dawn/src/common/StackContainer.h new file mode 100644 index 00000000000..be3cf32d0de --- /dev/null +++ b/chromium/third_party/dawn/src/common/StackContainer.h @@ -0,0 +1,262 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// This file is a modified copy of Chromium's /src/base/containers/stack_container.h + +#ifndef COMMON_STACKCONTAINER_H_ +#define COMMON_STACKCONTAINER_H_ + +#include "common/Compiler.h" + +#include <cstddef> +#include <vector> + +// This allocator can be used with STL containers to provide a stack buffer +// from which to allocate memory and overflows onto the heap. This stack buffer +// would be allocated on the stack and allows us to avoid heap operations in +// some situations. +// +// STL likes to make copies of allocators, so the allocator itself can't hold +// the data. Instead, we make the creator responsible for creating a +// StackAllocator::Source which contains the data. Copying the allocator +// merely copies the pointer to this shared source, so all allocators created +// based on our allocator will share the same stack buffer. +// +// This stack buffer implementation is very simple. The first allocation that +// fits in the stack buffer will use the stack buffer. Any subsequent +// allocations will not use the stack buffer, even if there is unused room. +// This makes it appropriate for array-like containers, but the caller should +// be sure to reserve() in the container up to the stack buffer size. Otherwise +// the container will allocate a small array which will "use up" the stack +// buffer. +template <typename T, size_t stack_capacity> +class StackAllocator : public std::allocator<T> { + public: + typedef typename std::allocator<T>::pointer pointer; + typedef typename std::allocator<T>::size_type size_type; + + // Backing store for the allocator. The container owner is responsible for + // maintaining this for as long as any containers using this allocator are + // live. + struct Source { + Source() : used_stack_buffer_(false) { + } + + // Casts the buffer in its right type. + T* stack_buffer() { + return reinterpret_cast<T*>(stack_buffer_); + } + const T* stack_buffer() const { + return reinterpret_cast<const T*>(&stack_buffer_); + } + + // The buffer itself. It is not of type T because we don't want the + // constructors and destructors to be automatically called. Define a POD + // buffer of the right size instead. + alignas(T) char stack_buffer_[sizeof(T[stack_capacity])]; +#if defined(DAWN_COMPILER_GCC) && !defined(__x86_64__) && !defined(__i386__) + static_assert(alignof(T) <= 16, "http://crbug.com/115612"); +#endif + + // Set when the stack buffer is used for an allocation. We do not track + // how much of the buffer is used, only that somebody is using it. + bool used_stack_buffer_; + }; + + // Used by containers when they want to refer to an allocator of type U. + template <typename U> + struct rebind { + typedef StackAllocator<U, stack_capacity> other; + }; + + // For the straight up copy c-tor, we can share storage. + StackAllocator(const StackAllocator<T, stack_capacity>& rhs) + : std::allocator<T>(), source_(rhs.source_) { + } + + // ISO C++ requires the following constructor to be defined, + // and std::vector in VC++2008SP1 Release fails with an error + // in the class _Container_base_aux_alloc_real (from <xutility>) + // if the constructor does not exist. + // For this constructor, we cannot share storage; there's + // no guarantee that the Source buffer of Ts is large enough + // for Us. + // TODO: If we were fancy pants, perhaps we could share storage + // iff sizeof(T) == sizeof(U). + template <typename U, size_t other_capacity> + StackAllocator(const StackAllocator<U, other_capacity>& other) : source_(nullptr) { + } + + // This constructor must exist. It creates a default allocator that doesn't + // actually have a stack buffer. glibc's std::string() will compare the + // current allocator against the default-constructed allocator, so this + // should be fast. + StackAllocator() : source_(nullptr) { + } + + explicit StackAllocator(Source* source) : source_(source) { + } + + // Actually do the allocation. Use the stack buffer if nobody has used it yet + // and the size requested fits. Otherwise, fall through to the standard + // allocator. + pointer allocate(size_type n) { + if (source_ && !source_->used_stack_buffer_ && n <= stack_capacity) { + source_->used_stack_buffer_ = true; + return source_->stack_buffer(); + } else { + return std::allocator<T>::allocate(n); + } + } + + // Free: when trying to free the stack buffer, just mark it as free. For + // non-stack-buffer pointers, just fall though to the standard allocator. + void deallocate(pointer p, size_type n) { + if (source_ && p == source_->stack_buffer()) + source_->used_stack_buffer_ = false; + else + std::allocator<T>::deallocate(p, n); + } + + private: + Source* source_; +}; + +// A wrapper around STL containers that maintains a stack-sized buffer that the +// initial capacity of the vector is based on. Growing the container beyond the +// stack capacity will transparently overflow onto the heap. The container must +// support reserve(). +// +// This will not work with std::string since some implementations allocate +// more bytes than requested in calls to reserve(), forcing the allocation onto +// the heap. http://crbug.com/709273 +// +// WATCH OUT: the ContainerType MUST use the proper StackAllocator for this +// type. This object is really intended to be used only internally. You'll want +// to use the wrappers below for different types. +template <typename TContainerType, size_t stack_capacity> +class StackContainer { + public: + typedef TContainerType ContainerType; + typedef typename ContainerType::value_type ContainedType; + typedef StackAllocator<ContainedType, stack_capacity> Allocator; + + // Allocator must be constructed before the container! + StackContainer() : allocator_(&stack_data_), container_(allocator_) { + // Make the container use the stack allocation by reserving our buffer size + // before doing anything else. + container_.reserve(stack_capacity); + } + + // Getters for the actual container. + // + // Danger: any copies of this made using the copy constructor must have + // shorter lifetimes than the source. The copy will share the same allocator + // and therefore the same stack buffer as the original. Use std::copy to + // copy into a "real" container for longer-lived objects. + ContainerType& container() { + return container_; + } + const ContainerType& container() const { + return container_; + } + + // Support operator-> to get to the container. This allows nicer syntax like: + // StackContainer<...> foo; + // std::sort(foo->begin(), foo->end()); + ContainerType* operator->() { + return &container_; + } + const ContainerType* operator->() const { + return &container_; + } + + // Retrieves the stack source so that that unit tests can verify that the + // buffer is being used properly. + const typename Allocator::Source& stack_data() const { + return stack_data_; + } + + protected: + typename Allocator::Source stack_data_; + Allocator allocator_; + ContainerType container_; + + private: + StackContainer(const StackContainer& rhs) = delete; + StackContainer& operator=(const StackContainer& rhs) = delete; + StackContainer(StackContainer&& rhs) = delete; + StackContainer& operator=(StackContainer&& rhs) = delete; +}; + +// Range-based iteration support for StackContainer. +template <typename TContainerType, size_t stack_capacity> +auto begin(const StackContainer<TContainerType, stack_capacity>& stack_container) + -> decltype(begin(stack_container.container())) { + return begin(stack_container.container()); +} + +template <typename TContainerType, size_t stack_capacity> +auto begin(StackContainer<TContainerType, stack_capacity>& stack_container) + -> decltype(begin(stack_container.container())) { + return begin(stack_container.container()); +} + +template <typename TContainerType, size_t stack_capacity> +auto end(StackContainer<TContainerType, stack_capacity>& stack_container) + -> decltype(end(stack_container.container())) { + return end(stack_container.container()); +} + +template <typename TContainerType, size_t stack_capacity> +auto end(const StackContainer<TContainerType, stack_capacity>& stack_container) + -> decltype(end(stack_container.container())) { + return end(stack_container.container()); +} + +// StackVector ----------------------------------------------------------------- + +// Example: +// StackVector<int, 16> foo; +// foo->push_back(22); // we have overloaded operator-> +// foo[0] = 10; // as well as operator[] +template <typename T, size_t stack_capacity> +class StackVector + : public StackContainer<std::vector<T, StackAllocator<T, stack_capacity>>, stack_capacity> { + public: + StackVector() + : StackContainer<std::vector<T, StackAllocator<T, stack_capacity>>, stack_capacity>() { + } + + // We need to put this in STL containers sometimes, which requires a copy + // constructor. We can't call the regular copy constructor because that will + // take the stack buffer from the original. Here, we create an empty object + // and make a stack buffer of its own. + StackVector(const StackVector<T, stack_capacity>& other) + : StackContainer<std::vector<T, StackAllocator<T, stack_capacity>>, stack_capacity>() { + this->container().assign(other->begin(), other->end()); + } + + StackVector<T, stack_capacity>& operator=(const StackVector<T, stack_capacity>& other) { + this->container().assign(other->begin(), other->end()); + return *this; + } + + // Vectors are commonly indexed, which isn't very convenient even with + // operator-> (using "->at()" does exception stuff we don't want). + T& operator[](size_t i) { + return this->container().operator[](i); + } + const T& operator[](size_t i) const { + return this->container().operator[](i); + } + + private: + // StackVector(const StackVector& rhs) = delete; + // StackVector& operator=(const StackVector& rhs) = delete; + StackVector(StackVector&& rhs) = delete; + StackVector& operator=(StackVector&& rhs) = delete; +}; + +#endif // COMMON_STACKCONTAINER_H_ diff --git a/chromium/third_party/dawn/src/common/SystemUtils.cpp b/chromium/third_party/dawn/src/common/SystemUtils.cpp index f8282eb4141..58aac01c443 100644 --- a/chromium/third_party/dawn/src/common/SystemUtils.cpp +++ b/chromium/third_party/dawn/src/common/SystemUtils.cpp @@ -108,6 +108,11 @@ std::string GetExecutablePath() { // TODO: Implement on Fuchsia return ""; } +#elif defined(DAWN_PLATFORM_EMSCRIPTEN) +std::string GetExecutablePath() { + UNREACHABLE(); + return ""; +} #else # error "Implement GetExecutablePath for your platform." #endif diff --git a/chromium/third_party/dawn/src/common/ityp_array.h b/chromium/third_party/dawn/src/common/ityp_array.h index d413ebc0ba0..fc772178ccc 100644 --- a/chromium/third_party/dawn/src/common/ityp_array.h +++ b/chromium/third_party/dawn/src/common/ityp_array.h @@ -65,19 +65,19 @@ namespace ityp { return Base::at(index); } - Value* begin() noexcept { + typename Base::iterator begin() noexcept { return Base::begin(); } - const Value* begin() const noexcept { + typename Base::const_iterator begin() const noexcept { return Base::begin(); } - Value* end() noexcept { + typename Base::iterator end() noexcept { return Base::end(); } - const Value* end() const noexcept { + typename Base::const_iterator end() const noexcept { return Base::end(); } diff --git a/chromium/third_party/dawn/src/common/ityp_bitset.h b/chromium/third_party/dawn/src/common/ityp_bitset.h index ef351d47d76..339cf182937 100644 --- a/chromium/third_party/dawn/src/common/ityp_bitset.h +++ b/chromium/third_party/dawn/src/common/ityp_bitset.h @@ -126,7 +126,7 @@ namespace ityp { return BitSetIterator<N, Index>(static_cast<const Base&>(bitset)); } - friend class std::hash<bitset>; + friend struct std::hash<bitset>; }; } // namespace ityp diff --git a/chromium/third_party/dawn/src/common/ityp_stack_vec.h b/chromium/third_party/dawn/src/common/ityp_stack_vec.h new file mode 100644 index 00000000000..b88888b7897 --- /dev/null +++ b/chromium/third_party/dawn/src/common/ityp_stack_vec.h @@ -0,0 +1,103 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef COMMON_ITYP_STACK_VEC_H_ +#define COMMON_ITYP_STACK_VEC_H_ + +#include "common/Assert.h" +#include "common/StackContainer.h" +#include "common/UnderlyingType.h" + +namespace ityp { + + template <typename Index, typename Value, size_t StaticCapacity> + class stack_vec : private StackVector<Value, StaticCapacity> { + using I = UnderlyingType<Index>; + using Base = StackVector<Value, StaticCapacity>; + using VectorBase = std::vector<Value, StackAllocator<Value, StaticCapacity>>; + static_assert(StaticCapacity <= std::numeric_limits<I>::max(), ""); + + public: + stack_vec() : Base() { + } + stack_vec(Index size) : Base() { + this->container().resize(static_cast<I>(size)); + } + + Value& operator[](Index i) { + ASSERT(i < size()); + return Base::operator[](static_cast<I>(i)); + } + + constexpr const Value& operator[](Index i) const { + ASSERT(i < size()); + return Base::operator[](static_cast<I>(i)); + } + + void resize(Index size) { + this->container().resize(static_cast<I>(size)); + } + + void reserve(Index size) { + this->container().reserve(static_cast<I>(size)); + } + + Value* data() { + return this->container().data(); + } + + const Value* data() const { + return this->container().data(); + } + + typename VectorBase::iterator begin() noexcept { + return this->container().begin(); + } + + typename VectorBase::const_iterator begin() const noexcept { + return this->container().begin(); + } + + typename VectorBase::iterator end() noexcept { + return this->container().end(); + } + + typename VectorBase::const_iterator end() const noexcept { + return this->container().end(); + } + + typename VectorBase::reference front() { + return this->container().front(); + } + + typename VectorBase::const_reference front() const { + return this->container().front(); + } + + typename VectorBase::reference back() { + return this->container().back(); + } + + typename VectorBase::const_reference back() const { + return this->container().back(); + } + + Index size() const { + return Index(static_cast<I>(this->container().size())); + } + }; + +} // namespace ityp + +#endif // COMMON_ITYP_STACK_VEC_H_ diff --git a/chromium/third_party/dawn/src/common/ityp_vector.h b/chromium/third_party/dawn/src/common/ityp_vector.h new file mode 100644 index 00000000000..a747d5aeb53 --- /dev/null +++ b/chromium/third_party/dawn/src/common/ityp_vector.h @@ -0,0 +1,108 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef COMMON_ITYP_VECTOR_H_ +#define COMMON_ITYP_VECTOR_H_ + +#include "common/TypedInteger.h" +#include "common/UnderlyingType.h" + +#include <type_traits> +#include <vector> + +namespace ityp { + + // ityp::vector is a helper class that wraps std::vector with the restriction that + // indices must be a particular type |Index|. + template <typename Index, typename Value> + class vector : public std::vector<Value> { + using I = UnderlyingType<Index>; + using Base = std::vector<Value>; + + private: + // Disallow access to base constructors and untyped index/size-related operators. + using Base::Base; + using Base::operator=; + using Base::operator[]; + using Base::at; + using Base::reserve; + using Base::resize; + using Base::size; + + public: + vector() : Base() { + } + + explicit vector(Index size) : Base(static_cast<I>(size)) { + } + + vector(Index size, const Value& init) : Base(static_cast<I>(size), init) { + } + + vector(const vector& rhs) : Base(static_cast<const Base&>(rhs)) { + } + + vector(vector&& rhs) : Base(static_cast<Base&&>(rhs)) { + } + + vector(std::initializer_list<Value> init) : Base(init) { + } + + vector& operator=(const vector& rhs) { + Base::operator=(static_cast<const Base&>(rhs)); + return *this; + } + + vector& operator=(vector&& rhs) noexcept { + Base::operator=(static_cast<Base&&>(rhs)); + return *this; + } + + Value& operator[](Index i) { + ASSERT(i >= Index(0) && i < size()); + return Base::operator[](static_cast<I>(i)); + } + + constexpr const Value& operator[](Index i) const { + ASSERT(i >= Index(0) && i < size()); + return Base::operator[](static_cast<I>(i)); + } + + Value& at(Index i) { + ASSERT(i >= Index(0) && i < size()); + return Base::at(static_cast<I>(i)); + } + + constexpr const Value& at(Index i) const { + ASSERT(i >= Index(0) && i < size()); + return Base::at(static_cast<I>(i)); + } + + constexpr Index size() const { + ASSERT(std::numeric_limits<I>::max() >= Base::size()); + return Index(static_cast<I>(Base::size())); + } + + void resize(Index size) { + Base::resize(static_cast<I>(size)); + } + + void reserve(Index size) { + Base::reserve(static_cast<I>(size)); + } + }; + +} // namespace ityp + +#endif // COMMON_ITYP_VECTOR_H_ diff --git a/chromium/third_party/dawn/src/common/vulkan_platform.h b/chromium/third_party/dawn/src/common/vulkan_platform.h index 113e831f432..236c68236c0 100644 --- a/chromium/third_party/dawn/src/common/vulkan_platform.h +++ b/chromium/third_party/dawn/src/common/vulkan_platform.h @@ -37,8 +37,18 @@ #if defined(DAWN_PLATFORM_64_BIT) # define DAWN_DEFINE_NATIVE_NON_DISPATCHABLE_HANDLE(object) using object = struct object##_T*; +// This function is needed because MSVC doesn't accept reinterpret_cast from uint64_t from uint64_t +// TODO(cwallez@chromium.org): Remove this once we rework vulkan_platform.h +template <typename T> +T NativeNonDispatachableHandleFromU64(uint64_t u64) { + return reinterpret_cast<T>(u64); +} #elif defined(DAWN_PLATFORM_32_BIT) # define DAWN_DEFINE_NATIVE_NON_DISPATCHABLE_HANDLE(object) using object = uint64_t; +template <typename T> +T NativeNonDispatachableHandleFromU64(uint64_t u64) { + return u64; +} #else # error "Unsupported platform" #endif diff --git a/chromium/third_party/dawn/src/dawn/BUILD.gn b/chromium/third_party/dawn/src/dawn/BUILD.gn index d095e9aab5f..9034be436f0 100644 --- a/chromium/third_party/dawn/src/dawn/BUILD.gn +++ b/chromium/third_party/dawn/src/dawn/BUILD.gn @@ -39,9 +39,7 @@ dawn_json_generator("emscripten_bits_gen") { source_set("dawn_headers") { all_dependent_configs = [ "${dawn_root}/src/common:dawn_public_include_dirs" ] - public_deps = [ - ":dawn_headers_gen", - ] + public_deps = [ ":dawn_headers_gen" ] sources = get_target_outputs(":dawn_headers_gen") sources += [ "${dawn_root}/src/include/dawn/dawn_wsi.h" ] @@ -53,9 +51,7 @@ source_set("dawn_headers") { dawn_json_generator("dawncpp_headers_gen") { target = "dawncpp_headers" - outputs = [ - "src/include/dawn/webgpu_cpp.h", - ] + outputs = [ "src/include/dawn/webgpu_cpp.h" ] } source_set("dawncpp_headers") { @@ -74,9 +70,7 @@ source_set("dawncpp_headers") { dawn_json_generator("dawncpp_gen") { target = "dawncpp" - outputs = [ - "src/dawn/webgpu_cpp.cpp", - ] + outputs = [ "src/dawn/webgpu_cpp.cpp" ] } source_set("dawncpp") { @@ -93,20 +87,14 @@ source_set("dawncpp") { dawn_json_generator("dawn_proc_gen") { target = "dawn_proc" - outputs = [ - "src/dawn/dawn_proc.c", - ] + outputs = [ "src/dawn/dawn_proc.c" ] } dawn_component("dawn_proc") { DEFINE_PREFIX = "WGPU" - public_deps = [ - ":dawn_headers", - ] - deps = [ - ":dawn_proc_gen", - ] + public_deps = [ ":dawn_headers" ] + deps = [ ":dawn_proc_gen" ] sources = get_target_outputs(":dawn_proc_gen") sources += [ "${dawn_root}/src/include/dawn/dawn_proc.h" ] } diff --git a/chromium/third_party/dawn/src/dawn_native/BUILD.gn b/chromium/third_party/dawn/src/dawn_native/BUILD.gn index f6974c8a3d0..cecfc1e903e 100644 --- a/chromium/third_party/dawn/src/dawn_native/BUILD.gn +++ b/chromium/third_party/dawn/src/dawn_native/BUILD.gn @@ -70,10 +70,7 @@ config("dawn_native_internal") { config("dawn_native_weak_framework") { if (is_mac && dawn_enable_metal) { - ldflags = [ - "-weak_framework", - "Metal", - ] + weak_frameworks = [ "Metal.framework" ] } } @@ -163,6 +160,7 @@ source_set("dawn_native_sources") { "BindGroupLayout.cpp", "BindGroupLayout.h", "BindGroupTracker.h", + "BindingInfo.cpp", "BindingInfo.h", "BuddyAllocator.cpp", "BuddyAllocator.h", @@ -194,6 +192,8 @@ source_set("dawn_native_sources") { "DynamicUploader.h", "EncodingContext.cpp", "EncodingContext.h", + "EnumClassBitmasks.h", + "EnumMaskIterator.h", "Error.cpp", "Error.h", "ErrorData.cpp", @@ -228,6 +228,8 @@ source_set("dawn_native_sources") { "Pipeline.h", "PipelineLayout.cpp", "PipelineLayout.h", + "PooledResourceMemoryAllocator.cpp", + "PooledResourceMemoryAllocator.h", "ProgrammablePassEncoder.cpp", "ProgrammablePassEncoder.h", "QuerySet.cpp", @@ -320,6 +322,8 @@ source_set("dawn_native_sources") { "d3d12/PipelineLayoutD3D12.h", "d3d12/PlatformFunctions.cpp", "d3d12/PlatformFunctions.h", + "d3d12/QuerySetD3D12.cpp", + "d3d12/QuerySetD3D12.h", "d3d12/QueueD3D12.cpp", "d3d12/QueueD3D12.h", "d3d12/RenderPassBuilderD3D12.cpp", @@ -357,7 +361,7 @@ source_set("dawn_native_sources") { } if (dawn_enable_metal) { - libs += [ + frameworks = [ "Cocoa.framework", "IOKit.framework", "IOSurface.framework", @@ -495,6 +499,8 @@ source_set("dawn_native_sources") { "vulkan/NativeSwapChainImplVk.h", "vulkan/PipelineLayoutVk.cpp", "vulkan/PipelineLayoutVk.h", + "vulkan/QuerySetVk.cpp", + "vulkan/QuerySetVk.h", "vulkan/QueueVk.cpp", "vulkan/QueueVk.h", "vulkan/RenderPassCache.cpp", diff --git a/chromium/third_party/dawn/src/dawn_native/BindGroup.cpp b/chromium/third_party/dawn/src/dawn_native/BindGroup.cpp index 4e4e67a447a..71e8a28d78d 100644 --- a/chromium/third_party/dawn/src/dawn_native/BindGroup.cpp +++ b/chromium/third_party/dawn/src/dawn_native/BindGroup.cpp @@ -32,7 +32,8 @@ namespace dawn_native { MaybeError ValidateBufferBinding(const DeviceBase* device, const BindGroupEntry& entry, wgpu::BufferUsage requiredUsage, - const BindingInfo& bindingInfo) { + const BindingInfo& bindingInfo, + const uint64_t maxBindingSize) { if (entry.buffer == nullptr || entry.sampler != nullptr || entry.textureView != nullptr) { return DAWN_VALIDATION_ERROR("expected buffer binding"); @@ -79,6 +80,14 @@ namespace dawn_native { " bytes"); } + if (bindingSize > maxBindingSize) { + return DAWN_VALIDATION_ERROR( + "Binding size bigger than maximum uniform buffer binding size: binding " + + std::to_string(entry.binding) + " given " + std::to_string(bindingSize) + + " bytes, maximum is " + std::to_string(kMaxUniformBufferBindingSize) + + " bytes"); + } + return {}; } @@ -169,8 +178,9 @@ namespace dawn_native { } const BindGroupLayoutBase::BindingMap& bindingMap = descriptor->layout->GetBindingMap(); + ASSERT(bindingMap.size() <= kMaxBindingsPerPipelineLayout); - ityp::bitset<BindingIndex, kMaxBindingsPerGroup> bindingsSet; + ityp::bitset<BindingIndex, kMaxBindingsPerPipelineLayout> bindingsSet; for (uint32_t i = 0; i < descriptor->entryCount; ++i) { const BindGroupEntry& entry = descriptor->entries[i]; @@ -192,12 +202,13 @@ namespace dawn_native { switch (bindingInfo.type) { case wgpu::BindingType::UniformBuffer: DAWN_TRY(ValidateBufferBinding(device, entry, wgpu::BufferUsage::Uniform, - bindingInfo)); + bindingInfo, kMaxUniformBufferBindingSize)); break; case wgpu::BindingType::StorageBuffer: case wgpu::BindingType::ReadonlyStorageBuffer: DAWN_TRY(ValidateBufferBinding(device, entry, wgpu::BufferUsage::Storage, - bindingInfo)); + bindingInfo, + std::numeric_limits<uint64_t>::max())); break; case wgpu::BindingType::SampledTexture: DAWN_TRY(ValidateTextureBinding(device, entry, wgpu::TextureUsage::Sampled, diff --git a/chromium/third_party/dawn/src/dawn_native/BindGroupAndStorageBarrierTracker.h b/chromium/third_party/dawn/src/dawn_native/BindGroupAndStorageBarrierTracker.h index e17d26349f1..e34a16511a2 100644 --- a/chromium/third_party/dawn/src/dawn_native/BindGroupAndStorageBarrierTracker.h +++ b/chromium/third_party/dawn/src/dawn_native/BindGroupAndStorageBarrierTracker.h @@ -16,6 +16,7 @@ #define DAWNNATIVE_BINDGROUPANDSTORAGEBARRIERTRACKER_H_ #include "common/ityp_bitset.h" +#include "common/ityp_stack_vec.h" #include "dawn_native/BindGroup.h" #include "dawn_native/BindGroupTracker.h" #include "dawn_native/Buffer.h" @@ -39,11 +40,12 @@ namespace dawn_native { ASSERT(index < kMaxBindGroupsTyped); if (this->mBindGroups[index] != bindGroup) { - mBindings[index] = {}; - mBindingsNeedingBarrier[index] = {}; - const BindGroupLayoutBase* layout = bindGroup->GetLayout(); + mBindings[index].resize(layout->GetBindingCount()); + mBindingTypes[index].resize(layout->GetBindingCount()); + mBindingsNeedingBarrier[index] = {}; + for (BindingIndex bindingIndex{0}; bindingIndex < layout->GetBindingCount(); ++bindingIndex) { const BindingInfo& bindingInfo = layout->GetBindingInfo(bindingIndex); @@ -91,15 +93,16 @@ namespace dawn_native { } protected: - ityp:: - array<BindGroupIndex, ityp::bitset<BindingIndex, kMaxBindingsPerGroup>, kMaxBindGroups> - mBindingsNeedingBarrier = {}; ityp::array<BindGroupIndex, - ityp::array<BindingIndex, wgpu::BindingType, kMaxBindingsPerGroup>, + ityp::bitset<BindingIndex, kMaxBindingsPerPipelineLayout>, + kMaxBindGroups> + mBindingsNeedingBarrier = {}; + ityp::array<BindGroupIndex, + ityp::stack_vec<BindingIndex, wgpu::BindingType, kMaxOptimalBindingsPerGroup>, kMaxBindGroups> mBindingTypes = {}; ityp::array<BindGroupIndex, - ityp::array<BindingIndex, ObjectBase*, kMaxBindingsPerGroup>, + ityp::stack_vec<BindingIndex, ObjectBase*, kMaxOptimalBindingsPerGroup>, kMaxBindGroups> mBindings = {}; }; diff --git a/chromium/third_party/dawn/src/dawn_native/BindGroupLayout.cpp b/chromium/third_party/dawn/src/dawn_native/BindGroupLayout.cpp index fce8b903293..9c5c8dca1fc 100644 --- a/chromium/third_party/dawn/src/dawn_native/BindGroupLayout.cpp +++ b/chromium/third_party/dawn/src/dawn_native/BindGroupLayout.cpp @@ -17,6 +17,7 @@ #include "common/BitSetIterator.h" #include "common/HashUtils.h" #include "dawn_native/Device.h" +#include "dawn_native/PerStage.h" #include "dawn_native/ValidationUtils_autogen.h" #include <algorithm> @@ -178,6 +179,8 @@ namespace dawn_native { return DAWN_VALIDATION_ERROR("3D texture bindings may not be multisampled"); case wgpu::TextureViewDimension::e1D: + return DAWN_VALIDATION_ERROR("1D texture bindings may not be multisampled"); + case wgpu::TextureViewDimension::Undefined: default: UNREACHABLE(); @@ -194,8 +197,7 @@ namespace dawn_native { } std::set<BindingNumber> bindingsSet; - uint32_t dynamicUniformBufferCount = 0; - uint32_t dynamicStorageBufferCount = 0; + BindingCounts bindingCounts = {}; for (uint32_t i = 0; i < descriptor->entryCount; ++i) { const BindGroupLayoutEntry& entry = descriptor->entries[i]; BindingNumber bindingNumber = BindingNumber(entry.binding); @@ -226,45 +228,39 @@ namespace dawn_native { switch (entry.type) { case wgpu::BindingType::UniformBuffer: - if (entry.hasDynamicOffset) { - ++dynamicUniformBufferCount; - } - break; case wgpu::BindingType::StorageBuffer: case wgpu::BindingType::ReadonlyStorageBuffer: + break; + case wgpu::BindingType::SampledTexture: if (entry.hasDynamicOffset) { - ++dynamicStorageBufferCount; + return DAWN_VALIDATION_ERROR("Sampled textures cannot be dynamic"); } break; - case wgpu::BindingType::SampledTexture: case wgpu::BindingType::Sampler: case wgpu::BindingType::ComparisonSampler: + if (entry.hasDynamicOffset) { + return DAWN_VALIDATION_ERROR("Samplers cannot be dynamic"); + } + break; case wgpu::BindingType::ReadonlyStorageTexture: case wgpu::BindingType::WriteonlyStorageTexture: if (entry.hasDynamicOffset) { - return DAWN_VALIDATION_ERROR("Samplers and textures cannot be dynamic"); + return DAWN_VALIDATION_ERROR("Storage textures cannot be dynamic"); } break; case wgpu::BindingType::StorageTexture: return DAWN_VALIDATION_ERROR("storage textures aren't supported (yet)"); + default: + UNREACHABLE(); + break; } - bindingsSet.insert(bindingNumber); - } - - if (bindingsSet.size() > kMaxBindingsPerGroup) { - return DAWN_VALIDATION_ERROR("The number of bindings exceeds kMaxBindingsPerGroup."); - } + IncrementBindingCounts(&bindingCounts, entry); - if (dynamicUniformBufferCount > kMaxDynamicUniformBufferCount) { - return DAWN_VALIDATION_ERROR( - "The number of dynamic uniform buffer exceeds the maximum value"); + bindingsSet.insert(bindingNumber); } - if (dynamicStorageBufferCount > kMaxDynamicStorageBufferCount) { - return DAWN_VALIDATION_ERROR( - "The number of dynamic storage buffer exceeds the maximum value"); - } + DAWN_TRY(ValidateBindingCounts(bindingCounts)); return {}; } @@ -360,8 +356,6 @@ namespace dawn_native { // This is a utility function to help ASSERT that the BGL-binding comparator places buffers // first. bool CheckBufferBindingsFirst(ityp::span<BindingIndex, const BindingInfo> bindings) { - ASSERT(bindings.size() <= BindingIndex(kMaxBindingsPerGroup)); - BindingIndex lastBufferIndex{0}; BindingIndex firstNonBufferIndex = std::numeric_limits<BindingIndex>::max(); for (BindingIndex i{0}; i < bindings.size(); ++i) { @@ -383,13 +377,12 @@ namespace dawn_native { BindGroupLayoutBase::BindGroupLayoutBase(DeviceBase* device, const BindGroupLayoutDescriptor* descriptor) - : CachedObject(device), mBindingCount(descriptor->entryCount) { + : CachedObject(device), mBindingInfo(BindingIndex(descriptor->entryCount)) { std::vector<BindGroupLayoutEntry> sortedBindings( descriptor->entries, descriptor->entries + descriptor->entryCount); - std::sort(sortedBindings.begin(), sortedBindings.end(), SortBindingsCompare); - for (BindingIndex i{0}; i < mBindingCount; ++i) { + for (BindingIndex i{0}; i < mBindingInfo.size(); ++i) { const BindGroupLayoutEntry& binding = sortedBindings[static_cast<uint32_t>(i)]; mBindingInfo[i].binding = BindingNumber(binding.binding); mBindingInfo[i].type = binding.type; @@ -399,53 +392,26 @@ namespace dawn_native { mBindingInfo[i].storageTextureFormat = binding.storageTextureFormat; mBindingInfo[i].minBufferBindingSize = binding.minBufferBindingSize; - switch (binding.type) { - case wgpu::BindingType::UniformBuffer: - case wgpu::BindingType::StorageBuffer: - case wgpu::BindingType::ReadonlyStorageBuffer: - // Buffers must be contiguously packed at the start of the binding info. - ASSERT(mBufferCount == i); - ++mBufferCount; - if (binding.minBufferBindingSize == 0) { - ++mUnverifiedBufferCount; - } - break; - default: - break; - } - if (binding.viewDimension == wgpu::TextureViewDimension::Undefined) { - mBindingInfo[i].viewDimension = wgpu::TextureViewDimension::e2D; + mBindingInfo[i].viewDimension = wgpu::TextureViewDimension::e2D; } else { mBindingInfo[i].viewDimension = binding.viewDimension; } mBindingInfo[i].multisampled = binding.multisampled; mBindingInfo[i].hasDynamicOffset = binding.hasDynamicOffset; - if (binding.hasDynamicOffset) { - switch (binding.type) { - case wgpu::BindingType::UniformBuffer: - ++mDynamicUniformBufferCount; - break; - case wgpu::BindingType::StorageBuffer: - case wgpu::BindingType::ReadonlyStorageBuffer: - ++mDynamicStorageBufferCount; - break; - case wgpu::BindingType::SampledTexture: - case wgpu::BindingType::Sampler: - case wgpu::BindingType::ComparisonSampler: - case wgpu::BindingType::StorageTexture: - case wgpu::BindingType::ReadonlyStorageTexture: - case wgpu::BindingType::WriteonlyStorageTexture: - UNREACHABLE(); - break; - } + + if (IsBufferBinding(binding.type)) { + // Buffers must be contiguously packed at the start of the binding info. + ASSERT(GetBufferCount() == i); } + IncrementBindingCounts(&mBindingCounts, binding); const auto& it = mBindingMap.emplace(BindingNumber(binding.binding), i); ASSERT(it.second); } - ASSERT(CheckBufferBindingsFirst({mBindingInfo.data(), mBindingCount})); + ASSERT(CheckBufferBindingsFirst({mBindingInfo.data(), GetBindingCount()})); + ASSERT(mBindingInfo.size() <= kMaxBindingsPerPipelineLayoutTyped); } BindGroupLayoutBase::BindGroupLayoutBase(DeviceBase* device, ObjectBase::ErrorTag tag) @@ -501,29 +467,26 @@ namespace dawn_native { } BindingIndex BindGroupLayoutBase::GetBindingCount() const { - return mBindingCount; + return mBindingInfo.size(); } BindingIndex BindGroupLayoutBase::GetBufferCount() const { - return mBufferCount; + return BindingIndex(mBindingCounts.bufferCount); } BindingIndex BindGroupLayoutBase::GetDynamicBufferCount() const { // This is a binding index because dynamic buffers are packed at the front of the binding // info. - return static_cast<BindingIndex>(mDynamicStorageBufferCount + mDynamicUniformBufferCount); + return static_cast<BindingIndex>(mBindingCounts.dynamicStorageBufferCount + + mBindingCounts.dynamicUniformBufferCount); } - uint32_t BindGroupLayoutBase::GetDynamicUniformBufferCount() const { - return mDynamicUniformBufferCount; - } - - uint32_t BindGroupLayoutBase::GetDynamicStorageBufferCount() const { - return mDynamicStorageBufferCount; + uint32_t BindGroupLayoutBase::GetUnverifiedBufferCount() const { + return mBindingCounts.unverifiedBufferCount; } - uint32_t BindGroupLayoutBase::GetUnverifiedBufferCount() const { - return mUnverifiedBufferCount; + const BindingCounts& BindGroupLayoutBase::GetBindingCountInfo() const { + return mBindingCounts; } size_t BindGroupLayoutBase::GetBindingDataSize() const { @@ -532,31 +495,29 @@ namespace dawn_native { // Followed by: // |---------buffer size array--------| // |-uint64_t[mUnverifiedBufferCount]-| - size_t objectPointerStart = static_cast<uint32_t>(mBufferCount) * sizeof(BufferBindingData); + size_t objectPointerStart = mBindingCounts.bufferCount * sizeof(BufferBindingData); ASSERT(IsAligned(objectPointerStart, alignof(Ref<ObjectBase>))); - size_t bufferSizeArrayStart = Align( - objectPointerStart + static_cast<uint32_t>(mBindingCount) * sizeof(Ref<ObjectBase>), - sizeof(uint64_t)); + size_t bufferSizeArrayStart = + Align(objectPointerStart + mBindingCounts.totalCount * sizeof(Ref<ObjectBase>), + sizeof(uint64_t)); ASSERT(IsAligned(bufferSizeArrayStart, alignof(uint64_t))); - return bufferSizeArrayStart + mUnverifiedBufferCount * sizeof(uint64_t); + return bufferSizeArrayStart + mBindingCounts.unverifiedBufferCount * sizeof(uint64_t); } BindGroupLayoutBase::BindingDataPointers BindGroupLayoutBase::ComputeBindingDataPointers( void* dataStart) const { BufferBindingData* bufferData = reinterpret_cast<BufferBindingData*>(dataStart); - auto bindings = - reinterpret_cast<Ref<ObjectBase>*>(bufferData + static_cast<uint32_t>(mBufferCount)); - uint64_t* unverifiedBufferSizes = - AlignPtr(reinterpret_cast<uint64_t*>(bindings + static_cast<uint32_t>(mBindingCount)), - sizeof(uint64_t)); + auto bindings = reinterpret_cast<Ref<ObjectBase>*>(bufferData + mBindingCounts.bufferCount); + uint64_t* unverifiedBufferSizes = AlignPtr( + reinterpret_cast<uint64_t*>(bindings + mBindingCounts.totalCount), sizeof(uint64_t)); ASSERT(IsPtrAligned(bufferData, alignof(BufferBindingData))); ASSERT(IsPtrAligned(bindings, alignof(Ref<ObjectBase>))); ASSERT(IsPtrAligned(unverifiedBufferSizes, alignof(uint64_t))); - return {{bufferData, mBufferCount}, - {bindings, mBindingCount}, - {unverifiedBufferSizes, mUnverifiedBufferCount}}; + return {{bufferData, GetBufferCount()}, + {bindings, GetBindingCount()}, + {unverifiedBufferSizes, mBindingCounts.unverifiedBufferCount}}; } } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/BindGroupLayout.h b/chromium/third_party/dawn/src/dawn_native/BindGroupLayout.h index 4c3c4c63ca6..958abf6f9fe 100644 --- a/chromium/third_party/dawn/src/dawn_native/BindGroupLayout.h +++ b/chromium/third_party/dawn/src/dawn_native/BindGroupLayout.h @@ -18,8 +18,8 @@ #include "common/Constants.h" #include "common/Math.h" #include "common/SlabAllocator.h" -#include "common/ityp_array.h" #include "common/ityp_span.h" +#include "common/ityp_vector.h" #include "dawn_native/BindingInfo.h" #include "dawn_native/CachedObject.h" #include "dawn_native/Error.h" @@ -64,7 +64,7 @@ namespace dawn_native { const BindingInfo& GetBindingInfo(BindingIndex bindingIndex) const { ASSERT(!IsError()); - ASSERT(bindingIndex < BindingIndex(kMaxBindingsPerGroup)); + ASSERT(bindingIndex < mBindingInfo.size()); return mBindingInfo[bindingIndex]; } const BindingMap& GetBindingMap() const; @@ -79,13 +79,16 @@ namespace dawn_native { }; BindingIndex GetBindingCount() const; + // Returns |BindingIndex| because buffers are packed at the front. BindingIndex GetBufferCount() const; // Returns |BindingIndex| because dynamic buffers are packed at the front. BindingIndex GetDynamicBufferCount() const; - uint32_t GetDynamicUniformBufferCount() const; - uint32_t GetDynamicStorageBufferCount() const; uint32_t GetUnverifiedBufferCount() const; + // Used to get counts and validate them in pipeline layout creation. Other getters + // should be used to get typed integer counts. + const BindingCounts& GetBindingCountInfo() const; + struct BufferBindingData { uint64_t offset; uint64_t size; @@ -120,13 +123,8 @@ namespace dawn_native { private: BindGroupLayoutBase(DeviceBase* device, ObjectBase::ErrorTag tag); - BindingIndex mBindingCount; - BindingIndex mBufferCount{0}; // |BindingIndex| because buffers are packed at the front. - uint32_t mUnverifiedBufferCount = 0; // Buffers with minimum buffer size unspecified - uint32_t mDynamicUniformBufferCount = 0; - uint32_t mDynamicStorageBufferCount = 0; - - ityp::array<BindingIndex, BindingInfo, kMaxBindingsPerGroup> mBindingInfo; + BindingCounts mBindingCounts = {}; + ityp::vector<BindingIndex, BindingInfo> mBindingInfo; // Map from BindGroupLayoutEntry.binding to packed indices. BindingMap mBindingMap; diff --git a/chromium/third_party/dawn/src/dawn_native/BindGroupTracker.h b/chromium/third_party/dawn/src/dawn_native/BindGroupTracker.h index 8d03ebf3614..a3addb27b19 100644 --- a/chromium/third_party/dawn/src/dawn_native/BindGroupTracker.h +++ b/chromium/third_party/dawn/src/dawn_native/BindGroupTracker.h @@ -103,7 +103,9 @@ namespace dawn_native { BindGroupLayoutMask mBindGroupLayoutsMask = 0; ityp::array<BindGroupIndex, BindGroupBase*, kMaxBindGroups> mBindGroups = {}; ityp::array<BindGroupIndex, uint32_t, kMaxBindGroups> mDynamicOffsetCounts = {}; - ityp::array<BindGroupIndex, std::array<DynamicOffset, kMaxBindingsPerGroup>, kMaxBindGroups> + ityp::array<BindGroupIndex, + std::array<DynamicOffset, kMaxDynamicBuffersPerPipelineLayout>, + kMaxBindGroups> mDynamicOffsets = {}; // |mPipelineLayout| is the current pipeline layout set on the command buffer. diff --git a/chromium/third_party/dawn/src/dawn_native/BindingInfo.cpp b/chromium/third_party/dawn/src/dawn_native/BindingInfo.cpp new file mode 100644 index 00000000000..6ade32bb6b2 --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_native/BindingInfo.cpp @@ -0,0 +1,137 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "dawn_native/BindingInfo.h" + +namespace dawn_native { + + void IncrementBindingCounts(BindingCounts* bindingCounts, const BindGroupLayoutEntry& entry) { + bindingCounts->totalCount += 1; + + uint32_t PerStageBindingCounts::*perStageBindingCountMember = nullptr; + switch (entry.type) { + case wgpu::BindingType::UniformBuffer: + ++bindingCounts->bufferCount; + if (entry.hasDynamicOffset) { + ++bindingCounts->dynamicUniformBufferCount; + } + if (entry.minBufferBindingSize == 0) { + ++bindingCounts->unverifiedBufferCount; + } + perStageBindingCountMember = &PerStageBindingCounts::uniformBufferCount; + break; + + case wgpu::BindingType::StorageBuffer: + case wgpu::BindingType::ReadonlyStorageBuffer: + ++bindingCounts->bufferCount; + if (entry.hasDynamicOffset) { + ++bindingCounts->dynamicStorageBufferCount; + } + if (entry.minBufferBindingSize == 0) { + ++bindingCounts->unverifiedBufferCount; + } + perStageBindingCountMember = &PerStageBindingCounts::storageBufferCount; + break; + + case wgpu::BindingType::SampledTexture: + perStageBindingCountMember = &PerStageBindingCounts::sampledTextureCount; + break; + + case wgpu::BindingType::Sampler: + case wgpu::BindingType::ComparisonSampler: + perStageBindingCountMember = &PerStageBindingCounts::samplerCount; + break; + + case wgpu::BindingType::ReadonlyStorageTexture: + case wgpu::BindingType::WriteonlyStorageTexture: + perStageBindingCountMember = &PerStageBindingCounts::storageTextureCount; + break; + + case wgpu::BindingType::StorageTexture: + default: + UNREACHABLE(); + break; + } + + ASSERT(perStageBindingCountMember != nullptr); + for (SingleShaderStage stage : IterateStages(entry.visibility)) { + ++(bindingCounts->perStage[stage].*perStageBindingCountMember); + } + } + + void AccumulateBindingCounts(BindingCounts* bindingCounts, const BindingCounts& rhs) { + bindingCounts->totalCount += rhs.totalCount; + bindingCounts->bufferCount += rhs.bufferCount; + bindingCounts->unverifiedBufferCount += rhs.unverifiedBufferCount; + bindingCounts->dynamicUniformBufferCount += rhs.dynamicUniformBufferCount; + bindingCounts->dynamicStorageBufferCount += rhs.dynamicStorageBufferCount; + + for (SingleShaderStage stage : IterateStages(kAllStages)) { + bindingCounts->perStage[stage].sampledTextureCount += + rhs.perStage[stage].sampledTextureCount; + bindingCounts->perStage[stage].samplerCount += rhs.perStage[stage].samplerCount; + bindingCounts->perStage[stage].storageBufferCount += + rhs.perStage[stage].storageBufferCount; + bindingCounts->perStage[stage].storageTextureCount += + rhs.perStage[stage].storageTextureCount; + bindingCounts->perStage[stage].uniformBufferCount += + rhs.perStage[stage].uniformBufferCount; + } + } + + MaybeError ValidateBindingCounts(const BindingCounts& bindingCounts) { + if (bindingCounts.dynamicUniformBufferCount > kMaxDynamicUniformBuffersPerPipelineLayout) { + return DAWN_VALIDATION_ERROR( + "The number of dynamic uniform buffers exceeds the maximum per-pipeline-layout " + "limit"); + } + + if (bindingCounts.dynamicStorageBufferCount > kMaxDynamicStorageBuffersPerPipelineLayout) { + return DAWN_VALIDATION_ERROR( + "The number of dynamic storage buffers exceeds the maximum per-pipeline-layout " + "limit"); + } + + for (SingleShaderStage stage : IterateStages(kAllStages)) { + if (bindingCounts.perStage[stage].sampledTextureCount > + kMaxSampledTexturesPerShaderStage) { + return DAWN_VALIDATION_ERROR( + "The number of sampled textures exceeds the maximum " + "per-stage limit."); + } + if (bindingCounts.perStage[stage].samplerCount > kMaxSamplersPerShaderStage) { + return DAWN_VALIDATION_ERROR( + "The number of samplers exceeds the maximum per-stage limit."); + } + if (bindingCounts.perStage[stage].storageBufferCount > + kMaxStorageBuffersPerShaderStage) { + return DAWN_VALIDATION_ERROR( + "The number of storage buffers exceeds the maximum per-stage limit."); + } + if (bindingCounts.perStage[stage].storageTextureCount > + kMaxStorageTexturesPerShaderStage) { + return DAWN_VALIDATION_ERROR( + "The number of storage textures exceeds the maximum per-stage limit."); + } + if (bindingCounts.perStage[stage].uniformBufferCount > + kMaxUniformBuffersPerShaderStage) { + return DAWN_VALIDATION_ERROR( + "The number of uniform buffers exceeds the maximum per-stage limit."); + } + } + + return {}; + } + +} // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/BindingInfo.h b/chromium/third_party/dawn/src/dawn_native/BindingInfo.h index cac4f4ccaab..b6cfad15618 100644 --- a/chromium/third_party/dawn/src/dawn_native/BindingInfo.h +++ b/chromium/third_party/dawn/src/dawn_native/BindingInfo.h @@ -18,7 +18,10 @@ #include "common/Constants.h" #include "common/TypedInteger.h" #include "common/ityp_array.h" +#include "dawn_native/Error.h" #include "dawn_native/Format.h" +#include "dawn_native/PerStage.h" + #include "dawn_native/dawn_platform.h" #include <cstdint> @@ -33,9 +36,28 @@ namespace dawn_native { using BindGroupIndex = TypedInteger<struct BindGroupIndexT, uint32_t>; - static constexpr BindingIndex kMaxBindingsPerGroupTyped = BindingIndex(kMaxBindingsPerGroup); static constexpr BindGroupIndex kMaxBindGroupsTyped = BindGroupIndex(kMaxBindGroups); + // Not a real WebGPU limit, but the sum of the two limits is useful for internal optimizations. + static constexpr uint32_t kMaxDynamicBuffersPerPipelineLayout = + kMaxDynamicUniformBuffersPerPipelineLayout + kMaxDynamicStorageBuffersPerPipelineLayout; + + static constexpr BindingIndex kMaxDynamicBuffersPerPipelineLayoutTyped = + BindingIndex(kMaxDynamicBuffersPerPipelineLayout); + + // Not a real WebGPU limit, but used to optimize parts of Dawn which expect valid usage of the + // API. There should never be more bindings than the max per stage, for each stage. + static constexpr uint32_t kMaxBindingsPerPipelineLayout = + 3 * (kMaxSampledTexturesPerShaderStage + kMaxSamplersPerShaderStage + + kMaxStorageBuffersPerShaderStage + kMaxStorageTexturesPerShaderStage + + kMaxUniformBuffersPerShaderStage); + + static constexpr BindingIndex kMaxBindingsPerPipelineLayoutTyped = + BindingIndex(kMaxBindingsPerPipelineLayout); + + // TODO(enga): Figure out a good number for this. + static constexpr uint32_t kMaxOptimalBindingsPerGroup = 32; + struct BindingInfo { BindingNumber binding; wgpu::ShaderStage visibility; @@ -48,6 +70,27 @@ namespace dawn_native { uint64_t minBufferBindingSize = 0; }; + struct PerStageBindingCounts { + uint32_t sampledTextureCount; + uint32_t samplerCount; + uint32_t storageBufferCount; + uint32_t storageTextureCount; + uint32_t uniformBufferCount; + }; + + struct BindingCounts { + uint32_t totalCount; + uint32_t bufferCount; + uint32_t unverifiedBufferCount; // Buffers with minimum buffer size unspecified + uint32_t dynamicUniformBufferCount; + uint32_t dynamicStorageBufferCount; + PerStage<PerStageBindingCounts> perStage; + }; + + void IncrementBindingCounts(BindingCounts* bindingCounts, const BindGroupLayoutEntry& entry); + void AccumulateBindingCounts(BindingCounts* bindingCounts, const BindingCounts& rhs); + MaybeError ValidateBindingCounts(const BindingCounts& bindingCounts); + // For buffer size validation using RequiredBufferSizes = ityp::array<BindGroupIndex, std::vector<uint64_t>, kMaxBindGroups>; diff --git a/chromium/third_party/dawn/src/dawn_native/Buffer.cpp b/chromium/third_party/dawn/src/dawn_native/Buffer.cpp index 3c97655fbaa..4dd1f3a0864 100644 --- a/chromium/third_party/dawn/src/dawn_native/Buffer.cpp +++ b/chromium/third_party/dawn/src/dawn_native/Buffer.cpp @@ -15,6 +15,7 @@ #include "dawn_native/Buffer.h" #include "common/Assert.h" +#include "dawn_native/Commands.h" #include "dawn_native/Device.h" #include "dawn_native/DynamicUploader.h" #include "dawn_native/ErrorData.h" @@ -32,20 +33,21 @@ namespace dawn_native { class ErrorBuffer final : public BufferBase { public: - ErrorBuffer(DeviceBase* device) : BufferBase(device, ObjectBase::kError) { - } - - static ErrorBuffer* MakeMapped(DeviceBase* device, - uint64_t size, - uint8_t** mappedPointer) { - ASSERT(mappedPointer != nullptr); - - ErrorBuffer* buffer = new ErrorBuffer(device); - buffer->mFakeMappedData = - std::unique_ptr<uint8_t[]>(new (std::nothrow) uint8_t[size]); - *mappedPointer = buffer->mFakeMappedData.get(); - - return buffer; + ErrorBuffer(DeviceBase* device, const BufferDescriptor* descriptor) + : BufferBase(device, descriptor, ObjectBase::kError) { + if (descriptor->mappedAtCreation) { + // Check that the size can be used to allocate an mFakeMappedData. A malloc(0) + // is invalid, and on 32bit systems we should avoid a narrowing conversion that + // would make size = 1 << 32 + 1 allocate one byte. + bool isValidSize = + descriptor->size != 0 && + descriptor->size < uint64_t(std::numeric_limits<size_t>::max()); + + if (isValidSize) { + mFakeMappedData = std::unique_ptr<uint8_t[]>(new (std::nothrow) + uint8_t[descriptor->size]); + } + } } void ClearMappedData() { @@ -53,21 +55,25 @@ namespace dawn_native { } private: - bool IsMapWritable() const override { + bool IsMappableAtCreation() const override { UNREACHABLE(); return false; } - MaybeError MapAtCreationImpl(uint8_t** mappedPointer) override { + MaybeError MapAtCreationImpl() override { UNREACHABLE(); return {}; } - MaybeError MapReadAsyncImpl(uint32_t serial) override { + MaybeError MapReadAsyncImpl() override { UNREACHABLE(); return {}; } - MaybeError MapWriteAsyncImpl(uint32_t serial) override { + MaybeError MapWriteAsyncImpl() override { + UNREACHABLE(); + return {}; + } + MaybeError MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) override { UNREACHABLE(); return {}; } @@ -107,6 +113,10 @@ namespace dawn_native { return DAWN_VALIDATION_ERROR("Only CopyDst is allowed with MapRead"); } + if (descriptor->mappedAtCreation && descriptor->size % 4 != 0) { + return DAWN_VALIDATION_ERROR("size must be aligned to 4 when mappedAtCreation is true"); + } + return {}; } @@ -125,8 +135,15 @@ namespace dawn_native { } } - BufferBase::BufferBase(DeviceBase* device, ObjectBase::ErrorTag tag) - : ObjectBase(device, tag), mState(BufferState::Unmapped) { + BufferBase::BufferBase(DeviceBase* device, + const BufferDescriptor* descriptor, + ObjectBase::ErrorTag tag) + : ObjectBase(device, tag), mSize(descriptor->size), mState(BufferState::Unmapped) { + if (descriptor->mappedAtCreation) { + mState = BufferState::MappedAtCreation; + mMapOffset = 0; + mMapSize = mSize; + } } BufferBase::~BufferBase() { @@ -134,19 +151,13 @@ namespace dawn_native { ASSERT(!IsError()); CallMapReadCallback(mMapSerial, WGPUBufferMapAsyncStatus_Unknown, nullptr, 0u); CallMapWriteCallback(mMapSerial, WGPUBufferMapAsyncStatus_Unknown, nullptr, 0u); + CallMapCallback(mMapSerial, WGPUBufferMapAsyncStatus_Unknown); } } // static - BufferBase* BufferBase::MakeError(DeviceBase* device) { - return new ErrorBuffer(device); - } - - // static - BufferBase* BufferBase::MakeErrorMapped(DeviceBase* device, - uint64_t size, - uint8_t** mappedPointer) { - return ErrorBuffer::MakeMapped(device, size, mappedPointer); + BufferBase* BufferBase::MakeError(DeviceBase* device, const BufferDescriptor* descriptor) { + return new ErrorBuffer(device, descriptor); } uint64_t BufferBase::GetSize() const { @@ -159,23 +170,21 @@ namespace dawn_native { return mUsage; } - MaybeError BufferBase::MapAtCreation(uint8_t** mappedPointer) { + MaybeError BufferBase::MapAtCreation() { ASSERT(!IsError()); - ASSERT(mappedPointer != nullptr); - - // Mappable buffers don't use a staging buffer and are just as if mapped through MapAsync. - if (IsMapWritable()) { - DAWN_TRY(MapAtCreationImpl(mappedPointer)); - mState = BufferState::Mapped; - ASSERT(*mappedPointer != nullptr); - return {}; - } - mState = BufferState::MappedAtCreation; + mMapOffset = 0; + mMapSize = mSize; // 0-sized buffers are not supposed to be written to, Return back any non-null pointer. + // Handle 0-sized buffers first so we don't try to map them in the backend. if (mSize == 0) { - *mappedPointer = reinterpret_cast<uint8_t*>(intptr_t(0xCAFED00D)); + return {}; + } + + // Mappable buffers don't use a staging buffer and are just as if mapped through MapAsync. + if (IsMappableAtCreation()) { + DAWN_TRY(MapAtCreationImpl()); return {}; } @@ -185,9 +194,6 @@ namespace dawn_native { // many small buffers. DAWN_TRY_ASSIGN(mStagingBuffer, GetDevice()->CreateStagingBuffer(GetSize())); - ASSERT(mStagingBuffer->GetMappedPointer() != nullptr); - *mappedPointer = reinterpret_cast<uint8_t*>(mStagingBuffer->GetMappedPointer()); - return {}; } @@ -249,14 +255,26 @@ namespace dawn_native { } } - void BufferBase::SetSubData(uint32_t start, uint32_t count, const void* data) { - Ref<QueueBase> queue = AcquireRef(GetDevice()->GetDefaultQueue()); - GetDevice()->EmitDeprecationWarning( - "Buffer::SetSubData is deprecated, use Queue::WriteBuffer instead"); - queue->WriteBuffer(this, start, data, count); + void BufferBase::CallMapCallback(uint32_t serial, WGPUBufferMapAsyncStatus status) { + ASSERT(!IsError()); + if (mMapCallback != nullptr && serial == mMapSerial) { + // Tag the callback as fired before firing it, otherwise it could fire a second time if + // for example buffer.Unmap() is called inside the application-provided callback. + WGPUBufferMapCallback callback = mMapCallback; + mMapCallback = nullptr; + + if (GetDevice()->IsLost()) { + callback(WGPUBufferMapAsyncStatus_DeviceLost, mMapUserdata); + } else { + callback(status, mMapUserdata); + } + } } void BufferBase::MapReadAsync(WGPUBufferMapReadCallback callback, void* userdata) { + GetDevice()->EmitDeprecationWarning( + "Buffer::MapReadAsync is deprecated. Use Buffer::MapAsync instead"); + WGPUBufferMapAsyncStatus status; if (GetDevice()->ConsumedError(ValidateMap(wgpu::BufferUsage::MapRead, &status))) { callback(status, nullptr, 0, userdata); @@ -270,18 +288,23 @@ namespace dawn_native { mMapSerial++; mMapReadCallback = callback; mMapUserdata = userdata; + mMapOffset = 0; + mMapSize = mSize; mState = BufferState::Mapped; - if (GetDevice()->ConsumedError(MapReadAsyncImpl(mMapSerial))) { + if (GetDevice()->ConsumedError(MapReadAsyncImpl())) { CallMapReadCallback(mMapSerial, WGPUBufferMapAsyncStatus_DeviceLost, nullptr, 0); return; } MapRequestTracker* tracker = GetDevice()->GetMapRequestTracker(); - tracker->Track(this, mMapSerial, false); + tracker->Track(this, mMapSerial, MapType::Read); } void BufferBase::MapWriteAsync(WGPUBufferMapWriteCallback callback, void* userdata) { + GetDevice()->EmitDeprecationWarning( + "Buffer::MapReadAsync is deprecated. Use Buffer::MapAsync instead"); + WGPUBufferMapAsyncStatus status; if (GetDevice()->ConsumedError(ValidateMap(wgpu::BufferUsage::MapWrite, &status))) { callback(status, nullptr, 0, userdata); @@ -295,22 +318,88 @@ namespace dawn_native { mMapSerial++; mMapWriteCallback = callback; mMapUserdata = userdata; + mMapOffset = 0; + mMapSize = mSize; mState = BufferState::Mapped; - if (GetDevice()->ConsumedError(MapWriteAsyncImpl(mMapSerial))) { + if (GetDevice()->ConsumedError(MapWriteAsyncImpl())) { CallMapWriteCallback(mMapSerial, WGPUBufferMapAsyncStatus_DeviceLost, nullptr, 0); return; } MapRequestTracker* tracker = GetDevice()->GetMapRequestTracker(); - tracker->Track(this, mMapSerial, true); + tracker->Track(this, mMapSerial, MapType::Write); + } + + void BufferBase::MapAsync(wgpu::MapMode mode, + size_t offset, + size_t size, + WGPUBufferMapCallback callback, + void* userdata) { + // Handle the defaulting of size required by WebGPU, even if in webgpu_cpp.h it is not + // possible to default the function argument (because there is the callback later in the + // argument list) + if (size == 0 && offset < mSize) { + size = mSize - offset; + } + + WGPUBufferMapAsyncStatus status; + if (GetDevice()->ConsumedError(ValidateMapAsync(mode, offset, size, &status))) { + if (callback) { + callback(status, userdata); + } + return; + } + ASSERT(!IsError()); + + // TODO(cwallez@chromium.org): what to do on wraparound? Could cause crashes. + mMapSerial++; + mMapMode = mode; + mMapOffset = offset; + mMapSize = size; + mMapCallback = callback; + mMapUserdata = userdata; + mState = BufferState::Mapped; + + if (GetDevice()->ConsumedError(MapAsyncImpl(mode, offset, size))) { + CallMapCallback(mMapSerial, WGPUBufferMapAsyncStatus_DeviceLost); + return; + } + + MapRequestTracker* tracker = GetDevice()->GetMapRequestTracker(); + tracker->Track(this, mMapSerial, MapType::Async); + } + + void* BufferBase::GetMappedRange(size_t offset, size_t size) { + return GetMappedRangeInternal(true, offset, size); + } + + const void* BufferBase::GetConstMappedRange(size_t offset, size_t size) { + return GetMappedRangeInternal(false, offset, size); + } + + // TODO(dawn:445): When CreateBufferMapped is removed, make GetMappedRangeInternal also take + // care of the validation of GetMappedRange. + void* BufferBase::GetMappedRangeInternal(bool writable, size_t offset, size_t size) { + if (!CanGetMappedRange(writable, offset, size)) { + return nullptr; + } + + if (mStagingBuffer != nullptr) { + return static_cast<uint8_t*>(mStagingBuffer->GetMappedPointer()) + offset; + } + if (mSize == 0) { + return reinterpret_cast<uint8_t*>(intptr_t(0xCAFED00D)); + } + return static_cast<uint8_t*>(GetMappedPointerImpl()) + offset; } void BufferBase::Destroy() { if (IsError()) { // It is an error to call Destroy() on an ErrorBuffer, but we still need to reclaim the // fake mapped staging data. - reinterpret_cast<ErrorBuffer*>(this)->ClearMappedData(); + static_cast<ErrorBuffer*>(this)->ClearMappedData(); + mState = BufferState::Destroyed; } if (GetDevice()->ConsumedError(ValidateDestroy())) { return; @@ -322,8 +411,9 @@ namespace dawn_native { } else if (mState == BufferState::MappedAtCreation) { if (mStagingBuffer != nullptr) { mStagingBuffer.reset(); - } else { - ASSERT(mSize == 0); + } else if (mSize != 0) { + ASSERT(IsMappableAtCreation()); + Unmap(); } } @@ -348,7 +438,8 @@ namespace dawn_native { if (IsError()) { // It is an error to call Unmap() on an ErrorBuffer, but we still need to reclaim the // fake mapped staging data. - reinterpret_cast<ErrorBuffer*>(this)->ClearMappedData(); + static_cast<ErrorBuffer*>(this)->ClearMappedData(); + mState = BufferState::Unmapped; } if (GetDevice()->ConsumedError(ValidateUnmap())) { return; @@ -362,17 +453,20 @@ namespace dawn_native { // CreateBufferMapped. CallMapReadCallback(mMapSerial, WGPUBufferMapAsyncStatus_Unknown, nullptr, 0u); CallMapWriteCallback(mMapSerial, WGPUBufferMapAsyncStatus_Unknown, nullptr, 0u); + CallMapCallback(mMapSerial, WGPUBufferMapAsyncStatus_Unknown); UnmapImpl(); mMapReadCallback = nullptr; mMapWriteCallback = nullptr; + mMapCallback = nullptr; mMapUserdata = 0; } else if (mState == BufferState::MappedAtCreation) { if (mStagingBuffer != nullptr) { GetDevice()->ConsumedError(CopyFromStagingBuffer()); - } else { - ASSERT(mSize == 0); + } else if (mSize != 0) { + ASSERT(IsMappableAtCreation()); + UnmapImpl(); } } @@ -390,7 +484,7 @@ namespace dawn_native { switch (mState) { case BufferState::Mapped: case BufferState::MappedAtCreation: - return DAWN_VALIDATION_ERROR("Buffer already mapped"); + return DAWN_VALIDATION_ERROR("Buffer is already mapped"); case BufferState::Destroyed: return DAWN_VALIDATION_ERROR("Buffer is destroyed"); case BufferState::Unmapped: @@ -405,6 +499,99 @@ namespace dawn_native { return {}; } + MaybeError BufferBase::ValidateMapAsync(wgpu::MapMode mode, + size_t offset, + size_t size, + WGPUBufferMapAsyncStatus* status) const { + *status = WGPUBufferMapAsyncStatus_DeviceLost; + DAWN_TRY(GetDevice()->ValidateIsAlive()); + + *status = WGPUBufferMapAsyncStatus_Error; + DAWN_TRY(GetDevice()->ValidateObject(this)); + + if (offset % 4 != 0) { + return DAWN_VALIDATION_ERROR("offset must be a multiple of 4"); + } + + if (size % 4 != 0) { + return DAWN_VALIDATION_ERROR("size must be a multiple of 4"); + } + + if (uint64_t(offset) > mSize || uint64_t(size) > mSize - uint64_t(offset)) { + return DAWN_VALIDATION_ERROR("size + offset must fit in the buffer"); + } + + switch (mState) { + case BufferState::Mapped: + case BufferState::MappedAtCreation: + return DAWN_VALIDATION_ERROR("Buffer is already mapped"); + case BufferState::Destroyed: + return DAWN_VALIDATION_ERROR("Buffer is destroyed"); + case BufferState::Unmapped: + break; + } + + bool isReadMode = mode & wgpu::MapMode::Read; + bool isWriteMode = mode & wgpu::MapMode::Write; + if (!(isReadMode ^ isWriteMode)) { + return DAWN_VALIDATION_ERROR("Exactly one of Read or Write mode must be set"); + } + + if (mode & wgpu::MapMode::Read) { + if (!(mUsage & wgpu::BufferUsage::MapRead)) { + return DAWN_VALIDATION_ERROR("The buffer must have the MapRead usage"); + } + } else { + ASSERT(mode & wgpu::MapMode::Write); + + if (!(mUsage & wgpu::BufferUsage::MapWrite)) { + return DAWN_VALIDATION_ERROR("The buffer must have the MapWrite usage"); + } + } + + *status = WGPUBufferMapAsyncStatus_Success; + return {}; + } + + bool BufferBase::CanGetMappedRange(bool writable, size_t offset, size_t size) const { + if (size > mMapSize || offset < mMapOffset) { + return false; + } + + size_t offsetInMappedRange = offset - mMapOffset; + if (offsetInMappedRange > mMapSize - size) { + return false; + } + + // Note that: + // + // - We don't check that the device is alive because the application can ask for the + // mapped pointer before it knows, and even Dawn knows, that the device was lost, and + // still needs to work properly. + // - We don't check that the object is alive because we need to return mapped pointers + // for error buffers too. + + switch (mState) { + // Writeable Buffer::GetMappedRange is always allowed when mapped at creation. + case BufferState::MappedAtCreation: + return true; + + case BufferState::Mapped: + // TODO(dawn:445): When mapRead/WriteAsync is removed, check against mMapMode + // instead of mUsage + ASSERT(bool(mUsage & wgpu::BufferUsage::MapRead) ^ + bool(mUsage & wgpu::BufferUsage::MapWrite)); + return !writable || (mUsage & wgpu::BufferUsage::MapWrite); + + case BufferState::Unmapped: + case BufferState::Destroyed: + return false; + + default: + UNREACHABLE(); + } + } + MaybeError BufferBase::ValidateUnmap() const { DAWN_TRY(GetDevice()->ValidateIsAlive()); DAWN_TRY(GetDevice()->ValidateObject(this)); @@ -439,17 +626,31 @@ namespace dawn_native { mState = BufferState::Destroyed; } - bool BufferBase::IsMapped() const { - return mState == BufferState::Mapped; + void BufferBase::OnMapCommandSerialFinished(uint32_t mapSerial, MapType type) { + switch (type) { + case MapType::Read: + CallMapReadCallback(mapSerial, WGPUBufferMapAsyncStatus_Success, + GetMappedRangeInternal(false, 0, mSize), GetSize()); + break; + case MapType::Write: + CallMapWriteCallback(mapSerial, WGPUBufferMapAsyncStatus_Success, + GetMappedRangeInternal(true, 0, mSize), GetSize()); + break; + case MapType::Async: + CallMapCallback(mapSerial, WGPUBufferMapAsyncStatus_Success); + break; + } + } + + bool BufferBase::IsDataInitialized() const { + return mIsDataInitialized; } - void BufferBase::OnMapCommandSerialFinished(uint32_t mapSerial, bool isWrite) { - void* data = GetMappedPointerImpl(); - if (isWrite) { - CallMapWriteCallback(mapSerial, WGPUBufferMapAsyncStatus_Success, data, GetSize()); - } else { - CallMapReadCallback(mapSerial, WGPUBufferMapAsyncStatus_Success, data, GetSize()); - } + void BufferBase::SetIsDataInitialized() { + mIsDataInitialized = true; } + bool BufferBase::IsFullBufferRange(uint64_t offset, uint64_t size) const { + return offset == 0 && size == GetSize(); + } } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/Buffer.h b/chromium/third_party/dawn/src/dawn_native/Buffer.h index 8a1feb755ef..1dd5278ce72 100644 --- a/chromium/third_party/dawn/src/dawn_native/Buffer.h +++ b/chromium/third_party/dawn/src/dawn_native/Buffer.h @@ -25,6 +25,10 @@ namespace dawn_native { + struct CopyTextureToBufferCmd; + + enum class MapType : uint32_t; + MaybeError ValidateBufferDescriptor(DeviceBase* device, const BufferDescriptor* descriptor); static constexpr wgpu::BufferUsage kReadOnlyBufferUsages = @@ -40,74 +44,92 @@ namespace dawn_native { }; public: - enum class ClearValue { Zero, NonZero }; - BufferBase(DeviceBase* device, const BufferDescriptor* descriptor); - static BufferBase* MakeError(DeviceBase* device); - static BufferBase* MakeErrorMapped(DeviceBase* device, - uint64_t size, - uint8_t** mappedPointer); + static BufferBase* MakeError(DeviceBase* device, const BufferDescriptor* descriptor); uint64_t GetSize() const; wgpu::BufferUsage GetUsage() const; - MaybeError MapAtCreation(uint8_t** mappedPointer); - void OnMapCommandSerialFinished(uint32_t mapSerial, bool isWrite); + MaybeError MapAtCreation(); + void OnMapCommandSerialFinished(uint32_t mapSerial, MapType type); MaybeError ValidateCanUseOnQueueNow() const; + bool IsFullBufferRange(uint64_t offset, uint64_t size) const; + bool IsDataInitialized() const; + void SetIsDataInitialized(); + // Dawn API - void SetSubData(uint32_t start, uint32_t count, const void* data); void MapReadAsync(WGPUBufferMapReadCallback callback, void* userdata); void MapWriteAsync(WGPUBufferMapWriteCallback callback, void* userdata); + void MapAsync(wgpu::MapMode mode, + size_t offset, + size_t size, + WGPUBufferMapCallback callback, + void* userdata); + void* GetMappedRange(size_t offset, size_t size); + const void* GetConstMappedRange(size_t offset, size_t size); void Unmap(); void Destroy(); protected: - BufferBase(DeviceBase* device, ObjectBase::ErrorTag tag); + BufferBase(DeviceBase* device, + const BufferDescriptor* descriptor, + ObjectBase::ErrorTag tag); ~BufferBase() override; - void CallMapReadCallback(uint32_t serial, - WGPUBufferMapAsyncStatus status, - const void* pointer, - uint64_t dataLength); - void CallMapWriteCallback(uint32_t serial, - WGPUBufferMapAsyncStatus status, - void* pointer, - uint64_t dataLength); - void DestroyInternal(); bool IsMapped() const; private: - virtual MaybeError MapAtCreationImpl(uint8_t** mappedPointer) = 0; - virtual MaybeError MapReadAsyncImpl(uint32_t serial) = 0; - virtual MaybeError MapWriteAsyncImpl(uint32_t serial) = 0; + virtual MaybeError MapAtCreationImpl() = 0; + virtual MaybeError MapReadAsyncImpl() = 0; + virtual MaybeError MapWriteAsyncImpl() = 0; + virtual MaybeError MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) = 0; virtual void UnmapImpl() = 0; virtual void DestroyImpl() = 0; virtual void* GetMappedPointerImpl() = 0; - virtual bool IsMapWritable() const = 0; + virtual bool IsMappableAtCreation() const = 0; MaybeError CopyFromStagingBuffer(); + void* GetMappedRangeInternal(bool writable, size_t offset, size_t size); + void CallMapReadCallback(uint32_t serial, + WGPUBufferMapAsyncStatus status, + const void* pointer, + uint64_t dataLength); + void CallMapWriteCallback(uint32_t serial, + WGPUBufferMapAsyncStatus status, + void* pointer, + uint64_t dataLength); + void CallMapCallback(uint32_t serial, WGPUBufferMapAsyncStatus status); MaybeError ValidateMap(wgpu::BufferUsage requiredUsage, WGPUBufferMapAsyncStatus* status) const; + MaybeError ValidateMapAsync(wgpu::MapMode mode, + size_t offset, + size_t size, + WGPUBufferMapAsyncStatus* status) const; MaybeError ValidateUnmap() const; MaybeError ValidateDestroy() const; + bool CanGetMappedRange(bool writable, size_t offset, size_t size) const; uint64_t mSize = 0; wgpu::BufferUsage mUsage = wgpu::BufferUsage::None; + BufferState mState; + bool mIsDataInitialized = false; + + std::unique_ptr<StagingBufferBase> mStagingBuffer; WGPUBufferMapReadCallback mMapReadCallback = nullptr; WGPUBufferMapWriteCallback mMapWriteCallback = nullptr; + WGPUBufferMapCallback mMapCallback = nullptr; void* mMapUserdata = 0; uint32_t mMapSerial = 0; - - std::unique_ptr<StagingBufferBase> mStagingBuffer; - - BufferState mState; + wgpu::MapMode mMapMode = wgpu::MapMode::None; + size_t mMapOffset = 0; + size_t mMapSize = 0; }; } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/CMakeLists.txt b/chromium/third_party/dawn/src/dawn_native/CMakeLists.txt index cabbc1a6c31..afac9c5cd97 100644 --- a/chromium/third_party/dawn/src/dawn_native/CMakeLists.txt +++ b/chromium/third_party/dawn/src/dawn_native/CMakeLists.txt @@ -35,6 +35,7 @@ target_sources(dawn_native PRIVATE "BindGroupLayout.cpp" "BindGroupLayout.h" "BindGroupTracker.h" + "BindingInfo.cpp" "BindingInfo.h" "BuddyAllocator.cpp" "BuddyAllocator.h" @@ -66,6 +67,8 @@ target_sources(dawn_native PRIVATE "DynamicUploader.h" "EncodingContext.cpp" "EncodingContext.h" + "EnumClassBitmasks.h" + "EnumMaskIterator.h" "Error.cpp" "Error.h" "ErrorData.cpp" @@ -100,6 +103,8 @@ target_sources(dawn_native PRIVATE "Pipeline.h" "PipelineLayout.cpp" "PipelineLayout.h" + "PooledResourceMemoryAllocator.cpp" + "PooledResourceMemoryAllocator.h" "ProgrammablePassEncoder.cpp" "ProgrammablePassEncoder.h" "QuerySet.cpp" @@ -201,6 +206,8 @@ if (DAWN_ENABLE_D3D12) "d3d12/PipelineLayoutD3D12.h" "d3d12/PlatformFunctions.cpp" "d3d12/PlatformFunctions.h" + "d3d12/QuerySetD3D12.cpp" + "d3d12/QuerySetD3D12.h" "d3d12/QueueD3D12.cpp" "d3d12/QueueD3D12.h" "d3d12/RenderPassBuilderD3D12.cpp" @@ -283,6 +290,7 @@ if (DAWN_ENABLE_METAL) "-framework IOKit" "-framework IOSurface" "-framework QuartzCore" + "-framework Metal" ) endif() @@ -388,6 +396,8 @@ if (DAWN_ENABLE_VULKAN) "vulkan/NativeSwapChainImplVk.h" "vulkan/PipelineLayoutVk.cpp" "vulkan/PipelineLayoutVk.h" + "vulkan/QuerySetVk.cpp" + "vulkan/QuerySetVk.h" "vulkan/QueueVk.cpp" "vulkan/QueueVk.h" "vulkan/RenderPassCache.cpp" diff --git a/chromium/third_party/dawn/src/dawn_native/CommandAllocator.cpp b/chromium/third_party/dawn/src/dawn_native/CommandAllocator.cpp index 553f8896577..7c94520e076 100644 --- a/chromium/third_party/dawn/src/dawn_native/CommandAllocator.cpp +++ b/chromium/third_party/dawn/src/dawn_native/CommandAllocator.cpp @@ -30,13 +30,7 @@ namespace dawn_native { } CommandIterator::~CommandIterator() { - ASSERT(mDataWasDestroyed); - - if (!IsEmpty()) { - for (auto& block : mBlocks) { - free(block.block); - } - } + ASSERT(IsEmpty()); } CommandIterator::CommandIterator(CommandIterator&& other) { @@ -44,18 +38,13 @@ namespace dawn_native { mBlocks = std::move(other.mBlocks); other.Reset(); } - other.DataWasDestroyed(); Reset(); } CommandIterator& CommandIterator::operator=(CommandIterator&& other) { - if (!other.IsEmpty()) { - mBlocks = std::move(other.mBlocks); - other.Reset(); - } else { - mBlocks.clear(); - } - other.DataWasDestroyed(); + ASSERT(IsEmpty()); + mBlocks = std::move(other.mBlocks); + other.Reset(); Reset(); return *this; } @@ -66,6 +55,7 @@ namespace dawn_native { } CommandIterator& CommandIterator::operator=(CommandAllocator&& allocator) { + ASSERT(IsEmpty()); mBlocks = allocator.AcquireBlocks(); Reset(); return *this; @@ -97,8 +87,17 @@ namespace dawn_native { } } - void CommandIterator::DataWasDestroyed() { - mDataWasDestroyed = true; + void CommandIterator::MakeEmptyAsDataWasDestroyed() { + if (IsEmpty()) { + return; + } + + for (auto& block : mBlocks) { + free(block.block); + } + mBlocks.clear(); + Reset(); + ASSERT(IsEmpty()); } bool CommandIterator::IsEmpty() const { diff --git a/chromium/third_party/dawn/src/dawn_native/CommandAllocator.h b/chromium/third_party/dawn/src/dawn_native/CommandAllocator.h index 82de05c1a45..0383dc1639a 100644 --- a/chromium/third_party/dawn/src/dawn_native/CommandAllocator.h +++ b/chromium/third_party/dawn/src/dawn_native/CommandAllocator.h @@ -91,10 +91,13 @@ namespace dawn_native { return static_cast<T*>(NextData(sizeof(T) * count, alignof(T))); } - // Needs to be called if iteration was stopped early. + // Sets iterator to the beginning of the commands without emptying the list. This method can + // be used if iteration was stopped early and the iterator needs to be restarted. void Reset(); - void DataWasDestroyed(); + // This method must to be called after commands have been deleted. This indicates that the + // commands have been submitted and they are no longer valid. + void MakeEmptyAsDataWasDestroyed(); private: bool IsEmpty() const; @@ -139,7 +142,6 @@ namespace dawn_native { size_t mCurrentBlock = 0; // Used to avoid a special case for empty iterators. uint32_t mEndOfBlock = detail::kEndOfBlock; - bool mDataWasDestroyed = false; }; class CommandAllocator { diff --git a/chromium/third_party/dawn/src/dawn_native/CommandBuffer.cpp b/chromium/third_party/dawn/src/dawn_native/CommandBuffer.cpp index 401451d4911..91fec29f180 100644 --- a/chromium/third_party/dawn/src/dawn_native/CommandBuffer.cpp +++ b/chromium/third_party/dawn/src/dawn_native/CommandBuffer.cpp @@ -15,6 +15,7 @@ #include "dawn_native/CommandBuffer.h" #include "common/BitSetIterator.h" +#include "dawn_native/Buffer.h" #include "dawn_native/CommandEncoder.h" #include "dawn_native/Commands.h" #include "dawn_native/Format.h" @@ -23,18 +24,39 @@ namespace dawn_native { CommandBufferBase::CommandBufferBase(CommandEncoder* encoder, const CommandBufferDescriptor*) - : ObjectBase(encoder->GetDevice()), mResourceUsages(encoder->AcquireResourceUsages()) { + : ObjectBase(encoder->GetDevice()), + mCommands(encoder->AcquireCommands()), + mResourceUsages(encoder->AcquireResourceUsages()) { } CommandBufferBase::CommandBufferBase(DeviceBase* device, ObjectBase::ErrorTag tag) : ObjectBase(device, tag) { } + CommandBufferBase::~CommandBufferBase() { + Destroy(); + } + // static CommandBufferBase* CommandBufferBase::MakeError(DeviceBase* device) { return new CommandBufferBase(device, ObjectBase::kError); } + MaybeError CommandBufferBase::ValidateCanUseInSubmitNow() const { + ASSERT(!IsError()); + + if (mDestroyed) { + return DAWN_VALIDATION_ERROR("Command buffer reused in submit"); + } + return {}; + } + + void CommandBufferBase::Destroy() { + FreeCommands(&mCommands); + mResourceUsages = {}; + mDestroyed = true; + } + const CommandBufferResourceUsage& CommandBufferBase::GetResourceUsages() const { return mResourceUsages; } @@ -51,6 +73,17 @@ namespace dawn_native { return false; } + SubresourceRange GetSubresourcesAffectedByCopy(const TextureCopy& copy, + const Extent3D& copySize) { + switch (copy.texture->GetDimension()) { + case wgpu::TextureDimension::e2D: + return {copy.mipLevel, 1, copy.origin.z, copySize.depth, copy.aspect}; + default: + UNREACHABLE(); + return {}; + } + } + void LazyClearRenderPassAttachments(BeginRenderPassCmd* renderPass) { for (uint32_t i : IterateBitSet(renderPass->attachmentState->GetColorAttachmentsMask())) { auto& attachmentInfo = renderPass->colorAttachments[i]; @@ -101,37 +134,75 @@ namespace dawn_native { ASSERT(view->GetLevelCount() == 1); SubresourceRange range = view->GetSubresourceRange(); + SubresourceRange depthRange = range; + depthRange.aspects = range.aspects & Aspect::Depth; + + SubresourceRange stencilRange = range; + stencilRange.aspects = range.aspects & Aspect::Stencil; + // If the depth stencil texture has not been initialized, we want to use loadop // clear to init the contents to 0's - if (!view->GetTexture()->IsSubresourceContentInitialized(range)) { - if (view->GetTexture()->GetFormat().HasDepth() && - attachmentInfo.depthLoadOp == wgpu::LoadOp::Load) { - attachmentInfo.clearDepth = 0.0f; - attachmentInfo.depthLoadOp = wgpu::LoadOp::Clear; - } - if (view->GetTexture()->GetFormat().HasStencil() && - attachmentInfo.stencilLoadOp == wgpu::LoadOp::Load) { - attachmentInfo.clearStencil = 0u; - attachmentInfo.stencilLoadOp = wgpu::LoadOp::Clear; - } + if (!view->GetTexture()->IsSubresourceContentInitialized(depthRange) && + attachmentInfo.depthLoadOp == wgpu::LoadOp::Load) { + attachmentInfo.clearDepth = 0.0f; + attachmentInfo.depthLoadOp = wgpu::LoadOp::Clear; } - // If these have different store ops, make them both Store because we can't track - // initialized state separately yet. TODO(crbug.com/dawn/145) - if (attachmentInfo.depthStoreOp != attachmentInfo.stencilStoreOp) { - attachmentInfo.depthStoreOp = wgpu::StoreOp::Store; - attachmentInfo.stencilStoreOp = wgpu::StoreOp::Store; + if (!view->GetTexture()->IsSubresourceContentInitialized(stencilRange) && + attachmentInfo.stencilLoadOp == wgpu::LoadOp::Load) { + attachmentInfo.clearStencil = 0u; + attachmentInfo.stencilLoadOp = wgpu::LoadOp::Clear; } - if (attachmentInfo.depthStoreOp == wgpu::StoreOp::Store && - attachmentInfo.stencilStoreOp == wgpu::StoreOp::Store) { - view->GetTexture()->SetIsSubresourceContentInitialized(true, range); - } else { - ASSERT(attachmentInfo.depthStoreOp == wgpu::StoreOp::Clear && - attachmentInfo.stencilStoreOp == wgpu::StoreOp::Clear); - view->GetTexture()->SetIsSubresourceContentInitialized(false, range); - } + view->GetTexture()->SetIsSubresourceContentInitialized( + attachmentInfo.depthStoreOp == wgpu::StoreOp::Store, depthRange); + + view->GetTexture()->SetIsSubresourceContentInitialized( + attachmentInfo.stencilStoreOp == wgpu::StoreOp::Store, stencilRange); } } + // TODO(jiawei.shao@intel.com): support copying with depth stencil textures + bool IsFullBufferOverwrittenInTextureToBufferCopy(const CopyTextureToBufferCmd* copy) { + ASSERT(copy != nullptr); + + if (copy->destination.offset > 0) { + return false; + } + + if (copy->destination.rowsPerImage > copy->copySize.height) { + return false; + } + + const TextureBase* texture = copy->source.texture.Get(); + const uint64_t copyTextureDataSizePerRow = copy->copySize.width / + texture->GetFormat().blockWidth * + texture->GetFormat().blockByteSize; + if (copy->destination.bytesPerRow > copyTextureDataSizePerRow) { + return false; + } + + const uint64_t overwrittenRangeSize = + copyTextureDataSizePerRow * (copy->copySize.height / texture->GetFormat().blockHeight) * + copy->copySize.depth; + if (copy->destination.buffer->GetSize() > overwrittenRangeSize) { + return false; + } + + return true; + } + + std::array<int32_t, 4> ConvertToSignedIntegerColor(dawn_native::Color color) { + const std::array<int32_t, 4> outputValue = { + static_cast<int32_t>(color.r), static_cast<int32_t>(color.g), + static_cast<int32_t>(color.b), static_cast<int32_t>(color.a)}; + return outputValue; + } + + std::array<uint32_t, 4> ConvertToUnsignedIntegerColor(dawn_native::Color color) { + const std::array<uint32_t, 4> outputValue = { + static_cast<uint32_t>(color.r), static_cast<uint32_t>(color.g), + static_cast<uint32_t>(color.b), static_cast<uint32_t>(color.a)}; + return outputValue; + } } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/CommandBuffer.h b/chromium/third_party/dawn/src/dawn_native/CommandBuffer.h index c1d2597ff84..54a8f7649f4 100644 --- a/chromium/third_party/dawn/src/dawn_native/CommandBuffer.h +++ b/chromium/third_party/dawn/src/dawn_native/CommandBuffer.h @@ -17,32 +17,55 @@ #include "dawn_native/dawn_platform.h" +#include "dawn_native/CommandAllocator.h" +#include "dawn_native/Error.h" #include "dawn_native/Forward.h" #include "dawn_native/ObjectBase.h" #include "dawn_native/PassResourceUsage.h" +#include "dawn_native/Texture.h" namespace dawn_native { struct BeginRenderPassCmd; + struct CopyTextureToBufferCmd; + struct TextureCopy; class CommandBufferBase : public ObjectBase { public: CommandBufferBase(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor); + static CommandBufferBase* MakeError(DeviceBase* device); + MaybeError ValidateCanUseInSubmitNow() const; + void Destroy(); + const CommandBufferResourceUsage& GetResourceUsages() const; + protected: + ~CommandBufferBase(); + + CommandIterator mCommands; + private: CommandBufferBase(DeviceBase* device, ObjectBase::ErrorTag tag); CommandBufferResourceUsage mResourceUsages; + bool mDestroyed = false; }; + bool IsCompleteSubresourceCopiedTo(const TextureBase* texture, const Extent3D copySize, const uint32_t mipLevel); + SubresourceRange GetSubresourcesAffectedByCopy(const TextureCopy& copy, + const Extent3D& copySize); void LazyClearRenderPassAttachments(BeginRenderPassCmd* renderPass); + bool IsFullBufferOverwrittenInTextureToBufferCopy(const CopyTextureToBufferCmd* copy); + + std::array<int32_t, 4> ConvertToSignedIntegerColor(dawn_native::Color color); + std::array<uint32_t, 4> ConvertToUnsignedIntegerColor(dawn_native::Color color); + } // namespace dawn_native #endif // DAWNNATIVE_COMMANDBUFFER_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/CommandEncoder.cpp b/chromium/third_party/dawn/src/dawn_native/CommandEncoder.cpp index 5b8bc640f83..db4968c3020 100644 --- a/chromium/third_party/dawn/src/dawn_native/CommandEncoder.cpp +++ b/chromium/third_party/dawn/src/dawn_native/CommandEncoder.cpp @@ -15,6 +15,7 @@ #include "dawn_native/CommandEncoder.h" #include "common/BitSetIterator.h" +#include "common/Math.h" #include "dawn_native/BindGroup.h" #include "dawn_native/Buffer.h" #include "dawn_native/CommandBuffer.h" @@ -24,6 +25,7 @@ #include "dawn_native/ComputePassEncoder.h" #include "dawn_native/Device.h" #include "dawn_native/ErrorData.h" +#include "dawn_native/QuerySet.h" #include "dawn_native/RenderPassEncoder.h" #include "dawn_native/RenderPipeline.h" #include "dawn_native/ValidationUtils_autogen.h" @@ -37,55 +39,6 @@ namespace dawn_native { namespace { - // TODO(jiawei.shao@intel.com): add validations on the texture-to-texture copies within the - // same texture. - MaybeError ValidateCopySizeFitsInTexture(const TextureCopyView& textureCopy, - const Extent3D& copySize) { - const TextureBase* texture = textureCopy.texture; - if (textureCopy.mipLevel >= texture->GetNumMipLevels()) { - return DAWN_VALIDATION_ERROR("Copy mipLevel out of range"); - } - - Extent3D mipSize = texture->GetMipLevelPhysicalSize(textureCopy.mipLevel); - // For 2D textures, include the array layer as depth so it can be checked with other - // dimensions. - ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D); - mipSize.depth = texture->GetArrayLayers(); - - // All texture dimensions are in uint32_t so by doing checks in uint64_t we avoid - // overflows. - if (static_cast<uint64_t>(textureCopy.origin.x) + - static_cast<uint64_t>(copySize.width) > - static_cast<uint64_t>(mipSize.width) || - static_cast<uint64_t>(textureCopy.origin.y) + - static_cast<uint64_t>(copySize.height) > - static_cast<uint64_t>(mipSize.height) || - static_cast<uint64_t>(textureCopy.origin.z) + - static_cast<uint64_t>(copySize.depth) > - static_cast<uint64_t>(mipSize.depth)) { - return DAWN_VALIDATION_ERROR("Copy would touch outside of the texture"); - } - - return {}; - } - - MaybeError ValidateCopySizeFitsInBuffer(const Ref<BufferBase>& buffer, - uint64_t offset, - uint64_t size) { - uint64_t bufferSize = buffer->GetSize(); - bool fitsInBuffer = offset <= bufferSize && (size <= (bufferSize - offset)); - if (!fitsInBuffer) { - return DAWN_VALIDATION_ERROR("Copy would overflow the buffer"); - } - - return {}; - } - - MaybeError ValidateCopySizeFitsInBuffer(const BufferCopyView& bufferCopy, - uint64_t dataSize) { - return ValidateCopySizeFitsInBuffer(bufferCopy.buffer, bufferCopy.offset, dataSize); - } - MaybeError ValidateB2BCopyAlignment(uint64_t dataSize, uint64_t srcOffset, uint64_t dstOffset) { @@ -103,31 +56,6 @@ namespace dawn_native { return {}; } - MaybeError ValidateTexelBufferOffset(const BufferCopyView& bufferCopy, - const Format& format) { - if (bufferCopy.offset % format.blockByteSize != 0) { - return DAWN_VALIDATION_ERROR( - "Buffer offset must be a multiple of the texel or block size"); - } - - return {}; - } - - MaybeError ValidateRowsPerImage(const Format& format, - uint32_t rowsPerImage, - uint32_t copyHeight) { - if (rowsPerImage < copyHeight) { - return DAWN_VALIDATION_ERROR("rowsPerImage must not be less than the copy height."); - } - - if (rowsPerImage % format.blockHeight != 0) { - return DAWN_VALIDATION_ERROR( - "rowsPerImage must be a multiple of compressed texture format block height"); - } - - return {}; - } - MaybeError ValidateTextureSampleCountInCopyCommands(const TextureBase* texture) { if (texture->GetSampleCount() > 1) { return DAWN_VALIDATION_ERROR("The sample count of textures must be 1"); @@ -173,6 +101,12 @@ namespace dawn_native { return DAWN_VALIDATION_ERROR("Source and destination texture formats must match."); } + if (src.aspect != wgpu::TextureAspect::All || dst.aspect != wgpu::TextureAspect::All) { + // Metal cannot select a single aspect for texture-to-texture copies + return DAWN_VALIDATION_ERROR( + "Texture aspect must be \"all\" for texture to texture copies"); + } + if (src.texture->GetFormat().HasDepthOrStencil()) { // D3D12 requires entire subresource to be copied when using CopyTextureRegion is // used with depth/stencil. @@ -192,72 +126,50 @@ namespace dawn_native { return {}; } - MaybeError ComputeTextureCopyBufferSize(const Format& textureFormat, - const Extent3D& copySize, - uint32_t bytesPerRow, - uint32_t rowsPerImage, - uint32_t* bufferSize) { - ASSERT(rowsPerImage >= copySize.height); - if (copySize.width == 0 || copySize.height == 0 || copySize.depth == 0) { - *bufferSize = 0; - return {}; - } - - uint32_t blockByteSize = textureFormat.blockByteSize; - uint32_t blockWidth = textureFormat.blockWidth; - uint32_t blockHeight = textureFormat.blockHeight; - - // TODO(cwallez@chromium.org): check for overflows - uint32_t slicePitch = bytesPerRow * rowsPerImage / blockWidth; - - ASSERT(copySize.height >= 1); - uint32_t sliceSize = bytesPerRow * (copySize.height / blockHeight - 1) + - (copySize.width / blockWidth) * blockByteSize; - - ASSERT(copySize.depth >= 1); - *bufferSize = (slicePitch * (copySize.depth - 1)) + sliceSize; - - return {}; - } - - MaybeError ValidateBytesPerRow(const Format& format, - const Extent3D& copySize, - uint32_t bytesPerRow) { - if (bytesPerRow % kTextureBytesPerRowAlignment != 0) { - return DAWN_VALIDATION_ERROR("bytesPerRow must be a multiple of 256"); - } - - if (bytesPerRow < copySize.width / format.blockWidth * format.blockByteSize) { - return DAWN_VALIDATION_ERROR( - "bytesPerRow must not be less than the number of bytes per row"); - } - - return {}; - } - - MaybeError ValidateImageOrigin(const Format& format, const Origin3D& offset) { - if (offset.x % format.blockWidth != 0) { - return DAWN_VALIDATION_ERROR( - "Offset.x must be a multiple of compressed texture format block width"); - } - - if (offset.y % format.blockHeight != 0) { - return DAWN_VALIDATION_ERROR( - "Offset.y must be a multiple of compressed texture format block height"); - } - - return {}; - } - - MaybeError ValidateImageCopySize(const Format& format, const Extent3D& extent) { - if (extent.width % format.blockWidth != 0) { - return DAWN_VALIDATION_ERROR( - "Extent.width must be a multiple of compressed texture format block width"); - } - - if (extent.height % format.blockHeight != 0) { - return DAWN_VALIDATION_ERROR( - "Extent.height must be a multiple of compressed texture format block height"); + MaybeError ValidateTextureToBufferCopyRestrictions(const TextureCopyView& src) { + const Format& format = src.texture->GetFormat(); + + bool depthSelected = false; + switch (src.aspect) { + case wgpu::TextureAspect::All: + switch (format.aspects) { + case Aspect::Color: + case Aspect::Stencil: + break; + case Aspect::Depth: + depthSelected = true; + break; + default: + return DAWN_VALIDATION_ERROR( + "A single aspect must be selected for multi planar formats in " + "texture to buffer copies"); + } + break; + case wgpu::TextureAspect::DepthOnly: + ASSERT(format.aspects & Aspect::Depth); + depthSelected = true; + break; + case wgpu::TextureAspect::StencilOnly: + ASSERT(format.aspects & Aspect::Stencil); + break; + default: + UNREACHABLE(); + } + + if (depthSelected) { + switch (format.format) { + case wgpu::TextureFormat::Depth24Plus: + case wgpu::TextureFormat::Depth24PlusStencil8: + return DAWN_VALIDATION_ERROR( + "The depth aspect of depth24plus texture cannot be selected in a " + "texture to buffer copy"); + break; + case wgpu::TextureFormat::Depth32Float: + break; + default: + UNREACHABLE(); + break; + } } return {}; @@ -442,15 +354,34 @@ namespace dawn_native { DAWN_TRY(ValidateStoreOp(depthStencilAttachment->depthStoreOp)); DAWN_TRY(ValidateStoreOp(depthStencilAttachment->stencilStoreOp)); - if (depthStencilAttachment->depthLoadOp == wgpu::LoadOp::Clear && - std::isnan(depthStencilAttachment->clearDepth)) { - return DAWN_VALIDATION_ERROR("Depth clear value cannot be NaN"); + if (attachment->GetAspect() == wgpu::TextureAspect::All && + attachment->GetFormat().HasStencil() && + depthStencilAttachment->depthReadOnly != depthStencilAttachment->stencilReadOnly) { + return DAWN_VALIDATION_ERROR( + "depthReadOnly and stencilReadOnly must be the same when texture aspect is " + "'all'"); } - // This validates that the depth storeOp and stencil storeOps are the same - if (depthStencilAttachment->depthStoreOp != depthStencilAttachment->stencilStoreOp) { + if (depthStencilAttachment->depthReadOnly && + (depthStencilAttachment->depthLoadOp != wgpu::LoadOp::Load || + depthStencilAttachment->depthStoreOp != wgpu::StoreOp::Store)) { return DAWN_VALIDATION_ERROR( - "The depth storeOp and stencil storeOp are not the same"); + "depthLoadOp must be load and depthStoreOp must be store when depthReadOnly " + "is true."); + } + + if (depthStencilAttachment->stencilReadOnly && + (depthStencilAttachment->stencilLoadOp != wgpu::LoadOp::Load || + depthStencilAttachment->stencilStoreOp != wgpu::StoreOp::Store)) { + return DAWN_VALIDATION_ERROR( + "stencilLoadOp must be load and stencilStoreOp must be store when " + "stencilReadOnly " + "is true."); + } + + if (depthStencilAttachment->depthLoadOp == wgpu::LoadOp::Clear && + std::isnan(depthStencilAttachment->clearDepth)) { + return DAWN_VALIDATION_ERROR("Depth clear value cannot be NaN"); } // *sampleCount == 0 must only happen when there is no color attachment. In that case we @@ -489,6 +420,10 @@ namespace dawn_native { device, descriptor->depthStencilAttachment, width, height, sampleCount)); } + if (descriptor->occlusionQuerySet != nullptr) { + return DAWN_VALIDATION_ERROR("occlusionQuerySet not implemented"); + } + if (descriptor->colorAttachmentCount == 0 && descriptor->depthStencilAttachment == nullptr) { return DAWN_VALIDATION_ERROR("Cannot use render pass with no attachments."); @@ -502,23 +437,42 @@ namespace dawn_native { return {}; } - ResultOrError<TextureCopyView> FixTextureCopyView(DeviceBase* device, - const TextureCopyView* view) { - TextureCopyView fixedView = *view; - - if (view->arrayLayer != 0) { - if (view->origin.z != 0) { - return DAWN_VALIDATION_ERROR("arrayLayer and origin.z cannot both be != 0"); - } else { - fixedView.origin.z = fixedView.arrayLayer; - fixedView.arrayLayer = 1; - device->EmitDeprecationWarning( - "wgpu::TextureCopyView::arrayLayer is deprecated in favor of " - "::origin::z"); - } + MaybeError ValidateQuerySetResolve(const QuerySetBase* querySet, + uint32_t firstQuery, + uint32_t queryCount, + const BufferBase* destination, + uint64_t destinationOffset) { + if (firstQuery >= querySet->GetQueryCount()) { + return DAWN_VALIDATION_ERROR("Query index out of bounds"); + } + + if (queryCount > querySet->GetQueryCount() - firstQuery) { + return DAWN_VALIDATION_ERROR( + "The sum of firstQuery and queryCount exceeds the number of queries in query " + "set"); + } + + // TODO(hao.x.li@intel.com): Validate that the queries between [firstQuery, firstQuery + + // queryCount - 1] must be available(written by query operations). + + // The destinationOffset must be a multiple of 8 bytes on D3D12 and Vulkan + if (destinationOffset % 8 != 0) { + return DAWN_VALIDATION_ERROR( + "The alignment offset into the destination buffer must be a multiple of 8 " + "bytes"); } - return fixedView; + uint64_t bufferSize = destination->GetSize(); + // The destination buffer must have enough storage, from destination offset, to contain + // the result of resolved queries + bool fitsInBuffer = destinationOffset <= bufferSize && + (static_cast<uint64_t>(queryCount) * sizeof(uint64_t) <= + (bufferSize - destinationOffset)); + if (!fitsInBuffer) { + return DAWN_VALIDATION_ERROR("The resolved query data would overflow the buffer"); + } + + return {}; } } // namespace @@ -529,14 +483,18 @@ namespace dawn_native { CommandBufferResourceUsage CommandEncoder::AcquireResourceUsages() { return CommandBufferResourceUsage{mEncodingContext.AcquirePassUsages(), - std::move(mTopLevelBuffers), - std::move(mTopLevelTextures)}; + std::move(mTopLevelBuffers), std::move(mTopLevelTextures), + std::move(mUsedQuerySets)}; } CommandIterator CommandEncoder::AcquireCommands() { return mEncodingContext.AcquireCommands(); } + void CommandEncoder::TrackUsedQuerySet(QuerySetBase* querySet) { + mUsedQuerySets.insert(querySet); + } + // Implementation of the API's command recording methods ComputePassEncoder* CommandEncoder::BeginComputePass(const ComputePassDescriptor* descriptor) { @@ -681,69 +639,56 @@ namespace dawn_native { const TextureCopyView* destination, const Extent3D* copySize) { mEncodingContext.TryEncode(this, [&](CommandAllocator* allocator) -> MaybeError { - // TODO(dawn:22): Remove once migration from GPUTextureCopyView.arrayLayer to - // GPUTextureCopyView.origin.z is done. - TextureCopyView fixedDest; - DAWN_TRY_ASSIGN(fixedDest, FixTextureCopyView(GetDevice(), destination)); - destination = &fixedDest; - - // Validate objects before doing the defaulting. if (GetDevice()->IsValidationEnabled()) { - DAWN_TRY(GetDevice()->ValidateObject(source->buffer)); - DAWN_TRY(GetDevice()->ValidateObject(destination->texture)); + DAWN_TRY(ValidateBufferCopyView(GetDevice(), *source)); + DAWN_TRY(ValidateCanUseAs(source->buffer, wgpu::BufferUsage::CopySrc)); + + DAWN_TRY(ValidateTextureCopyView(GetDevice(), *destination)); + DAWN_TRY(ValidateCanUseAs(destination->texture, wgpu::TextureUsage::CopyDst)); + DAWN_TRY(ValidateTextureSampleCountInCopyCommands(destination->texture)); + + // We validate texture copy range before validating linear texture data, + // because in the latter we divide copyExtent.width by blockWidth and + // copyExtent.height by blockHeight while the divisibility conditions are + // checked in validating texture copy range. + DAWN_TRY(ValidateTextureCopyRange(*destination, *copySize)); + DAWN_TRY(ValidateBufferToTextureCopyRestrictions(*destination)); + DAWN_TRY(ValidateLinearTextureData( + source->layout, source->buffer->GetSize(), + destination->texture->GetFormat().GetTexelBlockInfo(destination->aspect), + *copySize)); + + mTopLevelBuffers.insert(source->buffer); + mTopLevelTextures.insert(destination->texture); } // Compute default value for rowsPerImage - uint32_t defaultedRowsPerImage = source->rowsPerImage; + uint32_t defaultedRowsPerImage = source->layout.rowsPerImage; if (defaultedRowsPerImage == 0) { defaultedRowsPerImage = copySize->height; } - // Perform the rest of the validation using the default values. - if (GetDevice()->IsValidationEnabled()) { - DAWN_TRY(ValidateTextureSampleCountInCopyCommands(destination->texture)); - - DAWN_TRY(ValidateRowsPerImage(destination->texture->GetFormat(), - defaultedRowsPerImage, copySize->height)); - DAWN_TRY( - ValidateImageOrigin(destination->texture->GetFormat(), destination->origin)); - DAWN_TRY(ValidateImageCopySize(destination->texture->GetFormat(), *copySize)); - - uint32_t bufferCopySize = 0; - DAWN_TRY(ValidateBytesPerRow(destination->texture->GetFormat(), *copySize, - source->bytesPerRow)); - - DAWN_TRY(ComputeTextureCopyBufferSize(destination->texture->GetFormat(), *copySize, - source->bytesPerRow, defaultedRowsPerImage, - &bufferCopySize)); - - DAWN_TRY(ValidateCopySizeFitsInTexture(*destination, *copySize)); - DAWN_TRY(ValidateCopySizeFitsInBuffer(*source, bufferCopySize)); - DAWN_TRY(ValidateTexelBufferOffset(*source, destination->texture->GetFormat())); - - DAWN_TRY(ValidateCanUseAs(source->buffer, wgpu::BufferUsage::CopySrc)); - DAWN_TRY(ValidateCanUseAs(destination->texture, wgpu::TextureUsage::CopyDst)); - - mTopLevelBuffers.insert(source->buffer); - mTopLevelTextures.insert(destination->texture); + // In the case of one row copy bytesPerRow might not contain enough bytes + uint32_t bytesPerRow = source->layout.bytesPerRow; + if (copySize->height <= 1 && copySize->depth <= 1) { + bytesPerRow = + Align(copySize->width * destination->texture->GetFormat().blockByteSize, + kTextureBytesPerRowAlignment); } // Record the copy command. CopyBufferToTextureCmd* copy = allocator->Allocate<CopyBufferToTextureCmd>(Command::CopyBufferToTexture); copy->source.buffer = source->buffer; - copy->source.offset = source->offset; - copy->source.bytesPerRow = source->bytesPerRow; + copy->source.offset = source->layout.offset; + copy->source.bytesPerRow = bytesPerRow; copy->source.rowsPerImage = defaultedRowsPerImage; copy->destination.texture = destination->texture; copy->destination.origin = destination->origin; - copy->copySize = *copySize; copy->destination.mipLevel = destination->mipLevel; - copy->destination.arrayLayer = destination->arrayLayer; - - // TODO(cwallez@chromium.org): Make backends use origin.z instead of arrayLayer - copy->destination.arrayLayer = copy->destination.origin.z; - copy->destination.origin.z = 0; + copy->destination.aspect = + ConvertAspect(destination->texture->GetFormat(), destination->aspect); + copy->copySize = *copySize; return {}; }); @@ -753,49 +698,39 @@ namespace dawn_native { const BufferCopyView* destination, const Extent3D* copySize) { mEncodingContext.TryEncode(this, [&](CommandAllocator* allocator) -> MaybeError { - // TODO(dawn:22): Remove once migration from GPUTextureCopyView.arrayLayer to - // GPUTextureCopyView.origin.z is done. - TextureCopyView fixedSrc; - DAWN_TRY_ASSIGN(fixedSrc, FixTextureCopyView(GetDevice(), source)); - source = &fixedSrc; - - // Validate objects before doing the defaulting. if (GetDevice()->IsValidationEnabled()) { - DAWN_TRY(GetDevice()->ValidateObject(source->texture)); - DAWN_TRY(GetDevice()->ValidateObject(destination->buffer)); + DAWN_TRY(ValidateTextureCopyView(GetDevice(), *source)); + DAWN_TRY(ValidateCanUseAs(source->texture, wgpu::TextureUsage::CopySrc)); + DAWN_TRY(ValidateTextureSampleCountInCopyCommands(source->texture)); + + DAWN_TRY(ValidateBufferCopyView(GetDevice(), *destination)); + DAWN_TRY(ValidateCanUseAs(destination->buffer, wgpu::BufferUsage::CopyDst)); + + // We validate texture copy range before validating linear texture data, + // because in the latter we divide copyExtent.width by blockWidth and + // copyExtent.height by blockHeight while the divisibility conditions are + // checked in validating texture copy range. + DAWN_TRY(ValidateTextureCopyRange(*source, *copySize)); + DAWN_TRY(ValidateTextureToBufferCopyRestrictions(*source)); + DAWN_TRY(ValidateLinearTextureData( + destination->layout, destination->buffer->GetSize(), + source->texture->GetFormat().GetTexelBlockInfo(source->aspect), *copySize)); + + mTopLevelTextures.insert(source->texture); + mTopLevelBuffers.insert(destination->buffer); } // Compute default value for rowsPerImage - uint32_t defaultedRowsPerImage = destination->rowsPerImage; + uint32_t defaultedRowsPerImage = destination->layout.rowsPerImage; if (defaultedRowsPerImage == 0) { defaultedRowsPerImage = copySize->height; } - // Perform the rest of the validation using the default values. - if (GetDevice()->IsValidationEnabled()) { - DAWN_TRY(ValidateTextureSampleCountInCopyCommands(source->texture)); - - DAWN_TRY(ValidateRowsPerImage(source->texture->GetFormat(), defaultedRowsPerImage, - copySize->height)); - DAWN_TRY(ValidateImageOrigin(source->texture->GetFormat(), source->origin)); - DAWN_TRY(ValidateImageCopySize(source->texture->GetFormat(), *copySize)); - - uint32_t bufferCopySize = 0; - DAWN_TRY(ValidateBytesPerRow(source->texture->GetFormat(), *copySize, - destination->bytesPerRow)); - DAWN_TRY(ComputeTextureCopyBufferSize(source->texture->GetFormat(), *copySize, - destination->bytesPerRow, - defaultedRowsPerImage, &bufferCopySize)); - - DAWN_TRY(ValidateCopySizeFitsInTexture(*source, *copySize)); - DAWN_TRY(ValidateCopySizeFitsInBuffer(*destination, bufferCopySize)); - DAWN_TRY(ValidateTexelBufferOffset(*destination, source->texture->GetFormat())); - - DAWN_TRY(ValidateCanUseAs(source->texture, wgpu::TextureUsage::CopySrc)); - DAWN_TRY(ValidateCanUseAs(destination->buffer, wgpu::BufferUsage::CopyDst)); - - mTopLevelTextures.insert(source->texture); - mTopLevelBuffers.insert(destination->buffer); + // In the case of one row copy bytesPerRow might not contain enough bytes + uint32_t bytesPerRow = destination->layout.bytesPerRow; + if (copySize->height <= 1 && copySize->depth <= 1) { + bytesPerRow = Align(copySize->width * source->texture->GetFormat().blockByteSize, + kTextureBytesPerRowAlignment); } // Record the copy command. @@ -803,17 +738,13 @@ namespace dawn_native { allocator->Allocate<CopyTextureToBufferCmd>(Command::CopyTextureToBuffer); copy->source.texture = source->texture; copy->source.origin = source->origin; - copy->copySize = *copySize; copy->source.mipLevel = source->mipLevel; - copy->source.arrayLayer = source->arrayLayer; + copy->source.aspect = ConvertAspect(source->texture->GetFormat(), source->aspect); copy->destination.buffer = destination->buffer; - copy->destination.offset = destination->offset; - copy->destination.bytesPerRow = destination->bytesPerRow; + copy->destination.offset = destination->layout.offset; + copy->destination.bytesPerRow = bytesPerRow; copy->destination.rowsPerImage = defaultedRowsPerImage; - - // TODO(cwallez@chromium.org): Make backends use origin.z instead of arrayLayer - copy->source.arrayLayer = copy->source.origin.z; - copy->source.origin.z = 0; + copy->copySize = *copySize; return {}; }); @@ -823,15 +754,6 @@ namespace dawn_native { const TextureCopyView* destination, const Extent3D* copySize) { mEncodingContext.TryEncode(this, [&](CommandAllocator* allocator) -> MaybeError { - // TODO(dawn:22): Remove once migration from GPUTextureCopyView.arrayLayer to - // GPUTextureCopyView.origin.z is done. - TextureCopyView fixedSrc; - DAWN_TRY_ASSIGN(fixedSrc, FixTextureCopyView(GetDevice(), source)); - source = &fixedSrc; - TextureCopyView fixedDest; - DAWN_TRY_ASSIGN(fixedDest, FixTextureCopyView(GetDevice(), destination)); - destination = &fixedDest; - if (GetDevice()->IsValidationEnabled()) { DAWN_TRY(GetDevice()->ValidateObject(source->texture)); DAWN_TRY(GetDevice()->ValidateObject(destination->texture)); @@ -839,14 +761,11 @@ namespace dawn_native { DAWN_TRY( ValidateTextureToTextureCopyRestrictions(*source, *destination, *copySize)); - DAWN_TRY(ValidateImageOrigin(source->texture->GetFormat(), source->origin)); - DAWN_TRY(ValidateImageCopySize(source->texture->GetFormat(), *copySize)); - DAWN_TRY( - ValidateImageOrigin(destination->texture->GetFormat(), destination->origin)); - DAWN_TRY(ValidateImageCopySize(destination->texture->GetFormat(), *copySize)); + DAWN_TRY(ValidateTextureCopyRange(*source, *copySize)); + DAWN_TRY(ValidateTextureCopyRange(*destination, *copySize)); - DAWN_TRY(ValidateCopySizeFitsInTexture(*source, *copySize)); - DAWN_TRY(ValidateCopySizeFitsInTexture(*destination, *copySize)); + DAWN_TRY(ValidateTextureCopyView(GetDevice(), *source)); + DAWN_TRY(ValidateTextureCopyView(GetDevice(), *destination)); DAWN_TRY(ValidateCanUseAs(source->texture, wgpu::TextureUsage::CopySrc)); DAWN_TRY(ValidateCanUseAs(destination->texture, wgpu::TextureUsage::CopyDst)); @@ -860,19 +779,14 @@ namespace dawn_native { copy->source.texture = source->texture; copy->source.origin = source->origin; copy->source.mipLevel = source->mipLevel; - copy->source.arrayLayer = source->arrayLayer; + copy->source.aspect = ConvertAspect(source->texture->GetFormat(), source->aspect); copy->destination.texture = destination->texture; copy->destination.origin = destination->origin; copy->destination.mipLevel = destination->mipLevel; - copy->destination.arrayLayer = destination->arrayLayer; + copy->destination.aspect = + ConvertAspect(destination->texture->GetFormat(), destination->aspect); copy->copySize = *copySize; - // TODO(cwallez@chromium.org): Make backends use origin.z instead of arrayLayer - copy->source.arrayLayer = copy->source.origin.z; - copy->source.origin.z = 0; - copy->destination.arrayLayer = copy->destination.origin.z; - copy->destination.origin.z = 0; - return {}; }); } @@ -911,6 +825,54 @@ namespace dawn_native { }); } + void CommandEncoder::ResolveQuerySet(QuerySetBase* querySet, + uint32_t firstQuery, + uint32_t queryCount, + BufferBase* destination, + uint64_t destinationOffset) { + mEncodingContext.TryEncode(this, [&](CommandAllocator* allocator) -> MaybeError { + if (GetDevice()->IsValidationEnabled()) { + DAWN_TRY(GetDevice()->ValidateObject(querySet)); + DAWN_TRY(GetDevice()->ValidateObject(destination)); + + DAWN_TRY(ValidateQuerySetResolve(querySet, firstQuery, queryCount, destination, + destinationOffset)); + + DAWN_TRY(ValidateCanUseAs(destination, wgpu::BufferUsage::QueryResolve)); + + TrackUsedQuerySet(querySet); + mTopLevelBuffers.insert(destination); + } + + ResolveQuerySetCmd* cmd = + allocator->Allocate<ResolveQuerySetCmd>(Command::ResolveQuerySet); + cmd->querySet = querySet; + cmd->firstQuery = firstQuery; + cmd->queryCount = queryCount; + cmd->destination = destination; + cmd->destinationOffset = destinationOffset; + + return {}; + }); + } + + void CommandEncoder::WriteTimestamp(QuerySetBase* querySet, uint32_t queryIndex) { + mEncodingContext.TryEncode(this, [&](CommandAllocator* allocator) -> MaybeError { + if (GetDevice()->IsValidationEnabled()) { + DAWN_TRY(GetDevice()->ValidateObject(querySet)); + DAWN_TRY(ValidateTimestampQuery(querySet, queryIndex)); + TrackUsedQuerySet(querySet); + } + + WriteTimestampCmd* cmd = + allocator->Allocate<WriteTimestampCmd>(Command::WriteTimestamp); + cmd->querySet = querySet; + cmd->queryIndex = queryIndex; + + return {}; + }); + } + CommandBufferBase* CommandEncoder::Finish(const CommandBufferDescriptor* descriptor) { DeviceBase* device = GetDevice(); // Even if mEncodingContext.Finish() validation fails, calling it will mutate the internal @@ -994,6 +956,16 @@ namespace dawn_native { debugGroupStackSize++; break; } + + case Command::ResolveQuerySet: { + commands->NextCommand<ResolveQuerySetCmd>(); + break; + } + + case Command::WriteTimestamp: { + commands->NextCommand<WriteTimestampCmd>(); + break; + } default: return DAWN_VALIDATION_ERROR("Command disallowed outside of a pass"); } diff --git a/chromium/third_party/dawn/src/dawn_native/CommandEncoder.h b/chromium/third_party/dawn/src/dawn_native/CommandEncoder.h index 2c89c4bd777..2fe801042ce 100644 --- a/chromium/third_party/dawn/src/dawn_native/CommandEncoder.h +++ b/chromium/third_party/dawn/src/dawn_native/CommandEncoder.h @@ -35,6 +35,8 @@ namespace dawn_native { CommandIterator AcquireCommands(); CommandBufferResourceUsage AcquireResourceUsages(); + void TrackUsedQuerySet(QuerySetBase* querySet); + // Dawn API ComputePassEncoder* BeginComputePass(const ComputePassDescriptor* descriptor); RenderPassEncoder* BeginRenderPass(const RenderPassDescriptor* descriptor); @@ -58,6 +60,13 @@ namespace dawn_native { void PopDebugGroup(); void PushDebugGroup(const char* groupLabel); + void ResolveQuerySet(QuerySetBase* querySet, + uint32_t firstQuery, + uint32_t queryCount, + BufferBase* destination, + uint64_t destinationOffset); + void WriteTimestamp(QuerySetBase* querySet, uint32_t queryIndex); + CommandBufferBase* Finish(const CommandBufferDescriptor* descriptor); private: @@ -67,6 +76,7 @@ namespace dawn_native { EncodingContext mEncodingContext; std::set<BufferBase*> mTopLevelBuffers; std::set<TextureBase*> mTopLevelTextures; + std::set<QuerySetBase*> mUsedQuerySets; }; } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/CommandValidation.cpp b/chromium/third_party/dawn/src/dawn_native/CommandValidation.cpp index 7f8da9bcdf2..497a8dee0de 100644 --- a/chromium/third_party/dawn/src/dawn_native/CommandValidation.cpp +++ b/chromium/third_party/dawn/src/dawn_native/CommandValidation.cpp @@ -19,7 +19,9 @@ #include "dawn_native/Buffer.h" #include "dawn_native/CommandBufferStateTracker.h" #include "dawn_native/Commands.h" +#include "dawn_native/Device.h" #include "dawn_native/PassResourceUsage.h" +#include "dawn_native/QuerySet.h" #include "dawn_native/RenderBundle.h" #include "dawn_native/RenderPipeline.h" @@ -202,6 +204,11 @@ namespace dawn_native { break; } + case Command::WriteTimestamp: { + commands->NextCommand<WriteTimestampCmd>(); + break; + } + default: DAWN_TRY(ValidateRenderBundleCommand( commands, type, &commandBufferState, renderPass->attachmentState.Get(), @@ -274,6 +281,11 @@ namespace dawn_native { break; } + case Command::WriteTimestamp: { + commands->NextCommand<WriteTimestampCmd>(); + break; + } + default: return DAWN_VALIDATION_ERROR("Command disallowed inside a compute pass"); } @@ -339,6 +351,18 @@ namespace dawn_native { return {}; } + MaybeError ValidateTimestampQuery(QuerySetBase* querySet, uint32_t queryIndex) { + if (querySet->GetQueryType() != wgpu::QueryType::Timestamp) { + return DAWN_VALIDATION_ERROR("The query type of query set must be Timestamp"); + } + + if (queryIndex >= querySet->GetQueryCount()) { + return DAWN_VALIDATION_ERROR("Query index exceeds the number of queries in query set"); + } + + return {}; + } + bool IsRangeOverlapped(uint32_t startA, uint32_t startB, uint32_t length) { uint32_t maxStart = std::max(startA, startB); uint32_t minStart = std::min(startA, startB); @@ -346,4 +370,226 @@ namespace dawn_native { static_cast<uint64_t>(maxStart); } + ResultOrError<uint64_t> ComputeRequiredBytesInCopy(const TexelBlockInfo& blockInfo, + const Extent3D& copySize, + uint32_t bytesPerRow, + uint32_t rowsPerImage) { + // Default value for rowsPerImage + if (rowsPerImage == 0) { + rowsPerImage = copySize.height; + } + + ASSERT(rowsPerImage >= copySize.height); + if (copySize.height > 1 || copySize.depth > 1) { + ASSERT(bytesPerRow >= copySize.width / blockInfo.blockWidth * blockInfo.blockByteSize); + } + + if (copySize.width == 0 || copySize.height == 0 || copySize.depth == 0) { + return 0; + } + + ASSERT(copySize.height >= 1); + ASSERT(copySize.depth >= 1); + + uint32_t texelBlockRowsPerImage = rowsPerImage / blockInfo.blockHeight; + // bytesPerImage won't overflow since we're multiplying two uint32_t numbers + uint64_t bytesPerImage = uint64_t(texelBlockRowsPerImage) * bytesPerRow; + // Provided that copySize.height > 1: bytesInLastSlice won't overflow since it's at most + // bytesPerImage. Otherwise the result is a multiplication of two uint32_t numbers. + uint64_t bytesInLastSlice = + uint64_t(bytesPerRow) * (copySize.height / blockInfo.blockHeight - 1) + + (uint64_t(copySize.width) / blockInfo.blockWidth * blockInfo.blockByteSize); + + // This error cannot be thrown for copySize.depth = 1. + // For copySize.depth > 1 we know that: + // requiredBytesInCopy >= (copySize.depth * bytesPerImage) / 2, so if + // copySize.depth * bytesPerImage overflows uint64_t, then requiredBytesInCopy is definitely + // too large to fit in the available data size. + if (std::numeric_limits<uint64_t>::max() / copySize.depth < bytesPerImage) { + return DAWN_VALIDATION_ERROR("requiredBytesInCopy is too large"); + } + return bytesPerImage * (copySize.depth - 1) + bytesInLastSlice; + } + + MaybeError ValidateCopySizeFitsInBuffer(const Ref<BufferBase>& buffer, + uint64_t offset, + uint64_t size) { + uint64_t bufferSize = buffer->GetSize(); + bool fitsInBuffer = offset <= bufferSize && (size <= (bufferSize - offset)); + if (!fitsInBuffer) { + return DAWN_VALIDATION_ERROR("Copy would overflow the buffer"); + } + + return {}; + } + + MaybeError ValidateLinearTextureData(const TextureDataLayout& layout, + uint64_t byteSize, + const TexelBlockInfo& blockInfo, + const Extent3D& copyExtent) { + // Validation for the texel block alignments: + if (layout.rowsPerImage % blockInfo.blockHeight != 0) { + return DAWN_VALIDATION_ERROR( + "rowsPerImage must be a multiple of compressed texture format block height"); + } + + if (layout.offset % blockInfo.blockByteSize != 0) { + return DAWN_VALIDATION_ERROR("Offset must be a multiple of the texel or block size"); + } + + // Validation for other members in layout: + if ((copyExtent.height > 1 || copyExtent.depth > 1) && + layout.bytesPerRow < + copyExtent.width / blockInfo.blockWidth * blockInfo.blockByteSize) { + return DAWN_VALIDATION_ERROR( + "bytesPerRow must not be less than the number of bytes per row"); + } + + // TODO(tommek@google.com): to match the spec there should be another condition here + // on rowsPerImage >= copyExtent.height if copyExtent.depth > 1. + + // Validation for the copy being in-bounds: + if (layout.rowsPerImage != 0 && layout.rowsPerImage < copyExtent.height) { + return DAWN_VALIDATION_ERROR("rowsPerImage must not be less than the copy height."); + } + + // We compute required bytes in copy after validating texel block alignments + // because the divisibility conditions are necessary for the algorithm to be valid, + // also the bytesPerRow bound is necessary to avoid overflows. + uint64_t requiredBytesInCopy; + DAWN_TRY_ASSIGN(requiredBytesInCopy, + ComputeRequiredBytesInCopy(blockInfo, copyExtent, layout.bytesPerRow, + layout.rowsPerImage)); + + bool fitsInData = + layout.offset <= byteSize && (requiredBytesInCopy <= (byteSize - layout.offset)); + if (!fitsInData) { + return DAWN_VALIDATION_ERROR( + "Required size for texture data layout exceeds the given size"); + } + + return {}; + } + + MaybeError ValidateBufferCopyView(DeviceBase const* device, + const BufferCopyView& bufferCopyView) { + DAWN_TRY(device->ValidateObject(bufferCopyView.buffer)); + if (bufferCopyView.layout.bytesPerRow % kTextureBytesPerRowAlignment != 0) { + return DAWN_VALIDATION_ERROR("bytesPerRow must be a multiple of 256"); + } + + return {}; + } + + MaybeError ValidateTextureCopyView(DeviceBase const* device, + const TextureCopyView& textureCopy) { + DAWN_TRY(device->ValidateObject(textureCopy.texture)); + if (textureCopy.mipLevel >= textureCopy.texture->GetNumMipLevels()) { + return DAWN_VALIDATION_ERROR("mipLevel out of range"); + } + + if (textureCopy.origin.x % textureCopy.texture->GetFormat().blockWidth != 0) { + return DAWN_VALIDATION_ERROR( + "Offset.x must be a multiple of compressed texture format block width"); + } + + if (textureCopy.origin.y % textureCopy.texture->GetFormat().blockHeight != 0) { + return DAWN_VALIDATION_ERROR( + "Offset.y must be a multiple of compressed texture format block height"); + } + + switch (textureCopy.aspect) { + case wgpu::TextureAspect::All: + break; + case wgpu::TextureAspect::DepthOnly: + if ((textureCopy.texture->GetFormat().aspects & Aspect::Depth) == 0) { + return DAWN_VALIDATION_ERROR( + "Texture does not have depth aspect for texture copy"); + } + break; + case wgpu::TextureAspect::StencilOnly: + if ((textureCopy.texture->GetFormat().aspects & Aspect::Stencil) == 0) { + return DAWN_VALIDATION_ERROR( + "Texture does not have stencil aspect for texture copy"); + } + break; + default: + UNREACHABLE(); + break; + } + + return {}; + } + + MaybeError ValidateTextureCopyRange(const TextureCopyView& textureCopy, + const Extent3D& copySize) { + // TODO(jiawei.shao@intel.com): add validations on the texture-to-texture copies within the + // same texture. + const TextureBase* texture = textureCopy.texture; + + // Validation for the copy being in-bounds: + Extent3D mipSize = texture->GetMipLevelPhysicalSize(textureCopy.mipLevel); + // For 2D textures, include the array layer as depth so it can be checked with other + // dimensions. + ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D); + mipSize.depth = texture->GetArrayLayers(); + + // All texture dimensions are in uint32_t so by doing checks in uint64_t we avoid + // overflows. + if (static_cast<uint64_t>(textureCopy.origin.x) + static_cast<uint64_t>(copySize.width) > + static_cast<uint64_t>(mipSize.width) || + static_cast<uint64_t>(textureCopy.origin.y) + static_cast<uint64_t>(copySize.height) > + static_cast<uint64_t>(mipSize.height) || + static_cast<uint64_t>(textureCopy.origin.z) + static_cast<uint64_t>(copySize.depth) > + static_cast<uint64_t>(mipSize.depth)) { + return DAWN_VALIDATION_ERROR("Touching outside of the texture"); + } + + // Validation for the texel block alignments: + if (copySize.width % textureCopy.texture->GetFormat().blockWidth != 0) { + return DAWN_VALIDATION_ERROR( + "copySize.width must be a multiple of compressed texture format block width"); + } + + if (copySize.height % textureCopy.texture->GetFormat().blockHeight != 0) { + return DAWN_VALIDATION_ERROR( + "copySize.height must be a multiple of compressed texture format block height"); + } + + return {}; + } + + MaybeError ValidateBufferToTextureCopyRestrictions(const TextureCopyView& dst) { + const Format& format = dst.texture->GetFormat(); + + bool depthSelected = false; + switch (dst.aspect) { + case wgpu::TextureAspect::All: + switch (format.aspects) { + case Aspect::Color: + case Aspect::Stencil: + break; + case Aspect::Depth: + depthSelected = true; + break; + default: + return DAWN_VALIDATION_ERROR( + "A single aspect must be selected for multi planar formats in buffer " + "to texture copies"); + } + break; + case wgpu::TextureAspect::DepthOnly: + ASSERT(format.aspects & Aspect::Depth); + depthSelected = true; + break; + case wgpu::TextureAspect::StencilOnly: + ASSERT(format.aspects & Aspect::Stencil); + break; + } + if (depthSelected) { + return DAWN_VALIDATION_ERROR("Cannot copy into the depth aspect of a texture"); + } + return {}; + } + } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/CommandValidation.h b/chromium/third_party/dawn/src/dawn_native/CommandValidation.h index 53871ccdc5e..0aaa398ade0 100644 --- a/chromium/third_party/dawn/src/dawn_native/CommandValidation.h +++ b/chromium/third_party/dawn/src/dawn_native/CommandValidation.h @@ -17,14 +17,17 @@ #include "dawn_native/CommandAllocator.h" #include "dawn_native/Error.h" +#include "dawn_native/Texture.h" #include <vector> namespace dawn_native { class AttachmentState; + class QuerySetBase; struct BeginRenderPassCmd; struct PassResourceUsage; + struct TexelBlockInfo; MaybeError ValidateCanPopDebugGroup(uint64_t debugGroupStackSize); MaybeError ValidateFinalDebugGroupStackSize(uint64_t debugGroupStackSize); @@ -36,6 +39,36 @@ namespace dawn_native { MaybeError ValidatePassResourceUsage(const PassResourceUsage& usage); + MaybeError ValidateTimestampQuery(QuerySetBase* querySet, uint32_t queryIndex); + + ResultOrError<uint64_t> ComputeRequiredBytesInCopy(const TexelBlockInfo& blockInfo, + const Extent3D& copySize, + uint32_t bytesPerRow, + uint32_t rowsPerImage); + + MaybeError ValidateLinearTextureData(const TextureDataLayout& layout, + uint64_t byteSize, + const TexelBlockInfo& blockInfo, + const Extent3D& copyExtent); + MaybeError ValidateTextureCopyRange(const TextureCopyView& textureCopyView, + const Extent3D& copySize); + MaybeError ValidateBufferToTextureCopyRestrictions(const TextureCopyView& dst); + + MaybeError ValidateBufferCopyView(DeviceBase const* device, + const BufferCopyView& bufferCopyView); + MaybeError ValidateTextureCopyView(DeviceBase const* device, + const TextureCopyView& textureCopyView); + + MaybeError ValidateRowsPerImage(const Format& format, + uint32_t rowsPerImage, + uint32_t copyHeight); + MaybeError ValidateBytesPerRow(const Format& format, + const Extent3D& copySize, + uint32_t bytesPerRow); + MaybeError ValidateCopySizeFitsInBuffer(const Ref<BufferBase>& buffer, + uint64_t offset, + uint64_t size); + bool IsRangeOverlapped(uint32_t startA, uint32_t startB, uint32_t length); } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/Commands.cpp b/chromium/third_party/dawn/src/dawn_native/Commands.cpp index b4098108a32..621a7434abe 100644 --- a/chromium/third_party/dawn/src/dawn_native/Commands.cpp +++ b/chromium/third_party/dawn/src/dawn_native/Commands.cpp @@ -18,6 +18,7 @@ #include "dawn_native/Buffer.h" #include "dawn_native/CommandAllocator.h" #include "dawn_native/ComputePipeline.h" +#include "dawn_native/QuerySet.h" #include "dawn_native/RenderBundle.h" #include "dawn_native/RenderPipeline.h" #include "dawn_native/Texture.h" @@ -127,6 +128,11 @@ namespace dawn_native { cmd->~PushDebugGroupCmd(); break; } + case Command::ResolveQuerySet: { + ResolveQuerySetCmd* cmd = commands->NextCommand<ResolveQuerySetCmd>(); + cmd->~ResolveQuerySetCmd(); + break; + } case Command::SetComputePipeline: { SetComputePipelineCmd* cmd = commands->NextCommand<SetComputePipelineCmd>(); cmd->~SetComputePipelineCmd(); @@ -175,9 +181,15 @@ namespace dawn_native { cmd->~SetVertexBufferCmd(); break; } + case Command::WriteTimestamp: { + WriteTimestampCmd* cmd = commands->NextCommand<WriteTimestampCmd>(); + cmd->~WriteTimestampCmd(); + break; + } } } - commands->DataWasDestroyed(); + + commands->MakeEmptyAsDataWasDestroyed(); } void SkipCommand(CommandIterator* commands, Command type) { @@ -260,6 +272,11 @@ namespace dawn_native { break; } + case Command::ResolveQuerySet: { + commands->NextCommand<ResolveQuerySetCmd>(); + break; + } + case Command::SetComputePipeline: commands->NextCommand<SetComputePipelineCmd>(); break; @@ -300,6 +317,11 @@ namespace dawn_native { commands->NextCommand<SetVertexBufferCmd>(); break; } + + case Command::WriteTimestamp: { + commands->NextCommand<WriteTimestampCmd>(); + break; + } } } diff --git a/chromium/third_party/dawn/src/dawn_native/Commands.h b/chromium/third_party/dawn/src/dawn_native/Commands.h index 155a21d83be..094867839fd 100644 --- a/chromium/third_party/dawn/src/dawn_native/Commands.h +++ b/chromium/third_party/dawn/src/dawn_native/Commands.h @@ -51,6 +51,7 @@ namespace dawn_native { InsertDebugMarker, PopDebugGroup, PushDebugGroup, + ResolveQuerySet, SetComputePipeline, SetRenderPipeline, SetStencilReference, @@ -60,6 +61,7 @@ namespace dawn_native { SetBindGroup, SetIndexBuffer, SetVertexBuffer, + WriteTimestamp, }; struct BeginComputePassCmd {}; @@ -102,8 +104,8 @@ namespace dawn_native { struct TextureCopy { Ref<TextureBase> texture; uint32_t mipLevel; - uint32_t arrayLayer; - Origin3D origin; // Texels + Origin3D origin; // Texels / array layer + Aspect aspect; }; struct CopyBufferToBufferCmd { @@ -186,6 +188,14 @@ namespace dawn_native { uint32_t length; }; + struct ResolveQuerySetCmd { + Ref<QuerySetBase> querySet; + uint32_t firstQuery; + uint32_t queryCount; + Ref<BufferBase> destination; + uint64_t destinationOffset; + }; + struct SetComputePipelineCmd { Ref<ComputePipelineBase> pipeline; }; @@ -229,6 +239,11 @@ namespace dawn_native { uint64_t size; }; + struct WriteTimestampCmd { + Ref<QuerySetBase> querySet; + uint32_t queryIndex; + }; + // This needs to be called before the CommandIterator is freed so that the Ref<> present in // the commands have a chance to run their destructor and remove internal references. class CommandIterator; diff --git a/chromium/third_party/dawn/src/dawn_native/ComputePassEncoder.cpp b/chromium/third_party/dawn/src/dawn_native/ComputePassEncoder.cpp index 5e36601fcc5..2329c0008e2 100644 --- a/chromium/third_party/dawn/src/dawn_native/ComputePassEncoder.cpp +++ b/chromium/third_party/dawn/src/dawn_native/ComputePassEncoder.cpp @@ -16,9 +16,11 @@ #include "dawn_native/Buffer.h" #include "dawn_native/CommandEncoder.h" +#include "dawn_native/CommandValidation.h" #include "dawn_native/Commands.h" #include "dawn_native/ComputePipeline.h" #include "dawn_native/Device.h" +#include "dawn_native/QuerySet.h" namespace dawn_native { @@ -96,4 +98,21 @@ namespace dawn_native { }); } + void ComputePassEncoder::WriteTimestamp(QuerySetBase* querySet, uint32_t queryIndex) { + mEncodingContext->TryEncode(this, [&](CommandAllocator* allocator) -> MaybeError { + if (GetDevice()->IsValidationEnabled()) { + DAWN_TRY(GetDevice()->ValidateObject(querySet)); + DAWN_TRY(ValidateTimestampQuery(querySet, queryIndex)); + mCommandEncoder->TrackUsedQuerySet(querySet); + } + + WriteTimestampCmd* cmd = + allocator->Allocate<WriteTimestampCmd>(Command::WriteTimestamp); + cmd->querySet = querySet; + cmd->queryIndex = queryIndex; + + return {}; + }); + } + } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/ComputePassEncoder.h b/chromium/third_party/dawn/src/dawn_native/ComputePassEncoder.h index f790aad51ee..6ae796a2411 100644 --- a/chromium/third_party/dawn/src/dawn_native/ComputePassEncoder.h +++ b/chromium/third_party/dawn/src/dawn_native/ComputePassEncoder.h @@ -36,6 +36,8 @@ namespace dawn_native { void DispatchIndirect(BufferBase* indirectBuffer, uint64_t indirectOffset); void SetPipeline(ComputePipelineBase* pipeline); + void WriteTimestamp(QuerySetBase* querySet, uint32_t queryIndex); + protected: ComputePassEncoder(DeviceBase* device, CommandEncoder* commandEncoder, diff --git a/chromium/third_party/dawn/src/dawn_native/DawnNative.cpp b/chromium/third_party/dawn/src/dawn_native/DawnNative.cpp index 32061152bdf..b7bb94b32d1 100644 --- a/chromium/third_party/dawn/src/dawn_native/DawnNative.cpp +++ b/chromium/third_party/dawn/src/dawn_native/DawnNative.cpp @@ -149,6 +149,10 @@ namespace dawn_native { mImpl->EnableBackendValidation(enableBackendValidation); } + void Instance::EnableGPUBasedBackendValidation(bool enableGPUBasedBackendValidation) { + mImpl->EnableGPUBasedBackendValidation(enableGPUBasedBackendValidation); + } + void Instance::EnableBeginCaptureOnStartup(bool beginCaptureOnStartup) { mImpl->EnableBeginCaptureOnStartup(beginCaptureOnStartup); } @@ -175,10 +179,13 @@ namespace dawn_native { uint32_t baseMipLevel, uint32_t levelCount, uint32_t baseArrayLayer, - uint32_t layerCount) { + uint32_t layerCount, + WGPUTextureAspect aspect) { dawn_native::TextureBase* textureBase = reinterpret_cast<dawn_native::TextureBase*>(texture); - SubresourceRange range = {baseMipLevel, levelCount, baseArrayLayer, layerCount}; + SubresourceRange range = { + baseMipLevel, levelCount, baseArrayLayer, layerCount, + ConvertAspect(textureBase->GetFormat(), static_cast<wgpu::TextureAspect>(aspect))}; return textureBase->IsSubresourceContentInitialized(range); } diff --git a/chromium/third_party/dawn/src/dawn_native/Device.cpp b/chromium/third_party/dawn/src/dawn_native/Device.cpp index 1c30cd509c8..fcb2876ac9b 100644 --- a/chromium/third_party/dawn/src/dawn_native/Device.cpp +++ b/chromium/third_party/dawn/src/dawn_native/Device.cpp @@ -112,6 +112,8 @@ namespace dawn_native { // alive. mState = State::Alive; + DAWN_TRY_ASSIGN(mEmptyBindGroupLayout, CreateEmptyBindGroupLayout()); + return {}; } @@ -169,6 +171,8 @@ namespace dawn_native { mDynamicUploader = nullptr; mMapRequestTracker = nullptr; + mEmptyBindGroupLayout = nullptr; + AssumeCommandsComplete(); // Tell the backend that it can free all the objects now that the GPU timeline is empty. ShutDownImpl(); @@ -388,21 +392,22 @@ namespace dawn_native { return mFormatTable[index]; } - ResultOrError<BindGroupLayoutBase*> DeviceBase::GetOrCreateBindGroupLayout( + ResultOrError<Ref<BindGroupLayoutBase>> DeviceBase::GetOrCreateBindGroupLayout( const BindGroupLayoutDescriptor* descriptor) { BindGroupLayoutBase blueprint(this, descriptor); + Ref<BindGroupLayoutBase> result = nullptr; auto iter = mCaches->bindGroupLayouts.find(&blueprint); if (iter != mCaches->bindGroupLayouts.end()) { - (*iter)->Reference(); - return *iter; + result = *iter; + } else { + BindGroupLayoutBase* backendObj; + DAWN_TRY_ASSIGN(backendObj, CreateBindGroupLayoutImpl(descriptor)); + backendObj->SetIsCachedReference(); + mCaches->bindGroupLayouts.insert(backendObj); + result = AcquireRef(backendObj); } - - BindGroupLayoutBase* backendObj; - DAWN_TRY_ASSIGN(backendObj, CreateBindGroupLayoutImpl(descriptor)); - backendObj->SetIsCachedReference(); - mCaches->bindGroupLayouts.insert(backendObj); - return backendObj; + return std::move(result); } void DeviceBase::UncacheBindGroupLayout(BindGroupLayoutBase* obj) { @@ -411,6 +416,20 @@ namespace dawn_native { ASSERT(removedCount == 1); } + // Private function used at initialization + ResultOrError<Ref<BindGroupLayoutBase>> DeviceBase::CreateEmptyBindGroupLayout() { + BindGroupLayoutDescriptor desc = {}; + desc.entryCount = 0; + desc.entries = nullptr; + + return GetOrCreateBindGroupLayout(&desc); + } + + BindGroupLayoutBase* DeviceBase::GetEmptyBindGroupLayout() { + ASSERT(mEmptyBindGroupLayout); + return mEmptyBindGroupLayout.Get(); + } + ResultOrError<ComputePipelineBase*> DeviceBase::GetOrCreateComputePipeline( const ComputePipelineDescriptor* descriptor) { ComputePipelineBase blueprint(this, descriptor); @@ -585,43 +604,32 @@ namespace dawn_native { return result; } BufferBase* DeviceBase::CreateBuffer(const BufferDescriptor* descriptor) { - BufferBase* result = nullptr; + Ref<BufferBase> result = nullptr; if (ConsumedError(CreateBufferInternal(descriptor), &result)) { - ASSERT(result == nullptr); - return BufferBase::MakeError(this); + ASSERT(result.Get() == nullptr); + return BufferBase::MakeError(this, descriptor); } - return result; + return result.Detach(); } WGPUCreateBufferMappedResult DeviceBase::CreateBufferMapped( const BufferDescriptor* descriptor) { - BufferBase* buffer = nullptr; - uint8_t* data = nullptr; - - uint64_t size = descriptor->size; - if (ConsumedError(CreateBufferInternal(descriptor), &buffer) || - ConsumedError(buffer->MapAtCreation(&data))) { - // Map failed. Replace the buffer with an error buffer. - if (buffer != nullptr) { - buffer->Release(); - } - buffer = BufferBase::MakeErrorMapped(this, size, &data); - } + EmitDeprecationWarning( + "CreateBufferMapped is deprecated, use wgpu::BufferDescriptor::mappedAtCreation and " + "wgpu::Buffer::GetMappedRange instead"); - ASSERT(buffer != nullptr); - if (data == nullptr) { - // |data| may be nullptr if there was an OOM in MakeErrorMapped. - // Non-zero dataLength and nullptr data is used to indicate there should be - // mapped data but the allocation failed. - ASSERT(buffer->IsError()); - } else { - memset(data, 0, size); - } + BufferDescriptor fixedDesc = *descriptor; + fixedDesc.mappedAtCreation = true; + BufferBase* buffer = CreateBuffer(&fixedDesc); WGPUCreateBufferMappedResult result = {}; result.buffer = reinterpret_cast<WGPUBuffer>(buffer); - result.data = data; - result.dataLength = size; + result.data = buffer->GetMappedRange(0, descriptor->size); + result.dataLength = descriptor->size; + + if (result.data != nullptr) { + memset(result.data, 0, result.dataLength); + } return result; } @@ -734,7 +742,8 @@ namespace dawn_native { // For Dawn Wire BufferBase* DeviceBase::CreateErrorBuffer() { - return BufferBase::MakeError(this); + BufferDescriptor desc = {}; + return BufferBase::MakeError(this, &desc); } // Other Device API methods @@ -813,6 +822,10 @@ namespace dawn_native { return !IsToggleEnabled(Toggle::SkipValidation); } + bool DeviceBase::IsRobustnessEnabled() const { + return !IsToggleEnabled(Toggle::DisableRobustness); + } + size_t DeviceBase::GetLazyClearCountForTesting() { return mLazyClearCountForTesting; } @@ -851,17 +864,27 @@ namespace dawn_native { if (IsValidationEnabled()) { DAWN_TRY(ValidateBindGroupLayoutDescriptor(this, descriptor)); } - DAWN_TRY_ASSIGN(*result, GetOrCreateBindGroupLayout(descriptor)); + Ref<BindGroupLayoutBase> bgl; + DAWN_TRY_ASSIGN(bgl, GetOrCreateBindGroupLayout(descriptor)); + *result = bgl.Detach(); return {}; } - ResultOrError<BufferBase*> DeviceBase::CreateBufferInternal( + ResultOrError<Ref<BufferBase>> DeviceBase::CreateBufferInternal( const BufferDescriptor* descriptor) { DAWN_TRY(ValidateIsAlive()); if (IsValidationEnabled()) { DAWN_TRY(ValidateBufferDescriptor(this, descriptor)); } - return CreateBufferImpl(descriptor); + + Ref<BufferBase> buffer; + DAWN_TRY_ASSIGN(buffer, CreateBufferImpl(descriptor)); + + if (descriptor->mappedAtCreation) { + DAWN_TRY(buffer->MapAtCreation()); + } + + return std::move(buffer); } MaybeError DeviceBase::CreateComputePipelineInternal( @@ -1007,13 +1030,6 @@ namespace dawn_native { ResultOrError<Ref<TextureBase>> DeviceBase::CreateTextureInternal( const TextureDescriptor* descriptor) { DAWN_TRY(ValidateIsAlive()); - - // TODO(dawn:22): Remove once migration from GPUTextureDescriptor.arrayLayerCount to - // GPUTextureDescriptor.size.depth is done. - TextureDescriptor fixedDescriptor; - DAWN_TRY_ASSIGN(fixedDescriptor, FixTextureDescriptor(this, descriptor)); - descriptor = &fixedDescriptor; - if (IsValidationEnabled()) { DAWN_TRY(ValidateTextureDescriptor(this, descriptor)); } diff --git a/chromium/third_party/dawn/src/dawn_native/Device.h b/chromium/third_party/dawn/src/dawn_native/Device.h index 94d916f7c2b..16f076c3343 100644 --- a/chromium/third_party/dawn/src/dawn_native/Device.h +++ b/chromium/third_party/dawn/src/dawn_native/Device.h @@ -108,10 +108,12 @@ namespace dawn_native { // the created object will be, the "blueprint". The blueprint is just a FooBase object // instead of a backend Foo object. If the blueprint doesn't match an object in the // cache, then the descriptor is used to make a new object. - ResultOrError<BindGroupLayoutBase*> GetOrCreateBindGroupLayout( + ResultOrError<Ref<BindGroupLayoutBase>> GetOrCreateBindGroupLayout( const BindGroupLayoutDescriptor* descriptor); void UncacheBindGroupLayout(BindGroupLayoutBase* obj); + BindGroupLayoutBase* GetEmptyBindGroupLayout(); + ResultOrError<ComputePipelineBase*> GetOrCreateComputePipeline( const ComputePipelineDescriptor* descriptor); void UncacheComputePipeline(ComputePipelineBase* obj); @@ -214,6 +216,7 @@ namespace dawn_native { bool IsExtensionEnabled(Extension extension) const; bool IsToggleEnabled(Toggle toggle) const; bool IsValidationEnabled() const; + bool IsRobustnessEnabled() const; size_t GetLazyClearCountForTesting(); void IncrementLazyClearCountForTesting(); size_t GetDeprecationWarningCountForTesting(); @@ -238,7 +241,8 @@ namespace dawn_native { const BindGroupDescriptor* descriptor) = 0; virtual ResultOrError<BindGroupLayoutBase*> CreateBindGroupLayoutImpl( const BindGroupLayoutDescriptor* descriptor) = 0; - virtual ResultOrError<BufferBase*> CreateBufferImpl(const BufferDescriptor* descriptor) = 0; + virtual ResultOrError<Ref<BufferBase>> CreateBufferImpl( + const BufferDescriptor* descriptor) = 0; virtual ResultOrError<ComputePipelineBase*> CreateComputePipelineImpl( const ComputePipelineDescriptor* descriptor) = 0; virtual ResultOrError<PipelineLayoutBase*> CreatePipelineLayoutImpl( @@ -264,11 +268,13 @@ namespace dawn_native { TextureBase* texture, const TextureViewDescriptor* descriptor) = 0; + ResultOrError<Ref<BindGroupLayoutBase>> CreateEmptyBindGroupLayout(); + MaybeError CreateBindGroupInternal(BindGroupBase** result, const BindGroupDescriptor* descriptor); MaybeError CreateBindGroupLayoutInternal(BindGroupLayoutBase** result, const BindGroupLayoutDescriptor* descriptor); - ResultOrError<BufferBase*> CreateBufferInternal(const BufferDescriptor* descriptor); + ResultOrError<Ref<BufferBase>> CreateBufferInternal(const BufferDescriptor* descriptor); MaybeError CreateComputePipelineInternal(ComputePipelineBase** result, const ComputePipelineDescriptor* descriptor); MaybeError CreatePipelineLayoutInternal(PipelineLayoutBase** result, @@ -340,6 +346,8 @@ namespace dawn_native { struct Caches; std::unique_ptr<Caches> mCaches; + Ref<BindGroupLayoutBase> mEmptyBindGroupLayout; + std::unique_ptr<DynamicUploader> mDynamicUploader; std::unique_ptr<ErrorScopeTracker> mErrorScopeTracker; std::unique_ptr<FenceSignalTracker> mFenceSignalTracker; diff --git a/chromium/third_party/dawn/src/dawn_native/DynamicUploader.cpp b/chromium/third_party/dawn/src/dawn_native/DynamicUploader.cpp index be92ea09591..6fb8eef6851 100644 --- a/chromium/third_party/dawn/src/dawn_native/DynamicUploader.cpp +++ b/chromium/third_party/dawn/src/dawn_native/DynamicUploader.cpp @@ -28,7 +28,8 @@ namespace dawn_native { mDevice->GetPendingCommandSerial()); } - ResultOrError<UploadHandle> DynamicUploader::Allocate(uint64_t allocationSize, Serial serial) { + ResultOrError<UploadHandle> DynamicUploader::AllocateInternal(uint64_t allocationSize, + Serial serial) { // Disable further sub-allocation should the request be too large. if (allocationSize > kRingBufferSize) { std::unique_ptr<StagingBufferBase> stagingBuffer; @@ -108,4 +109,21 @@ namespace dawn_native { } mReleasedStagingBuffers.ClearUpTo(lastCompletedSerial); } -} // namespace dawn_native
\ No newline at end of file + + // TODO(dawn:512): Optimize this function so that it doesn't allocate additional memory + // when it's not necessary. + ResultOrError<UploadHandle> DynamicUploader::Allocate(uint64_t allocationSize, + Serial serial, + uint64_t offsetAlignment) { + ASSERT(offsetAlignment > 0); + UploadHandle uploadHandle; + DAWN_TRY_ASSIGN(uploadHandle, + AllocateInternal(allocationSize + offsetAlignment - 1, serial)); + uint64_t additionalOffset = + Align(uploadHandle.startOffset, offsetAlignment) - uploadHandle.startOffset; + uploadHandle.mappedBuffer = + static_cast<uint8_t*>(uploadHandle.mappedBuffer) + additionalOffset; + uploadHandle.startOffset += additionalOffset; + return uploadHandle; + } +} // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/DynamicUploader.h b/chromium/third_party/dawn/src/dawn_native/DynamicUploader.h index 8210b035b22..a652bd8853f 100644 --- a/chromium/third_party/dawn/src/dawn_native/DynamicUploader.h +++ b/chromium/third_party/dawn/src/dawn_native/DynamicUploader.h @@ -40,7 +40,9 @@ namespace dawn_native { // implemented. void ReleaseStagingBuffer(std::unique_ptr<StagingBufferBase> stagingBuffer); - ResultOrError<UploadHandle> Allocate(uint64_t allocationSize, Serial serial); + ResultOrError<UploadHandle> Allocate(uint64_t allocationSize, + Serial serial, + uint64_t offsetAlignment); void Deallocate(Serial lastCompletedSerial); private: @@ -51,6 +53,8 @@ namespace dawn_native { RingBufferAllocator mAllocator; }; + ResultOrError<UploadHandle> AllocateInternal(uint64_t allocationSize, Serial serial); + std::vector<std::unique_ptr<RingBuffer>> mRingBuffers; SerialQueue<std::unique_ptr<StagingBufferBase>> mReleasedStagingBuffers; DeviceBase* mDevice; diff --git a/chromium/third_party/dawn/src/dawn_native/EnumClassBitmasks.h b/chromium/third_party/dawn/src/dawn_native/EnumClassBitmasks.h new file mode 100644 index 00000000000..3227cd2db0f --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_native/EnumClassBitmasks.h @@ -0,0 +1,48 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef DAWNNATIVE_ENUMCLASSBITMASK_H_ +#define DAWNNATIVE_ENUMCLASSBITMASK_H_ + +#include "dawn/EnumClassBitmasks.h" + +namespace dawn_native { + + // EnumClassBitmmasks is a WebGPU helper in the wgpu:: namespace. + // Re-export it in the dawn_native namespace. + + // Specify this for usage with EnumMaskIterator + template <typename T> + struct EnumBitmaskSize { + static constexpr unsigned value = 0; + }; + + using wgpu::operator|; + using wgpu::operator&; + using wgpu::operator^; + using wgpu::operator~; + using wgpu::operator&=; + using wgpu::operator|=; + using wgpu::operator^=; + + using wgpu::HasZeroOrOneBits; + + template <typename T> + constexpr bool HasOneBit(T value) { + return HasZeroOrOneBits(value) && value != T(0); + } + +} // namespace dawn_native + +#endif // DAWNNATIVE_ENUMCLASSBITMASK_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/EnumMaskIterator.h b/chromium/third_party/dawn/src/dawn_native/EnumMaskIterator.h new file mode 100644 index 00000000000..85fde1032dc --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_native/EnumMaskIterator.h @@ -0,0 +1,80 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef DAWNNATIVE_ENUMMASKITERATOR_H_ +#define DAWNNATIVE_ENUMMASKITERATOR_H_ + +#include "common/BitSetIterator.h" +#include "dawn_native/EnumClassBitmasks.h" + +namespace dawn_native { + + template <typename T> + class EnumMaskIterator final { + static constexpr size_t N = EnumBitmaskSize<T>::value; + static_assert(N > 0, ""); + + using U = std::underlying_type_t<T>; + + public: + EnumMaskIterator(const T& mask) : mBitSetIterator(std::bitset<N>(static_cast<U>(mask))) { + } + + class Iterator final { + public: + Iterator(const typename BitSetIterator<N, U>::Iterator& iter) : mIter(iter) { + } + + Iterator& operator++() { + ++mIter; + return *this; + } + + bool operator==(const Iterator& other) const { + return mIter == other.mIter; + } + + bool operator!=(const Iterator& other) const { + return mIter != other.mIter; + } + + T operator*() const { + U value = *mIter; + return static_cast<T>(U(1) << value); + } + + private: + typename BitSetIterator<N, U>::Iterator mIter; + }; + + Iterator begin() const { + return Iterator(mBitSetIterator.begin()); + } + + Iterator end() const { + return Iterator(mBitSetIterator.end()); + } + + private: + BitSetIterator<N, U> mBitSetIterator; + }; + + template <typename T> + EnumMaskIterator<T> IterateEnumMask(const T& mask) { + return EnumMaskIterator<T>(mask); + } + +} // namespace dawn_native + +#endif // DAWNNATIVE_ENUMMASKITERATOR_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/Format.cpp b/chromium/third_party/dawn/src/dawn_native/Format.cpp index 9c7fa58a3eb..11b71d269b8 100644 --- a/chromium/third_party/dawn/src/dawn_native/Format.cpp +++ b/chromium/third_party/dawn/src/dawn_native/Format.cpp @@ -13,8 +13,10 @@ // limitations under the License. #include "dawn_native/Format.h" + #include "dawn_native/Device.h" #include "dawn_native/Extensions.h" +#include "dawn_native/Texture.h" #include <bitset> @@ -58,25 +60,96 @@ namespace dawn_native { } bool Format::IsColor() const { - return aspect == Aspect::Color; + return aspects == Aspect::Color; } bool Format::HasDepth() const { - return aspect == Depth || aspect == DepthStencil; + return (aspects & Aspect::Depth) != 0; } bool Format::HasStencil() const { - return aspect == Stencil || aspect == DepthStencil; + return (aspects & Aspect::Stencil) != 0; } bool Format::HasDepthOrStencil() const { - return aspect != Color; + return (aspects & (Aspect::Depth | Aspect::Stencil)) != 0; } bool Format::HasComponentType(Type componentType) const { return componentType == type; } + TexelBlockInfo Format::GetTexelBlockInfo(wgpu::TextureAspect aspect) const { + switch (aspect) { + case wgpu::TextureAspect::All: + switch (aspects) { + case Aspect::Color: + case Aspect::Depth: + case Aspect::Stencil: + break; + default: + UNREACHABLE(); + } + return *this; + + case wgpu::TextureAspect::DepthOnly: + ASSERT(HasDepth()); + switch (format) { + case wgpu::TextureFormat::Depth32Float: + return *this; + default: + UNREACHABLE(); + break; + } + break; + + case wgpu::TextureAspect::StencilOnly: + ASSERT(HasStencil()); + switch (format) { + case wgpu::TextureFormat::Depth24PlusStencil8: + return {1, 1, 1}; + default: + UNREACHABLE(); + break; + } + break; + + default: + UNREACHABLE(); + break; + } + } + + TexelBlockInfo Format::GetTexelBlockInfo(Aspect aspect) const { + ASSERT(HasOneBit(aspect)); + ASSERT(aspects & aspect); + switch (aspect) { + case Aspect::Color: + ASSERT(aspects == aspect); + return *this; + case Aspect::Depth: + switch (format) { + case wgpu::TextureFormat::Depth32Float: + return *this; + default: + UNREACHABLE(); + break; + } + case Aspect::Stencil: + switch (format) { + case wgpu::TextureFormat::Depth24PlusStencil8: + return {1, 1, 1}; + default: + UNREACHABLE(); + break; + } + break; + default: + UNREACHABLE(); + break; + } + } + size_t Format::GetIndex() const { return ComputeFormatIndex(format); } @@ -98,7 +171,6 @@ namespace dawn_native { std::bitset<kKnownFormatCount> formatsSet; using Type = Format::Type; - using Aspect = Format::Aspect; auto AddFormat = [&table, &formatsSet](Format format) { size_t index = ComputeFormatIndex(format.format); @@ -121,7 +193,7 @@ namespace dawn_native { internalFormat.isCompressed = false; internalFormat.isSupported = true; internalFormat.supportsStorageUsage = supportsStorageUsage; - internalFormat.aspect = Aspect::Color; + internalFormat.aspects = Aspect::Color; internalFormat.type = type; internalFormat.blockByteSize = byteSize; internalFormat.blockWidth = 1; @@ -129,7 +201,7 @@ namespace dawn_native { AddFormat(internalFormat); }; - auto AddDepthStencilFormat = [&AddFormat](wgpu::TextureFormat format, Format::Aspect aspect, + auto AddDepthStencilFormat = [&AddFormat](wgpu::TextureFormat format, Aspect aspects, uint32_t byteSize) { Format internalFormat; internalFormat.format = format; @@ -137,7 +209,7 @@ namespace dawn_native { internalFormat.isCompressed = false; internalFormat.isSupported = true; internalFormat.supportsStorageUsage = false; - internalFormat.aspect = aspect; + internalFormat.aspects = aspects; internalFormat.type = Type::Other; internalFormat.blockByteSize = byteSize; internalFormat.blockWidth = 1; @@ -153,7 +225,7 @@ namespace dawn_native { internalFormat.isCompressed = false; internalFormat.isSupported = true; internalFormat.supportsStorageUsage = false; - internalFormat.aspect = Aspect::Depth; + internalFormat.aspects = Aspect::Depth; internalFormat.type = type; internalFormat.blockByteSize = byteSize; internalFormat.blockWidth = 1; @@ -169,7 +241,7 @@ namespace dawn_native { internalFormat.isCompressed = true; internalFormat.isSupported = isSupported; internalFormat.supportsStorageUsage = false; - internalFormat.aspect = Aspect::Color; + internalFormat.aspects = Aspect::Color; internalFormat.type = Type::Float; internalFormat.blockByteSize = byteSize; internalFormat.blockWidth = width; @@ -210,7 +282,8 @@ namespace dawn_native { AddColorFormat(wgpu::TextureFormat::BGRA8UnormSrgb, true, false, 4, Type::Float); AddColorFormat(wgpu::TextureFormat::RGB10A2Unorm, true, false, 4, Type::Float); - AddColorFormat(wgpu::TextureFormat::RG11B10Float, false, false, 4, Type::Float); + AddColorFormat(wgpu::TextureFormat::RG11B10Ufloat, false, false, 4, Type::Float); + AddColorFormat(wgpu::TextureFormat::RGB9E5Ufloat, false, false, 4, Type::Float); // 8 bytes color formats AddColorFormat(wgpu::TextureFormat::RG32Uint, true, true, 8, Type::Uint); @@ -232,7 +305,8 @@ namespace dawn_native { AddDepthStencilFormat(wgpu::TextureFormat::Depth24Plus, Aspect::Depth, 4); // TODO(cwallez@chromium.org): It isn't clear if this format should be copyable // because its size isn't well defined, is it 4, 5 or 8? - AddDepthStencilFormat(wgpu::TextureFormat::Depth24PlusStencil8, Aspect::DepthStencil, 4); + AddDepthStencilFormat(wgpu::TextureFormat::Depth24PlusStencil8, + Aspect::Depth | Aspect::Stencil, 4); // BC compressed formats bool isBCFormatSupported = device->IsExtensionEnabled(Extension::TextureCompressionBC); diff --git a/chromium/third_party/dawn/src/dawn_native/Format.h b/chromium/third_party/dawn/src/dawn_native/Format.h index 82b40d81cce..15c6c311a32 100644 --- a/chromium/third_party/dawn/src/dawn_native/Format.h +++ b/chromium/third_party/dawn/src/dawn_native/Format.h @@ -17,28 +17,31 @@ #include "dawn_native/dawn_platform.h" +#include "common/ityp_bitset.h" #include "dawn_native/Error.h" +#include "dawn_native/EnumClassBitmasks.h" + #include <array> namespace dawn_native { + enum class Aspect : uint8_t; class DeviceBase; + struct TexelBlockInfo { + uint32_t blockByteSize; + uint32_t blockWidth; + uint32_t blockHeight; + }; + // The number of formats Dawn knows about. Asserts in BuildFormatTable ensure that this is the // exact number of known format. - static constexpr size_t kKnownFormatCount = 52; + static constexpr size_t kKnownFormatCount = 53; // A wgpu::TextureFormat along with all the information about it necessary for validation. - struct Format { - enum Aspect { - Color, - Depth, - Stencil, - DepthStencil, - }; - - enum Type { + struct Format : TexelBlockInfo { + enum class Type { Float, Sint, Uint, @@ -51,12 +54,8 @@ namespace dawn_native { // A format can be known but not supported because it is part of a disabled extension. bool isSupported; bool supportsStorageUsage; - Aspect aspect; Type type; - - uint32_t blockByteSize; - uint32_t blockWidth; - uint32_t blockHeight; + Aspect aspects; static Type TextureComponentTypeToFormatType(wgpu::TextureComponentType componentType); static wgpu::TextureComponentType FormatTypeToTextureComponentType(Type type); @@ -67,6 +66,9 @@ namespace dawn_native { bool HasDepthOrStencil() const; bool HasComponentType(Type componentType) const; + TexelBlockInfo GetTexelBlockInfo(wgpu::TextureAspect aspect) const; + TexelBlockInfo GetTexelBlockInfo(Aspect aspect) const; + // The index of the format in the list of all known formats: a unique number for each format // in [0, kKnownFormatCount) size_t GetIndex() const; diff --git a/chromium/third_party/dawn/src/dawn_native/Instance.cpp b/chromium/third_party/dawn/src/dawn_native/Instance.cpp index c7a1e310f30..49db4b60fc5 100644 --- a/chromium/third_party/dawn/src/dawn_native/Instance.cpp +++ b/chromium/third_party/dawn/src/dawn_native/Instance.cpp @@ -145,7 +145,7 @@ namespace dawn_native { # if defined(DAWN_ENABLE_SWIFTSHADER) Register(vulkan::Connect(this, true), wgpu::BackendType::Vulkan); # endif // defined(DAWN_ENABLE_SWIFTSHADER) -#endif // defined(DAWN_ENABLE_BACKEND_VULKAN) +#endif // defined(DAWN_ENABLE_BACKEND_VULKAN) #if defined(DAWN_ENABLE_BACKEND_OPENGL) Register(opengl::Connect(this), wgpu::BackendType::OpenGL); #endif // defined(DAWN_ENABLE_BACKEND_OPENGL) @@ -202,6 +202,14 @@ namespace dawn_native { return mEnableBackendValidation; } + void InstanceBase::EnableGPUBasedBackendValidation(bool enableGPUBasedBackendValidation) { + mEnableGPUValidation = enableGPUBasedBackendValidation; + } + + bool InstanceBase::IsGPUBasedBackendValidationEnabled() const { + return mEnableGPUValidation; + } + void InstanceBase::EnableBeginCaptureOnStartup(bool beginCaptureOnStartup) { mBeginCaptureOnStartup = beginCaptureOnStartup; } diff --git a/chromium/third_party/dawn/src/dawn_native/Instance.h b/chromium/third_party/dawn/src/dawn_native/Instance.h index 0ade98b4e60..14d4eef2ff1 100644 --- a/chromium/third_party/dawn/src/dawn_native/Instance.h +++ b/chromium/third_party/dawn/src/dawn_native/Instance.h @@ -60,6 +60,9 @@ namespace dawn_native { void EnableBackendValidation(bool enableBackendValidation); bool IsBackendValidationEnabled() const; + void EnableGPUBasedBackendValidation(bool enableGPUBasedBackendValidation); + bool IsGPUBasedBackendValidationEnabled() const; + void EnableBeginCaptureOnStartup(bool beginCaptureOnStartup); bool IsBeginCaptureOnStartupEnabled() const; @@ -88,6 +91,7 @@ namespace dawn_native { bool mEnableBackendValidation = false; bool mBeginCaptureOnStartup = false; + bool mEnableGPUValidation = false; dawn_platform::Platform* mPlatform = nullptr; diff --git a/chromium/third_party/dawn/src/dawn_native/MapRequestTracker.cpp b/chromium/third_party/dawn/src/dawn_native/MapRequestTracker.cpp index 8f33e023110..91af10bcece 100644 --- a/chromium/third_party/dawn/src/dawn_native/MapRequestTracker.cpp +++ b/chromium/third_party/dawn/src/dawn_native/MapRequestTracker.cpp @@ -27,11 +27,11 @@ namespace dawn_native { ASSERT(mInflightRequests.Empty()); } - void MapRequestTracker::Track(BufferBase* buffer, uint32_t mapSerial, bool isWrite) { + void MapRequestTracker::Track(BufferBase* buffer, uint32_t mapSerial, MapType type) { Request request; request.buffer = buffer; request.mapSerial = mapSerial; - request.isWrite = isWrite; + request.type = type; mInflightRequests.Enqueue(std::move(request), mDevice->GetPendingCommandSerial()); mDevice->AddFutureCallbackSerial(mDevice->GetPendingCommandSerial()); @@ -39,8 +39,8 @@ namespace dawn_native { void MapRequestTracker::Tick(Serial finishedSerial) { for (auto& request : mInflightRequests.IterateUpTo(finishedSerial)) { - request.buffer->OnMapCommandSerialFinished(request.mapSerial, request.isWrite); + request.buffer->OnMapCommandSerialFinished(request.mapSerial, request.type); } mInflightRequests.ClearUpTo(finishedSerial); } -} // namespace dawn_native
\ No newline at end of file +} // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/MapRequestTracker.h b/chromium/third_party/dawn/src/dawn_native/MapRequestTracker.h index 0dffca18f3d..1daf47dad92 100644 --- a/chromium/third_party/dawn/src/dawn_native/MapRequestTracker.h +++ b/chromium/third_party/dawn/src/dawn_native/MapRequestTracker.h @@ -20,12 +20,15 @@ namespace dawn_native { + // TODO(dawn:22) remove this enum once MapReadAsync/MapWriteAsync are removed. + enum class MapType : uint32_t { Read, Write, Async }; + class MapRequestTracker { public: MapRequestTracker(DeviceBase* device); ~MapRequestTracker(); - void Track(BufferBase* buffer, uint32_t mapSerial, bool isWrite); + void Track(BufferBase* buffer, uint32_t mapSerial, MapType type); void Tick(Serial finishedSerial); private: @@ -34,11 +37,11 @@ namespace dawn_native { struct Request { Ref<BufferBase> buffer; uint32_t mapSerial; - bool isWrite; + MapType type; }; SerialQueue<Request> mInflightRequests; }; } // namespace dawn_native -#endif // DAWNNATIVE_MAPREQUESTTRACKER_H
\ No newline at end of file +#endif // DAWNNATIVE_MAPREQUESTTRACKER_H diff --git a/chromium/third_party/dawn/src/dawn_native/PassResourceUsage.h b/chromium/third_party/dawn/src/dawn_native/PassResourceUsage.h index 9271114bc58..60dcd219a6d 100644 --- a/chromium/third_party/dawn/src/dawn_native/PassResourceUsage.h +++ b/chromium/third_party/dawn/src/dawn_native/PassResourceUsage.h @@ -23,6 +23,7 @@ namespace dawn_native { class BufferBase; + class QuerySetBase; class TextureBase; enum class PassType { Render, Compute }; @@ -45,8 +46,8 @@ namespace dawn_native { // the vector to record every single subresource's Usages. The texture usage is enough. And we // can decompress texture usage to a vector if necessary. struct PassTextureUsage { - wgpu::TextureUsage usage; - bool sameUsagesAcrossSubresources; + wgpu::TextureUsage usage = wgpu::TextureUsage::None; + bool sameUsagesAcrossSubresources = true; std::vector<wgpu::TextureUsage> subresourceUsages; }; @@ -68,6 +69,7 @@ namespace dawn_native { PerPassUsages perPass; std::set<BufferBase*> topLevelBuffers; std::set<TextureBase*> topLevelTextures; + std::set<QuerySetBase*> usedQuerySets; }; } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/PassResourceUsageTracker.cpp b/chromium/third_party/dawn/src/dawn_native/PassResourceUsageTracker.cpp index f5e4a5664e0..3f5a9159dd8 100644 --- a/chromium/third_party/dawn/src/dawn_native/PassResourceUsageTracker.cpp +++ b/chromium/third_party/dawn/src/dawn_native/PassResourceUsageTracker.cpp @@ -15,6 +15,8 @@ #include "dawn_native/PassResourceUsageTracker.h" #include "dawn_native/Buffer.h" +#include "dawn_native/EnumMaskIterator.h" +#include "dawn_native/Format.h" #include "dawn_native/Texture.h" namespace dawn_native { @@ -30,10 +32,7 @@ namespace dawn_native { void PassResourceUsageTracker::TextureViewUsedAs(TextureViewBase* view, wgpu::TextureUsage usage) { TextureBase* texture = view->GetTexture(); - uint32_t baseMipLevel = view->GetBaseMipLevel(); - uint32_t levelCount = view->GetLevelCount(); - uint32_t baseArrayLayer = view->GetBaseArrayLayer(); - uint32_t layerCount = view->GetLayerCount(); + const SubresourceRange& range = view->GetSubresourceRange(); // std::map's operator[] will create the key and return a PassTextureUsage with usage = 0 // and an empty vector for subresourceUsages. @@ -42,20 +41,25 @@ namespace dawn_native { // Set parameters for the whole texture textureUsage.usage |= usage; - uint32_t subresourceCount = texture->GetSubresourceCount(); - textureUsage.sameUsagesAcrossSubresources = levelCount * layerCount == subresourceCount; + textureUsage.sameUsagesAcrossSubresources &= + (range.levelCount == texture->GetNumMipLevels() && // + range.layerCount == texture->GetArrayLayers() && // + range.aspects == texture->GetFormat().aspects); // Set usages for subresources if (!textureUsage.subresourceUsages.size()) { - textureUsage.subresourceUsages = - std::vector<wgpu::TextureUsage>(subresourceCount, wgpu::TextureUsage::None); + textureUsage.subresourceUsages = std::vector<wgpu::TextureUsage>( + texture->GetSubresourceCount(), wgpu::TextureUsage::None); } - for (uint32_t arrayLayer = baseArrayLayer; arrayLayer < baseArrayLayer + layerCount; - ++arrayLayer) { - for (uint32_t mipLevel = baseMipLevel; mipLevel < baseMipLevel + levelCount; - ++mipLevel) { - uint32_t subresourceIndex = texture->GetSubresourceIndex(mipLevel, arrayLayer); - textureUsage.subresourceUsages[subresourceIndex] |= usage; + for (Aspect aspect : IterateEnumMask(range.aspects)) { + for (uint32_t arrayLayer = range.baseArrayLayer; + arrayLayer < range.baseArrayLayer + range.layerCount; ++arrayLayer) { + for (uint32_t mipLevel = range.baseMipLevel; + mipLevel < range.baseMipLevel + range.levelCount; ++mipLevel) { + uint32_t subresourceIndex = + texture->GetSubresourceIndex(mipLevel, arrayLayer, aspect); + textureUsage.subresourceUsages[subresourceIndex] |= usage; + } } } } diff --git a/chromium/third_party/dawn/src/dawn_native/Pipeline.cpp b/chromium/third_party/dawn/src/dawn_native/Pipeline.cpp index df344416a56..ae05c02f4b4 100644 --- a/chromium/third_party/dawn/src/dawn_native/Pipeline.cpp +++ b/chromium/third_party/dawn/src/dawn_native/Pipeline.cpp @@ -91,25 +91,13 @@ namespace dawn_native { } BindGroupIndex groupIndex(groupIndexIn); + + BindGroupLayoutBase* bgl = nullptr; if (!mLayout->GetBindGroupLayoutsMask()[groupIndex]) { - // Get or create an empty bind group layout. - // TODO(enga): Consider caching this object on the Device and reusing it. - // Today, this can't be done correctly because of the order of Device destruction. - // For example, vulkan::~Device will be called before ~DeviceBase. If DeviceBase owns - // a Ref<BindGroupLayoutBase>, then the VkDevice will be destroyed before the - // VkDescriptorSetLayout. - BindGroupLayoutDescriptor desc = {}; - desc.entryCount = 0; - desc.entries = nullptr; - - BindGroupLayoutBase* bgl = nullptr; - if (GetDevice()->ConsumedError(GetDevice()->GetOrCreateBindGroupLayout(&desc), &bgl)) { - return BindGroupLayoutBase::MakeError(GetDevice()); - } - return bgl; + bgl = GetDevice()->GetEmptyBindGroupLayout(); + } else { + bgl = mLayout->GetBindGroupLayout(groupIndex); } - - BindGroupLayoutBase* bgl = mLayout->GetBindGroupLayout(groupIndex); bgl->Reference(); return bgl; } diff --git a/chromium/third_party/dawn/src/dawn_native/PipelineLayout.cpp b/chromium/third_party/dawn/src/dawn_native/PipelineLayout.cpp index def4875c445..f34003a5b7a 100644 --- a/chromium/third_party/dawn/src/dawn_native/PipelineLayout.cpp +++ b/chromium/third_party/dawn/src/dawn_native/PipelineLayout.cpp @@ -17,6 +17,7 @@ #include "common/Assert.h" #include "common/BitSetIterator.h" #include "common/HashUtils.h" +#include "common/ityp_stack_vec.h" #include "dawn_native/BindGroupLayout.h" #include "dawn_native/Device.h" #include "dawn_native/ShaderModule.h" @@ -70,24 +71,14 @@ namespace dawn_native { return DAWN_VALIDATION_ERROR("too many bind group layouts"); } - uint32_t totalDynamicUniformBufferCount = 0; - uint32_t totalDynamicStorageBufferCount = 0; + BindingCounts bindingCounts = {}; for (uint32_t i = 0; i < descriptor->bindGroupLayoutCount; ++i) { DAWN_TRY(device->ValidateObject(descriptor->bindGroupLayouts[i])); - totalDynamicUniformBufferCount += - descriptor->bindGroupLayouts[i]->GetDynamicUniformBufferCount(); - totalDynamicStorageBufferCount += - descriptor->bindGroupLayouts[i]->GetDynamicStorageBufferCount(); - } - - if (totalDynamicUniformBufferCount > kMaxDynamicUniformBufferCount) { - return DAWN_VALIDATION_ERROR("too many dynamic uniform buffers in pipeline layout"); - } - - if (totalDynamicStorageBufferCount > kMaxDynamicStorageBufferCount) { - return DAWN_VALIDATION_ERROR("too many dynamic storage buffers in pipeline layout"); + AccumulateBindingCounts(&bindingCounts, + descriptor->bindGroupLayouts[i]->GetBindingCountInfo()); } + DAWN_TRY(ValidateBindingCounts(bindingCounts)); return {}; } @@ -128,9 +119,10 @@ namespace dawn_native { ASSERT(count > 0); // Data which BindGroupLayoutDescriptor will point to for creation - ityp::array<BindGroupIndex, - ityp::array<BindingIndex, BindGroupLayoutEntry, kMaxBindingsPerGroup>, - kMaxBindGroups> + ityp::array< + BindGroupIndex, + ityp::stack_vec<BindingIndex, BindGroupLayoutEntry, kMaxOptimalBindingsPerGroup>, + kMaxBindGroups> entryData = {}; // A map of bindings to the index in |entryData| @@ -140,6 +132,7 @@ namespace dawn_native { // A counter of how many bindings we've populated in |entryData| ityp::array<BindGroupIndex, BindingIndex, kMaxBindGroups> entryCounts = {}; + BindingCounts bindingCounts = {}; BindGroupIndex bindGroupLayoutCount(0); for (uint32_t moduleIndex = 0; moduleIndex < count; ++moduleIndex) { const ShaderModuleBase* module = modules[moduleIndex]; @@ -201,7 +194,9 @@ namespace dawn_native { } } + IncrementBindingCounts(&bindingCounts, bindingSlot); BindingIndex currentBindingCount = entryCounts[group]; + entryData[group].resize(currentBindingCount + BindingIndex(1)); entryData[group][currentBindingCount] = bindingSlot; usedBindingsMap[group][bindingNumber] = currentBindingCount; @@ -214,6 +209,8 @@ namespace dawn_native { } } + DAWN_TRY(ValidateBindingCounts(bindingCounts)); + ityp::array<BindGroupIndex, BindGroupLayoutBase*, kMaxBindGroups> bindGroupLayouts = {}; for (BindGroupIndex group(0); group < bindGroupLayoutCount; ++group) { BindGroupLayoutDescriptor desc = {}; @@ -222,7 +219,10 @@ namespace dawn_native { // We should never produce a bad descriptor. ASSERT(!ValidateBindGroupLayoutDescriptor(device, &desc).IsError()); - DAWN_TRY_ASSIGN(bindGroupLayouts[group], device->GetOrCreateBindGroupLayout(&desc)); + + Ref<BindGroupLayoutBase> bgl; + DAWN_TRY_ASSIGN(bgl, device->GetOrCreateBindGroupLayout(&desc)); + bindGroupLayouts[group] = bgl.Detach(); } PipelineLayoutDescriptor desc = {}; diff --git a/chromium/third_party/dawn/src/dawn_native/PooledResourceMemoryAllocator.cpp b/chromium/third_party/dawn/src/dawn_native/PooledResourceMemoryAllocator.cpp new file mode 100644 index 00000000000..5ba502e8fbf --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_native/PooledResourceMemoryAllocator.cpp @@ -0,0 +1,60 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "dawn_native/PooledResourceMemoryAllocator.h" +#include "dawn_native/Device.h" + +namespace dawn_native { + + PooledResourceMemoryAllocator::PooledResourceMemoryAllocator( + ResourceHeapAllocator* heapAllocator) + : mHeapAllocator(heapAllocator) { + } + + void PooledResourceMemoryAllocator::DestroyPool() { + for (auto& resourceHeap : mPool) { + ASSERT(resourceHeap != nullptr); + mHeapAllocator->DeallocateResourceHeap(std::move(resourceHeap)); + } + + mPool.clear(); + } + + ResultOrError<std::unique_ptr<ResourceHeapBase>> + PooledResourceMemoryAllocator::AllocateResourceHeap(uint64_t size) { + // Pooled memory is LIFO because memory can be evicted by LRU. However, this means + // pooling is disabled in-frame when the memory is still pending. For high in-frame + // memory users, FIFO might be preferable when memory consumption is a higher priority. + std::unique_ptr<ResourceHeapBase> memory; + if (!mPool.empty()) { + memory = std::move(mPool.front()); + mPool.pop_front(); + } + + if (memory == nullptr) { + DAWN_TRY_ASSIGN(memory, mHeapAllocator->AllocateResourceHeap(size)); + } + + return std::move(memory); + } + + void PooledResourceMemoryAllocator::DeallocateResourceHeap( + std::unique_ptr<ResourceHeapBase> allocation) { + mPool.push_front(std::move(allocation)); + } + + uint64_t PooledResourceMemoryAllocator::GetPoolSizeForTesting() const { + return mPool.size(); + } +} // namespace dawn_native
\ No newline at end of file diff --git a/chromium/third_party/dawn/src/dawn_native/PooledResourceMemoryAllocator.h b/chromium/third_party/dawn/src/dawn_native/PooledResourceMemoryAllocator.h new file mode 100644 index 00000000000..5b6b816ee6a --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_native/PooledResourceMemoryAllocator.h @@ -0,0 +1,53 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef DAWNNATIVE_POOLEDRESOURCEMEMORYALLOCATOR_H_ +#define DAWNNATIVE_POOLEDRESOURCEMEMORYALLOCATOR_H_ + +#include "common/SerialQueue.h" +#include "dawn_native/ResourceHeapAllocator.h" + +#include <deque> + +namespace dawn_native { + + class DeviceBase; + + // |PooledResourceMemoryAllocator| allocates a fixed-size resource memory from a resource memory + // pool. Internally, it manages a list of heaps using LIFO (newest heaps are recycled first). + // The heap is in one of two states: AVAILABLE or not. Upon de-allocate, the heap is returned + // the pool and made AVAILABLE. + class PooledResourceMemoryAllocator : public ResourceHeapAllocator { + public: + PooledResourceMemoryAllocator(ResourceHeapAllocator* heapAllocator); + ~PooledResourceMemoryAllocator() override = default; + + ResultOrError<std::unique_ptr<ResourceHeapBase>> AllocateResourceHeap( + uint64_t size) override; + void DeallocateResourceHeap(std::unique_ptr<ResourceHeapBase> allocation) override; + + void DestroyPool(); + + // For testing purposes. + uint64_t GetPoolSizeForTesting() const; + + private: + ResourceHeapAllocator* mHeapAllocator = nullptr; + + std::deque<std::unique_ptr<ResourceHeapBase>> mPool; + }; + +} // namespace dawn_native + +#endif // DAWNNATIVE_POOLEDRESOURCEMEMORYALLOCATOR_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/QuerySet.cpp b/chromium/third_party/dawn/src/dawn_native/QuerySet.cpp index 513658db4c7..bd65ed05baa 100644 --- a/chromium/third_party/dawn/src/dawn_native/QuerySet.cpp +++ b/chromium/third_party/dawn/src/dawn_native/QuerySet.cpp @@ -65,11 +65,11 @@ namespace dawn_native { "PipelineStatistics"); } - std::set<wgpu::PipelineStatisticsName> pipelineStatisticsSet; + std::set<wgpu::PipelineStatisticName> pipelineStatisticsSet; for (uint32_t i = 0; i < descriptor->pipelineStatisticsCount; i++) { - DAWN_TRY(ValidatePipelineStatisticsName(descriptor->pipelineStatistics[i])); + DAWN_TRY(ValidatePipelineStatisticName(descriptor->pipelineStatistics[i])); - std::pair<std::set<wgpu::PipelineStatisticsName>::iterator, bool> res = + std::pair<std::set<wgpu::PipelineStatisticName>::iterator, bool> res = pipelineStatisticsSet.insert((descriptor->pipelineStatistics[i])); if (!res.second) { return DAWN_VALIDATION_ERROR("Duplicate pipeline statistics found"); @@ -127,10 +127,18 @@ namespace dawn_native { return mQueryCount; } - const std::vector<wgpu::PipelineStatisticsName>& QuerySetBase::GetPipelineStatistics() const { + const std::vector<wgpu::PipelineStatisticName>& QuerySetBase::GetPipelineStatistics() const { return mPipelineStatistics; } + MaybeError QuerySetBase::ValidateCanUseInSubmitNow() const { + ASSERT(!IsError()); + if (mState == QuerySetState::Destroyed) { + return DAWN_VALIDATION_ERROR("Destroyed query set used in a submit"); + } + return {}; + } + void QuerySetBase::Destroy() { if (GetDevice()->ConsumedError(ValidateDestroy())) { return; diff --git a/chromium/third_party/dawn/src/dawn_native/QuerySet.h b/chromium/third_party/dawn/src/dawn_native/QuerySet.h index 7883678ffb0..a8f4deace6a 100644 --- a/chromium/third_party/dawn/src/dawn_native/QuerySet.h +++ b/chromium/third_party/dawn/src/dawn_native/QuerySet.h @@ -33,7 +33,9 @@ namespace dawn_native { wgpu::QueryType GetQueryType() const; uint32_t GetQueryCount() const; - const std::vector<wgpu::PipelineStatisticsName>& GetPipelineStatistics() const; + const std::vector<wgpu::PipelineStatisticName>& GetPipelineStatistics() const; + + MaybeError ValidateCanUseInSubmitNow() const; void Destroy(); @@ -50,7 +52,7 @@ namespace dawn_native { wgpu::QueryType mQueryType; uint32_t mQueryCount; - std::vector<wgpu::PipelineStatisticsName> mPipelineStatistics; + std::vector<wgpu::PipelineStatisticName> mPipelineStatistics; enum class QuerySetState { Unavailable, Available, Destroyed }; QuerySetState mState = QuerySetState::Unavailable; diff --git a/chromium/third_party/dawn/src/dawn_native/Queue.cpp b/chromium/third_party/dawn/src/dawn_native/Queue.cpp index 3dcf2b0e469..eb0b053e7d3 100644 --- a/chromium/third_party/dawn/src/dawn_native/Queue.cpp +++ b/chromium/third_party/dawn/src/dawn_native/Queue.cpp @@ -14,14 +14,17 @@ #include "dawn_native/Queue.h" +#include "common/Constants.h" #include "dawn_native/Buffer.h" #include "dawn_native/CommandBuffer.h" +#include "dawn_native/CommandValidation.h" #include "dawn_native/Device.h" #include "dawn_native/DynamicUploader.h" #include "dawn_native/ErrorScope.h" #include "dawn_native/ErrorScopeTracker.h" #include "dawn_native/Fence.h" #include "dawn_native/FenceSignalTracker.h" +#include "dawn_native/QuerySet.h" #include "dawn_native/Texture.h" #include "dawn_platform/DawnPlatform.h" #include "dawn_platform/tracing/TraceEvent.h" @@ -49,24 +52,11 @@ namespace dawn_native { } void QueueBase::Submit(uint32_t commandCount, CommandBufferBase* const* commands) { - DeviceBase* device = GetDevice(); - if (device->ConsumedError(device->ValidateIsAlive())) { - // If device is lost, don't let any commands be submitted - return; - } + SubmitInternal(commandCount, commands); - TRACE_EVENT0(device->GetPlatform(), General, "Queue::Submit"); - if (device->IsValidationEnabled() && - device->ConsumedError(ValidateSubmit(commandCount, commands))) { - return; - } - ASSERT(!IsError()); - - if (device->ConsumedError(SubmitImpl(commandCount, commands))) { - return; + for (uint32_t i = 0; i < commandCount; ++i) { + commands[i]->Destroy(); } - device->GetErrorScopeTracker()->TrackUntilLastSubmitComplete( - device->GetCurrentErrorScope()); } void QueueBase::Signal(Fence* fence, uint64_t signalValue) { @@ -121,7 +111,8 @@ namespace dawn_native { UploadHandle uploadHandle; DAWN_TRY_ASSIGN(uploadHandle, device->GetDynamicUploader()->Allocate( - size, device->GetPendingCommandSerial())); + size, device->GetPendingCommandSerial(), + kCopyBufferToBufferOffsetAlignment)); ASSERT(uploadHandle.mappedBuffer != nullptr); memcpy(uploadHandle.mappedBuffer, data, size); @@ -130,6 +121,37 @@ namespace dawn_native { buffer, bufferOffset, size); } + void QueueBase::WriteTexture(const TextureCopyView* destination, + const void* data, + size_t dataSize, + const TextureDataLayout* dataLayout, + const Extent3D* writeSize) { + GetDevice()->ConsumedError( + WriteTextureInternal(destination, data, dataSize, dataLayout, writeSize)); + } + + MaybeError QueueBase::WriteTextureInternal(const TextureCopyView* destination, + const void* data, + size_t dataSize, + const TextureDataLayout* dataLayout, + const Extent3D* writeSize) { + DAWN_TRY(ValidateWriteTexture(destination, dataSize, dataLayout, writeSize)); + + if (writeSize->width == 0 || writeSize->height == 0 || writeSize->depth == 0) { + return {}; + } + + return WriteTextureImpl(*destination, data, *dataLayout, *writeSize); + } + + MaybeError QueueBase::WriteTextureImpl(const TextureCopyView& destination, + const void* data, + const TextureDataLayout& dataLayout, + const Extent3D& writeSize) { + // TODO(tommek@google.com): This should be implemented. + return {}; + } + MaybeError QueueBase::ValidateSubmit(uint32_t commandCount, CommandBufferBase* const* commands) const { TRACE_EVENT0(GetDevice()->GetPlatform(), Validation, "Queue::ValidateSubmit"); @@ -137,6 +159,7 @@ namespace dawn_native { for (uint32_t i = 0; i < commandCount; ++i) { DAWN_TRY(GetDevice()->ValidateObject(commands[i])); + DAWN_TRY(commands[i]->ValidateCanUseInSubmitNow()); const CommandBufferResourceUsage& usages = commands[i]->GetResourceUsages(); @@ -155,6 +178,9 @@ namespace dawn_native { for (const TextureBase* texture : usages.topLevelTextures) { DAWN_TRY(texture->ValidateCanUseInSubmitNow()); } + for (const QuerySetBase* querySet : usages.usedQuerySets) { + DAWN_TRY(querySet->ValidateCanUseInSubmitNow()); + } } return {}; @@ -208,7 +234,102 @@ namespace dawn_native { return DAWN_VALIDATION_ERROR("Buffer needs the CopyDst usage bit"); } - return buffer->ValidateCanUseOnQueueNow(); + DAWN_TRY(buffer->ValidateCanUseOnQueueNow()); + + return {}; } + MaybeError QueueBase::ValidateWriteTexture(const TextureCopyView* destination, + size_t dataSize, + const TextureDataLayout* dataLayout, + const Extent3D* writeSize) const { + DAWN_TRY(GetDevice()->ValidateIsAlive()); + DAWN_TRY(GetDevice()->ValidateObject(this)); + DAWN_TRY(GetDevice()->ValidateObject(destination->texture)); + + DAWN_TRY(ValidateTextureCopyView(GetDevice(), *destination)); + + if (dataLayout->offset > dataSize) { + return DAWN_VALIDATION_ERROR("Queue::WriteTexture out of range"); + } + + if (!(destination->texture->GetUsage() & wgpu::TextureUsage::CopyDst)) { + return DAWN_VALIDATION_ERROR("Texture needs the CopyDst usage bit"); + } + + if (destination->texture->GetSampleCount() > 1) { + return DAWN_VALIDATION_ERROR("The sample count of textures must be 1"); + } + + // We validate texture copy range before validating linear texture data, + // because in the latter we divide copyExtent.width by blockWidth and + // copyExtent.height by blockHeight while the divisibility conditions are + // checked in validating texture copy range. + DAWN_TRY(ValidateTextureCopyRange(*destination, *writeSize)); + DAWN_TRY(ValidateBufferToTextureCopyRestrictions(*destination)); + DAWN_TRY(ValidateLinearTextureData( + *dataLayout, dataSize, + destination->texture->GetFormat().GetTexelBlockInfo(destination->aspect), *writeSize)); + + DAWN_TRY(destination->texture->ValidateCanUseInSubmitNow()); + + return {}; + } + + void QueueBase::SubmitInternal(uint32_t commandCount, CommandBufferBase* const* commands) { + DeviceBase* device = GetDevice(); + if (device->ConsumedError(device->ValidateIsAlive())) { + // If device is lost, don't let any commands be submitted + return; + } + + TRACE_EVENT0(device->GetPlatform(), General, "Queue::Submit"); + if (device->IsValidationEnabled() && + device->ConsumedError(ValidateSubmit(commandCount, commands))) { + return; + } + ASSERT(!IsError()); + + if (device->ConsumedError(SubmitImpl(commandCount, commands))) { + return; + } + + device->GetErrorScopeTracker()->TrackUntilLastSubmitComplete( + device->GetCurrentErrorScope()); + } + + void CopyTextureData(uint8_t* dstPointer, + const uint8_t* srcPointer, + uint32_t depth, + uint32_t rowsPerImageInBlock, + uint64_t imageAdditionalStride, + uint32_t actualBytesPerRow, + uint32_t dstBytesPerRow, + uint32_t srcBytesPerRow) { + bool copyWholeLayer = + actualBytesPerRow == dstBytesPerRow && dstBytesPerRow == srcBytesPerRow; + bool copyWholeData = copyWholeLayer && imageAdditionalStride == 0; + + if (!copyWholeLayer) { // copy row by row + for (uint32_t d = 0; d < depth; ++d) { + for (uint32_t h = 0; h < rowsPerImageInBlock; ++h) { + memcpy(dstPointer, srcPointer, actualBytesPerRow); + dstPointer += dstBytesPerRow; + srcPointer += srcBytesPerRow; + } + srcPointer += imageAdditionalStride; + } + } else { + uint64_t layerSize = uint64_t(rowsPerImageInBlock) * actualBytesPerRow; + if (!copyWholeData) { // copy layer by layer + for (uint32_t d = 0; d < depth; ++d) { + memcpy(dstPointer, srcPointer, layerSize); + dstPointer += layerSize; + srcPointer += layerSize + imageAdditionalStride; + } + } else { // do a single copy + memcpy(dstPointer, srcPointer, layerSize * depth); + } + } + } } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/Queue.h b/chromium/third_party/dawn/src/dawn_native/Queue.h index 5fd722ddf5e..be1ce62b573 100644 --- a/chromium/third_party/dawn/src/dawn_native/Queue.h +++ b/chromium/third_party/dawn/src/dawn_native/Queue.h @@ -34,6 +34,11 @@ namespace dawn_native { void Signal(Fence* fence, uint64_t signalValue); Fence* CreateFence(const FenceDescriptor* descriptor); void WriteBuffer(BufferBase* buffer, uint64_t bufferOffset, const void* data, size_t size); + void WriteTexture(const TextureCopyView* destination, + const void* data, + size_t dataSize, + const TextureDataLayout* dataLayout, + const Extent3D* writeSize); private: QueueBase(DeviceBase* device, ObjectBase::ErrorTag tag); @@ -42,12 +47,21 @@ namespace dawn_native { uint64_t bufferOffset, const void* data, size_t size); + MaybeError WriteTextureInternal(const TextureCopyView* destination, + const void* data, + size_t dataSize, + const TextureDataLayout* dataLayout, + const Extent3D* writeSize); virtual MaybeError SubmitImpl(uint32_t commandCount, CommandBufferBase* const* commands); virtual MaybeError WriteBufferImpl(BufferBase* buffer, uint64_t bufferOffset, const void* data, size_t size); + virtual MaybeError WriteTextureImpl(const TextureCopyView& destination, + const void* data, + const TextureDataLayout& dataLayout, + const Extent3D& writeSize); MaybeError ValidateSubmit(uint32_t commandCount, CommandBufferBase* const* commands) const; MaybeError ValidateSignal(const Fence* fence, uint64_t signalValue) const; @@ -55,8 +69,24 @@ namespace dawn_native { MaybeError ValidateWriteBuffer(const BufferBase* buffer, uint64_t bufferOffset, size_t size) const; + MaybeError ValidateWriteTexture(const TextureCopyView* destination, + size_t dataSize, + const TextureDataLayout* dataLayout, + const Extent3D* writeSize) const; + + void SubmitInternal(uint32_t commandCount, CommandBufferBase* const* commands); }; + // A helper function used in Queue::WriteTexture. The destination data layout must not + // contain any additional rows per image. + void CopyTextureData(uint8_t* dstPointer, + const uint8_t* srcPointer, + uint32_t depth, + uint32_t rowsPerImageInBlock, + uint64_t imageAdditionalStride, + uint32_t actualBytesPerRow, + uint32_t dstBytesPerRow, + uint32_t srcBytesPerRow); } // namespace dawn_native #endif // DAWNNATIVE_QUEUE_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/RenderPassEncoder.cpp b/chromium/third_party/dawn/src/dawn_native/RenderPassEncoder.cpp index e2dc4db7fce..e51b52f7662 100644 --- a/chromium/third_party/dawn/src/dawn_native/RenderPassEncoder.cpp +++ b/chromium/third_party/dawn/src/dawn_native/RenderPassEncoder.cpp @@ -17,8 +17,10 @@ #include "common/Constants.h" #include "dawn_native/Buffer.h" #include "dawn_native/CommandEncoder.h" +#include "dawn_native/CommandValidation.h" #include "dawn_native/Commands.h" #include "dawn_native/Device.h" +#include "dawn_native/QuerySet.h" #include "dawn_native/RenderBundle.h" #include "dawn_native/RenderPipeline.h" @@ -163,4 +165,21 @@ namespace dawn_native { }); } + void RenderPassEncoder::WriteTimestamp(QuerySetBase* querySet, uint32_t queryIndex) { + mEncodingContext->TryEncode(this, [&](CommandAllocator* allocator) -> MaybeError { + if (GetDevice()->IsValidationEnabled()) { + DAWN_TRY(GetDevice()->ValidateObject(querySet)); + DAWN_TRY(ValidateTimestampQuery(querySet, queryIndex)); + mCommandEncoder->TrackUsedQuerySet(querySet); + } + + WriteTimestampCmd* cmd = + allocator->Allocate<WriteTimestampCmd>(Command::WriteTimestamp); + cmd->querySet = querySet; + cmd->queryIndex = queryIndex; + + return {}; + }); + } + } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/RenderPassEncoder.h b/chromium/third_party/dawn/src/dawn_native/RenderPassEncoder.h index cd9ac017fb7..d5a2f7aebbc 100644 --- a/chromium/third_party/dawn/src/dawn_native/RenderPassEncoder.h +++ b/chromium/third_party/dawn/src/dawn_native/RenderPassEncoder.h @@ -46,6 +46,8 @@ namespace dawn_native { void SetScissorRect(uint32_t x, uint32_t y, uint32_t width, uint32_t height); void ExecuteBundles(uint32_t count, RenderBundleBase* const* renderBundles); + void WriteTimestamp(QuerySetBase* querySet, uint32_t queryIndex); + protected: RenderPassEncoder(DeviceBase* device, CommandEncoder* commandEncoder, diff --git a/chromium/third_party/dawn/src/dawn_native/RenderPipeline.cpp b/chromium/third_party/dawn/src/dawn_native/RenderPipeline.cpp index ceeb1eb2fc9..87c9f62e65e 100644 --- a/chromium/third_party/dawn/src/dawn_native/RenderPipeline.cpp +++ b/chromium/third_party/dawn/src/dawn_native/RenderPipeline.cpp @@ -366,12 +366,8 @@ namespace dawn_native { DAWN_TRY(ValidateDepthStencilStateDescriptor(device, descriptor->depthStencilState)); } - if (descriptor->sampleMask != 0xFFFFFFFF) { - return DAWN_VALIDATION_ERROR("sampleMask must be 0xFFFFFFFF (for now)"); - } - - if (descriptor->alphaToCoverageEnabled) { - return DAWN_VALIDATION_ERROR("alphaToCoverageEnabled isn't supported (yet)"); + if (descriptor->alphaToCoverageEnabled && descriptor->sampleCount <= 1) { + return DAWN_VALIDATION_ERROR("Enabling alphaToCoverage requires sampleCount > 1"); } return {}; @@ -571,6 +567,16 @@ namespace dawn_native { return mAttachmentState->GetSampleCount(); } + uint32_t RenderPipelineBase::GetSampleMask() const { + ASSERT(!IsError()); + return mSampleMask; + } + + bool RenderPipelineBase::IsAlphaToCoverageEnabled() const { + ASSERT(!IsError()); + return mAlphaToCoverageEnabled; + } + const AttachmentState* RenderPipelineBase::GetAttachmentState() const { ASSERT(!IsError()); diff --git a/chromium/third_party/dawn/src/dawn_native/RenderPipeline.h b/chromium/third_party/dawn/src/dawn_native/RenderPipeline.h index 9cc76dea82d..bdc3af0fd9a 100644 --- a/chromium/third_party/dawn/src/dawn_native/RenderPipeline.h +++ b/chromium/third_party/dawn/src/dawn_native/RenderPipeline.h @@ -76,6 +76,8 @@ namespace dawn_native { wgpu::TextureFormat GetColorAttachmentFormat(uint32_t attachment) const; wgpu::TextureFormat GetDepthStencilFormat() const; uint32_t GetSampleCount() const; + uint32_t GetSampleMask() const; + bool IsAlphaToCoverageEnabled() const; const AttachmentState* GetAttachmentState() const; diff --git a/chromium/third_party/dawn/src/dawn_native/ResourceHeap.h b/chromium/third_party/dawn/src/dawn_native/ResourceHeap.h index 6e6af795122..e9a4a672263 100644 --- a/chromium/third_party/dawn/src/dawn_native/ResourceHeap.h +++ b/chromium/third_party/dawn/src/dawn_native/ResourceHeap.h @@ -28,4 +28,4 @@ namespace dawn_native { } // namespace dawn_native -#endif // DAWNNATIVE_RESOURCEHEAP_H_
\ No newline at end of file +#endif // DAWNNATIVE_RESOURCEHEAP_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/ResourceMemoryAllocation.cpp b/chromium/third_party/dawn/src/dawn_native/ResourceMemoryAllocation.cpp index 0d726c9d66c..b1c35d41727 100644 --- a/chromium/third_party/dawn/src/dawn_native/ResourceMemoryAllocation.cpp +++ b/chromium/third_party/dawn/src/dawn_native/ResourceMemoryAllocation.cpp @@ -50,4 +50,4 @@ namespace dawn_native { mResourceHeap = nullptr; mInfo = {}; } -} // namespace dawn_native
\ No newline at end of file +} // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/RingBufferAllocator.h b/chromium/third_party/dawn/src/dawn_native/RingBufferAllocator.h index e437632d7c4..50b0f15d001 100644 --- a/chromium/third_party/dawn/src/dawn_native/RingBufferAllocator.h +++ b/chromium/third_party/dawn/src/dawn_native/RingBufferAllocator.h @@ -56,4 +56,4 @@ namespace dawn_native { }; } // namespace dawn_native -#endif // DAWNNATIVE_RINGBUFFERALLOCATOR_H_
\ No newline at end of file +#endif // DAWNNATIVE_RINGBUFFERALLOCATOR_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/ShaderModule.cpp b/chromium/third_party/dawn/src/dawn_native/ShaderModule.cpp index 04b6dfc2079..6057e70d6eb 100644 --- a/chromium/third_party/dawn/src/dawn_native/ShaderModule.cpp +++ b/chromium/third_party/dawn/src/dawn_native/ShaderModule.cpp @@ -39,14 +39,14 @@ namespace dawn_native { Format::Type SpirvCrossBaseTypeToFormatType(spirv_cross::SPIRType::BaseType spirvBaseType) { switch (spirvBaseType) { case spirv_cross::SPIRType::Float: - return Format::Float; + return Format::Type::Float; case spirv_cross::SPIRType::Int: - return Format::Sint; + return Format::Type::Sint; case spirv_cross::SPIRType::UInt: - return Format::Uint; + return Format::Type::Uint; default: UNREACHABLE(); - return Format::Other; + return Format::Type::Other; } } @@ -194,7 +194,7 @@ namespace dawn_native { case spv::ImageFormatRgb10A2: return wgpu::TextureFormat::RGB10A2Unorm; case spv::ImageFormatR11fG11fB10f: - return wgpu::TextureFormat::RG11B10Float; + return wgpu::TextureFormat::RG11B10Ufloat; case spv::ImageFormatRg32f: return wgpu::TextureFormat::RG32Float; case spv::ImageFormatRg32ui: @@ -265,7 +265,7 @@ namespace dawn_native { case shaderc_spvc_storage_texture_format_rgb10a2unorm: return wgpu::TextureFormat::RGB10A2Unorm; case shaderc_spvc_storage_texture_format_rg11b10float: - return wgpu::TextureFormat::RG11B10Float; + return wgpu::TextureFormat::RG11B10Ufloat; case shaderc_spvc_storage_texture_format_rg32float: return wgpu::TextureFormat::RG32Float; case shaderc_spvc_storage_texture_format_rg32uint: @@ -295,6 +295,82 @@ namespace dawn_native { << " binding " << static_cast<uint32_t>(binding); return ostream.str(); } + +#ifdef DAWN_ENABLE_WGSL + tint::ast::transform::VertexFormat ToTintVertexFormat(wgpu::VertexFormat format) { + switch (format) { + case wgpu::VertexFormat::UChar2: + return tint::ast::transform::VertexFormat::kVec2U8; + case wgpu::VertexFormat::UChar4: + return tint::ast::transform::VertexFormat::kVec4U8; + case wgpu::VertexFormat::Char2: + return tint::ast::transform::VertexFormat::kVec2I8; + case wgpu::VertexFormat::Char4: + return tint::ast::transform::VertexFormat::kVec4I8; + case wgpu::VertexFormat::UChar2Norm: + return tint::ast::transform::VertexFormat::kVec2U8Norm; + case wgpu::VertexFormat::UChar4Norm: + return tint::ast::transform::VertexFormat::kVec4U8Norm; + case wgpu::VertexFormat::Char2Norm: + return tint::ast::transform::VertexFormat::kVec2I8Norm; + case wgpu::VertexFormat::Char4Norm: + return tint::ast::transform::VertexFormat::kVec4I8Norm; + case wgpu::VertexFormat::UShort2: + return tint::ast::transform::VertexFormat::kVec2U16; + case wgpu::VertexFormat::UShort4: + return tint::ast::transform::VertexFormat::kVec4U16; + case wgpu::VertexFormat::Short2: + return tint::ast::transform::VertexFormat::kVec2I16; + case wgpu::VertexFormat::Short4: + return tint::ast::transform::VertexFormat::kVec4I16; + case wgpu::VertexFormat::UShort2Norm: + return tint::ast::transform::VertexFormat::kVec2U16Norm; + case wgpu::VertexFormat::UShort4Norm: + return tint::ast::transform::VertexFormat::kVec4U16Norm; + case wgpu::VertexFormat::Short2Norm: + return tint::ast::transform::VertexFormat::kVec2I16Norm; + case wgpu::VertexFormat::Short4Norm: + return tint::ast::transform::VertexFormat::kVec4I16Norm; + case wgpu::VertexFormat::Half2: + return tint::ast::transform::VertexFormat::kVec2F16; + case wgpu::VertexFormat::Half4: + return tint::ast::transform::VertexFormat::kVec4F16; + case wgpu::VertexFormat::Float: + return tint::ast::transform::VertexFormat::kF32; + case wgpu::VertexFormat::Float2: + return tint::ast::transform::VertexFormat::kVec2F32; + case wgpu::VertexFormat::Float3: + return tint::ast::transform::VertexFormat::kVec3F32; + case wgpu::VertexFormat::Float4: + return tint::ast::transform::VertexFormat::kVec4F32; + case wgpu::VertexFormat::UInt: + return tint::ast::transform::VertexFormat::kU32; + case wgpu::VertexFormat::UInt2: + return tint::ast::transform::VertexFormat::kVec2U32; + case wgpu::VertexFormat::UInt3: + return tint::ast::transform::VertexFormat::kVec3U32; + case wgpu::VertexFormat::UInt4: + return tint::ast::transform::VertexFormat::kVec4U32; + case wgpu::VertexFormat::Int: + return tint::ast::transform::VertexFormat::kI32; + case wgpu::VertexFormat::Int2: + return tint::ast::transform::VertexFormat::kVec2I32; + case wgpu::VertexFormat::Int3: + return tint::ast::transform::VertexFormat::kVec3I32; + case wgpu::VertexFormat::Int4: + return tint::ast::transform::VertexFormat::kVec4I32; + } + } + + tint::ast::transform::InputStepMode ToTintInputStepMode(wgpu::InputStepMode mode) { + switch (mode) { + case wgpu::InputStepMode::Vertex: + return tint::ast::transform::InputStepMode::kVertex; + case wgpu::InputStepMode::Instance: + return tint::ast::transform::InputStepMode::kInstance; + } + } +#endif } // anonymous namespace MaybeError ValidateSpirv(DeviceBase*, const uint32_t* code, uint32_t codeSize) { @@ -357,7 +433,7 @@ namespace dawn_native { } tint::Validator validator; - if (!validator.Validate(module)) { + if (!validator.Validate(&module)) { errorStream << "Validation: " << validator.error() << std::endl; return DAWN_VALIDATION_ERROR(errorStream.str().c_str()); } @@ -400,6 +476,75 @@ namespace dawn_native { std::vector<uint32_t> spirv = generator.result(); return std::move(spirv); } + + ResultOrError<std::vector<uint32_t>> ConvertWGSLToSPIRVWithPulling( + const char* source, + const VertexStateDescriptor& vertexState, + const std::string& entryPoint, + uint32_t pullingBufferBindingSet) { + std::ostringstream errorStream; + errorStream << "Tint WGSL->SPIR-V failure:" << std::endl; + + tint::Context context; + tint::reader::wgsl::Parser parser(&context, source); + + // TODO: This is a duplicate parse with ValidateWGSL, need to store + // state between calls to avoid this. + if (!parser.Parse()) { + errorStream << "Parser: " << parser.error() << std::endl; + return DAWN_VALIDATION_ERROR(errorStream.str().c_str()); + } + + tint::ast::Module module = parser.module(); + if (!module.IsValid()) { + errorStream << "Invalid module generated..." << std::endl; + return DAWN_VALIDATION_ERROR(errorStream.str().c_str()); + } + + tint::ast::transform::VertexPullingTransform transform(&context, &module); + auto state = std::make_unique<tint::ast::transform::VertexStateDescriptor>(); + for (uint32_t i = 0; i < vertexState.vertexBufferCount; ++i) { + auto& vertexBuffer = vertexState.vertexBuffers[i]; + tint::ast::transform::VertexBufferLayoutDescriptor layout; + layout.array_stride = vertexBuffer.arrayStride; + layout.step_mode = ToTintInputStepMode(vertexBuffer.stepMode); + + for (uint32_t j = 0; j < vertexBuffer.attributeCount; ++j) { + auto& attribute = vertexBuffer.attributes[j]; + tint::ast::transform::VertexAttributeDescriptor attr; + attr.format = ToTintVertexFormat(attribute.format); + attr.offset = attribute.offset; + attr.shader_location = attribute.shaderLocation; + + layout.attributes.push_back(std::move(attr)); + } + + state->vertex_buffers.push_back(std::move(layout)); + } + transform.SetVertexState(std::move(state)); + transform.SetEntryPoint(entryPoint); + transform.SetPullingBufferBindingSet(pullingBufferBindingSet); + + if (!transform.Run()) { + errorStream << "Vertex pulling transform: " << transform.GetError(); + return DAWN_VALIDATION_ERROR(errorStream.str().c_str()); + } + + tint::TypeDeterminer type_determiner(&context, &module); + if (!type_determiner.Determine()) { + errorStream << "Type Determination: " << type_determiner.error(); + return DAWN_VALIDATION_ERROR(errorStream.str().c_str()); + } + + tint::writer::spirv::Generator generator(std::move(module)); + if (!generator.Generate()) { + errorStream << "Generator: " << generator.error() << std::endl; + return DAWN_VALIDATION_ERROR(errorStream.str().c_str()); + } + + std::vector<uint32_t> spirv = generator.result(); + return std::move(spirv); + } #endif // DAWN_ENABLE_WGSL MaybeError ValidateShaderModuleDescriptor(DeviceBase* device, @@ -463,7 +608,7 @@ namespace dawn_native { UNREACHABLE(); } - mFragmentOutputFormatBaseTypes.fill(Format::Other); + mFragmentOutputFormatBaseTypes.fill(Format::Type::Other); if (GetDevice()->IsToggleEnabled(Toggle::UseSpvcParser)) { mSpvcContext.SetUseSpvcParser(true); } @@ -1094,9 +1239,22 @@ namespace dawn_native { return mSpirv; } +#ifdef DAWN_ENABLE_WGSL + ResultOrError<std::vector<uint32_t>> ShaderModuleBase::GeneratePullingSpirv( + const VertexStateDescriptor& vertexState, + const std::string& entryPoint, + uint32_t pullingBufferBindingSet) const { + return ConvertWGSLToSPIRVWithPulling(mWgsl.c_str(), vertexState, entryPoint, + pullingBufferBindingSet); + } +#endif + shaderc_spvc::CompileOptions ShaderModuleBase::GetCompileOptions() const { shaderc_spvc::CompileOptions options; options.SetValidate(GetDevice()->IsValidationEnabled()); + options.SetRobustBufferAccessPass(GetDevice()->IsRobustnessEnabled()); + options.SetSourceEnvironment(shaderc_target_env_vulkan, shaderc_env_version_vulkan_1_1); + options.SetTargetEnvironment(shaderc_target_env_vulkan, shaderc_env_version_vulkan_1_1); return options; } diff --git a/chromium/third_party/dawn/src/dawn_native/ShaderModule.h b/chromium/third_party/dawn/src/dawn_native/ShaderModule.h index 4e771aa8335..336551ee85e 100644 --- a/chromium/third_party/dawn/src/dawn_native/ShaderModule.h +++ b/chromium/third_party/dawn/src/dawn_native/ShaderModule.h @@ -91,6 +91,13 @@ namespace dawn_native { shaderc_spvc::Context* GetContext(); const std::vector<uint32_t>& GetSpirv() const; +#ifdef DAWN_ENABLE_WGSL + ResultOrError<std::vector<uint32_t>> GeneratePullingSpirv( + const VertexStateDescriptor& vertexState, + const std::string& entryPoint, + uint32_t pullingBufferBindingSet) const; +#endif + protected: static MaybeError CheckSpvcSuccess(shaderc_spvc_status status, const char* error_msg); shaderc_spvc::CompileOptions GetCompileOptions() const; diff --git a/chromium/third_party/dawn/src/dawn_native/StagingBuffer.cpp b/chromium/third_party/dawn/src/dawn_native/StagingBuffer.cpp index 51f5fa8c5e5..63dd65e9bf7 100644 --- a/chromium/third_party/dawn/src/dawn_native/StagingBuffer.cpp +++ b/chromium/third_party/dawn/src/dawn_native/StagingBuffer.cpp @@ -26,4 +26,4 @@ namespace dawn_native { void* StagingBufferBase::GetMappedPointer() const { return mMappedPointer; } -} // namespace dawn_native
\ No newline at end of file +} // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/StagingBuffer.h b/chromium/third_party/dawn/src/dawn_native/StagingBuffer.h index 1da8900a128..4d195488bc9 100644 --- a/chromium/third_party/dawn/src/dawn_native/StagingBuffer.h +++ b/chromium/third_party/dawn/src/dawn_native/StagingBuffer.h @@ -38,4 +38,4 @@ namespace dawn_native { } // namespace dawn_native -#endif // DAWNNATIVE_STAGINGBUFFER_H_
\ No newline at end of file +#endif // DAWNNATIVE_STAGINGBUFFER_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/Surface.cpp b/chromium/third_party/dawn/src/dawn_native/Surface.cpp index 382bb71c6a5..ccd240c9dfc 100644 --- a/chromium/third_party/dawn/src/dawn_native/Surface.cpp +++ b/chromium/third_party/dawn/src/dawn_native/Surface.cpp @@ -99,7 +99,7 @@ namespace dawn_native { } #endif // defined(DAWN_USE_X11) - case wgpu::SType::SurfaceDescriptorFromHTMLCanvasId: + case wgpu::SType::SurfaceDescriptorFromCanvasHTMLSelector: default: return DAWN_VALIDATION_ERROR("Unsupported sType"); } diff --git a/chromium/third_party/dawn/src/dawn_native/Texture.cpp b/chromium/third_party/dawn/src/dawn_native/Texture.cpp index f029981fb04..8f93b46ddf9 100644 --- a/chromium/third_party/dawn/src/dawn_native/Texture.cpp +++ b/chromium/third_party/dawn/src/dawn_native/Texture.cpp @@ -20,6 +20,7 @@ #include "common/Constants.h" #include "common/Math.h" #include "dawn_native/Device.h" +#include "dawn_native/EnumMaskIterator.h" #include "dawn_native/PassResourceUsage.h" #include "dawn_native/ValidationUtils_autogen.h" @@ -200,6 +201,11 @@ namespace dawn_native { return {}; } + uint8_t GetPlaneIndex(Aspect aspect) { + ASSERT(HasOneBit(aspect)); + return static_cast<uint8_t>(Log2(static_cast<uint32_t>(aspect))); + } + } // anonymous namespace MaybeError ValidateTextureDescriptor(const DeviceBase* device, @@ -327,25 +333,6 @@ namespace dawn_native { return desc; } - ResultOrError<TextureDescriptor> FixTextureDescriptor(DeviceBase* device, - const TextureDescriptor* desc) { - TextureDescriptor fixedDesc = *desc; - - if (desc->arrayLayerCount != 1) { - if (desc->size.depth != 1) { - return DAWN_VALIDATION_ERROR("arrayLayerCount and size.depth cannot both be != 1"); - } else { - fixedDesc.size.depth = fixedDesc.arrayLayerCount; - fixedDesc.arrayLayerCount = 1; - device->EmitDeprecationWarning( - "wgpu::TextureDescriptor::arrayLayerCount is deprecated in favor of " - "::size::depth"); - } - } - - return {std::move(fixedDesc)}; - } - bool IsValidSampleCount(uint32_t sampleCount) { switch (sampleCount) { case 1: @@ -357,10 +344,33 @@ namespace dawn_native { } } + Aspect ConvertSingleAspect(const Format& format, wgpu::TextureAspect aspect) { + Aspect aspectMask = ConvertAspect(format, aspect); + ASSERT(HasOneBit(aspectMask)); + return aspectMask; + } + + Aspect ConvertAspect(const Format& format, wgpu::TextureAspect aspect) { + switch (aspect) { + case wgpu::TextureAspect::All: + return format.aspects; + case wgpu::TextureAspect::DepthOnly: + ASSERT(format.aspects & Aspect::Depth); + return Aspect::Depth; + case wgpu::TextureAspect::StencilOnly: + ASSERT(format.aspects & Aspect::Stencil); + return Aspect::Stencil; + default: + UNREACHABLE(); + break; + } + } + // static - SubresourceRange SubresourceRange::SingleSubresource(uint32_t baseMipLevel, - uint32_t baseArrayLayer) { - return {baseMipLevel, 1, baseArrayLayer, 1}; + SubresourceRange SubresourceRange::SingleMipAndLayer(uint32_t baseMipLevel, + uint32_t baseArrayLayer, + Aspect aspects) { + return {baseMipLevel, 1, baseArrayLayer, 1, aspects}; } // TextureBase @@ -376,7 +386,13 @@ namespace dawn_native { mSampleCount(descriptor->sampleCount), mUsage(descriptor->usage), mState(state) { - uint32_t subresourceCount = GetSubresourceCount(); + uint8_t planeIndex = 0; + for (Aspect aspect : IterateEnumMask(mFormat.aspects)) { + mPlaneIndices[GetPlaneIndex(aspect)] = planeIndex++; + } + + uint32_t subresourceCount = + mMipLevelCount * mSize.depth * static_cast<uint32_t>(planeIndex); mIsSubresourceContentInitializedAtIndex = std::vector<bool>(subresourceCount, false); // Add readonly storage usage if the texture has a storage usage. The validation rules in @@ -438,7 +454,7 @@ namespace dawn_native { } SubresourceRange TextureBase::GetAllSubresources() const { ASSERT(!IsError()); - return {0, mMipLevelCount, 0, GetArrayLayers()}; + return {0, mMipLevelCount, 0, GetArrayLayers(), mFormat.aspects}; } uint32_t TextureBase::GetSampleCount() const { ASSERT(!IsError()); @@ -446,7 +462,7 @@ namespace dawn_native { } uint32_t TextureBase::GetSubresourceCount() const { ASSERT(!IsError()); - return mMipLevelCount * mSize.depth; + return static_cast<uint32_t>(mIsSubresourceContentInitializedAtIndex.size()); } wgpu::TextureUsage TextureBase::GetUsage() const { ASSERT(!IsError()); @@ -458,25 +474,32 @@ namespace dawn_native { return mState; } - uint32_t TextureBase::GetSubresourceIndex(uint32_t mipLevel, uint32_t arraySlice) const { + uint32_t TextureBase::GetSubresourceIndex(uint32_t mipLevel, + uint32_t arraySlice, + Aspect aspect) const { ASSERT(arraySlice <= kMaxTexture2DArrayLayers); ASSERT(mipLevel <= kMaxTexture2DMipLevels); + ASSERT(HasOneBit(aspect)); static_assert(kMaxTexture2DMipLevels <= std::numeric_limits<uint32_t>::max() / kMaxTexture2DArrayLayers, "texture size overflows uint32_t"); - return GetNumMipLevels() * arraySlice + mipLevel; + return mipLevel + + GetNumMipLevels() * + (arraySlice + GetArrayLayers() * mPlaneIndices[GetPlaneIndex(aspect)]); } bool TextureBase::IsSubresourceContentInitialized(const SubresourceRange& range) const { ASSERT(!IsError()); - for (uint32_t arrayLayer = range.baseArrayLayer; - arrayLayer < range.baseArrayLayer + range.layerCount; ++arrayLayer) { - for (uint32_t mipLevel = range.baseMipLevel; - mipLevel < range.baseMipLevel + range.levelCount; ++mipLevel) { - uint32_t subresourceIndex = GetSubresourceIndex(mipLevel, arrayLayer); - ASSERT(subresourceIndex < mIsSubresourceContentInitializedAtIndex.size()); - if (!mIsSubresourceContentInitializedAtIndex[subresourceIndex]) { - return false; + for (Aspect aspect : IterateEnumMask(range.aspects)) { + for (uint32_t arrayLayer = range.baseArrayLayer; + arrayLayer < range.baseArrayLayer + range.layerCount; ++arrayLayer) { + for (uint32_t mipLevel = range.baseMipLevel; + mipLevel < range.baseMipLevel + range.levelCount; ++mipLevel) { + uint32_t subresourceIndex = GetSubresourceIndex(mipLevel, arrayLayer, aspect); + ASSERT(subresourceIndex < mIsSubresourceContentInitializedAtIndex.size()); + if (!mIsSubresourceContentInitializedAtIndex[subresourceIndex]) { + return false; + } } } } @@ -486,13 +509,15 @@ namespace dawn_native { void TextureBase::SetIsSubresourceContentInitialized(bool isInitialized, const SubresourceRange& range) { ASSERT(!IsError()); - for (uint32_t arrayLayer = range.baseArrayLayer; - arrayLayer < range.baseArrayLayer + range.layerCount; ++arrayLayer) { - for (uint32_t mipLevel = range.baseMipLevel; - mipLevel < range.baseMipLevel + range.levelCount; ++mipLevel) { - uint32_t subresourceIndex = GetSubresourceIndex(mipLevel, arrayLayer); - ASSERT(subresourceIndex < mIsSubresourceContentInitializedAtIndex.size()); - mIsSubresourceContentInitializedAtIndex[subresourceIndex] = isInitialized; + for (Aspect aspect : IterateEnumMask(range.aspects)) { + for (uint32_t arrayLayer = range.baseArrayLayer; + arrayLayer < range.baseArrayLayer + range.layerCount; ++arrayLayer) { + for (uint32_t mipLevel = range.baseMipLevel; + mipLevel < range.baseMipLevel + range.levelCount; ++mipLevel) { + uint32_t subresourceIndex = GetSubresourceIndex(mipLevel, arrayLayer, aspect); + ASSERT(subresourceIndex < mIsSubresourceContentInitializedAtIndex.size()); + mIsSubresourceContentInitializedAtIndex[subresourceIndex] = isInitialized; + } } } } @@ -541,6 +566,19 @@ namespace dawn_native { return extent; } + Extent3D TextureBase::ClampToMipLevelVirtualSize(uint32_t level, + const Origin3D& origin, + const Extent3D& extent) const { + const Extent3D virtualSizeAtLevel = GetMipLevelVirtualSize(level); + uint32_t clampedCopyExtentWidth = (origin.x + extent.width > virtualSizeAtLevel.width) + ? (virtualSizeAtLevel.width - origin.x) + : extent.width; + uint32_t clampedCopyExtentHeight = (origin.y + extent.height > virtualSizeAtLevel.height) + ? (virtualSizeAtLevel.height - origin.y) + : extent.height; + return {clampedCopyExtentWidth, clampedCopyExtentHeight, extent.depth}; + } + TextureViewBase* TextureBase::CreateView(const TextureViewDescriptor* descriptor) { return GetDevice()->CreateTextureView(this, descriptor); } @@ -571,10 +609,13 @@ namespace dawn_native { TextureViewBase::TextureViewBase(TextureBase* texture, const TextureViewDescriptor* descriptor) : ObjectBase(texture->GetDevice()), mTexture(texture), + mAspect(descriptor->aspect), mFormat(GetDevice()->GetValidInternalFormat(descriptor->format)), mDimension(descriptor->dimension), mRange({descriptor->baseMipLevel, descriptor->mipLevelCount, descriptor->baseArrayLayer, - descriptor->arrayLayerCount}) { + descriptor->arrayLayerCount, ConvertAspect(mFormat, mAspect)}) { + // TODO(crbug.com/dawn/439): Current validation only allows texture views with aspect "all". + ASSERT(mAspect == wgpu::TextureAspect::All); } TextureViewBase::TextureViewBase(DeviceBase* device, ObjectBase::ErrorTag tag) @@ -596,6 +637,11 @@ namespace dawn_native { return mTexture.Get(); } + wgpu::TextureAspect TextureViewBase::GetAspect() const { + ASSERT(!IsError()); + return mAspect; + } + const Format& TextureViewBase::GetFormat() const { ASSERT(!IsError()); return mFormat; @@ -630,4 +676,5 @@ namespace dawn_native { ASSERT(!IsError()); return mRange; } + } // namespace dawn_native diff --git a/chromium/third_party/dawn/src/dawn_native/Texture.h b/chromium/third_party/dawn/src/dawn_native/Texture.h index 29007941344..252f11f64a2 100644 --- a/chromium/third_party/dawn/src/dawn_native/Texture.h +++ b/chromium/third_party/dawn/src/dawn_native/Texture.h @@ -15,6 +15,9 @@ #ifndef DAWNNATIVE_TEXTURE_H_ #define DAWNNATIVE_TEXTURE_H_ +#include "common/ityp_array.h" +#include "common/ityp_bitset.h" +#include "dawn_native/EnumClassBitmasks.h" #include "dawn_native/Error.h" #include "dawn_native/Forward.h" #include "dawn_native/ObjectBase.h" @@ -24,6 +27,35 @@ #include <vector> namespace dawn_native { + + // Note: Subresource indices are computed by iterating the aspects in increasing order. + // D3D12 uses these directly, so the order much match D3D12's indices. + // - Depth/Stencil textures have Depth as Plane 0, and Stencil as Plane 1. + enum class Aspect : uint8_t { + None = 0x0, + Color = 0x1, + Depth = 0x2, + Stencil = 0x4, + }; + + template <> + struct EnumBitmaskSize<Aspect> { + static constexpr unsigned value = 3; + }; + +} // namespace dawn_native + +namespace wgpu { + + template <> + struct IsDawnBitmask<dawn_native::Aspect> { + static constexpr bool enable = true; + }; + +} // namespace wgpu + +namespace dawn_native { + MaybeError ValidateTextureDescriptor(const DeviceBase* device, const TextureDescriptor* descriptor); MaybeError ValidateTextureViewDescriptor(const TextureBase* texture, @@ -32,11 +64,6 @@ namespace dawn_native { const TextureBase* texture, const TextureViewDescriptor* descriptor); - // TODO(dawn:22): Remove once migration from GPUTextureDescriptor.arrayLayerCount to - // GPUTextureDescriptor.size.depth is done. - ResultOrError<TextureDescriptor> FixTextureDescriptor(DeviceBase* device, - const TextureDescriptor* desc); - bool IsValidSampleCount(uint32_t sampleCount); static constexpr wgpu::TextureUsage kReadOnlyTextureUsages = @@ -46,13 +73,19 @@ namespace dawn_native { wgpu::TextureUsage::CopyDst | wgpu::TextureUsage::Storage | wgpu::TextureUsage::OutputAttachment; + Aspect ConvertSingleAspect(const Format& format, wgpu::TextureAspect aspect); + Aspect ConvertAspect(const Format& format, wgpu::TextureAspect aspect); + struct SubresourceRange { uint32_t baseMipLevel; uint32_t levelCount; uint32_t baseArrayLayer; uint32_t layerCount; + Aspect aspects; - static SubresourceRange SingleSubresource(uint32_t baseMipLevel, uint32_t baseArrayLayer); + static SubresourceRange SingleMipAndLayer(uint32_t baseMipLevel, + uint32_t baseArrayLayer, + Aspect aspects); }; class TextureBase : public ObjectBase { @@ -76,7 +109,7 @@ namespace dawn_native { uint32_t GetSubresourceCount() const; wgpu::TextureUsage GetUsage() const; TextureState GetTextureState() const; - uint32_t GetSubresourceIndex(uint32_t mipLevel, uint32_t arraySlice) const; + uint32_t GetSubresourceIndex(uint32_t mipLevel, uint32_t arraySlice, Aspect aspect) const; bool IsSubresourceContentInitialized(const SubresourceRange& range) const; void SetIsSubresourceContentInitialized(bool isInitialized, const SubresourceRange& range); @@ -91,6 +124,9 @@ namespace dawn_native { // required to be a multiple of the block size and used in texture sampling. Extent3D GetMipLevelPhysicalSize(uint32_t level) const; Extent3D GetMipLevelVirtualSize(uint32_t level) const; + Extent3D ClampToMipLevelVirtualSize(uint32_t level, + const Origin3D& origin, + const Extent3D& extent) const; // Dawn API TextureViewBase* CreateView(const TextureViewDescriptor* descriptor); @@ -115,6 +151,7 @@ namespace dawn_native { // TODO(natlee@microsoft.com): Use a more optimized data structure to save space std::vector<bool> mIsSubresourceContentInitializedAtIndex; + std::array<uint8_t, EnumBitmaskSize<Aspect>::value> mPlaneIndices; }; class TextureViewBase : public ObjectBase { @@ -126,6 +163,7 @@ namespace dawn_native { const TextureBase* GetTexture() const; TextureBase* GetTexture(); + wgpu::TextureAspect GetAspect() const; const Format& GetFormat() const; wgpu::TextureViewDimension GetDimension() const; uint32_t GetBaseMipLevel() const; @@ -139,6 +177,7 @@ namespace dawn_native { Ref<TextureBase> mTexture; + wgpu::TextureAspect mAspect; // TODO(cwallez@chromium.org): This should be deduplicated in the Device const Format& mFormat; wgpu::TextureViewDimension mDimension; diff --git a/chromium/third_party/dawn/src/dawn_native/Toggles.cpp b/chromium/third_party/dawn/src/dawn_native/Toggles.cpp index b4ec0db0e31..d54dc094bac 100644 --- a/chromium/third_party/dawn/src/dawn_native/Toggles.cpp +++ b/chromium/third_party/dawn/src/dawn_native/Toggles.cpp @@ -29,109 +29,120 @@ namespace dawn_native { using ToggleEnumAndInfoList = std::array<ToggleEnumAndInfo, static_cast<size_t>(Toggle::EnumCount)>; - static constexpr ToggleEnumAndInfoList kToggleNameAndInfoList = {{ - {Toggle::EmulateStoreAndMSAAResolve, - {"emulate_store_and_msaa_resolve", - "Emulate storing into multisampled color attachments and doing MSAA resolve " - "simultaneously. This workaround is enabled by default on the Metal drivers that do " - "not support MTLStoreActionStoreAndMultisampleResolve. To support StoreOp::Store on " - "those platforms, we should do MSAA resolve in another render pass after ending the " - "previous one.", - "https://crbug.com/dawn/56"}}, - {Toggle::NonzeroClearResourcesOnCreationForTesting, - {"nonzero_clear_resources_on_creation_for_testing", - "Clears texture to full 1 bits as soon as they are created, but doesn't update " - "the tracking state of the texture. This way we can test the logic of clearing " - "textures that use recycled memory.", - "https://crbug.com/dawn/145"}}, - {Toggle::AlwaysResolveIntoZeroLevelAndLayer, - {"always_resolve_into_zero_level_and_layer", - "When the resolve target is a texture view that is created on the non-zero level or " - "layer of a texture, we first resolve into a temporarily 2D texture with only one " - "mipmap level and one array layer, and copy the result of MSAA resolve into the " - "true resolve target. This workaround is enabled by default on the Metal drivers " - "that have bugs when setting non-zero resolveLevel or resolveSlice.", - "https://crbug.com/dawn/56"}}, - {Toggle::LazyClearResourceOnFirstUse, - {"lazy_clear_resource_on_first_use", - "Clears resource to zero on first usage. This initializes the resource " - "so that no dirty bits from recycled memory is present in the new resource.", - "https://crbug.com/dawn/145"}}, - {Toggle::TurnOffVsync, - {"turn_off_vsync", - "Turn off vsync when rendering. In order to do performance test or run perf tests, " - "turn off vsync so that the fps can exeed 60.", - "https://crbug.com/dawn/237"}}, - {Toggle::UseTemporaryBufferInCompressedTextureToTextureCopy, - {"use_temporary_buffer_in_texture_to_texture_copy", - "Split texture-to-texture copy into two copies: copy from source texture into a " - "temporary buffer, and copy from the temporary buffer into the destination texture " - "when copying between compressed textures that don't have block-aligned sizes. This " - "workaround is enabled by default on all Vulkan drivers to solve an issue in the " - "Vulkan SPEC about the texture-to-texture copies with compressed formats. See #1005 " - "(https://github.com/KhronosGroup/Vulkan-Docs/issues/1005) for more details.", - "https://crbug.com/dawn/42"}}, - {Toggle::UseD3D12ResourceHeapTier2, - {"use_d3d12_resource_heap_tier2", - "Enable support for resource heap tier 2. Resource heap tier 2 allows mixing of " - "texture and buffers in the same heap. This allows better heap re-use and reduces " - "fragmentation.", - "https://crbug.com/dawn/27"}}, - {Toggle::UseD3D12RenderPass, - {"use_d3d12_render_pass", - "Use the D3D12 render pass API introduced in Windows build 1809 by default. On " - "versions of Windows prior to build 1809, or when this toggle is turned off, Dawn " - "will emulate a render pass.", - "https://crbug.com/dawn/36"}}, - {Toggle::UseD3D12ResidencyManagement, - {"use_d3d12_residency_management", - "Enable residency management. This allows page-in and page-out of resource heaps in " - "GPU memory. This component improves overcommitted performance by keeping the most " - "recently used resources local to the GPU. Turning this component off can cause " - "allocation failures when application memory exceeds physical device memory.", - "https://crbug.com/dawn/193"}}, - {Toggle::SkipValidation, - {"skip_validation", "Skip expensive validation of Dawn commands.", - "https://crbug.com/dawn/271"}}, - {Toggle::UseSpvc, - {"use_spvc", - "Enable use of spvc for shader compilation, instead of accessing spirv_cross " - "directly.", - "https://crbug.com/dawn/288"}}, - {Toggle::UseSpvcParser, - {"use_spvc_parser", - "Enable usage of spvc's internal parsing and IR generation code, instead of " - "spirv_cross's.", - "https://crbug.com/dawn/288"}}, - {Toggle::VulkanUseD32S8, - {"vulkan_use_d32s8", - "Vulkan mandates support of either D32_FLOAT_S8 or D24_UNORM_S8. When available the " - "backend will use D32S8 (toggle to on) but setting the toggle to off will make it" - "use the D24S8 format when possible.", - "https://crbug.com/dawn/286"}}, - {Toggle::MetalDisableSamplerCompare, - {"metal_disable_sampler_compare", - "Disables the use of sampler compare on Metal. This is unsupported before A9 " - "processors.", - "https://crbug.com/dawn/342"}}, - {Toggle::DisableBaseVertex, - {"disable_base_vertex", - "Disables the use of non-zero base vertex which is unsupported on some platforms.", - "https://crbug.com/dawn/343"}}, - {Toggle::DisableBaseInstance, - {"disable_base_instance", - "Disables the use of non-zero base instance which is unsupported on some " - "platforms.", - "https://crbug.com/dawn/343"}}, - {Toggle::UseD3D12SmallShaderVisibleHeapForTesting, - {"use_d3d12_small_shader_visible_heap", - "Enable use of a small D3D12 shader visible heap, instead of using a large one by " - "default. This setting is used to test bindgroup encoding.", - "https://crbug.com/dawn/155"}}, - {Toggle::UseDXC, - {"use_dxc", "Use DXC instead of FXC for compiling HLSL", - "https://crbug.com/dawn/402"}}, - }}; + static constexpr ToggleEnumAndInfoList kToggleNameAndInfoList = { + {{Toggle::EmulateStoreAndMSAAResolve, + {"emulate_store_and_msaa_resolve", + "Emulate storing into multisampled color attachments and doing MSAA resolve " + "simultaneously. This workaround is enabled by default on the Metal drivers that do " + "not support MTLStoreActionStoreAndMultisampleResolve. To support StoreOp::Store on " + "those platforms, we should do MSAA resolve in another render pass after ending the " + "previous one.", + "https://crbug.com/dawn/56"}}, + {Toggle::NonzeroClearResourcesOnCreationForTesting, + {"nonzero_clear_resources_on_creation_for_testing", + "Clears texture to full 1 bits as soon as they are created, but doesn't update " + "the tracking state of the texture. This way we can test the logic of clearing " + "textures that use recycled memory.", + "https://crbug.com/dawn/145"}}, + {Toggle::AlwaysResolveIntoZeroLevelAndLayer, + {"always_resolve_into_zero_level_and_layer", + "When the resolve target is a texture view that is created on the non-zero level or " + "layer of a texture, we first resolve into a temporarily 2D texture with only one " + "mipmap level and one array layer, and copy the result of MSAA resolve into the " + "true resolve target. This workaround is enabled by default on the Metal drivers " + "that have bugs when setting non-zero resolveLevel or resolveSlice.", + "https://crbug.com/dawn/56"}}, + {Toggle::LazyClearResourceOnFirstUse, + {"lazy_clear_resource_on_first_use", + "Clears resource to zero on first usage. This initializes the resource " + "so that no dirty bits from recycled memory is present in the new resource.", + "https://crbug.com/dawn/145"}}, + {Toggle::TurnOffVsync, + {"turn_off_vsync", + "Turn off vsync when rendering. In order to do performance test or run perf tests, " + "turn off vsync so that the fps can exeed 60.", + "https://crbug.com/dawn/237"}}, + {Toggle::UseTemporaryBufferInCompressedTextureToTextureCopy, + {"use_temporary_buffer_in_texture_to_texture_copy", + "Split texture-to-texture copy into two copies: copy from source texture into a " + "temporary buffer, and copy from the temporary buffer into the destination texture " + "when copying between compressed textures that don't have block-aligned sizes. This " + "workaround is enabled by default on all Vulkan drivers to solve an issue in the " + "Vulkan SPEC about the texture-to-texture copies with compressed formats. See #1005 " + "(https://github.com/KhronosGroup/Vulkan-Docs/issues/1005) for more details.", + "https://crbug.com/dawn/42"}}, + {Toggle::UseD3D12ResourceHeapTier2, + {"use_d3d12_resource_heap_tier2", + "Enable support for resource heap tier 2. Resource heap tier 2 allows mixing of " + "texture and buffers in the same heap. This allows better heap re-use and reduces " + "fragmentation.", + "https://crbug.com/dawn/27"}}, + {Toggle::UseD3D12RenderPass, + {"use_d3d12_render_pass", + "Use the D3D12 render pass API introduced in Windows build 1809 by default. On " + "versions of Windows prior to build 1809, or when this toggle is turned off, Dawn " + "will emulate a render pass.", + "https://crbug.com/dawn/36"}}, + {Toggle::UseD3D12ResidencyManagement, + {"use_d3d12_residency_management", + "Enable residency management. This allows page-in and page-out of resource heaps in " + "GPU memory. This component improves overcommitted performance by keeping the most " + "recently used resources local to the GPU. Turning this component off can cause " + "allocation failures when application memory exceeds physical device memory.", + "https://crbug.com/dawn/193"}}, + {Toggle::SkipValidation, + {"skip_validation", "Skip expensive validation of Dawn commands.", + "https://crbug.com/dawn/271"}}, + {Toggle::UseSpvc, + {"use_spvc", + "Enable use of spvc for shader compilation, instead of accessing spirv_cross " + "directly.", + "https://crbug.com/dawn/288"}}, + {Toggle::UseSpvcParser, + {"use_spvc_parser", + "Enable usage of spvc's internal parsing and IR generation code, instead of " + "spirv_cross's.", + "https://crbug.com/dawn/288"}}, + {Toggle::VulkanUseD32S8, + {"vulkan_use_d32s8", + "Vulkan mandates support of either D32_FLOAT_S8 or D24_UNORM_S8. When available the " + "backend will use D32S8 (toggle to on) but setting the toggle to off will make it" + "use the D24S8 format when possible.", + "https://crbug.com/dawn/286"}}, + {Toggle::MetalDisableSamplerCompare, + {"metal_disable_sampler_compare", + "Disables the use of sampler compare on Metal. This is unsupported before A9 " + "processors.", + "https://crbug.com/dawn/342"}}, + {Toggle::DisableBaseVertex, + {"disable_base_vertex", + "Disables the use of non-zero base vertex which is unsupported on some platforms.", + "https://crbug.com/dawn/343"}}, + {Toggle::DisableBaseInstance, + {"disable_base_instance", + "Disables the use of non-zero base instance which is unsupported on some " + "platforms.", + "https://crbug.com/dawn/343"}}, + {Toggle::UseD3D12SmallShaderVisibleHeapForTesting, + {"use_d3d12_small_shader_visible_heap", + "Enable use of a small D3D12 shader visible heap, instead of using a large one by " + "default. This setting is used to test bindgroup encoding.", + "https://crbug.com/dawn/155"}}, + {Toggle::UseDXC, + {"use_dxc", "Use DXC instead of FXC for compiling HLSL", + "https://crbug.com/dawn/402"}}, + {Toggle::DisableRobustness, + {"disable_robustness", "Disable robust buffer access", "https://crbug.com/dawn/480"}}, + {Toggle::LazyClearBufferOnFirstUse, + {"lazy_clear_buffer_on_first_use", + "Clear buffers on their first use. This is a temporary toggle only for the " + "development of buffer lazy initialization and will be removed after buffer lazy " + "initialization is completely implemented.", + "https://crbug.com/dawn/414"}}, + {Toggle::MetalEnableVertexPulling, + {"metal_enable_vertex_pulling", + "Uses vertex pulling to protect out-of-bounds reads on Metal", + "https://crbug.com/dawn/480"}}}}; } // anonymous namespace diff --git a/chromium/third_party/dawn/src/dawn_native/Toggles.h b/chromium/third_party/dawn/src/dawn_native/Toggles.h index 18a26a9a63d..5186cb56394 100644 --- a/chromium/third_party/dawn/src/dawn_native/Toggles.h +++ b/chromium/third_party/dawn/src/dawn_native/Toggles.h @@ -42,6 +42,9 @@ namespace dawn_native { DisableBaseInstance, UseD3D12SmallShaderVisibleHeapForTesting, UseDXC, + DisableRobustness, + LazyClearBufferOnFirstUse, + MetalEnableVertexPulling, EnumCount, InvalidEnum = EnumCount, diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/AdapterD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/AdapterD3D12.cpp index 68fa27c2e5e..364e6c13e8f 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/AdapterD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/AdapterD3D12.cpp @@ -102,6 +102,9 @@ namespace dawn_native { namespace d3d12 { mSupportedExtensions.EnableExtension(Extension::TextureCompressionBC); mSupportedExtensions.EnableExtension(Extension::PipelineStatisticsQuery); mSupportedExtensions.EnableExtension(Extension::TimestampQuery); + if (mDeviceInfo.supportsShaderFloat16 && GetBackend()->GetFunctions()->IsDXCAvailable()) { + mSupportedExtensions.EnableExtension(Extension::ShaderFloat16); + } } MaybeError Adapter::InitializeDebugLayerFilters() { @@ -141,12 +144,13 @@ namespace dawn_native { namespace d3d12 { // // Remove after warning have been addressed - // https://crbug.com/dawn/419 - D3D12_MESSAGE_ID_UNMAP_RANGE_NOT_EMPTY, - - // Remove after warning have been addressed // https://crbug.com/dawn/421 D3D12_MESSAGE_ID_GPU_BASED_VALIDATION_INCOMPATIBLE_RESOURCE_STATE, + + // For small placed resource alignment, we first request the small alignment, which may + // get rejected and generate a debug error. Then, we request 0 to get the allowed + // allowed alignment. + D3D12_MESSAGE_ID_CREATERESOURCE_INVALIDALIGNMENT, }; // Create a retrieval filter with a deny list to suppress messages. diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/BackendD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/BackendD3D12.cpp index 46dc1b67921..946ce7bd98a 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/BackendD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/BackendD3D12.cpp @@ -26,7 +26,8 @@ namespace dawn_native { namespace d3d12 { ResultOrError<ComPtr<IDXGIFactory4>> CreateFactory(const PlatformFunctions* functions, bool enableBackendValidation, - bool beginCaptureOnStartup) { + bool beginCaptureOnStartup, + bool enableGPUBasedBackendValidation) { ComPtr<IDXGIFactory4> factory; uint32_t dxgiFactoryFlags = 0; @@ -39,7 +40,8 @@ namespace dawn_native { namespace d3d12 { functions->d3d12GetDebugInterface(IID_PPV_ARGS(&debugController)))) { ASSERT(debugController != nullptr); debugController->EnableDebugLayer(); - debugController->SetEnableGPUBasedValidation(true); + debugController->SetEnableGPUBasedValidation( + enableGPUBasedBackendValidation); // Enable additional debug layers. dxgiFactoryFlags |= DXGI_CREATE_FACTORY_DEBUG; @@ -70,6 +72,18 @@ namespace dawn_native { namespace d3d12 { return std::move(factory); } + ResultOrError<std::unique_ptr<AdapterBase>> CreateAdapterFromIDXGIAdapter( + Backend* backend, + ComPtr<IDXGIAdapter> dxgiAdapter) { + ComPtr<IDXGIAdapter3> dxgiAdapter3; + DAWN_TRY(CheckHRESULT(dxgiAdapter.As(&dxgiAdapter3), "DXGIAdapter retrieval")); + std::unique_ptr<Adapter> adapter = + std::make_unique<Adapter>(backend, std::move(dxgiAdapter3)); + DAWN_TRY(adapter->Initialize()); + + return {std::move(adapter)}; + } + } // anonymous namespace Backend::Backend(InstanceBase* instance) @@ -84,7 +98,8 @@ namespace dawn_native { namespace d3d12 { DAWN_TRY_ASSIGN(mFactory, CreateFactory(mFunctions.get(), instance->IsBackendValidationEnabled(), - instance->IsBeginCaptureOnStartupEnabled())); + instance->IsBeginCaptureOnStartupEnabled(), + instance->IsGPUBasedBackendValidationEnabled())); return {}; } @@ -127,23 +142,34 @@ namespace dawn_native { namespace d3d12 { } ASSERT(dxgiAdapter != nullptr); - - ComPtr<IDXGIAdapter3> dxgiAdapter3; - HRESULT result = dxgiAdapter.As(&dxgiAdapter3); - ASSERT(SUCCEEDED(result)); - - std::unique_ptr<Adapter> adapter = - std::make_unique<Adapter>(this, std::move(dxgiAdapter3)); - if (GetInstance()->ConsumedError(adapter->Initialize())) { + ResultOrError<std::unique_ptr<AdapterBase>> adapter = + CreateAdapterFromIDXGIAdapter(this, dxgiAdapter); + if (adapter.IsError()) { + adapter.AcquireError(); continue; } - adapters.push_back(std::move(adapter)); + adapters.push_back(std::move(adapter.AcquireSuccess())); } return adapters; } + ResultOrError<std::vector<std::unique_ptr<AdapterBase>>> Backend::DiscoverAdapters( + const AdapterDiscoveryOptionsBase* optionsBase) { + ASSERT(optionsBase->backendType == WGPUBackendType_D3D12); + const AdapterDiscoveryOptions* options = + static_cast<const AdapterDiscoveryOptions*>(optionsBase); + + ASSERT(options->dxgiAdapter != nullptr); + + std::unique_ptr<AdapterBase> adapter; + DAWN_TRY_ASSIGN(adapter, CreateAdapterFromIDXGIAdapter(this, options->dxgiAdapter)); + std::vector<std::unique_ptr<AdapterBase>> adapters; + adapters.push_back(std::move(adapter)); + return std::move(adapters); + } + BackendConnection* Connect(InstanceBase* instance) { Backend* backend = new Backend(instance); diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/BackendD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/BackendD3D12.h index 27ef1d16a36..87c2d13f04b 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/BackendD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/BackendD3D12.h @@ -35,6 +35,8 @@ namespace dawn_native { namespace d3d12 { const PlatformFunctions* GetFunctions() const; std::vector<std::unique_ptr<AdapterBase>> DiscoverDefaultAdapters() override; + ResultOrError<std::vector<std::unique_ptr<AdapterBase>>> DiscoverAdapters( + const AdapterDiscoveryOptionsBase* optionsBase) override; private: // Keep mFunctions as the first member so that in the destructor it is freed last. Otherwise @@ -47,4 +49,4 @@ namespace dawn_native { namespace d3d12 { }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_BACKENDD3D12_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_BACKENDD3D12_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupD3D12.cpp index 76fb0289796..21d2743e94f 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupD3D12.cpp @@ -88,7 +88,7 @@ namespace dawn_native { namespace d3d12 { desc.Buffer.Flags = D3D12_BUFFER_UAV_FLAG_RAW; d3d12Device->CreateUnorderedAccessView( - ToBackend(binding.buffer)->GetD3D12Resource().Get(), nullptr, &desc, + ToBackend(binding.buffer)->GetD3D12Resource(), nullptr, &desc, viewAllocation.OffsetFrom(viewSizeIncrement, bindingOffsets[bindingIndex])); break; } @@ -108,7 +108,7 @@ namespace dawn_native { namespace d3d12 { desc.Buffer.StructureByteStride = 0; desc.Buffer.Flags = D3D12_BUFFER_SRV_FLAG_RAW; d3d12Device->CreateShaderResourceView( - ToBackend(binding.buffer)->GetD3D12Resource().Get(), &desc, + ToBackend(binding.buffer)->GetD3D12Resource(), &desc, viewAllocation.OffsetFrom(viewSizeIncrement, bindingOffsets[bindingIndex])); break; } diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupLayoutD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupLayoutD3D12.cpp index 9280f8c5f4f..62722f6bd6d 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupLayoutD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupLayoutD3D12.cpp @@ -46,6 +46,7 @@ namespace dawn_native { namespace d3d12 { BindGroupLayout::BindGroupLayout(Device* device, const BindGroupLayoutDescriptor* descriptor) : BindGroupLayoutBase(device, descriptor), + mBindingOffsets(GetBindingCount()), mDescriptorCounts{}, mBindGroupAllocator(MakeFrontendBindGroupAllocator<BindGroup>(4096)) { for (BindingIndex bindingIndex = GetDynamicBufferCount(); bindingIndex < GetBindingCount(); @@ -170,9 +171,8 @@ namespace dawn_native { namespace d3d12 { mBindGroupAllocator.Deallocate(bindGroup); } - const ityp::array<BindingIndex, uint32_t, kMaxBindingsPerGroup>& - BindGroupLayout::GetBindingOffsets() const { - return mBindingOffsets; + ityp::span<BindingIndex, const uint32_t> BindGroupLayout::GetBindingOffsets() const { + return {mBindingOffsets.data(), mBindingOffsets.size()}; } uint32_t BindGroupLayout::GetCbvUavSrvDescriptorTableSize() const { diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupLayoutD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupLayoutD3D12.h index 5a5ba8bce33..c5f32f54f3b 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupLayoutD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/BindGroupLayoutD3D12.h @@ -18,6 +18,7 @@ #include "dawn_native/BindGroupLayout.h" #include "common/SlabAllocator.h" +#include "common/ityp_stack_vec.h" #include "dawn_native/d3d12/d3d12_platform.h" namespace dawn_native { namespace d3d12 { @@ -44,7 +45,7 @@ namespace dawn_native { namespace d3d12 { Count, }; - const ityp::array<BindingIndex, uint32_t, kMaxBindingsPerGroup>& GetBindingOffsets() const; + ityp::span<BindingIndex, const uint32_t> GetBindingOffsets() const; uint32_t GetCbvUavSrvDescriptorTableSize() const; uint32_t GetSamplerDescriptorTableSize() const; uint32_t GetCbvUavSrvDescriptorCount() const; @@ -54,7 +55,7 @@ namespace dawn_native { namespace d3d12 { private: ~BindGroupLayout() override = default; - ityp::array<BindingIndex, uint32_t, kMaxBindingsPerGroup> mBindingOffsets; + ityp::stack_vec<BindingIndex, uint32_t, kMaxOptimalBindingsPerGroup> mBindingOffsets; std::array<uint32_t, DescriptorType::Count> mDescriptorCounts; D3D12_DESCRIPTOR_RANGE mRanges[DescriptorType::Count]; diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/BufferD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/BufferD3D12.cpp index 879709f4bae..66991109990 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/BufferD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/BufferD3D12.cpp @@ -17,6 +17,7 @@ #include "common/Assert.h" #include "common/Constants.h" #include "common/Math.h" +#include "dawn_native/CommandBuffer.h" #include "dawn_native/DynamicUploader.h" #include "dawn_native/d3d12/CommandRecordingContext.h" #include "dawn_native/d3d12/D3D12Error.h" @@ -62,6 +63,13 @@ namespace dawn_native { namespace d3d12 { if (usage & wgpu::BufferUsage::Indirect) { resourceState |= D3D12_RESOURCE_STATE_INDIRECT_ARGUMENT; } + if (usage & wgpu::BufferUsage::QueryResolve) { + // D3D12_RESOURCE_STATE_COPY_DEST is required by ResolveQueryData but we also add + // D3D12_RESOURCE_STATE_UNORDERED_ACCESS because the queries will be post-processed + // by a compute shader and written to this buffer via a UAV. + resourceState |= + (D3D12_RESOURCE_STATE_UNORDERED_ACCESS | D3D12_RESOURCE_STATE_COPY_DEST); + } return resourceState; } @@ -123,7 +131,11 @@ namespace dawn_native { namespace d3d12 { ToBackend(GetDevice())->AllocateMemory(heapType, resourceDescriptor, bufferUsage)); if (GetDevice()->IsToggleEnabled(Toggle::NonzeroClearResourcesOnCreationForTesting)) { - DAWN_TRY(ClearBuffer(ClearValue::NonZero)); + CommandRecordingContext* commandRecordingContext; + DAWN_TRY_ASSIGN(commandRecordingContext, + ToBackend(GetDevice())->GetPendingCommandContext()); + + DAWN_TRY(ClearBuffer(commandRecordingContext, uint8_t(1u))); } return {}; @@ -133,7 +145,7 @@ namespace dawn_native { namespace d3d12 { DestroyInternal(); } - ComPtr<ID3D12Resource> Buffer::GetD3D12Resource() const { + ID3D12Resource* Buffer::GetD3D12Resource() const { return mResourceAllocation.GetD3D12Resource(); } @@ -183,7 +195,7 @@ namespace dawn_native { namespace d3d12 { if (needsUAVBarrier) { barrier->Type = D3D12_RESOURCE_BARRIER_TYPE_UAV; barrier->Flags = D3D12_RESOURCE_BARRIER_FLAG_NONE; - barrier->UAV.pResource = GetD3D12Resource().Get(); + barrier->UAV.pResource = GetD3D12Resource(); mLastUsage = newUsage; return true; @@ -227,7 +239,7 @@ namespace dawn_native { namespace d3d12 { barrier->Type = D3D12_RESOURCE_BARRIER_TYPE_TRANSITION; barrier->Flags = D3D12_RESOURCE_BARRIER_FLAG_NONE; - barrier->Transition.pResource = GetD3D12Resource().Get(); + barrier->Transition.pResource = GetD3D12Resource(); barrier->Transition.StateBefore = lastState; barrier->Transition.StateAfter = newState; barrier->Transition.Subresource = D3D12_RESOURCE_BARRIER_ALL_SUBRESOURCES; @@ -239,79 +251,88 @@ namespace dawn_native { namespace d3d12 { return mResourceAllocation.GetGPUPointer(); } - bool Buffer::IsMapWritable() const { + bool Buffer::IsMappableAtCreation() const { // TODO(enga): Handle CPU-visible memory on UMA return (GetUsage() & (wgpu::BufferUsage::MapRead | wgpu::BufferUsage::MapWrite)) != 0; } - MaybeError Buffer::MapBufferInternal(D3D12_RANGE mappedRange, - void** mappedPointer, - const char* contextInfo) { + MaybeError Buffer::MapInternal(bool isWrite, + size_t offset, + size_t size, + const char* contextInfo) { // The mapped buffer can be accessed at any time, so it must be locked to ensure it is never // evicted. This buffer should already have been made resident when it was created. Heap* heap = ToBackend(mResourceAllocation.GetResourceHeap()); DAWN_TRY(ToBackend(GetDevice())->GetResidencyManager()->LockAllocation(heap)); - DAWN_TRY( - CheckHRESULT(GetD3D12Resource()->Map(0, &mappedRange, mappedPointer), contextInfo)); + D3D12_RANGE range = {offset, offset + size}; + // mMappedData is the pointer to the start of the resource, irrespective of offset. + // MSDN says (note the weird use of "never"): + // + // When ppData is not NULL, the pointer returned is never offset by any values in + // pReadRange. + // + // https://docs.microsoft.com/en-us/windows/win32/api/d3d12/nf-d3d12-id3d12resource-map + DAWN_TRY(CheckHRESULT(GetD3D12Resource()->Map(0, &range, &mMappedData), contextInfo)); + + if (isWrite) { + mWrittenMappedRange = range; + } + return {}; } - void Buffer::UnmapBufferInternal(D3D12_RANGE mappedRange) { - GetD3D12Resource()->Unmap(0, &mappedRange); - - // When buffers are mapped, they are locked to keep them in resident memory. We must unlock - // them when they are unmapped. - Heap* heap = ToBackend(mResourceAllocation.GetResourceHeap()); - ToBackend(GetDevice())->GetResidencyManager()->UnlockAllocation(heap); - } + MaybeError Buffer::MapAtCreationImpl() { + CommandRecordingContext* commandContext; + DAWN_TRY_ASSIGN(commandContext, ToBackend(GetDevice())->GetPendingCommandContext()); + DAWN_TRY(EnsureDataInitialized(commandContext)); - MaybeError Buffer::MapAtCreationImpl(uint8_t** mappedPointer) { - mWrittenMappedRange = {0, static_cast<size_t>(GetSize())}; - DAWN_TRY(MapBufferInternal(mWrittenMappedRange, reinterpret_cast<void**>(mappedPointer), - "D3D12 map at creation")); - mMappedData = reinterpret_cast<char*>(mappedPointer); + // Setting isMapWrite to false on MapRead buffers to silence D3D12 debug layer warning. + bool isMapWrite = (GetUsage() & wgpu::BufferUsage::MapWrite) != 0; + DAWN_TRY(MapInternal(isMapWrite, 0, size_t(GetSize()), "D3D12 map at creation")); return {}; } - MaybeError Buffer::MapReadAsyncImpl(uint32_t serial) { - mWrittenMappedRange = {}; - D3D12_RANGE readRange = {0, static_cast<size_t>(GetSize())}; - DAWN_TRY(MapBufferInternal(readRange, reinterpret_cast<void**>(&mMappedData), - "D3D12 map read async")); + MaybeError Buffer::MapReadAsyncImpl() { + return MapInternal(false, 0, size_t(GetSize()), "D3D12 map read async"); + } - // There is no need to transition the resource to a new state: D3D12 seems to make the GPU - // writes available when the fence is passed. - return {}; + MaybeError Buffer::MapWriteAsyncImpl() { + return MapInternal(true, 0, size_t(GetSize()), "D3D12 map write async"); } - MaybeError Buffer::MapWriteAsyncImpl(uint32_t serial) { - mWrittenMappedRange = {0, static_cast<size_t>(GetSize())}; - DAWN_TRY(MapBufferInternal(mWrittenMappedRange, reinterpret_cast<void**>(&mMappedData), - "D3D12 map write async")); + MaybeError Buffer::MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) { + CommandRecordingContext* commandContext; + DAWN_TRY_ASSIGN(commandContext, ToBackend(GetDevice())->GetPendingCommandContext()); + DAWN_TRY(EnsureDataInitialized(commandContext)); - // There is no need to transition the resource to a new state: D3D12 seems to make the CPU - // writes available on queue submission. - return {}; + return MapInternal(mode & wgpu::MapMode::Write, offset, size, "D3D12 map async"); } void Buffer::UnmapImpl() { - UnmapBufferInternal(mWrittenMappedRange); - - mWrittenMappedRange = {}; + GetD3D12Resource()->Unmap(0, &mWrittenMappedRange); mMappedData = nullptr; + mWrittenMappedRange = {0, 0}; + + // When buffers are mapped, they are locked to keep them in resident memory. We must unlock + // them when they are unmapped. + Heap* heap = ToBackend(mResourceAllocation.GetResourceHeap()); + ToBackend(GetDevice())->GetResidencyManager()->UnlockAllocation(heap); } void* Buffer::GetMappedPointerImpl() { + // The frontend asks that the pointer returned is from the start of the resource + // irrespective of the offset passed in MapAsyncImpl, which is what mMappedData is. return mMappedData; } void Buffer::DestroyImpl() { - // We must ensure that if a mapped buffer is destroyed, it does not leave a dangling lock - // reference on its heap. - if (IsMapped()) { - Heap* heap = ToBackend(mResourceAllocation.GetResourceHeap()); - ToBackend(GetDevice())->GetResidencyManager()->UnlockAllocation(heap); + if (mMappedData != nullptr) { + // If the buffer is currently mapped, unmap without flushing the writes to the GPU + // since the buffer cannot be used anymore. UnmapImpl checks mWrittenRange to know + // which parts to flush, so we set it to an empty range to prevent flushes. + mWrittenMappedRange = {0, 0}; + UnmapImpl(); } ToBackend(GetDevice())->DeallocateMemory(mResourceAllocation); @@ -326,37 +347,91 @@ namespace dawn_native { namespace d3d12 { return mResourceAllocation.GetInfo().mMethod == allocationMethod; } - MaybeError Buffer::ClearBuffer(ClearValue clearValue) { - // TODO(jiawei.shao@intel.com): support buffer lazy-initialization to 0. - ASSERT(clearValue == BufferBase::ClearValue::NonZero); - constexpr uint8_t kClearBufferValue = 1u; + MaybeError Buffer::EnsureDataInitialized(CommandRecordingContext* commandContext) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return {}; + } + + DAWN_TRY(InitializeToZero(commandContext)); + + return {}; + } + MaybeError Buffer::EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext, + uint64_t offset, + uint64_t size) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return {}; + } + + if (IsFullBufferRange(offset, size)) { + SetIsDataInitialized(); + } else { + DAWN_TRY(InitializeToZero(commandContext)); + } + + return {}; + } + + MaybeError Buffer::EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext, + const CopyTextureToBufferCmd* copy) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return {}; + } + + if (IsFullBufferOverwrittenInTextureToBufferCopy(copy)) { + SetIsDataInitialized(); + } else { + DAWN_TRY(InitializeToZero(commandContext)); + } + + return {}; + } + + MaybeError Buffer::InitializeToZero(CommandRecordingContext* commandContext) { + ASSERT(GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)); + ASSERT(!IsDataInitialized()); + + // TODO(jiawei.shao@intel.com): skip initializing the buffer when it is created on a heap + // that has already been zero initialized. + DAWN_TRY(ClearBuffer(commandContext, uint8_t(0u))); + SetIsDataInitialized(); + GetDevice()->IncrementLazyClearCountForTesting(); + + return {}; + } + + MaybeError Buffer::ClearBuffer(CommandRecordingContext* commandContext, uint8_t clearValue) { Device* device = ToBackend(GetDevice()); // The state of the buffers on UPLOAD heap must always be GENERIC_READ and cannot be // changed away, so we can only clear such buffer with buffer mapping. if (D3D12HeapType(GetUsage()) == D3D12_HEAP_TYPE_UPLOAD) { - uint8_t* mappedData = nullptr; - D3D12_RANGE writeRange = {0, static_cast<size_t>(GetSize())}; - DAWN_TRY(MapBufferInternal(writeRange, reinterpret_cast<void**>(&mappedData), - "D3D12 map at clear buffer")); - - memset(mappedData, kClearBufferValue, GetSize()); - - UnmapBufferInternal(writeRange); - mappedData = nullptr; + DAWN_TRY(MapInternal(true, 0, size_t(GetSize()), "D3D12 map at clear buffer")); + memset(mMappedData, clearValue, GetSize()); + UnmapImpl(); } else { // TODO(jiawei.shao@intel.com): use ClearUnorderedAccessView*() when the buffer usage // includes STORAGE. DynamicUploader* uploader = device->GetDynamicUploader(); UploadHandle uploadHandle; DAWN_TRY_ASSIGN(uploadHandle, - uploader->Allocate(GetSize(), device->GetPendingCommandSerial())); + uploader->Allocate(GetSize(), device->GetPendingCommandSerial(), + kCopyBufferToBufferOffsetAlignment)); - memset(uploadHandle.mappedBuffer, kClearBufferValue, GetSize()); + memset(uploadHandle.mappedBuffer, clearValue, GetSize()); - DAWN_TRY(device->CopyFromStagingToBuffer(uploadHandle.stagingBuffer, - uploadHandle.startOffset, this, 0, GetSize())); + device->CopyFromStagingToBufferImpl(commandContext, uploadHandle.stagingBuffer, + uploadHandle.startOffset, this, 0, GetSize()); } return {}; diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/BufferD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/BufferD3D12.h index 3c51b743946..a2af099a79e 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/BufferD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/BufferD3D12.h @@ -32,7 +32,7 @@ namespace dawn_native { namespace d3d12 { MaybeError Initialize(); - ComPtr<ID3D12Resource> GetD3D12Resource() const; + ID3D12Resource* GetD3D12Resource() const; D3D12_GPU_VIRTUAL_ADDRESS GetVA() const; bool TrackUsageAndGetResourceBarrier(CommandRecordingContext* commandContext, @@ -44,34 +44,41 @@ namespace dawn_native { namespace d3d12 { bool CheckAllocationMethodForTesting(AllocationMethod allocationMethod) const; bool CheckIsResidentForTesting() const; + MaybeError EnsureDataInitialized(CommandRecordingContext* commandContext); + MaybeError EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext, + uint64_t offset, + uint64_t size); + MaybeError EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext, + const CopyTextureToBufferCmd* copy); + private: ~Buffer() override; // Dawn API - MaybeError MapReadAsyncImpl(uint32_t serial) override; - MaybeError MapWriteAsyncImpl(uint32_t serial) override; + MaybeError MapReadAsyncImpl() override; + MaybeError MapWriteAsyncImpl() override; + MaybeError MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) override; void UnmapImpl() override; void DestroyImpl() override; - bool IsMapWritable() const override; - virtual MaybeError MapAtCreationImpl(uint8_t** mappedPointer) override; + bool IsMappableAtCreation() const override; + virtual MaybeError MapAtCreationImpl() override; void* GetMappedPointerImpl() override; - MaybeError MapBufferInternal(D3D12_RANGE mappedRange, - void** mappedPointer, - const char* contextInfo); - void UnmapBufferInternal(D3D12_RANGE mappedRange); + MaybeError MapInternal(bool isWrite, size_t start, size_t end, const char* contextInfo); bool TransitionUsageAndGetResourceBarrier(CommandRecordingContext* commandContext, D3D12_RESOURCE_BARRIER* barrier, wgpu::BufferUsage newUsage); - MaybeError ClearBuffer(ClearValue clearValue); + MaybeError InitializeToZero(CommandRecordingContext* commandContext); + MaybeError ClearBuffer(CommandRecordingContext* commandContext, uint8_t clearValue); ResourceHeapAllocation mResourceAllocation; bool mFixedResourceState = false; wgpu::BufferUsage mLastUsage = wgpu::BufferUsage::None; Serial mLastUsedSerial = UINT64_MAX; - D3D12_RANGE mWrittenMappedRange; - char* mMappedData = nullptr; + + D3D12_RANGE mWrittenMappedRange = {0, 0}; + void* mMappedData = nullptr; }; }} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/CPUDescriptorHeapAllocationD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/CPUDescriptorHeapAllocationD3D12.cpp index 021c47ab87c..d92398584d9 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/CPUDescriptorHeapAllocationD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/CPUDescriptorHeapAllocationD3D12.cpp @@ -50,4 +50,4 @@ namespace dawn_native { namespace d3d12 { mBaseDescriptor = {0}; } -}} // namespace dawn_native::d3d12
\ No newline at end of file +}} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/CPUDescriptorHeapAllocationD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/CPUDescriptorHeapAllocationD3D12.h index 05aaf5185bd..51ae2fdb61f 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/CPUDescriptorHeapAllocationD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/CPUDescriptorHeapAllocationD3D12.h @@ -44,4 +44,4 @@ namespace dawn_native { namespace d3d12 { }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_CPUDESCRIPTORHEAPALLOCATION_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_CPUDESCRIPTORHEAPALLOCATION_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/CommandBufferD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/CommandBufferD3D12.cpp index a0df3b68730..8ca8fc27168 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/CommandBufferD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/CommandBufferD3D12.cpp @@ -19,6 +19,7 @@ #include "dawn_native/CommandEncoder.h" #include "dawn_native/CommandValidation.h" #include "dawn_native/Commands.h" +#include "dawn_native/EnumMaskIterator.h" #include "dawn_native/RenderBundle.h" #include "dawn_native/d3d12/BindGroupD3D12.h" #include "dawn_native/d3d12/BindGroupLayoutD3D12.h" @@ -28,6 +29,7 @@ #include "dawn_native/d3d12/DeviceD3D12.h" #include "dawn_native/d3d12/PipelineLayoutD3D12.h" #include "dawn_native/d3d12/PlatformFunctions.h" +#include "dawn_native/d3d12/QuerySetD3D12.h" #include "dawn_native/d3d12/RenderPassBuilderD3D12.h" #include "dawn_native/d3d12/RenderPipelineD3D12.h" #include "dawn_native/d3d12/SamplerD3D12.h" @@ -55,6 +57,19 @@ namespace dawn_native { namespace d3d12 { } } + D3D12_QUERY_TYPE D3D12QueryType(wgpu::QueryType type) { + switch (type) { + case wgpu::QueryType::Occlusion: + return D3D12_QUERY_TYPE_OCCLUSION; + case wgpu::QueryType::PipelineStatistics: + return D3D12_QUERY_TYPE_PIPELINE_STATISTICS; + case wgpu::QueryType::Timestamp: + return D3D12_QUERY_TYPE_TIMESTAMP; + default: + UNREACHABLE(); + } + } + bool CanUseCopyResource(const Texture* src, const Texture* dst, const Extent3D& copySize) { // Checked by validation ASSERT(src->GetSampleCount() == dst->GetSampleCount()); @@ -80,9 +95,51 @@ namespace dawn_native { namespace d3d12 { copySize.depth == srcSize.depth; } + void RecordCopyTextureToBufferFromTextureCopySplit(ID3D12GraphicsCommandList* commandList, + const Texture2DCopySplit& baseCopySplit, + Buffer* buffer, + uint64_t baseOffset, + uint64_t bufferBytesPerRow, + Texture* texture, + uint32_t textureMiplevel, + uint32_t textureSlice, + Aspect aspect) { + const D3D12_TEXTURE_COPY_LOCATION textureLocation = + ComputeTextureCopyLocationForTexture(texture, textureMiplevel, textureSlice, + aspect); + + const uint64_t offset = baseCopySplit.offset + baseOffset; + + for (uint32_t i = 0; i < baseCopySplit.count; ++i) { + const Texture2DCopySplit::CopyInfo& info = baseCopySplit.copies[i]; + + // TODO(jiawei.shao@intel.com): pre-compute bufferLocation and sourceRegion as + // members in Texture2DCopySplit::CopyInfo. + const D3D12_TEXTURE_COPY_LOCATION bufferLocation = + ComputeBufferLocationForCopyTextureRegion(texture, buffer->GetD3D12Resource(), + info.bufferSize, offset, + bufferBytesPerRow, aspect); + const D3D12_BOX sourceRegion = + ComputeD3D12BoxFromOffsetAndSize(info.textureOffset, info.copySize); + + commandList->CopyTextureRegion(&bufferLocation, info.bufferOffset.x, + info.bufferOffset.y, info.bufferOffset.z, + &textureLocation, &sourceRegion); + } + } + + void RecordWriteTimestampCmd(ID3D12GraphicsCommandList* commandList, + WriteTimestampCmd* cmd) { + QuerySet* querySet = ToBackend(cmd->querySet.Get()); + ASSERT(D3D12QueryType(querySet->GetQueryType()) == D3D12_QUERY_TYPE_TIMESTAMP); + commandList->EndQuery(querySet->GetQueryHeap(), D3D12_QUERY_TYPE_TIMESTAMP, + cmd->queryIndex); + } } // anonymous namespace class BindGroupStateTracker : public BindGroupAndStorageBarrierTrackerBase<false, uint64_t> { + using Base = BindGroupAndStorageBarrierTrackerBase; + public: BindGroupStateTracker(Device* device) : BindGroupAndStorageBarrierTrackerBase(), @@ -95,6 +152,17 @@ namespace dawn_native { namespace d3d12 { mInCompute = inCompute_; } + void OnSetPipeline(PipelineBase* pipeline) { + // Invalidate the root sampler tables previously set in the root signature. + // This is because changing the pipeline layout also changes the root signature. + const PipelineLayout* pipelineLayout = ToBackend(pipeline->GetLayout()); + if (mLastAppliedPipelineLayout != pipelineLayout) { + mBoundRootSamplerTables = {}; + } + + Base::OnSetPipeline(pipeline); + } + MaybeError Apply(CommandRecordingContext* commandContext) { // Bindgroups are allocated in shader-visible descriptor heaps which are managed by a // ringbuffer. There can be a single shader-visible descriptor heap of each type bound @@ -309,10 +377,16 @@ namespace dawn_native { namespace d3d12 { uint32_t parameterIndex = pipelineLayout->GetSamplerRootParameterIndex(index); const D3D12_GPU_DESCRIPTOR_HANDLE baseDescriptor = group->GetBaseSamplerDescriptor(); - if (mInCompute) { - commandList->SetComputeRootDescriptorTable(parameterIndex, baseDescriptor); - } else { - commandList->SetGraphicsRootDescriptorTable(parameterIndex, baseDescriptor); + // Check if the group requires its sampler table to be set in the pipeline. + // This because sampler heap allocations could be cached and use the same table. + if (mBoundRootSamplerTables[index].ptr != baseDescriptor.ptr) { + if (mInCompute) { + commandList->SetComputeRootDescriptorTable(parameterIndex, baseDescriptor); + } else { + commandList->SetGraphicsRootDescriptorTable(parameterIndex, baseDescriptor); + } + + mBoundRootSamplerTables[index] = baseDescriptor; } } } @@ -321,6 +395,9 @@ namespace dawn_native { namespace d3d12 { bool mInCompute = false; + ityp::array<BindGroupIndex, D3D12_GPU_DESCRIPTOR_HANDLE, kMaxBindGroups> + mBoundRootSamplerTables = {}; + ShaderVisibleDescriptorAllocator* mViewAllocator; ShaderVisibleDescriptorAllocator* mSamplerAllocator; }; @@ -446,7 +523,8 @@ namespace dawn_native { namespace d3d12 { ID3D12Resource* colorTextureHandle = colorTexture->GetD3D12Resource(); ID3D12Resource* resolveTextureHandle = resolveTexture->GetD3D12Resource(); const uint32_t resolveTextureSubresourceIndex = resolveTexture->GetSubresourceIndex( - resolveTarget->GetBaseMipLevel(), resolveTarget->GetBaseArrayLayer()); + resolveTarget->GetBaseMipLevel(), resolveTarget->GetBaseArrayLayer(), + Aspect::Color); constexpr uint32_t kColorTextureSubresourceIndex = 0; commandContext->GetCommandList()->ResolveSubresource( resolveTextureHandle, resolveTextureSubresourceIndex, colorTextureHandle, @@ -457,11 +535,7 @@ namespace dawn_native { namespace d3d12 { } // anonymous namespace CommandBuffer::CommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor) - : CommandBufferBase(encoder, descriptor), mCommands(encoder->AcquireCommands()) { - } - - CommandBuffer::~CommandBuffer() { - FreeCommands(&mCommands); + : CommandBufferBase(encoder, descriptor) { } MaybeError CommandBuffer::RecordCommands(CommandRecordingContext* commandContext) { @@ -484,10 +558,15 @@ namespace dawn_native { namespace d3d12 { wgpu::BufferUsage bufferUsages = wgpu::BufferUsage::None; for (size_t i = 0; i < usages.buffers.size(); ++i) { + Buffer* buffer = ToBackend(usages.buffers[i]); + + // TODO(jiawei.shao@intel.com): clear storage buffers with + // ClearUnorderedAccessView*(). + buffer->GetDevice()->ConsumedError(buffer->EnsureDataInitialized(commandContext)); + D3D12_RESOURCE_BARRIER barrier; - if (ToBackend(usages.buffers[i]) - ->TrackUsageAndGetResourceBarrier(commandContext, &barrier, - usages.bufferUsages[i])) { + if (buffer->TrackUsageAndGetResourceBarrier(commandContext, &barrier, + usages.bufferUsages[i])) { barriers.push_back(barrier); } bufferUsages |= usages.bufferUsages[i]; @@ -560,14 +639,18 @@ namespace dawn_native { namespace d3d12 { Buffer* srcBuffer = ToBackend(copy->source.Get()); Buffer* dstBuffer = ToBackend(copy->destination.Get()); + DAWN_TRY(srcBuffer->EnsureDataInitialized(commandContext)); + DAWN_TRY(dstBuffer->EnsureDataInitializedAsDestination( + commandContext, copy->destinationOffset, copy->size)); + srcBuffer->TrackUsageAndTransitionNow(commandContext, wgpu::BufferUsage::CopySrc); dstBuffer->TrackUsageAndTransitionNow(commandContext, wgpu::BufferUsage::CopyDst); commandList->CopyBufferRegion( - dstBuffer->GetD3D12Resource().Get(), copy->destinationOffset, - srcBuffer->GetD3D12Resource().Get(), copy->sourceOffset, copy->size); + dstBuffer->GetD3D12Resource(), copy->destinationOffset, + srcBuffer->GetD3D12Resource(), copy->sourceOffset, copy->size); break; } @@ -576,13 +659,12 @@ namespace dawn_native { namespace d3d12 { Buffer* buffer = ToBackend(copy->source.buffer.Get()); Texture* texture = ToBackend(copy->destination.texture.Get()); + DAWN_TRY(buffer->EnsureDataInitialized(commandContext)); + ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D); - // TODO(jiawei.shao@intel.com): use copy->destination.origin.z instead of - // copy->destination.arrayLayer once GPUTextureCopyView.arrayLayer to - // GPUTextureCopyView.origin.z is done. - SubresourceRange subresources = {copy->destination.mipLevel, 1, - copy->destination.arrayLayer, - copy->copySize.depth}; + SubresourceRange subresources = + GetSubresourcesAffectedByCopy(copy->destination, copy->copySize); + if (IsCompleteSubresourceCopiedTo(texture, copy->copySize, copy->destination.mipLevel)) { texture->SetIsSubresourceContentInitialized(true, subresources); @@ -594,43 +676,12 @@ namespace dawn_native { namespace d3d12 { texture->TrackUsageAndTransitionNow(commandContext, wgpu::TextureUsage::CopyDst, subresources); - const uint64_t bytesPerSlice = - copy->source.bytesPerRow * copy->source.rowsPerImage; - - const dawn_native::Extent3D copyOneLayerSize = {copy->copySize.width, - copy->copySize.height, 1}; - uint64_t bufferOffsetForNextSlice = 0; - for (uint32_t copySlice = copy->destination.arrayLayer; - copySlice < copy->destination.arrayLayer + copy->copySize.depth; - ++copySlice) { - // TODO(jiawei.shao@intel.com): compute copySplit once for all texture array - // layers when possible. - auto copySplit = ComputeTextureCopySplit( - copy->destination.origin, copyOneLayerSize, texture->GetFormat(), - bufferOffsetForNextSlice + copy->source.offset, - copy->source.bytesPerRow, copy->source.rowsPerImage); - - D3D12_TEXTURE_COPY_LOCATION textureLocation = - ComputeTextureCopyLocationForTexture( - texture, copy->destination.mipLevel, copySlice); - - for (uint32_t i = 0; i < copySplit.count; ++i) { - const TextureCopySplit::CopyInfo& info = copySplit.copies[i]; - - D3D12_TEXTURE_COPY_LOCATION bufferLocation = - ComputeBufferLocationForCopyTextureRegion( - texture, buffer->GetD3D12Resource().Get(), info.bufferSize, - copySplit.offset, copy->source.bytesPerRow); - D3D12_BOX sourceRegion = - ComputeD3D12BoxFromOffsetAndSize(info.bufferOffset, info.copySize); - - commandList->CopyTextureRegion( - &textureLocation, info.textureOffset.x, info.textureOffset.y, - info.textureOffset.z, &bufferLocation, &sourceRegion); - } + // compute the copySplits and record the CopyTextureRegion commands + CopyBufferToTextureWithCopySplit( + commandContext, copy->destination, copy->copySize, texture, + buffer->GetD3D12Resource(), copy->source.offset, copy->source.bytesPerRow, + copy->source.rowsPerImage, subresources.aspects); - bufferOffsetForNextSlice += bytesPerSlice; - } break; } @@ -639,55 +690,57 @@ namespace dawn_native { namespace d3d12 { Texture* texture = ToBackend(copy->source.texture.Get()); Buffer* buffer = ToBackend(copy->destination.buffer.Get()); + DAWN_TRY(buffer->EnsureDataInitializedAsDestination(commandContext, copy)); + ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D); - // TODO(jiawei.shao@intel.com): use copy->destination.origin.z instead of - // copy->destination.arrayLayer once GPUTextureCopyView.arrayLayer to - // GPUTextureCopyView.origin.z is done. - SubresourceRange subresources = {copy->source.mipLevel, 1, - copy->source.arrayLayer, copy->copySize.depth}; + SubresourceRange subresources = + GetSubresourcesAffectedByCopy(copy->source, copy->copySize); + texture->EnsureSubresourceContentInitialized(commandContext, subresources); texture->TrackUsageAndTransitionNow(commandContext, wgpu::TextureUsage::CopySrc, subresources); buffer->TrackUsageAndTransitionNow(commandContext, wgpu::BufferUsage::CopyDst); - const uint64_t bytesPerSlice = - copy->destination.bytesPerRow * copy->destination.rowsPerImage; - - const dawn_native::Extent3D copyOneLayerSize = {copy->copySize.width, - copy->copySize.height, 1}; - uint64_t bufferOffsetForNextSlice = 0; - for (uint32_t copySlice = copy->source.arrayLayer; - copySlice < copy->source.arrayLayer + copy->copySize.depth; ++copySlice) { - // TODO(jiawei.shao@intel.com): compute copySplit once for all texture array - // layers when possible. - TextureCopySplit copySplit = ComputeTextureCopySplit( - copy->source.origin, copyOneLayerSize, texture->GetFormat(), - bufferOffsetForNextSlice + copy->destination.offset, - copy->destination.bytesPerRow, copy->destination.rowsPerImage); - - D3D12_TEXTURE_COPY_LOCATION textureLocation = - ComputeTextureCopyLocationForTexture(texture, copy->source.mipLevel, - copySlice); - - for (uint32_t i = 0; i < copySplit.count; ++i) { - const TextureCopySplit::CopyInfo& info = copySplit.copies[i]; - - D3D12_TEXTURE_COPY_LOCATION bufferLocation = - ComputeBufferLocationForCopyTextureRegion( - texture, buffer->GetD3D12Resource().Get(), info.bufferSize, - copySplit.offset, copy->destination.bytesPerRow); - - D3D12_BOX sourceRegion = - ComputeD3D12BoxFromOffsetAndSize(info.textureOffset, info.copySize); - - commandList->CopyTextureRegion(&bufferLocation, info.bufferOffset.x, - info.bufferOffset.y, info.bufferOffset.z, - &textureLocation, &sourceRegion); - } + const TexelBlockInfo& blockInfo = + texture->GetFormat().GetTexelBlockInfo(copy->source.aspect); - bufferOffsetForNextSlice += bytesPerSlice; + // See comments around ComputeTextureCopySplits() for more details. + const TextureCopySplits copySplits = ComputeTextureCopySplits( + copy->source.origin, copy->copySize, blockInfo, copy->destination.offset, + copy->destination.bytesPerRow, copy->destination.rowsPerImage); + + const uint64_t bytesPerSlice = + copy->destination.bytesPerRow * + (copy->destination.rowsPerImage / blockInfo.blockHeight); + + // copySplits.copies2D[1] is always calculated for the second copy slice with + // extra "bytesPerSlice" copy offset compared with the first copy slice. So + // here we use an array bufferOffsetsForNextSlice to record the extra offsets + // for each copy slice: bufferOffsetsForNextSlice[0] is the extra offset for + // the next copy slice that uses copySplits.copies2D[0], and + // bufferOffsetsForNextSlice[1] is the extra offset for the next copy slice + // that uses copySplits.copies2D[1]. + std::array<uint64_t, TextureCopySplits::kMaxTextureCopySplits> + bufferOffsetsForNextSlice = {{0u, 0u}}; + for (uint32_t copySlice = 0; copySlice < copy->copySize.depth; ++copySlice) { + const uint32_t splitIndex = copySlice % copySplits.copies2D.size(); + + const Texture2DCopySplit& copySplitPerLayerBase = + copySplits.copies2D[splitIndex]; + const uint64_t bufferOffsetForNextSlice = + bufferOffsetsForNextSlice[splitIndex]; + const uint32_t copyTextureLayer = copySlice + copy->source.origin.z; + + RecordCopyTextureToBufferFromTextureCopySplit( + commandList, copySplitPerLayerBase, buffer, bufferOffsetForNextSlice, + copy->destination.bytesPerRow, texture, copy->source.mipLevel, + copyTextureLayer, subresources.aspects); + + bufferOffsetsForNextSlice[splitIndex] += + bytesPerSlice * copySplits.copies2D.size(); } + break; } @@ -697,11 +750,11 @@ namespace dawn_native { namespace d3d12 { Texture* source = ToBackend(copy->source.texture.Get()); Texture* destination = ToBackend(copy->destination.texture.Get()); - SubresourceRange srcRange = {copy->source.mipLevel, 1, copy->source.arrayLayer, - copy->copySize.depth}; - SubresourceRange dstRange = {copy->destination.mipLevel, 1, - copy->destination.arrayLayer, - copy->copySize.depth}; + + SubresourceRange srcRange = + GetSubresourcesAffectedByCopy(copy->source, copy->copySize); + SubresourceRange dstRange = + GetSubresourcesAffectedByCopy(copy->destination, copy->copySize); source->EnsureSubresourceContentInitialized(commandContext, srcRange); if (IsCompleteSubresourceCopiedTo(destination, copy->copySize, @@ -717,8 +770,7 @@ namespace dawn_native { namespace d3d12 { // subresources should all be COMMON instead of what we set now. Currently // it is not allowed to copy with overlapped subresources, but we still // add the ASSERT here as a reminder for this possible misuse. - ASSERT(!IsRangeOverlapped(copy->source.arrayLayer, - copy->destination.arrayLayer, + ASSERT(!IsRangeOverlapped(copy->source.origin.z, copy->destination.origin.z, copy->copySize.depth)); } source->TrackUsageAndTransitionNow(commandContext, wgpu::TextureUsage::CopySrc, @@ -726,6 +778,7 @@ namespace dawn_native { namespace d3d12 { destination->TrackUsageAndTransitionNow(commandContext, wgpu::TextureUsage::CopyDst, dstRange); + ASSERT(srcRange.aspects == dstRange.aspects); if (CanUseCopyResource(source, destination, copy->copySize)) { commandList->CopyResource(destination->GetD3D12Resource(), source->GetD3D12Resource()); @@ -735,28 +788,56 @@ namespace dawn_native { namespace d3d12 { destination->GetDimension() == wgpu::TextureDimension::e2D); const dawn_native::Extent3D copyExtentOneSlice = { copy->copySize.width, copy->copySize.height, 1u}; - for (uint32_t slice = 0; slice < copy->copySize.depth; ++slice) { - D3D12_TEXTURE_COPY_LOCATION srcLocation = - ComputeTextureCopyLocationForTexture( - source, copy->source.mipLevel, copy->source.arrayLayer + slice); - - D3D12_TEXTURE_COPY_LOCATION dstLocation = - ComputeTextureCopyLocationForTexture( - destination, copy->destination.mipLevel, - copy->destination.arrayLayer + slice); - - D3D12_BOX sourceRegion = ComputeD3D12BoxFromOffsetAndSize( - copy->source.origin, copyExtentOneSlice); - - commandList->CopyTextureRegion(&dstLocation, copy->destination.origin.x, - copy->destination.origin.y, - copy->destination.origin.z, &srcLocation, - &sourceRegion); + + for (Aspect aspect : IterateEnumMask(srcRange.aspects)) { + for (uint32_t slice = 0; slice < copy->copySize.depth; ++slice) { + D3D12_TEXTURE_COPY_LOCATION srcLocation = + ComputeTextureCopyLocationForTexture( + source, copy->source.mipLevel, + copy->source.origin.z + slice, aspect); + + D3D12_TEXTURE_COPY_LOCATION dstLocation = + ComputeTextureCopyLocationForTexture( + destination, copy->destination.mipLevel, + copy->destination.origin.z + slice, aspect); + + Origin3D sourceOriginInSubresource = copy->source.origin; + sourceOriginInSubresource.z = 0; + D3D12_BOX sourceRegion = ComputeD3D12BoxFromOffsetAndSize( + sourceOriginInSubresource, copyExtentOneSlice); + + commandList->CopyTextureRegion( + &dstLocation, copy->destination.origin.x, + copy->destination.origin.y, 0, &srcLocation, &sourceRegion); + } } } break; } + case Command::ResolveQuerySet: { + ResolveQuerySetCmd* cmd = mCommands.NextCommand<ResolveQuerySetCmd>(); + QuerySet* querySet = ToBackend(cmd->querySet.Get()); + Buffer* destination = ToBackend(cmd->destination.Get()); + + commandList->ResolveQueryData( + querySet->GetQueryHeap(), D3D12QueryType(querySet->GetQueryType()), + cmd->firstQuery, cmd->queryCount, destination->GetD3D12Resource(), + cmd->destinationOffset); + + // TODO(hao.x.li@intel.com): Add compute shader to convert the query result + // (ticks) to timestamp (ns) + + break; + } + + case Command::WriteTimestamp: { + WriteTimestampCmd* cmd = mCommands.NextCommand<WriteTimestampCmd>(); + + RecordWriteTimestampCmd(commandList, cmd); + break; + } + default: { UNREACHABLE(); break; @@ -790,8 +871,7 @@ namespace dawn_native { namespace d3d12 { Buffer* buffer = ToBackend(dispatch->indirectBuffer.Get()); ComPtr<ID3D12CommandSignature> signature = ToBackend(GetDevice())->GetDispatchIndirectSignature(); - commandList->ExecuteIndirect(signature.Get(), 1, - buffer->GetD3D12Resource().Get(), + commandList->ExecuteIndirect(signature.Get(), 1, buffer->GetD3D12Resource(), dispatch->indirectOffset, nullptr, 0); break; } @@ -868,6 +948,13 @@ namespace dawn_native { namespace d3d12 { break; } + case Command::WriteTimestamp: { + WriteTimestampCmd* cmd = mCommands.NextCommand<WriteTimestampCmd>(); + + RecordWriteTimestampCmd(commandList, cmd); + break; + } + default: { UNREACHABLE(); break; @@ -1106,8 +1193,7 @@ namespace dawn_native { namespace d3d12 { Buffer* buffer = ToBackend(draw->indirectBuffer.Get()); ComPtr<ID3D12CommandSignature> signature = ToBackend(GetDevice())->GetDrawIndirectSignature(); - commandList->ExecuteIndirect(signature.Get(), 1, - buffer->GetD3D12Resource().Get(), + commandList->ExecuteIndirect(signature.Get(), 1, buffer->GetD3D12Resource(), draw->indirectOffset, nullptr, 0); break; } @@ -1121,8 +1207,7 @@ namespace dawn_native { namespace d3d12 { Buffer* buffer = ToBackend(draw->indirectBuffer.Get()); ComPtr<ID3D12CommandSignature> signature = ToBackend(GetDevice())->GetDrawIndexedIndirectSignature(); - commandList->ExecuteIndirect(signature.Get(), 1, - buffer->GetD3D12Resource().Get(), + commandList->ExecuteIndirect(signature.Get(), 1, buffer->GetD3D12Resource(), draw->indirectOffset, nullptr, 0); break; } @@ -1286,6 +1371,13 @@ namespace dawn_native { namespace d3d12 { break; } + case Command::WriteTimestamp: { + WriteTimestampCmd* cmd = mCommands.NextCommand<WriteTimestampCmd>(); + + RecordWriteTimestampCmd(commandList, cmd); + break; + } + default: { DAWN_TRY(EncodeRenderBundleCommand(&mCommands, type)); break; diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/CommandBufferD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/CommandBufferD3D12.h index cc53fd54cf9..f4d858c8585 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/CommandBufferD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/CommandBufferD3D12.h @@ -16,7 +16,6 @@ #define DAWNNATIVE_D3D12_COMMANDBUFFERD3D12_H_ #include "common/Constants.h" -#include "dawn_native/CommandAllocator.h" #include "dawn_native/CommandBuffer.h" #include "dawn_native/Error.h" #include "dawn_native/d3d12/Forward.h" @@ -43,7 +42,6 @@ namespace dawn_native { namespace d3d12 { MaybeError RecordCommands(CommandRecordingContext* commandContext); private: - ~CommandBuffer() override; MaybeError RecordComputePass(CommandRecordingContext* commandContext, BindGroupStateTracker* bindingTracker); MaybeError RecordRenderPass(CommandRecordingContext* commandContext, @@ -55,8 +53,6 @@ namespace dawn_native { namespace d3d12 { RenderPassBuilder* renderPassBuilder); void EmulateBeginRenderPass(CommandRecordingContext* commandContext, const RenderPassBuilder* renderPassBuilder) const; - - CommandIterator mCommands; }; }} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Backend.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Backend.cpp index f78975ef7a9..71965994265 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Backend.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Backend.cpp @@ -69,4 +69,7 @@ namespace dawn_native { namespace d3d12 { return reinterpret_cast<WGPUTexture>(texture.Detach()); } + AdapterDiscoveryOptions::AdapterDiscoveryOptions(ComPtr<IDXGIAdapter> adapter) + : AdapterDiscoveryOptionsBase(WGPUBackendType_D3D12), dxgiAdapter(std::move(adapter)) { + } }} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Error.h b/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Error.h index 16ea4c4e834..ade06fe62e6 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Error.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Error.h @@ -42,4 +42,4 @@ namespace dawn_native { namespace d3d12 { }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_D3D12ERROR_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_D3D12ERROR_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Info.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Info.cpp index 6505e44aef5..e19dd8179be 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Info.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Info.cpp @@ -50,15 +50,59 @@ namespace dawn_native { namespace d3d12 { D3D12_FEATURE_DATA_D3D12_OPTIONS5 featureOptions5 = {}; if (SUCCEEDED(adapter.GetDevice()->CheckFeatureSupport( D3D12_FEATURE_D3D12_OPTIONS5, &featureOptions5, sizeof(featureOptions5)))) { - // Performance regressions been observed when using a render pass on Intel graphics with - // RENDER_PASS_TIER_1 available, so fall back to a software emulated render pass on - // these platforms. + // Performance regressions been observed when using a render pass on Intel graphics + // with RENDER_PASS_TIER_1 available, so fall back to a software emulated render + // pass on these platforms. if (featureOptions5.RenderPassesTier < D3D12_RENDER_PASS_TIER_1 || !gpu_info::IsIntel(adapter.GetPCIInfo().vendorId)) { info.supportsRenderPass = true; } } - return info; + D3D12_FEATURE_DATA_SHADER_MODEL knownShaderModels[] = {{D3D_SHADER_MODEL_6_2}, + {D3D_SHADER_MODEL_6_1}, + {D3D_SHADER_MODEL_6_0}, + {D3D_SHADER_MODEL_5_1}}; + uint32_t driverShaderModel = 0; + for (D3D12_FEATURE_DATA_SHADER_MODEL shaderModel : knownShaderModels) { + if (SUCCEEDED(adapter.GetDevice()->CheckFeatureSupport( + D3D12_FEATURE_SHADER_MODEL, &shaderModel, sizeof(shaderModel)))) { + driverShaderModel = shaderModel.HighestShaderModel; + break; + } + } + + if (driverShaderModel < D3D_SHADER_MODEL_5_1) { + return DAWN_INTERNAL_ERROR("Driver doesn't support Shader Model 5.1 or higher"); + } + + // D3D_SHADER_MODEL is encoded as 0xMm with M the major version and m the minor version + ASSERT(driverShaderModel <= 0xFF); + uint32_t shaderModelMajor = (driverShaderModel & 0xF0) >> 4; + uint32_t shaderModelMinor = (driverShaderModel & 0xF); + + ASSERT(shaderModelMajor < 10); + ASSERT(shaderModelMinor < 10); + info.shaderModel = 10 * shaderModelMajor + shaderModelMinor; + + // Profiles are always <stage>s_<minor>_<major> so we build the s_<minor>_major and add + // it to each of the stage's suffix. + std::wstring profileSuffix = L"s_M_n"; + profileSuffix[2] = wchar_t('0' + shaderModelMajor); + profileSuffix[4] = wchar_t('0' + shaderModelMinor); + + info.shaderProfiles[SingleShaderStage::Vertex] = L"v" + profileSuffix; + info.shaderProfiles[SingleShaderStage::Fragment] = L"p" + profileSuffix; + info.shaderProfiles[SingleShaderStage::Compute] = L"c" + profileSuffix; + + D3D12_FEATURE_DATA_D3D12_OPTIONS4 featureData4 = {}; + if (SUCCEEDED(adapter.GetDevice()->CheckFeatureSupport( + D3D12_FEATURE_D3D12_OPTIONS4, &featureData4, sizeof(featureData4)))) { + info.supportsShaderFloat16 = info.shaderModel >= D3D_SHADER_MODEL_6_2 && + featureData4.Native16BitShaderOpsSupported; + } + + return std::move(info); } + }} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Info.h b/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Info.h index 78d38208169..46b2d09eba2 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Info.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/D3D12Info.h @@ -16,6 +16,7 @@ #define DAWNNATIVE_D3D12_D3D12INFO_H_ #include "dawn_native/Error.h" +#include "dawn_native/PerStage.h" #include "dawn_native/d3d12/d3d12_platform.h" namespace dawn_native { namespace d3d12 { @@ -26,9 +27,14 @@ namespace dawn_native { namespace d3d12 { bool isUMA; uint32_t resourceHeapTier; bool supportsRenderPass; + bool supportsShaderFloat16; + // shaderModel indicates the maximum supported shader model, for example, the value 62 + // indicates that current driver supports the maximum shader model is shader model 6.2. + uint32_t shaderModel; + PerStage<std::wstring> shaderProfiles; }; ResultOrError<D3D12DeviceInfo> GatherDeviceInfo(const Adapter& adapter); }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_D3D12INFO_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_D3D12INFO_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/DeviceD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/DeviceD3D12.cpp index 9f180450c5c..145a3829c84 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/DeviceD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/DeviceD3D12.cpp @@ -17,6 +17,7 @@ #include "common/Assert.h" #include "dawn_native/BackendConnection.h" #include "dawn_native/ErrorData.h" +#include "dawn_native/Format.h" #include "dawn_native/Instance.h" #include "dawn_native/d3d12/AdapterD3D12.h" #include "dawn_native/d3d12/BackendD3D12.h" @@ -29,6 +30,7 @@ #include "dawn_native/d3d12/D3D12Error.h" #include "dawn_native/d3d12/PipelineLayoutD3D12.h" #include "dawn_native/d3d12/PlatformFunctions.h" +#include "dawn_native/d3d12/QuerySetD3D12.h" #include "dawn_native/d3d12/QueueD3D12.h" #include "dawn_native/d3d12/RenderPipelineD3D12.h" #include "dawn_native/d3d12/ResidencyManagerD3D12.h" @@ -41,6 +43,7 @@ #include "dawn_native/d3d12/StagingDescriptorAllocatorD3D12.h" #include "dawn_native/d3d12/SwapChainD3D12.h" #include "dawn_native/d3d12/TextureD3D12.h" +#include "dawn_native/d3d12/UtilsD3D12.h" #include <sstream> @@ -90,14 +93,15 @@ namespace dawn_native { namespace d3d12 { mCommandAllocatorManager = std::make_unique<CommandAllocatorManager>(this); // Zero sized allocator is never requested and does not need to exist. - for (uint32_t countIndex = 1; countIndex < kNumOfStagingDescriptorAllocators; - countIndex++) { - mViewAllocators[countIndex] = std::make_unique<StagingDescriptorAllocator>( - this, countIndex, kShaderVisibleDescriptorHeapSize, + for (uint32_t countIndex = 0; countIndex < kNumViewDescriptorAllocators; countIndex++) { + mViewAllocators[countIndex + 1] = std::make_unique<StagingDescriptorAllocator>( + this, 1u << countIndex, kShaderVisibleDescriptorHeapSize, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV); + } - mSamplerAllocators[countIndex] = std::make_unique<StagingDescriptorAllocator>( - this, countIndex, kShaderVisibleDescriptorHeapSize, + for (uint32_t countIndex = 0; countIndex < kNumSamplerDescriptorAllocators; countIndex++) { + mSamplerAllocators[countIndex + 1] = std::make_unique<StagingDescriptorAllocator>( + this, 1u << countIndex, kShaderVisibleDescriptorHeapSize, D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER); } @@ -270,10 +274,10 @@ namespace dawn_native { namespace d3d12 { const BindGroupLayoutDescriptor* descriptor) { return new BindGroupLayout(this, descriptor); } - ResultOrError<BufferBase*> Device::CreateBufferImpl(const BufferDescriptor* descriptor) { + ResultOrError<Ref<BufferBase>> Device::CreateBufferImpl(const BufferDescriptor* descriptor) { Ref<Buffer> buffer = AcquireRef(new Buffer(this, descriptor)); DAWN_TRY(buffer->Initialize()); - return buffer.Detach(); + return std::move(buffer); } CommandBufferBase* Device::CreateCommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor) { @@ -288,7 +292,7 @@ namespace dawn_native { namespace d3d12 { return PipelineLayout::Create(this, descriptor); } ResultOrError<QuerySetBase*> Device::CreateQuerySetImpl(const QuerySetDescriptor* descriptor) { - return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation"); + return QuerySet::Create(this, descriptor); } ResultOrError<RenderPipelineBase*> Device::CreateRenderPipelineImpl( const RenderPipelineDescriptor* descriptor) { @@ -336,12 +340,55 @@ namespace dawn_native { namespace d3d12 { DAWN_TRY_ASSIGN(commandRecordingContext, GetPendingCommandContext()); Buffer* dstBuffer = ToBackend(destination); + + DAWN_TRY(dstBuffer->EnsureDataInitializedAsDestination(commandRecordingContext, + destinationOffset, size)); + + CopyFromStagingToBufferImpl(commandRecordingContext, source, sourceOffset, destination, + destinationOffset, size); + + return {}; + } + + void Device::CopyFromStagingToBufferImpl(CommandRecordingContext* commandContext, + StagingBufferBase* source, + uint64_t sourceOffset, + BufferBase* destination, + uint64_t destinationOffset, + uint64_t size) { + ASSERT(commandContext != nullptr); + Buffer* dstBuffer = ToBackend(destination); StagingBuffer* srcBuffer = ToBackend(source); - dstBuffer->TrackUsageAndTransitionNow(commandRecordingContext, wgpu::BufferUsage::CopyDst); + dstBuffer->TrackUsageAndTransitionNow(commandContext, wgpu::BufferUsage::CopyDst); - commandRecordingContext->GetCommandList()->CopyBufferRegion( - dstBuffer->GetD3D12Resource().Get(), destinationOffset, srcBuffer->GetResource(), + commandContext->GetCommandList()->CopyBufferRegion( + dstBuffer->GetD3D12Resource(), destinationOffset, srcBuffer->GetResource(), sourceOffset, size); + } + + MaybeError Device::CopyFromStagingToTexture(const StagingBufferBase* source, + const TextureDataLayout& src, + TextureCopy* dst, + const Extent3D& copySizePixels) { + CommandRecordingContext* commandContext; + DAWN_TRY_ASSIGN(commandContext, GetPendingCommandContext()); + Texture* texture = ToBackend(dst->texture.Get()); + ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D); + + SubresourceRange range = GetSubresourcesAffectedByCopy(*dst, copySizePixels); + + if (IsCompleteSubresourceCopiedTo(texture, copySizePixels, dst->mipLevel)) { + texture->SetIsSubresourceContentInitialized(true, range); + } else { + texture->EnsureSubresourceContentInitialized(commandContext, range); + } + + texture->TrackUsageAndTransitionNow(commandContext, wgpu::TextureUsage::CopyDst, range); + + // compute the copySplits and record the CopyTextureRegion commands + CopyBufferToTextureWithCopySplit(commandContext, *dst, copySizePixels, texture, + ToBackend(source)->GetResource(), src.offset, + src.bytesPerRow, src.rowsPerImage, range.aspects); return {}; } @@ -523,8 +570,11 @@ namespace dawn_native { namespace d3d12 { ::CloseHandle(mFenceEvent); } + // Release recycled resource heaps. + mResourceAllocatorManager->DestroyPool(); + // We need to handle clearing up com object refs that were enqeued after TickImpl - mUsedComObjectRefs.ClearUpTo(GetCompletedCommandSerial()); + mUsedComObjectRefs.ClearUpTo(std::numeric_limits<Serial>::max()); ASSERT(mUsedComObjectRefs.Empty()); ASSERT(!mPendingCommands.IsOpen()); @@ -540,14 +590,18 @@ namespace dawn_native { namespace d3d12 { StagingDescriptorAllocator* Device::GetViewStagingDescriptorAllocator( uint32_t descriptorCount) const { - ASSERT(descriptorCount < kNumOfStagingDescriptorAllocators); - return mViewAllocators[descriptorCount].get(); + ASSERT(descriptorCount <= kMaxViewDescriptorsPerBindGroup); + // This is Log2 of the next power of two, plus 1. + uint32_t allocatorIndex = descriptorCount == 0 ? 0 : Log2Ceil(descriptorCount) + 1; + return mViewAllocators[allocatorIndex].get(); } StagingDescriptorAllocator* Device::GetSamplerStagingDescriptorAllocator( uint32_t descriptorCount) const { - ASSERT(descriptorCount < kNumOfStagingDescriptorAllocators); - return mSamplerAllocators[descriptorCount].get(); + ASSERT(descriptorCount <= kMaxSamplerDescriptorsPerBindGroup); + // This is Log2 of the next power of two, plus 1. + uint32_t allocatorIndex = descriptorCount == 0 ? 0 : Log2Ceil(descriptorCount) + 1; + return mSamplerAllocators[allocatorIndex].get(); } StagingDescriptorAllocator* Device::GetRenderTargetViewAllocator() const { diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/DeviceD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/DeviceD3D12.h index 1ee40927dfb..3ab03934add 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/DeviceD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/DeviceD3D12.h @@ -19,6 +19,8 @@ #include "common/Constants.h" #include "common/SerialQueue.h" +#include "dawn_native/BindingInfo.h" +#include "dawn_native/Commands.h" #include "dawn_native/Device.h" #include "dawn_native/d3d12/CommandRecordingContext.h" #include "dawn_native/d3d12/D3D12Info.h" @@ -91,6 +93,18 @@ namespace dawn_native { namespace d3d12 { uint64_t destinationOffset, uint64_t size) override; + void CopyFromStagingToBufferImpl(CommandRecordingContext* commandContext, + StagingBufferBase* source, + uint64_t sourceOffset, + BufferBase* destination, + uint64_t destinationOffset, + uint64_t size); + + MaybeError CopyFromStagingToTexture(const StagingBufferBase* source, + const TextureDataLayout& src, + TextureCopy* dst, + const Extent3D& copySizePixels); + ResultOrError<ResourceHeapAllocation> AllocateMemory( D3D12_HEAP_TYPE heapType, const D3D12_RESOURCE_DESC& resourceDescriptor, @@ -131,7 +145,8 @@ namespace dawn_native { namespace d3d12 { const BindGroupDescriptor* descriptor) override; ResultOrError<BindGroupLayoutBase*> CreateBindGroupLayoutImpl( const BindGroupLayoutDescriptor* descriptor) override; - ResultOrError<BufferBase*> CreateBufferImpl(const BufferDescriptor* descriptor) override; + ResultOrError<Ref<BufferBase>> CreateBufferImpl( + const BufferDescriptor* descriptor) override; ResultOrError<ComputePipelineBase*> CreateComputePipelineImpl( const ComputePipelineDescriptor* descriptor) override; ResultOrError<PipelineLayoutBase*> CreatePipelineLayoutImpl( @@ -184,13 +199,24 @@ namespace dawn_native { namespace d3d12 { std::unique_ptr<ResourceAllocatorManager> mResourceAllocatorManager; std::unique_ptr<ResidencyManager> mResidencyManager; - // Index corresponds to the descriptor count in the range [0, kMaxBindingsPerGroup]. - static constexpr uint32_t kNumOfStagingDescriptorAllocators = kMaxBindingsPerGroup + 1; + static constexpr uint32_t kMaxSamplerDescriptorsPerBindGroup = + 3 * kMaxSamplersPerShaderStage; + static constexpr uint32_t kMaxViewDescriptorsPerBindGroup = + kMaxBindingsPerPipelineLayout - kMaxSamplerDescriptorsPerBindGroup; + + static constexpr uint32_t kNumSamplerDescriptorAllocators = + ConstexprLog2Ceil(kMaxSamplerDescriptorsPerBindGroup) + 1; + static constexpr uint32_t kNumViewDescriptorAllocators = + ConstexprLog2Ceil(kMaxViewDescriptorsPerBindGroup) + 1; - std::array<std::unique_ptr<StagingDescriptorAllocator>, kNumOfStagingDescriptorAllocators> + // Index corresponds to Log2Ceil(descriptorCount) where descriptorCount is in + // the range [0, kMaxSamplerDescriptorsPerBindGroup]. + std::array<std::unique_ptr<StagingDescriptorAllocator>, kNumViewDescriptorAllocators + 1> mViewAllocators; - std::array<std::unique_ptr<StagingDescriptorAllocator>, kNumOfStagingDescriptorAllocators> + // Index corresponds to Log2Ceil(descriptorCount) where descriptorCount is in + // the range [0, kMaxViewDescriptorsPerBindGroup]. + std::array<std::unique_ptr<StagingDescriptorAllocator>, kNumSamplerDescriptorAllocators + 1> mSamplerAllocators; std::unique_ptr<StagingDescriptorAllocator> mRenderTargetViewAllocator; diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/GPUDescriptorHeapAllocationD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/GPUDescriptorHeapAllocationD3D12.cpp index c72605a58b6..3d3ba20fad3 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/GPUDescriptorHeapAllocationD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/GPUDescriptorHeapAllocationD3D12.cpp @@ -36,4 +36,4 @@ namespace dawn_native { namespace d3d12 { Serial GPUDescriptorHeapAllocation::GetHeapSerial() const { return mHeapSerial; } -}} // namespace dawn_native::d3d12
\ No newline at end of file +}} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/GPUDescriptorHeapAllocationD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/GPUDescriptorHeapAllocationD3D12.h index c18d2662c3a..08930d79fcf 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/GPUDescriptorHeapAllocationD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/GPUDescriptorHeapAllocationD3D12.h @@ -40,4 +40,4 @@ namespace dawn_native { namespace d3d12 { }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_CPUDESCRIPTORHEAPALLOCATION_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_CPUDESCRIPTORHEAPALLOCATION_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/HeapD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/HeapD3D12.cpp index ade5d4a5c1c..7bb4e323a67 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/HeapD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/HeapD3D12.cpp @@ -28,4 +28,4 @@ namespace dawn_native { namespace d3d12 { return mD3d12Heap.Get(); } -}} // namespace dawn_native::d3d12
\ No newline at end of file +}} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/HeapD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/HeapD3D12.h index 715ffcdaf10..b59c6449dec 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/HeapD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/HeapD3D12.h @@ -37,4 +37,4 @@ namespace dawn_native { namespace d3d12 { }; }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_HEAPD3D12_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_HEAPD3D12_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/PageableD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/PageableD3D12.cpp index 58847808644..5c5ef9c0f0d 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/PageableD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/PageableD3D12.cpp @@ -73,4 +73,4 @@ namespace dawn_native { namespace d3d12 { bool Pageable::IsResidencyLocked() const { return mResidencyLockRefCount != 0; } -}} // namespace dawn_native::d3d12
\ No newline at end of file +}} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/PipelineLayoutD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/PipelineLayoutD3D12.cpp index 5e3460bb0f5..8eed07ed2c9 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/PipelineLayoutD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/PipelineLayoutD3D12.cpp @@ -194,7 +194,7 @@ namespace dawn_native { namespace d3d12 { uint32_t PipelineLayout::GetDynamicRootParameterIndex(BindGroupIndex group, BindingIndex bindingIndex) const { ASSERT(group < kMaxBindGroupsTyped); - ASSERT(bindingIndex < kMaxBindingsPerGroupTyped); + ASSERT(bindingIndex < kMaxDynamicBuffersPerPipelineLayoutTyped); ASSERT(GetBindGroupLayout(group)->GetBindingInfo(bindingIndex).hasDynamicOffset); ASSERT(GetBindGroupLayout(group)->GetBindingInfo(bindingIndex).visibility != wgpu::ShaderStage::None); diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/PipelineLayoutD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/PipelineLayoutD3D12.h index 8de41b06773..6116cd81988 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/PipelineLayoutD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/PipelineLayoutD3D12.h @@ -45,7 +45,7 @@ namespace dawn_native { namespace d3d12 { ityp::array<BindGroupIndex, uint32_t, kMaxBindGroups> mCbvUavSrvRootParameterInfo; ityp::array<BindGroupIndex, uint32_t, kMaxBindGroups> mSamplerRootParameterInfo; ityp::array<BindGroupIndex, - ityp::array<BindingIndex, uint32_t, kMaxBindingsPerGroup>, + ityp::array<BindingIndex, uint32_t, kMaxDynamicBuffersPerPipelineLayout>, kMaxBindGroups> mDynamicRootParameterIndices; ComPtr<ID3D12RootSignature> mRootSignature; diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/QuerySetD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/QuerySetD3D12.cpp new file mode 100644 index 00000000000..2bea5263167 --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/QuerySetD3D12.cpp @@ -0,0 +1,68 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "dawn_native/d3d12/QuerySetD3D12.h" + +#include "dawn_native/d3d12/D3D12Error.h" +#include "dawn_native/d3d12/DeviceD3D12.h" + +namespace dawn_native { namespace d3d12 { + + namespace { + D3D12_QUERY_HEAP_TYPE D3D12QueryHeapType(wgpu::QueryType type) { + switch (type) { + case wgpu::QueryType::Occlusion: + return D3D12_QUERY_HEAP_TYPE_OCCLUSION; + case wgpu::QueryType::PipelineStatistics: + return D3D12_QUERY_HEAP_TYPE_PIPELINE_STATISTICS; + case wgpu::QueryType::Timestamp: + return D3D12_QUERY_HEAP_TYPE_TIMESTAMP; + default: + UNREACHABLE(); + } + } + } // anonymous namespace + + // static + ResultOrError<QuerySet*> QuerySet::Create(Device* device, + const QuerySetDescriptor* descriptor) { + Ref<QuerySet> querySet = AcquireRef(new QuerySet(device, descriptor)); + DAWN_TRY(querySet->Initialize()); + return querySet.Detach(); + } + + MaybeError QuerySet::Initialize() { + D3D12_QUERY_HEAP_DESC queryHeapDesc = {}; + queryHeapDesc.Type = D3D12QueryHeapType(GetQueryType()); + queryHeapDesc.Count = GetQueryCount(); + + ID3D12Device* d3d12Device = ToBackend(GetDevice())->GetD3D12Device(); + return CheckOutOfMemoryHRESULT( + d3d12Device->CreateQueryHeap(&queryHeapDesc, IID_PPV_ARGS(&mQueryHeap)), + "ID3D12Device::CreateQueryHeap"); + } + + ID3D12QueryHeap* QuerySet::GetQueryHeap() const { + return mQueryHeap.Get(); + } + + QuerySet::~QuerySet() { + DestroyInternal(); + } + + void QuerySet::DestroyImpl() { + ToBackend(GetDevice())->ReferenceUntilUnused(mQueryHeap); + } + +}} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/QuerySetD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/QuerySetD3D12.h new file mode 100644 index 00000000000..7b24cceb67d --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/QuerySetD3D12.h @@ -0,0 +1,45 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef DAWNNATIVE_D3D12_QUERYSETD3D12_H_ +#define DAWNNATIVE_D3D12_QUERYSETD3D12_H_ + +#include "dawn_native/QuerySet.h" +#include "dawn_native/d3d12/d3d12_platform.h" + +namespace dawn_native { namespace d3d12 { + + class Device; + + class QuerySet : public QuerySetBase { + public: + static ResultOrError<QuerySet*> Create(Device* device, + const QuerySetDescriptor* descriptor); + + ID3D12QueryHeap* GetQueryHeap() const; + + private: + ~QuerySet() override; + using QuerySetBase::QuerySetBase; + MaybeError Initialize(); + + // Dawn API + void DestroyImpl() override; + + ComPtr<ID3D12QueryHeap> mQueryHeap; + }; + +}} // namespace dawn_native::d3d12 + +#endif // DAWNNATIVE_D3D12_QUERYSETD3D12_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/QueueD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/QueueD3D12.cpp index 710e41f052c..b17919eec4b 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/QueueD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/QueueD3D12.cpp @@ -14,6 +14,10 @@ #include "dawn_native/d3d12/QueueD3D12.h" +#include "common/Math.h" +#include "dawn_native/CommandValidation.h" +#include "dawn_native/Commands.h" +#include "dawn_native/DynamicUploader.h" #include "dawn_native/d3d12/CommandBufferD3D12.h" #include "dawn_native/d3d12/D3D12Error.h" #include "dawn_native/d3d12/DeviceD3D12.h" @@ -22,6 +26,50 @@ namespace dawn_native { namespace d3d12 { + namespace { + ResultOrError<UploadHandle> UploadTextureDataAligningBytesPerRow( + DeviceBase* device, + const void* data, + uint32_t alignedBytesPerRow, + uint32_t optimallyAlignedBytesPerRow, + uint32_t alignedRowsPerImage, + const TextureDataLayout& dataLayout, + const Format& textureFormat, + const Extent3D& writeSizePixel) { + uint64_t newDataSizeBytes; + DAWN_TRY_ASSIGN( + newDataSizeBytes, + ComputeRequiredBytesInCopy(textureFormat, writeSizePixel, + optimallyAlignedBytesPerRow, alignedRowsPerImage)); + + UploadHandle uploadHandle; + DAWN_TRY_ASSIGN(uploadHandle, device->GetDynamicUploader()->Allocate( + newDataSizeBytes, device->GetPendingCommandSerial(), + textureFormat.blockByteSize)); + ASSERT(uploadHandle.mappedBuffer != nullptr); + + uint8_t* dstPointer = static_cast<uint8_t*>(uploadHandle.mappedBuffer); + const uint8_t* srcPointer = static_cast<const uint8_t*>(data); + srcPointer += dataLayout.offset; + + uint32_t alignedRowsPerImageInBlock = alignedRowsPerImage / textureFormat.blockHeight; + uint32_t dataRowsPerImageInBlock = dataLayout.rowsPerImage / textureFormat.blockHeight; + if (dataRowsPerImageInBlock == 0) { + dataRowsPerImageInBlock = writeSizePixel.height / textureFormat.blockHeight; + } + + ASSERT(dataRowsPerImageInBlock >= alignedRowsPerImageInBlock); + uint64_t imageAdditionalStride = + dataLayout.bytesPerRow * (dataRowsPerImageInBlock - alignedRowsPerImageInBlock); + + CopyTextureData(dstPointer, srcPointer, writeSizePixel.depth, + alignedRowsPerImageInBlock, imageAdditionalStride, alignedBytesPerRow, + optimallyAlignedBytesPerRow, dataLayout.bytesPerRow); + + return uploadHandle; + } + } // namespace + Queue::Queue(Device* device) : QueueBase(device) { } @@ -47,4 +95,43 @@ namespace dawn_native { namespace d3d12 { return {}; } + MaybeError Queue::WriteTextureImpl(const TextureCopyView& destination, + const void* data, + const TextureDataLayout& dataLayout, + const Extent3D& writeSizePixel) { + const TexelBlockInfo& blockInfo = + destination.texture->GetFormat().GetTexelBlockInfo(destination.aspect); + + // We are only copying the part of the data that will appear in the texture. + // Note that validating texture copy range ensures that writeSizePixel->width and + // writeSizePixel->height are multiples of blockWidth and blockHeight respectively. + uint32_t alignedBytesPerRow = + (writeSizePixel.width) / blockInfo.blockWidth * blockInfo.blockByteSize; + uint32_t alignedRowsPerImage = writeSizePixel.height; + uint32_t optimallyAlignedBytesPerRow = + Align(alignedBytesPerRow, D3D12_TEXTURE_DATA_PITCH_ALIGNMENT); + + UploadHandle uploadHandle; + DAWN_TRY_ASSIGN( + uploadHandle, + UploadTextureDataAligningBytesPerRow( + GetDevice(), data, alignedBytesPerRow, optimallyAlignedBytesPerRow, + alignedRowsPerImage, dataLayout, destination.texture->GetFormat(), writeSizePixel)); + + TextureDataLayout passDataLayout = dataLayout; + passDataLayout.offset = uploadHandle.startOffset; + passDataLayout.bytesPerRow = optimallyAlignedBytesPerRow; + passDataLayout.rowsPerImage = alignedRowsPerImage; + + TextureCopy textureCopy; + textureCopy.texture = destination.texture; + textureCopy.mipLevel = destination.mipLevel; + textureCopy.origin = destination.origin; + textureCopy.aspect = ConvertAspect(destination.texture->GetFormat(), destination.aspect); + + return ToBackend(GetDevice()) + ->CopyFromStagingToTexture(uploadHandle.stagingBuffer, passDataLayout, &textureCopy, + writeSizePixel); + } + }} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/QueueD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/QueueD3D12.h index f211d0bf587..6b23a4569eb 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/QueueD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/QueueD3D12.h @@ -31,6 +31,10 @@ namespace dawn_native { namespace d3d12 { private: MaybeError SubmitImpl(uint32_t commandCount, CommandBufferBase* const* commands) override; + MaybeError WriteTextureImpl(const TextureCopyView& destination, + const void* data, + const TextureDataLayout& dataLayout, + const Extent3D& writeSizePixel) override; }; }} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPassBuilderD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPassBuilderD3D12.cpp index 19657849a19..587f48a9aec 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPassBuilderD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPassBuilderD3D12.cpp @@ -86,12 +86,14 @@ namespace dawn_native { namespace d3d12 { D3D12EndingAccessResolveSubresourceParameters(TextureView* resolveDestination) { D3D12_RENDER_PASS_ENDING_ACCESS_RESOLVE_SUBRESOURCE_PARAMETERS subresourceParameters; Texture* resolveDestinationTexture = ToBackend(resolveDestination->GetTexture()); + ASSERT(resolveDestinationTexture->GetFormat().aspects == Aspect::Color); subresourceParameters.DstX = 0; subresourceParameters.DstY = 0; subresourceParameters.SrcSubresource = 0; subresourceParameters.DstSubresource = resolveDestinationTexture->GetSubresourceIndex( - resolveDestination->GetBaseMipLevel(), resolveDestination->GetBaseArrayLayer()); + resolveDestination->GetBaseMipLevel(), resolveDestination->GetBaseArrayLayer(), + Aspect::Color); // Resolving a specified sub-rect is only valid on hardware that supports sample // positions. This means even {0, 0, width, height} would be invalid if unsupported. To // avoid this, we assume sub-rect resolves never work by setting them to all zeros or @@ -234,4 +236,4 @@ namespace dawn_native { namespace d3d12 { D3D12_RENDER_PASS_ENDING_ACCESS_TYPE_NO_ACCESS; } -}} // namespace dawn_native::d3d12
\ No newline at end of file +}} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPassBuilderD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPassBuilderD3D12.h index aadb2e8226d..cf61a182a09 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPassBuilderD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPassBuilderD3D12.h @@ -88,4 +88,4 @@ namespace dawn_native { namespace d3d12 { }; }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_RENDERPASSBUILDERD3D12_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_RENDERPASSBUILDERD3D12_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPipelineD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPipelineD3D12.cpp index f629e76f261..165315469fe 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPipelineD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/RenderPipelineD3D12.cpp @@ -327,7 +327,6 @@ namespace dawn_native { namespace d3d12 { wgpu::ShaderStage renderStages = wgpu::ShaderStage::Vertex | wgpu::ShaderStage::Fragment; for (auto stage : IterateStages(renderStages)) { - std::string hlslSource; DAWN_TRY_ASSIGN(hlslSource, modules[stage]->GetHLSLSource(ToBackend(GetLayout()))); @@ -384,13 +383,13 @@ namespace dawn_native { namespace d3d12 { } descriptorD3D12.NumRenderTargets = static_cast<uint32_t>(GetColorAttachmentsMask().count()); - descriptorD3D12.BlendState.AlphaToCoverageEnable = FALSE; + descriptorD3D12.BlendState.AlphaToCoverageEnable = descriptor->alphaToCoverageEnabled; descriptorD3D12.BlendState.IndependentBlendEnable = TRUE; descriptorD3D12.DepthStencilState = ComputeDepthStencilDesc(GetDepthStencilStateDescriptor()); - descriptorD3D12.SampleMask = UINT_MAX; + descriptorD3D12.SampleMask = GetSampleMask(); descriptorD3D12.PrimitiveTopologyType = D3D12PrimitiveTopologyType(GetPrimitiveTopology()); descriptorD3D12.SampleDesc.Count = GetSampleCount(); descriptorD3D12.SampleDesc.Quality = 0; diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/ResidencyManagerD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/ResidencyManagerD3D12.cpp index 053a1690fdb..e66296e4e26 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/ResidencyManagerD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/ResidencyManagerD3D12.cpp @@ -367,4 +367,4 @@ namespace dawn_native { namespace d3d12 { } } -}} // namespace dawn_native::d3d12
\ No newline at end of file +}} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/ResidencyManagerD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/ResidencyManagerD3D12.h index 304a211b084..a79a4fca6a2 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/ResidencyManagerD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/ResidencyManagerD3D12.h @@ -80,4 +80,4 @@ namespace dawn_native { namespace d3d12 { }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_RESIDENCYMANAGERD3D12_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_RESIDENCYMANAGERD3D12_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceAllocatorManagerD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceAllocatorManagerD3D12.cpp index 4e48f2edafc..8090b7db468 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceAllocatorManagerD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceAllocatorManagerD3D12.cpp @@ -19,6 +19,7 @@ #include "dawn_native/d3d12/HeapAllocatorD3D12.h" #include "dawn_native/d3d12/HeapD3D12.h" #include "dawn_native/d3d12/ResidencyManagerD3D12.h" +#include "dawn_native/d3d12/UtilsD3D12.h" namespace dawn_native { namespace d3d12 { namespace { @@ -148,6 +149,17 @@ namespace dawn_native { namespace d3d12 { } } + bool IsClearValueOptimizable(const D3D12_RESOURCE_DESC& resourceDescriptor) { + // Optimized clear color cannot be set on buffers, non-render-target/depth-stencil + // textures, or typeless resources + // https://docs.microsoft.com/en-us/windows/win32/api/d3d12/nf-d3d12-id3d12device-createcommittedresource + // https://docs.microsoft.com/en-us/windows/win32/api/d3d12/nf-d3d12-id3d12device-createplacedresource + return !IsTypeless(resourceDescriptor.Format) && + resourceDescriptor.Dimension != D3D12_RESOURCE_DIMENSION_BUFFER && + (resourceDescriptor.Flags & (D3D12_RESOURCE_FLAG_ALLOW_RENDER_TARGET | + D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL)) != 0; + } + } // namespace ResourceAllocatorManager::ResourceAllocatorManager(Device* device) : mDevice(device) { @@ -160,8 +172,10 @@ namespace dawn_native { namespace d3d12 { mHeapAllocators[i] = std::make_unique<HeapAllocator>( mDevice, GetD3D12HeapType(resourceHeapKind), GetD3D12HeapFlags(resourceHeapKind), GetMemorySegment(device, GetD3D12HeapType(resourceHeapKind))); + mPooledHeapAllocators[i] = + std::make_unique<PooledResourceMemoryAllocator>(mHeapAllocators[i].get()); mSubAllocatedResourceAllocators[i] = std::make_unique<BuddyMemoryAllocator>( - kMaxHeapSize, kMinHeapSize, mHeapAllocators[i].get()); + kMaxHeapSize, kMinHeapSize, mPooledHeapAllocators[i].get()); } } @@ -169,14 +183,25 @@ namespace dawn_native { namespace d3d12 { D3D12_HEAP_TYPE heapType, const D3D12_RESOURCE_DESC& resourceDescriptor, D3D12_RESOURCE_STATES initialUsage) { + // In order to suppress a warning in the D3D12 debug layer, we need to specify an + // optimized clear value. As there are no negative consequences when picking a mismatched + // clear value, we use zero as the optimized clear value. This also enables fast clears on + // some architectures. + D3D12_CLEAR_VALUE zero{}; + D3D12_CLEAR_VALUE* optimizedClearValue = nullptr; + if (IsClearValueOptimizable(resourceDescriptor)) { + zero.Format = resourceDescriptor.Format; + optimizedClearValue = &zero; + } + // TODO(bryan.bernhart@intel.com): Conditionally disable sub-allocation. // For very large resources, there is no benefit to suballocate. // For very small resources, it is inefficent to suballocate given the min. heap // size could be much larger then the resource allocation. // Attempt to satisfy the request using sub-allocation (placed resource in a heap). ResourceHeapAllocation subAllocation; - DAWN_TRY_ASSIGN(subAllocation, - CreatePlacedResource(heapType, resourceDescriptor, initialUsage)); + DAWN_TRY_ASSIGN(subAllocation, CreatePlacedResource(heapType, resourceDescriptor, + optimizedClearValue, initialUsage)); if (subAllocation.GetInfo().mMethod != AllocationMethod::kInvalid) { return std::move(subAllocation); } @@ -184,7 +209,8 @@ namespace dawn_native { namespace d3d12 { // If sub-allocation fails, fall-back to direct allocation (committed resource). ResourceHeapAllocation directAllocation; DAWN_TRY_ASSIGN(directAllocation, - CreateCommittedResource(heapType, resourceDescriptor, initialUsage)); + CreateCommittedResource(heapType, resourceDescriptor, optimizedClearValue, + initialUsage)); if (directAllocation.GetInfo().mMethod != AllocationMethod::kInvalid) { return std::move(directAllocation); } @@ -221,7 +247,7 @@ namespace dawn_native { namespace d3d12 { // calls DeallocateMemory again using the same allocation. allocation.Invalidate(); - ASSERT(allocation.GetD3D12Resource().Get() == nullptr); + ASSERT(allocation.GetD3D12Resource() == nullptr); } void ResourceAllocatorManager::FreeMemory(ResourceHeapAllocation& allocation) { @@ -242,6 +268,7 @@ namespace dawn_native { namespace d3d12 { ResultOrError<ResourceHeapAllocation> ResourceAllocatorManager::CreatePlacedResource( D3D12_HEAP_TYPE heapType, const D3D12_RESOURCE_DESC& requestedResourceDescriptor, + const D3D12_CLEAR_VALUE* optimizedClearValue, D3D12_RESOURCE_STATES initialUsage) { const ResourceHeapKind resourceHeapKind = GetResourceHeapKind(requestedResourceDescriptor.Dimension, heapType, @@ -252,12 +279,15 @@ namespace dawn_native { namespace d3d12 { resourceHeapKind, requestedResourceDescriptor.SampleDesc.Count, requestedResourceDescriptor.Alignment); + // TODO(bryan.bernhart): Figure out how to compute the alignment without calling this + // twice. D3D12_RESOURCE_ALLOCATION_INFO resourceInfo = mDevice->GetD3D12Device()->GetResourceAllocationInfo(0, 1, &resourceDescriptor); // If the requested resource alignment was rejected, let D3D tell us what the // required alignment is for this resource. - if (resourceDescriptor.Alignment != resourceInfo.Alignment) { + if (resourceDescriptor.Alignment != 0 && + resourceDescriptor.Alignment != resourceInfo.Alignment) { resourceDescriptor.Alignment = 0; resourceInfo = mDevice->GetD3D12Device()->GetResourceAllocationInfo(0, 1, &resourceDescriptor); @@ -299,7 +329,7 @@ namespace dawn_native { namespace d3d12 { DAWN_TRY(CheckOutOfMemoryHRESULT( mDevice->GetD3D12Device()->CreatePlacedResource( heap->GetD3D12Heap(), allocation.GetOffset(), &resourceDescriptor, initialUsage, - nullptr, IID_PPV_ARGS(&placedResource)), + optimizedClearValue, IID_PPV_ARGS(&placedResource)), "ID3D12Device::CreatePlacedResource")); // After CreatePlacedResource has finished, the heap can be unlocked from residency. This @@ -313,6 +343,7 @@ namespace dawn_native { namespace d3d12 { ResultOrError<ResourceHeapAllocation> ResourceAllocatorManager::CreateCommittedResource( D3D12_HEAP_TYPE heapType, const D3D12_RESOURCE_DESC& resourceDescriptor, + const D3D12_CLEAR_VALUE* optimizedClearValue, D3D12_RESOURCE_STATES initialUsage) { D3D12_HEAP_PROPERTIES heapProperties; heapProperties.Type = heapType; @@ -345,11 +376,11 @@ namespace dawn_native { namespace d3d12 { // Note: Heap flags are inferred by the resource descriptor and do not need to be explicitly // provided to CreateCommittedResource. ComPtr<ID3D12Resource> committedResource; - DAWN_TRY( - CheckOutOfMemoryHRESULT(mDevice->GetD3D12Device()->CreateCommittedResource( - &heapProperties, D3D12_HEAP_FLAG_NONE, &resourceDescriptor, - initialUsage, nullptr, IID_PPV_ARGS(&committedResource)), - "ID3D12Device::CreateCommittedResource")); + DAWN_TRY(CheckOutOfMemoryHRESULT( + mDevice->GetD3D12Device()->CreateCommittedResource( + &heapProperties, D3D12_HEAP_FLAG_NONE, &resourceDescriptor, initialUsage, + optimizedClearValue, IID_PPV_ARGS(&committedResource)), + "ID3D12Device::CreateCommittedResource")); // When using CreateCommittedResource, D3D12 creates an implicit heap that contains the // resource allocation. Because Dawn's memory residency management occurs at the resource @@ -370,4 +401,10 @@ namespace dawn_native { namespace d3d12 { /*offset*/ 0, std::move(committedResource), heap}; } + void ResourceAllocatorManager::DestroyPool() { + for (auto& alloc : mPooledHeapAllocators) { + alloc->DestroyPool(); + } + } + }} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceAllocatorManagerD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceAllocatorManagerD3D12.h index b60a024b27c..0bf2a02c61e 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceAllocatorManagerD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceAllocatorManagerD3D12.h @@ -17,6 +17,7 @@ #include "common/SerialQueue.h" #include "dawn_native/BuddyMemoryAllocator.h" +#include "dawn_native/PooledResourceMemoryAllocator.h" #include "dawn_native/d3d12/HeapAllocatorD3D12.h" #include "dawn_native/d3d12/ResourceHeapAllocationD3D12.h" @@ -67,17 +68,21 @@ namespace dawn_native { namespace d3d12 { void Tick(Serial lastCompletedSerial); + void DestroyPool(); + private: void FreeMemory(ResourceHeapAllocation& allocation); ResultOrError<ResourceHeapAllocation> CreatePlacedResource( D3D12_HEAP_TYPE heapType, const D3D12_RESOURCE_DESC& requestedResourceDescriptor, + const D3D12_CLEAR_VALUE* optimizedClearValue, D3D12_RESOURCE_STATES initialUsage); ResultOrError<ResourceHeapAllocation> CreateCommittedResource( D3D12_HEAP_TYPE heapType, const D3D12_RESOURCE_DESC& resourceDescriptor, + const D3D12_CLEAR_VALUE* optimizedClearValue, D3D12_RESOURCE_STATES initialUsage); Device* mDevice; @@ -90,6 +95,9 @@ namespace dawn_native { namespace d3d12 { mSubAllocatedResourceAllocators; std::array<std::unique_ptr<HeapAllocator>, ResourceHeapKind::EnumCount> mHeapAllocators; + std::array<std::unique_ptr<PooledResourceMemoryAllocator>, ResourceHeapKind::EnumCount> + mPooledHeapAllocators; + SerialQueue<ResourceHeapAllocation> mAllocationsToDelete; }; diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceHeapAllocationD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceHeapAllocationD3D12.cpp index c3a89f0590d..edf21c1e06d 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceHeapAllocationD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceHeapAllocationD3D12.cpp @@ -32,8 +32,8 @@ namespace dawn_native { namespace d3d12 { mResource.Reset(); } - ComPtr<ID3D12Resource> ResourceHeapAllocation::GetD3D12Resource() const { - return mResource; + ID3D12Resource* ResourceHeapAllocation::GetD3D12Resource() const { + return mResource.Get(); } D3D12_GPU_VIRTUAL_ADDRESS ResourceHeapAllocation::GetGPUPointer() const { diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceHeapAllocationD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceHeapAllocationD3D12.h index 71b00fd5fd7..ace1a7efa00 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceHeapAllocationD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/ResourceHeapAllocationD3D12.h @@ -33,7 +33,7 @@ namespace dawn_native { namespace d3d12 { void Invalidate() override; - ComPtr<ID3D12Resource> GetD3D12Resource() const; + ID3D12Resource* GetD3D12Resource() const; D3D12_GPU_VIRTUAL_ADDRESS GetGPUPointer() const; private: diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/SamplerHeapCacheD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/SamplerHeapCacheD3D12.cpp index 224051a1c42..8ba3aab3197 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/SamplerHeapCacheD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/SamplerHeapCacheD3D12.cpp @@ -164,4 +164,4 @@ namespace dawn_native { namespace d3d12 { const SamplerHeapCacheEntry* b) const { return a->mSamplers == b->mSamplers; } -}} // namespace dawn_native::d3d12
\ No newline at end of file +}} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/SamplerHeapCacheD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/SamplerHeapCacheD3D12.h index 2f41086c8b2..2e5a2d4237a 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/SamplerHeapCacheD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/SamplerHeapCacheD3D12.h @@ -105,4 +105,4 @@ namespace dawn_native { namespace d3d12 { }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_SAMPLERHEAPCACHE_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_SAMPLERHEAPCACHE_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderModuleD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderModuleD3D12.cpp index 770a006fb42..0bdbafa12cb 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderModuleD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderModuleD3D12.cpp @@ -31,7 +31,7 @@ namespace dawn_native { namespace d3d12 { namespace { - std::vector<const wchar_t*> GetDXCArguments(uint32_t compileFlags) { + std::vector<const wchar_t*> GetDXCArguments(uint32_t compileFlags, bool enable16BitTypes) { std::vector<const wchar_t*> arguments; if (compileFlags & D3DCOMPILE_ENABLE_BACKWARDS_COMPATIBILITY) { arguments.push_back(L"/Gec"); @@ -70,9 +70,15 @@ namespace dawn_native { namespace d3d12 { if (compileFlags & D3DCOMPILE_RESOURCES_MAY_ALIAS) { arguments.push_back(L"/res_may_alias"); } - // Enable FXC backward compatibility by setting the language version to 2016 - arguments.push_back(L"-HV"); - arguments.push_back(L"2016"); + + if (enable16BitTypes) { + // enable-16bit-types are only allowed in -HV 2018 (default) + arguments.push_back(L"/enable-16bit-types"); + } else { + // Enable FXC backward compatibility by setting the language version to 2016 + arguments.push_back(L"-HV"); + arguments.push_back(L"2016"); + } return arguments; } @@ -98,7 +104,12 @@ namespace dawn_native { namespace d3d12 { shaderc_spvc::CompileOptions options = GetCompileOptions(); options.SetForceZeroInitializedVariables(true); - options.SetHLSLShaderModel(51); + if (GetDevice()->IsExtensionEnabled(Extension::ShaderFloat16)) { + options.SetHLSLShaderModel(ToBackend(GetDevice())->GetDeviceInfo().shaderModel); + options.SetHLSLEnable16BitTypes(true); + } else { + options.SetHLSLShaderModel(51); + } // PointCoord and PointSize are not supported in HLSL // TODO (hao.x.li@intel.com): The point_coord_compat and point_size_compat are // required temporarily for https://bugs.chromium.org/p/dawn/issues/detail?id=146, @@ -138,7 +149,12 @@ namespace dawn_native { namespace d3d12 { options_glsl.force_zero_initialized_variables = true; spirv_cross::CompilerHLSL::Options options_hlsl; - options_hlsl.shader_model = 51; + if (GetDevice()->IsExtensionEnabled(Extension::ShaderFloat16)) { + options_hlsl.shader_model = ToBackend(GetDevice())->GetDeviceInfo().shaderModel; + options_hlsl.enable_16bit_types = true; + } else { + options_hlsl.shader_model = 51; + } // PointCoord and PointSize are not supported in HLSL // TODO (hao.x.li@intel.com): The point_coord_compat and point_size_compat are // required temporarily for https://bugs.chromium.org/p/dawn/issues/detail?id=146, @@ -210,19 +226,6 @@ namespace dawn_native { namespace d3d12 { const std::string& hlslSource, const char* entryPoint, uint32_t compileFlags) { - const wchar_t* targetProfile = nullptr; - switch (stage) { - case SingleShaderStage::Vertex: - targetProfile = L"vs_6_0"; - break; - case SingleShaderStage::Fragment: - targetProfile = L"ps_6_0"; - break; - case SingleShaderStage::Compute: - targetProfile = L"cs_6_0"; - break; - } - IDxcLibrary* dxcLibrary; DAWN_TRY_ASSIGN(dxcLibrary, ToBackend(GetDevice())->GetOrCreateDxcLibrary()); @@ -237,13 +240,16 @@ namespace dawn_native { namespace d3d12 { std::wstring entryPointW; DAWN_TRY_ASSIGN(entryPointW, ConvertStringToWstring(entryPoint)); - std::vector<const wchar_t*> arguments = GetDXCArguments(compileFlags); + std::vector<const wchar_t*> arguments = GetDXCArguments( + compileFlags, GetDevice()->IsExtensionEnabled(Extension::ShaderFloat16)); ComPtr<IDxcOperationResult> result; - DAWN_TRY(CheckHRESULT( - dxcCompiler->Compile(sourceBlob.Get(), nullptr, entryPointW.c_str(), targetProfile, - arguments.data(), arguments.size(), nullptr, 0, nullptr, &result), - "DXC compile")); + DAWN_TRY( + CheckHRESULT(dxcCompiler->Compile( + sourceBlob.Get(), nullptr, entryPointW.c_str(), + ToBackend(GetDevice())->GetDeviceInfo().shaderProfiles[stage].c_str(), + arguments.data(), arguments.size(), nullptr, 0, nullptr, &result), + "DXC compile")); HRESULT hr; DAWN_TRY(CheckHRESULT(result->GetStatus(&hr), "DXC get status")); diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderVisibleDescriptorAllocatorD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderVisibleDescriptorAllocatorD3D12.cpp index 9039d8ab9d7..ba1b4939a27 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderVisibleDescriptorAllocatorD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderVisibleDescriptorAllocatorD3D12.cpp @@ -20,10 +20,32 @@ namespace dawn_native { namespace d3d12 { + // Limits the min/max heap size to always be some known value for testing. // Thresholds should be adjusted (lower == faster) to avoid tests taking too long to complete. static constexpr const uint32_t kShaderVisibleSmallHeapSizes[] = {1024, 512}; - uint32_t GetD3D12ShaderVisibleHeapSize(D3D12_DESCRIPTOR_HEAP_TYPE heapType, bool useSmallSize) { + uint32_t GetD3D12ShaderVisibleHeapMinSize(D3D12_DESCRIPTOR_HEAP_TYPE heapType, + bool useSmallSize) { + if (useSmallSize) { + return kShaderVisibleSmallHeapSizes[heapType]; + } + + // Minimum heap size must be large enough to satisfy the largest descriptor allocation + // request and to amortize the cost of sub-allocation. But small enough to avoid wasting + // memory should only a tiny fraction ever be used. + // TODO(dawn:155): Figure out these values. + switch (heapType) { + case D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV: + return 4096; + case D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER: + return 256; + default: + UNREACHABLE(); + } + } + + uint32_t GetD3D12ShaderVisibleHeapMaxSize(D3D12_DESCRIPTOR_HEAP_TYPE heapType, + bool useSmallSize) { if (useSmallSize) { return kShaderVisibleSmallHeapSizes[heapType]; } @@ -62,7 +84,10 @@ namespace dawn_native { namespace d3d12 { D3D12_DESCRIPTOR_HEAP_TYPE heapType) : mHeapType(heapType), mDevice(device), - mSizeIncrement(device->GetD3D12Device()->GetDescriptorHandleIncrementSize(heapType)) { + mSizeIncrement(device->GetD3D12Device()->GetDescriptorHandleIncrementSize(heapType)), + mDescriptorCount(GetD3D12ShaderVisibleHeapMinSize( + heapType, + mDevice->IsToggleEnabled(Toggle::UseD3D12SmallShaderVisibleHeapForTesting))) { ASSERT(heapType == D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV || heapType == D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER); } @@ -107,60 +132,74 @@ namespace dawn_native { namespace d3d12 { mAllocator.Deallocate(completedSerial); } + ResultOrError<std::unique_ptr<ShaderVisibleDescriptorHeap>> + ShaderVisibleDescriptorAllocator::AllocateHeap(uint32_t descriptorCount) const { + // The size in bytes of a descriptor heap is best calculated by the increment size + // multiplied by the number of descriptors. In practice, this is only an estimate and + // the actual size may vary depending on the driver. + const uint64_t kSize = mSizeIncrement * descriptorCount; + + DAWN_TRY(mDevice->GetResidencyManager()->EnsureCanAllocate(kSize, MemorySegment::Local)); + + ComPtr<ID3D12DescriptorHeap> d3d12DescriptorHeap; + D3D12_DESCRIPTOR_HEAP_DESC heapDescriptor; + heapDescriptor.Type = mHeapType; + heapDescriptor.NumDescriptors = descriptorCount; + heapDescriptor.Flags = GetD3D12HeapFlags(mHeapType); + heapDescriptor.NodeMask = 0; + DAWN_TRY(CheckOutOfMemoryHRESULT(mDevice->GetD3D12Device()->CreateDescriptorHeap( + &heapDescriptor, IID_PPV_ARGS(&d3d12DescriptorHeap)), + "ID3D12Device::CreateDescriptorHeap")); + + std::unique_ptr<ShaderVisibleDescriptorHeap> descriptorHeap = + std::make_unique<ShaderVisibleDescriptorHeap>(std::move(d3d12DescriptorHeap), kSize); + + // We must track the allocation in the LRU when it is created, otherwise the residency + // manager will see the allocation as non-resident in the later call to LockAllocation. + mDevice->GetResidencyManager()->TrackResidentAllocation(descriptorHeap.get()); + + return std::move(descriptorHeap); + } + // Creates a GPU descriptor heap that manages descriptors in a FIFO queue. MaybeError ShaderVisibleDescriptorAllocator::AllocateAndSwitchShaderVisibleHeap() { std::unique_ptr<ShaderVisibleDescriptorHeap> descriptorHeap; - // Return the switched out heap to the pool and retrieve the oldest heap that is no longer - // used by GPU. This maintains a heap buffer to avoid frequently re-creating heaps for heavy - // users. - // TODO(dawn:256): Consider periodically triming to avoid OOM. + // Dynamically allocate using a two-phase allocation strategy. + // The first phase increasingly grows a small heap in binary sizes for light users while the + // second phase pool-allocates largest sized heaps for heavy users. if (mHeap != nullptr) { mDevice->GetResidencyManager()->UnlockAllocation(mHeap.get()); - mPool.push_back({mDevice->GetPendingCommandSerial(), std::move(mHeap)}); - } - // Recycle existing heap if possible. - if (!mPool.empty() && mPool.front().heapSerial <= mDevice->GetCompletedCommandSerial()) { - descriptorHeap = std::move(mPool.front().heap); - mPool.pop_front(); + const uint32_t maxDescriptorCount = GetD3D12ShaderVisibleHeapMaxSize( + mHeapType, + mDevice->IsToggleEnabled(Toggle::UseD3D12SmallShaderVisibleHeapForTesting)); + if (mDescriptorCount < maxDescriptorCount) { + // Phase #1. Grow the heaps in powers-of-two. + mDevice->ReferenceUntilUnused(mHeap->GetD3D12DescriptorHeap()); + mDescriptorCount = std::min(mDescriptorCount * 2, maxDescriptorCount); + } else { + // Phase #2. Pool-allocate heaps. + // Return the switched out heap to the pool and retrieve the oldest heap that is no + // longer used by GPU. This maintains a heap buffer to avoid frequently re-creating + // heaps for heavy users. + // TODO(dawn:256): Consider periodically triming to avoid OOM. + mPool.push_back({mDevice->GetPendingCommandSerial(), std::move(mHeap)}); + if (mPool.front().heapSerial <= mDevice->GetCompletedCommandSerial()) { + descriptorHeap = std::move(mPool.front().heap); + mPool.pop_front(); + } + } } - // TODO(bryan.bernhart@intel.com): Allocating to max heap size wastes memory - // should the developer not allocate any bindings for the heap type. - // Consider dynamically re-sizing GPU heaps. - const uint32_t descriptorCount = GetD3D12ShaderVisibleHeapSize( - mHeapType, mDevice->IsToggleEnabled(Toggle::UseD3D12SmallShaderVisibleHeapForTesting)); - if (descriptorHeap == nullptr) { - // The size in bytes of a descriptor heap is best calculated by the increment size - // multiplied by the number of descriptors. In practice, this is only an estimate and - // the actual size may vary depending on the driver. - const uint64_t kSize = mSizeIncrement * descriptorCount; - - DAWN_TRY( - mDevice->GetResidencyManager()->EnsureCanAllocate(kSize, MemorySegment::Local)); - - ComPtr<ID3D12DescriptorHeap> d3d12DescriptorHeap; - D3D12_DESCRIPTOR_HEAP_DESC heapDescriptor; - heapDescriptor.Type = mHeapType; - heapDescriptor.NumDescriptors = descriptorCount; - heapDescriptor.Flags = GetD3D12HeapFlags(mHeapType); - heapDescriptor.NodeMask = 0; - DAWN_TRY( - CheckOutOfMemoryHRESULT(mDevice->GetD3D12Device()->CreateDescriptorHeap( - &heapDescriptor, IID_PPV_ARGS(&d3d12DescriptorHeap)), - "ID3D12Device::CreateDescriptorHeap")); - descriptorHeap = std::make_unique<ShaderVisibleDescriptorHeap>( - std::move(d3d12DescriptorHeap), kSize); - // We must track the allocation in the LRU when it is created, otherwise the residency - // manager will see the allocation as non-resident in the later call to LockAllocation. - mDevice->GetResidencyManager()->TrackResidentAllocation(descriptorHeap.get()); + DAWN_TRY_ASSIGN(descriptorHeap, AllocateHeap(mDescriptorCount)); } DAWN_TRY(mDevice->GetResidencyManager()->LockAllocation(descriptorHeap.get())); + // Create a FIFO buffer from the recently created heap. mHeap = std::move(descriptorHeap); - mAllocator = RingBufferAllocator(descriptorCount); + mAllocator = RingBufferAllocator(mDescriptorCount); // Invalidate all bindgroup allocations on previously bound heaps by incrementing the heap // serial. When a bindgroup attempts to re-populate, it will compare with its recorded @@ -209,4 +248,4 @@ namespace dawn_native { namespace d3d12 { ID3D12DescriptorHeap* ShaderVisibleDescriptorHeap::GetD3D12DescriptorHeap() const { return mD3d12DescriptorHeap.Get(); } -}} // namespace dawn_native::d3d12
\ No newline at end of file +}} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderVisibleDescriptorAllocatorD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderVisibleDescriptorAllocatorD3D12.h index aec20a3b449..d93e57a0730 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderVisibleDescriptorAllocatorD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/ShaderVisibleDescriptorAllocatorD3D12.h @@ -78,6 +78,9 @@ namespace dawn_native { namespace d3d12 { std::unique_ptr<ShaderVisibleDescriptorHeap> heap; }; + ResultOrError<std::unique_ptr<ShaderVisibleDescriptorHeap>> AllocateHeap( + uint32_t descriptorCount) const; + std::unique_ptr<ShaderVisibleDescriptorHeap> mHeap; RingBufferAllocator mAllocator; std::list<SerialDescriptorHeap> mPool; @@ -91,7 +94,11 @@ namespace dawn_native { namespace d3d12 { Serial mHeapSerial = 0; uint32_t mSizeIncrement; + + // The descriptor count is the current size of the heap in number of descriptors. + // This is stored on the allocator to avoid extra conversions. + uint32_t mDescriptorCount = 0; }; }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_SHADERVISIBLEDESCRIPTORALLOCATOR_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_SHADERVISIBLEDESCRIPTORALLOCATOR_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/StagingBufferD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/StagingBufferD3D12.cpp index b3aec3f1d9d..11c612ba82f 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/StagingBufferD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/StagingBufferD3D12.cpp @@ -69,6 +69,6 @@ namespace dawn_native { namespace d3d12 { } ID3D12Resource* StagingBuffer::GetResource() const { - return mUploadHeap.GetD3D12Resource().Get(); + return mUploadHeap.GetD3D12Resource(); } }} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/StagingDescriptorAllocatorD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/StagingDescriptorAllocatorD3D12.cpp index 84fd7880ca8..92a47412e96 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/StagingDescriptorAllocatorD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/StagingDescriptorAllocatorD3D12.cpp @@ -149,4 +149,4 @@ namespace dawn_native { namespace d3d12 { mAllocationsToDelete.ClearUpTo(completedSerial); } -}} // namespace dawn_native::d3d12
\ No newline at end of file +}} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/StagingDescriptorAllocatorD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/StagingDescriptorAllocatorD3D12.h index eebbe2a1649..519920b4655 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/StagingDescriptorAllocatorD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/StagingDescriptorAllocatorD3D12.h @@ -82,4 +82,4 @@ namespace dawn_native { namespace d3d12 { }} // namespace dawn_native::d3d12 -#endif // DAWNNATIVE_D3D12_STAGINGDESCRIPTORALLOCATOR_H_
\ No newline at end of file +#endif // DAWNNATIVE_D3D12_STAGINGDESCRIPTORALLOCATOR_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/TextureCopySplitter.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/TextureCopySplitter.cpp index f2a8388220e..e1d380c5fe9 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/TextureCopySplitter.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/TextureCopySplitter.cpp @@ -21,7 +21,7 @@ namespace dawn_native { namespace d3d12 { namespace { - Origin3D ComputeTexelOffsets(const Format& format, + Origin3D ComputeTexelOffsets(const TexelBlockInfo& blockInfo, uint32_t offset, uint32_t bytesPerRow, uint32_t slicePitch) { @@ -32,20 +32,20 @@ namespace dawn_native { namespace d3d12 { uint32_t byteOffsetY = offset % slicePitch; uint32_t byteOffsetZ = offset - byteOffsetY; - return {byteOffsetX / format.blockByteSize * format.blockWidth, - byteOffsetY / bytesPerRow * format.blockHeight, byteOffsetZ / slicePitch}; + return {byteOffsetX / blockInfo.blockByteSize * blockInfo.blockWidth, + byteOffsetY / bytesPerRow * blockInfo.blockHeight, byteOffsetZ / slicePitch}; } } // namespace - TextureCopySplit ComputeTextureCopySplit(Origin3D origin, - Extent3D copySize, - const Format& format, - uint64_t offset, - uint32_t bytesPerRow, - uint32_t rowsPerImage) { - TextureCopySplit copy; + Texture2DCopySplit ComputeTextureCopySplit(Origin3D origin, + Extent3D copySize, + const TexelBlockInfo& blockInfo, + uint64_t offset, + uint32_t bytesPerRow, + uint32_t rowsPerImage) { + Texture2DCopySplit copy; - ASSERT(bytesPerRow % format.blockByteSize == 0); + ASSERT(bytesPerRow % blockInfo.blockByteSize == 0); uint64_t alignedOffset = offset & ~static_cast<uint64_t>(D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT - 1); @@ -71,12 +71,14 @@ namespace dawn_native { namespace d3d12 { ASSERT(alignedOffset < offset); ASSERT(offset - alignedOffset < D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT); - uint32_t slicePitch = bytesPerRow * (rowsPerImage / format.blockHeight); + uint32_t slicePitch = bytesPerRow * (rowsPerImage / blockInfo.blockHeight); Origin3D texelOffset = ComputeTexelOffsets( - format, static_cast<uint32_t>(offset - alignedOffset), bytesPerRow, slicePitch); + blockInfo, static_cast<uint32_t>(offset - alignedOffset), bytesPerRow, slicePitch); - uint32_t copyBytesPerRowPitch = copySize.width / format.blockWidth * format.blockByteSize; - uint32_t byteOffsetInRowPitch = texelOffset.x / format.blockWidth * format.blockByteSize; + uint32_t copyBytesPerRowPitch = + copySize.width / blockInfo.blockWidth * blockInfo.blockByteSize; + uint32_t byteOffsetInRowPitch = + texelOffset.x / blockInfo.blockWidth * blockInfo.blockByteSize; if (copyBytesPerRowPitch + byteOffsetInRowPitch <= bytesPerRow) { // The region's rows fit inside the bytes per row. In this case, extend the width of the // PlacedFootprint and copy the buffer with an offset location @@ -154,7 +156,7 @@ namespace dawn_native { namespace d3d12 { copy.copies[0].textureOffset = origin; ASSERT(bytesPerRow > byteOffsetInRowPitch); - uint32_t texelsPerRow = bytesPerRow / format.blockByteSize * format.blockWidth; + uint32_t texelsPerRow = bytesPerRow / blockInfo.blockByteSize * blockInfo.blockWidth; copy.copies[0].copySize.width = texelsPerRow - texelOffset.x; copy.copies[0].copySize.height = copySize.height; copy.copies[0].copySize.depth = copySize.depth; @@ -174,13 +176,59 @@ namespace dawn_native { namespace d3d12 { copy.copies[1].copySize.depth = copySize.depth; copy.copies[1].bufferOffset.x = 0; - copy.copies[1].bufferOffset.y = texelOffset.y + format.blockHeight; + copy.copies[1].bufferOffset.y = texelOffset.y + blockInfo.blockHeight; copy.copies[1].bufferOffset.z = texelOffset.z; copy.copies[1].bufferSize.width = copy.copies[1].copySize.width; - copy.copies[1].bufferSize.height = rowsPerImage + texelOffset.y + format.blockHeight; + copy.copies[1].bufferSize.height = rowsPerImage + texelOffset.y + blockInfo.blockHeight; copy.copies[1].bufferSize.depth = copySize.depth + texelOffset.z; return copy; } + TextureCopySplits ComputeTextureCopySplits(Origin3D origin, + Extent3D copySize, + const TexelBlockInfo& blockInfo, + uint64_t offset, + uint32_t bytesPerRow, + uint32_t rowsPerImage) { + TextureCopySplits copies; + + const uint64_t bytesPerSlice = bytesPerRow * (rowsPerImage / blockInfo.blockHeight); + + // The function ComputeTextureCopySplit() decides how to split the copy based on: + // - the alignment of the buffer offset with D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT (512) + // - the alignment of the buffer offset with D3D12_TEXTURE_DATA_PITCH_ALIGNMENT (256) + // Each slice of a 2D array or 3D copy might need to be split, but because of the WebGPU + // constraint that "bytesPerRow" must be a multiple of 256, all odd (resp. all even) slices + // will be at an offset multiple of 512 of each other, which means they will all result in + // the same 2D split. Thus we can just compute the copy splits for the first and second + // slices, and reuse them for the remaining slices by adding the related offset of each + // slice. Moreover, if "rowsPerImage" is even, both the first and second copy layers can + // share the same copy split, so in this situation we just need to compute copy split once + // and reuse it for all the slices. + const dawn_native::Extent3D copyOneLayerSize = {copySize.width, copySize.height, 1}; + const dawn_native::Origin3D copyFirstLayerOrigin = {origin.x, origin.y, 0}; + + copies.copies2D[0] = ComputeTextureCopySplit(copyFirstLayerOrigin, copyOneLayerSize, + blockInfo, offset, bytesPerRow, rowsPerImage); + + // When the copy only refers one texture 2D array layer copies.copies2D[1] will never be + // used so we can safely early return here. + if (copySize.depth == 1) { + return copies; + } + + if (bytesPerSlice % D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT == 0) { + copies.copies2D[1] = copies.copies2D[0]; + copies.copies2D[1].offset += bytesPerSlice; + } else { + const uint64_t bufferOffsetNextLayer = offset + bytesPerSlice; + copies.copies2D[1] = + ComputeTextureCopySplit(copyFirstLayerOrigin, copyOneLayerSize, blockInfo, + bufferOffsetNextLayer, bytesPerRow, rowsPerImage); + } + + return copies; + } + }} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/TextureCopySplitter.h b/chromium/third_party/dawn/src/dawn_native/d3d12/TextureCopySplitter.h index cd7037bc0b4..962c33239df 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/TextureCopySplitter.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/TextureCopySplitter.h @@ -21,13 +21,13 @@ namespace dawn_native { - struct Format; + struct TexelBlockInfo; } // namespace dawn_native namespace dawn_native { namespace d3d12 { - struct TextureCopySplit { + struct Texture2DCopySplit { static constexpr unsigned int kMaxTextureCopyRegions = 2; struct CopyInfo { @@ -43,12 +43,25 @@ namespace dawn_native { namespace d3d12 { std::array<CopyInfo, kMaxTextureCopyRegions> copies; }; - TextureCopySplit ComputeTextureCopySplit(Origin3D origin, - Extent3D copySize, - const Format& format, - uint64_t offset, - uint32_t bytesPerRow, - uint32_t rowsPerImage); + struct TextureCopySplits { + static constexpr uint32_t kMaxTextureCopySplits = 2; + + std::array<Texture2DCopySplit, kMaxTextureCopySplits> copies2D; + }; + + Texture2DCopySplit ComputeTextureCopySplit(Origin3D origin, + Extent3D copySize, + const TexelBlockInfo& blockInfo, + uint64_t offset, + uint32_t bytesPerRow, + uint32_t rowsPerImage); + + TextureCopySplits ComputeTextureCopySplits(Origin3D origin, + Extent3D copySize, + const TexelBlockInfo& blockInfo, + uint64_t offset, + uint32_t bytesPerRow, + uint32_t rowsPerImage); }} // namespace dawn_native::d3d12 #endif // DAWNNATIVE_D3D12_TEXTURECOPYSPLITTER_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/TextureD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/TextureD3D12.cpp index 7b86367c62c..3d87186e4ac 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/TextureD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/TextureD3D12.cpp @@ -17,6 +17,7 @@ #include "common/Constants.h" #include "common/Math.h" #include "dawn_native/DynamicUploader.h" +#include "dawn_native/EnumMaskIterator.h" #include "dawn_native/Error.h" #include "dawn_native/d3d12/BufferD3D12.h" #include "dawn_native/d3d12/CommandRecordingContext.h" @@ -78,13 +79,7 @@ namespace dawn_native { namespace d3d12 { // A multisampled resource must have either D3D12_RESOURCE_FLAG_ALLOW_RENDER_TARGET or // D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL set in D3D12_RESOURCE_DESC::Flags. // https://docs.microsoft.com/en-us/windows/desktop/api/d3d12/ns-d3d12-d3d12_resource_desc - // Currently all textures are zero-initialized via the render-target path so always add - // the render target flag, except for compressed textures for which the render-target - // flag is invalid. - // TODO(natlee@microsoft.com, jiawei.shao@intel.com): do not require render target for - // lazy clearing. - if ((usage & wgpu::TextureUsage::OutputAttachment) || isMultisampledTexture || - !format.isCompressed) { + if ((usage & wgpu::TextureUsage::OutputAttachment) != 0 || isMultisampledTexture) { if (format.HasDepthOrStencil()) { flags |= D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL; } else { @@ -149,8 +144,10 @@ namespace dawn_native { namespace d3d12 { case wgpu::TextureFormat::RGB10A2Unorm: return DXGI_FORMAT_R10G10B10A2_TYPELESS; - case wgpu::TextureFormat::RG11B10Float: + case wgpu::TextureFormat::RG11B10Ufloat: return DXGI_FORMAT_R11G11B10_FLOAT; + case wgpu::TextureFormat::RGB9E5Ufloat: + return DXGI_FORMAT_R9G9B9E5_SHAREDEXP; case wgpu::TextureFormat::RG32Uint: case wgpu::TextureFormat::RG32Sint: @@ -263,8 +260,10 @@ namespace dawn_native { namespace d3d12 { return DXGI_FORMAT_B8G8R8A8_UNORM_SRGB; case wgpu::TextureFormat::RGB10A2Unorm: return DXGI_FORMAT_R10G10B10A2_UNORM; - case wgpu::TextureFormat::RG11B10Float: + case wgpu::TextureFormat::RG11B10Ufloat: return DXGI_FORMAT_R11G11B10_FLOAT; + case wgpu::TextureFormat::RGB9E5Ufloat: + return DXGI_FORMAT_R9G9B9E5_SHAREDEXP; case wgpu::TextureFormat::RG32Uint: return DXGI_FORMAT_R32G32_UINT; @@ -393,12 +392,6 @@ namespace dawn_native { namespace d3d12 { const TextureDescriptor* textureDescriptor = reinterpret_cast<const TextureDescriptor*>(descriptor->cTextureDescriptor); - // TODO(dawn:22): Remove once migration from GPUTextureDescriptor.arrayLayerCount to - // GPUTextureDescriptor.size.depth is done. - TextureDescriptor fixedDescriptor; - DAWN_TRY_ASSIGN(fixedDescriptor, FixTextureDescriptor(device, textureDescriptor)); - textureDescriptor = &fixedDescriptor; - Ref<Texture> dawnTexture = AcquireRef(new Texture(device, textureDescriptor, TextureState::OwnedExternal)); DAWN_TRY(dawnTexture->InitializeAsExternalTexture(textureDescriptor, sharedHandle, @@ -527,7 +520,7 @@ namespace dawn_native { namespace d3d12 { if (mSwapChainTexture) { ID3D12SharingContract* d3dSharingContract = device->GetSharingContract(); if (d3dSharingContract != nullptr) { - d3dSharingContract->Present(mResourceAllocation.GetD3D12Resource().Get(), 0, 0); + d3dSharingContract->Present(mResourceAllocation.GetD3D12Resource(), 0, 0); } } @@ -544,7 +537,27 @@ namespace dawn_native { namespace d3d12 { } ID3D12Resource* Texture::GetD3D12Resource() const { - return mResourceAllocation.GetD3D12Resource().Get(); + return mResourceAllocation.GetD3D12Resource(); + } + + DXGI_FORMAT Texture::GetD3D12CopyableSubresourceFormat(Aspect aspect) const { + ASSERT(GetFormat().aspects & aspect); + + switch (GetFormat().format) { + case wgpu::TextureFormat::Depth24PlusStencil8: + switch (aspect) { + case Aspect::Depth: + return DXGI_FORMAT_R32_FLOAT; + case Aspect::Stencil: + return DXGI_FORMAT_R8_UINT; + default: + UNREACHABLE(); + return GetD3D12Format(); + } + default: + ASSERT(HasOneBit(GetFormat().aspects)); + return GetD3D12Format(); + } } void Texture::TrackUsageAndTransitionNow(CommandRecordingContext* commandContext, @@ -574,7 +587,15 @@ namespace dawn_native { namespace d3d12 { } std::vector<D3D12_RESOURCE_BARRIER> barriers; - barriers.reserve(range.levelCount * range.layerCount); + + // TODO(enga): Consider adding a Count helper. + uint32_t aspectCount = 0; + for (Aspect aspect : IterateEnumMask(range.aspects)) { + aspectCount++; + DAWN_UNUSED(aspect); + } + + barriers.reserve(range.levelCount * range.layerCount * aspectCount); TransitionUsageAndGetResourceBarrier(commandContext, &barriers, newState, range); if (barriers.size()) { @@ -652,19 +673,6 @@ namespace dawn_native { namespace d3d12 { barrier.Transition.Subresource = allSubresources ? D3D12_RESOURCE_BARRIER_ALL_SUBRESOURCES : index; barriers->push_back(barrier); - // TODO(yunchao.he@intel.com): support subresource for depth/stencil. Depth stencil - // texture has different plane slices. While the current implementation only has differernt - // mip slices and array slices for subresources. - // This is a hack because Dawn doesn't handle subresource of multiplanar resources - // correctly. We force the transition to be the same for all planes to match what the - // frontend validation checks for. This hack might be incorrect for stencil-only texture - // because we always set transition barrier for depth plane. - if (!allSubresources && newState == D3D12_RESOURCE_STATE_DEPTH_WRITE && - GetFormat().HasStencil()) { - D3D12_RESOURCE_BARRIER barrierStencil = barrier; - barrierStencil.Transition.Subresource += GetArrayLayers() * GetNumMipLevels(); - barriers->push_back(barrierStencil); - } state->isValidToDecay = false; } @@ -686,7 +694,6 @@ namespace dawn_native { namespace d3d12 { HandleTransitionSpecialCases(commandContext); const Serial pendingCommandSerial = ToBackend(GetDevice())->GetPendingCommandSerial(); - uint32_t subresourceCount = GetSubresourceCount(); // This transitions assume it is a 2D texture ASSERT(GetDimension() == wgpu::TextureDimension::e2D); @@ -695,25 +702,29 @@ namespace dawn_native { namespace d3d12 { // are the same, then we can use one barrier to do state transition for all subresources. // Note that if the texture has only one mip level and one array slice, it will fall into // this category. - bool areAllSubresourcesCovered = range.levelCount * range.layerCount == subresourceCount; + bool areAllSubresourcesCovered = (range.levelCount == GetNumMipLevels() && // + range.layerCount == GetArrayLayers() && // + range.aspects == GetFormat().aspects); if (mSameLastUsagesAcrossSubresources && areAllSubresourcesCovered) { TransitionSingleOrAllSubresources(barriers, 0, newState, pendingCommandSerial, true); // TODO(yunchao.he@intel.com): compress and decompress if all subresources have the // same states. We may need to retain mSubresourceStateAndDecay[0] only. - for (uint32_t i = 1; i < subresourceCount; ++i) { + for (uint32_t i = 1; i < GetSubresourceCount(); ++i) { mSubresourceStateAndDecay[i] = mSubresourceStateAndDecay[0]; } return; } - for (uint32_t arrayLayer = 0; arrayLayer < range.layerCount; ++arrayLayer) { - for (uint32_t mipLevel = 0; mipLevel < range.levelCount; ++mipLevel) { - uint32_t index = GetSubresourceIndex(range.baseMipLevel + mipLevel, - range.baseArrayLayer + arrayLayer); - - TransitionSingleOrAllSubresources(barriers, index, newState, pendingCommandSerial, - false); + for (Aspect aspect : IterateEnumMask(range.aspects)) { + for (uint32_t arrayLayer = 0; arrayLayer < range.layerCount; ++arrayLayer) { + for (uint32_t mipLevel = 0; mipLevel < range.levelCount; ++mipLevel) { + uint32_t index = GetSubresourceIndex(range.baseMipLevel + mipLevel, + range.baseArrayLayer + arrayLayer, aspect); + + TransitionSingleOrAllSubresources(barriers, index, newState, + pendingCommandSerial, false); + } } } mSameLastUsagesAcrossSubresources = areAllSubresourcesCovered; @@ -723,6 +734,12 @@ namespace dawn_native { namespace d3d12 { CommandRecordingContext* commandContext, std::vector<D3D12_RESOURCE_BARRIER>* barriers, const PassTextureUsage& textureUsages) { + if (mResourceAllocation.GetInfo().mMethod != AllocationMethod::kExternal) { + // Track the underlying heap to ensure residency. + Heap* heap = ToBackend(mResourceAllocation.GetResourceHeap()); + commandContext->TrackHeapUsage(heap, GetDevice()->GetPendingCommandSerial()); + } + HandleTransitionSpecialCases(commandContext); const Serial pendingCommandSerial = ToBackend(GetDevice())->GetPendingCommandSerial(); @@ -748,20 +765,22 @@ namespace dawn_native { namespace d3d12 { return; } - for (uint32_t arrayLayer = 0; arrayLayer < GetArrayLayers(); ++arrayLayer) { - for (uint32_t mipLevel = 0; mipLevel < GetNumMipLevels(); ++mipLevel) { - uint32_t index = GetSubresourceIndex(mipLevel, arrayLayer); + for (Aspect aspect : IterateEnumMask(GetFormat().aspects)) { + for (uint32_t arrayLayer = 0; arrayLayer < GetArrayLayers(); ++arrayLayer) { + for (uint32_t mipLevel = 0; mipLevel < GetNumMipLevels(); ++mipLevel) { + uint32_t index = GetSubresourceIndex(mipLevel, arrayLayer, aspect); - // Skip if this subresource is not used during the current pass - if (textureUsages.subresourceUsages[index] == wgpu::TextureUsage::None) { - continue; - } + // Skip if this subresource is not used during the current pass + if (textureUsages.subresourceUsages[index] == wgpu::TextureUsage::None) { + continue; + } - D3D12_RESOURCE_STATES newState = - D3D12TextureUsage(textureUsages.subresourceUsages[index], GetFormat()); + D3D12_RESOURCE_STATES newState = + D3D12TextureUsage(textureUsages.subresourceUsages[index], GetFormat()); - TransitionSingleOrAllSubresources(barriers, index, newState, pendingCommandSerial, - false); + TransitionSingleOrAllSubresources(barriers, index, newState, + pendingCommandSerial, false); + } } } mSameLastUsagesAcrossSubresources = textureUsages.sameUsagesAcrossSubresources; @@ -834,20 +853,38 @@ namespace dawn_native { namespace d3d12 { uint8_t clearColor = (clearValue == TextureBase::ClearValue::Zero) ? 0 : 1; float fClearColor = (clearValue == TextureBase::ClearValue::Zero) ? 0.f : 1.f; - if (GetFormat().isRenderable) { + if ((GetUsage() & wgpu::TextureUsage::OutputAttachment) != 0) { if (GetFormat().HasDepthOrStencil()) { TrackUsageAndTransitionNow(commandContext, D3D12_RESOURCE_STATE_DEPTH_WRITE, range); - D3D12_CLEAR_FLAGS clearFlags = {}; - for (uint32_t level = range.baseMipLevel; level < range.baseMipLevel + range.levelCount; ++level) { for (uint32_t layer = range.baseArrayLayer; layer < range.baseArrayLayer + range.layerCount; ++layer) { - if (clearValue == TextureBase::ClearValue::Zero && - IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, layer))) { - // Skip lazy clears if already initialized. + // Iterate the aspects individually to determine which clear flags to use. + D3D12_CLEAR_FLAGS clearFlags = {}; + for (Aspect aspect : IterateEnumMask(range.aspects)) { + if (clearValue == TextureBase::ClearValue::Zero && + IsSubresourceContentInitialized( + SubresourceRange::SingleMipAndLayer(level, layer, aspect))) { + // Skip lazy clears if already initialized. + continue; + } + + switch (aspect) { + case Aspect::Depth: + clearFlags |= D3D12_CLEAR_FLAG_DEPTH; + break; + case Aspect::Stencil: + clearFlags |= D3D12_CLEAR_FLAG_STENCIL; + break; + default: + UNREACHABLE(); + break; + } + } + + if (clearFlags == 0) { continue; } @@ -860,13 +897,6 @@ namespace dawn_native { namespace d3d12 { device->GetD3D12Device()->CreateDepthStencilView(GetD3D12Resource(), &dsvDesc, baseDescriptor); - if (GetFormat().HasDepth()) { - clearFlags |= D3D12_CLEAR_FLAG_DEPTH; - } - if (GetFormat().HasStencil()) { - clearFlags |= D3D12_CLEAR_FLAG_STENCIL; - } - commandList->ClearDepthStencilView(baseDescriptor, clearFlags, fClearColor, clearColor, 0, nullptr); } @@ -878,13 +908,14 @@ namespace dawn_native { namespace d3d12 { const float clearColorRGBA[4] = {fClearColor, fClearColor, fClearColor, fClearColor}; + ASSERT(range.aspects == Aspect::Color); for (uint32_t level = range.baseMipLevel; level < range.baseMipLevel + range.levelCount; ++level) { for (uint32_t layer = range.baseArrayLayer; layer < range.baseArrayLayer + range.layerCount; ++layer) { if (clearValue == TextureBase::ClearValue::Zero && IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, layer))) { + SubresourceRange::SingleMipAndLayer(level, layer, Aspect::Color))) { // Skip lazy clears if already initialized. continue; } @@ -902,59 +933,64 @@ namespace dawn_native { namespace d3d12 { } } } else { - // TODO(natlee@microsoft.com): test compressed textures are cleared // create temp buffer with clear color to copy to the texture image - uint32_t bytesPerRow = - Align((GetWidth() / GetFormat().blockWidth) * GetFormat().blockByteSize, - kTextureBytesPerRowAlignment); - uint64_t bufferSize64 = bytesPerRow * (GetHeight() / GetFormat().blockHeight); - if (bufferSize64 > std::numeric_limits<uint32_t>::max()) { - return DAWN_OUT_OF_MEMORY_ERROR("Unable to allocate buffer."); - } - uint32_t bufferSize = static_cast<uint32_t>(bufferSize64); - DynamicUploader* uploader = device->GetDynamicUploader(); - UploadHandle uploadHandle; - DAWN_TRY_ASSIGN(uploadHandle, - uploader->Allocate(bufferSize, device->GetPendingCommandSerial())); - memset(uploadHandle.mappedBuffer, clearColor, bufferSize); - TrackUsageAndTransitionNow(commandContext, D3D12_RESOURCE_STATE_COPY_DEST, range); - for (uint32_t level = range.baseMipLevel; level < range.baseMipLevel + range.levelCount; - ++level) { - // compute d3d12 texture copy locations for texture and buffer - Extent3D copySize = GetMipLevelVirtualSize(level); - - uint32_t rowsPerImage = GetHeight(); - TextureCopySplit copySplit = - ComputeTextureCopySplit({0, 0, 0}, copySize, GetFormat(), - uploadHandle.startOffset, bytesPerRow, rowsPerImage); - - for (uint32_t layer = range.baseArrayLayer; - layer < range.baseArrayLayer + range.layerCount; ++layer) { - if (clearValue == TextureBase::ClearValue::Zero && - IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, layer))) { - // Skip lazy clears if already initialized. - continue; - } + for (Aspect aspect : IterateEnumMask(range.aspects)) { + const TexelBlockInfo& blockInfo = GetFormat().GetTexelBlockInfo(aspect); + + uint32_t bytesPerRow = + Align((GetWidth() / blockInfo.blockWidth) * blockInfo.blockByteSize, + kTextureBytesPerRowAlignment); + uint64_t bufferSize64 = bytesPerRow * (GetHeight() / blockInfo.blockHeight); + if (bufferSize64 > std::numeric_limits<uint32_t>::max()) { + return DAWN_OUT_OF_MEMORY_ERROR("Unable to allocate buffer."); + } + uint32_t bufferSize = static_cast<uint32_t>(bufferSize64); + + DynamicUploader* uploader = device->GetDynamicUploader(); + UploadHandle uploadHandle; + DAWN_TRY_ASSIGN(uploadHandle, + uploader->Allocate(bufferSize, device->GetPendingCommandSerial(), + GetFormat().blockByteSize)); + memset(uploadHandle.mappedBuffer, clearColor, bufferSize); - D3D12_TEXTURE_COPY_LOCATION textureLocation = - ComputeTextureCopyLocationForTexture(this, level, layer); - for (uint32_t i = 0; i < copySplit.count; ++i) { - TextureCopySplit::CopyInfo& info = copySplit.copies[i]; - - D3D12_TEXTURE_COPY_LOCATION bufferLocation = - ComputeBufferLocationForCopyTextureRegion( - this, ToBackend(uploadHandle.stagingBuffer)->GetResource(), - info.bufferSize, copySplit.offset, bytesPerRow); - D3D12_BOX sourceRegion = - ComputeD3D12BoxFromOffsetAndSize(info.bufferOffset, info.copySize); - - // copy the buffer filled with clear color to the texture - commandList->CopyTextureRegion(&textureLocation, info.textureOffset.x, - info.textureOffset.y, info.textureOffset.z, - &bufferLocation, &sourceRegion); + for (uint32_t level = range.baseMipLevel; + level < range.baseMipLevel + range.levelCount; ++level) { + // compute d3d12 texture copy locations for texture and buffer + Extent3D copySize = GetMipLevelVirtualSize(level); + + uint32_t rowsPerImage = GetHeight(); + Texture2DCopySplit copySplit = ComputeTextureCopySplit( + {0, 0, 0}, copySize, blockInfo, uploadHandle.startOffset, bytesPerRow, + rowsPerImage); + + for (uint32_t layer = range.baseArrayLayer; + layer < range.baseArrayLayer + range.layerCount; ++layer) { + if (clearValue == TextureBase::ClearValue::Zero && + IsSubresourceContentInitialized( + SubresourceRange::SingleMipAndLayer(level, layer, aspect))) { + // Skip lazy clears if already initialized. + continue; + } + + D3D12_TEXTURE_COPY_LOCATION textureLocation = + ComputeTextureCopyLocationForTexture(this, level, layer, aspect); + for (uint32_t i = 0; i < copySplit.count; ++i) { + Texture2DCopySplit::CopyInfo& info = copySplit.copies[i]; + + D3D12_TEXTURE_COPY_LOCATION bufferLocation = + ComputeBufferLocationForCopyTextureRegion( + this, ToBackend(uploadHandle.stagingBuffer)->GetResource(), + info.bufferSize, copySplit.offset, bytesPerRow, aspect); + D3D12_BOX sourceRegion = + ComputeD3D12BoxFromOffsetAndSize(info.bufferOffset, info.copySize); + + // copy the buffer filled with clear color to the texture + commandList->CopyTextureRegion( + &textureLocation, info.textureOffset.x, info.textureOffset.y, + info.textureOffset.z, &bufferLocation, &sourceRegion); + } } } } diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/TextureD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/TextureD3D12.h index ef17bf40628..d2df6bea69b 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/TextureD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/TextureD3D12.h @@ -48,6 +48,7 @@ namespace dawn_native { namespace d3d12 { DXGI_FORMAT GetD3D12Format() const; ID3D12Resource* GetD3D12Resource() const; + DXGI_FORMAT GetD3D12CopyableSubresourceFormat(Aspect aspect) const; D3D12_RENDER_TARGET_VIEW_DESC GetRTVDescriptor(uint32_t mipLevel, uint32_t baseArrayLayer, diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/UtilsD3D12.cpp b/chromium/third_party/dawn/src/dawn_native/d3d12/UtilsD3D12.cpp index d8c20ef1613..283727486b8 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/UtilsD3D12.cpp +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/UtilsD3D12.cpp @@ -15,6 +15,9 @@ #include "dawn_native/d3d12/UtilsD3D12.h" #include "common/Assert.h" +#include "dawn_native/Format.h" +#include "dawn_native/d3d12/BufferD3D12.h" +#include "dawn_native/d3d12/CommandRecordingContext.h" #include <stringapiset.h> @@ -64,11 +67,12 @@ namespace dawn_native { namespace d3d12 { D3D12_TEXTURE_COPY_LOCATION ComputeTextureCopyLocationForTexture(const Texture* texture, uint32_t level, - uint32_t slice) { + uint32_t slice, + Aspect aspect) { D3D12_TEXTURE_COPY_LOCATION copyLocation; copyLocation.pResource = texture->GetD3D12Resource(); copyLocation.Type = D3D12_TEXTURE_COPY_TYPE_SUBRESOURCE_INDEX; - copyLocation.SubresourceIndex = texture->GetSubresourceIndex(level, slice); + copyLocation.SubresourceIndex = texture->GetSubresourceIndex(level, slice, aspect); return copyLocation; } @@ -78,12 +82,14 @@ namespace dawn_native { namespace d3d12 { ID3D12Resource* bufferResource, const Extent3D& bufferSize, const uint64_t offset, - const uint32_t rowPitch) { + const uint32_t rowPitch, + Aspect aspect) { D3D12_TEXTURE_COPY_LOCATION bufferLocation; bufferLocation.pResource = bufferResource; bufferLocation.Type = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT; bufferLocation.PlacedFootprint.Offset = offset; - bufferLocation.PlacedFootprint.Footprint.Format = texture->GetD3D12Format(); + bufferLocation.PlacedFootprint.Footprint.Format = + texture->GetD3D12CopyableSubresourceFormat(aspect); bufferLocation.PlacedFootprint.Footprint.Width = bufferSize.width; bufferLocation.PlacedFootprint.Footprint.Height = bufferSize.height; bufferLocation.PlacedFootprint.Footprint.Depth = bufferSize.depth; @@ -102,4 +108,112 @@ namespace dawn_native { namespace d3d12 { return sourceRegion; } + bool IsTypeless(DXGI_FORMAT format) { + // List generated from <dxgiformat.h> + switch (format) { + case DXGI_FORMAT_R32G32B32A32_TYPELESS: + case DXGI_FORMAT_R32G32B32_TYPELESS: + case DXGI_FORMAT_R16G16B16A16_TYPELESS: + case DXGI_FORMAT_R32G32_TYPELESS: + case DXGI_FORMAT_R32G8X24_TYPELESS: + case DXGI_FORMAT_R32_FLOAT_X8X24_TYPELESS: + case DXGI_FORMAT_R10G10B10A2_TYPELESS: + case DXGI_FORMAT_R8G8B8A8_TYPELESS: + case DXGI_FORMAT_R16G16_TYPELESS: + case DXGI_FORMAT_R32_TYPELESS: + case DXGI_FORMAT_R24G8_TYPELESS: + case DXGI_FORMAT_R24_UNORM_X8_TYPELESS: + case DXGI_FORMAT_R8G8_TYPELESS: + case DXGI_FORMAT_R16_TYPELESS: + case DXGI_FORMAT_R8_TYPELESS: + case DXGI_FORMAT_BC1_TYPELESS: + case DXGI_FORMAT_BC2_TYPELESS: + case DXGI_FORMAT_BC3_TYPELESS: + case DXGI_FORMAT_BC4_TYPELESS: + case DXGI_FORMAT_BC5_TYPELESS: + case DXGI_FORMAT_B8G8R8A8_TYPELESS: + case DXGI_FORMAT_B8G8R8X8_TYPELESS: + case DXGI_FORMAT_BC6H_TYPELESS: + case DXGI_FORMAT_BC7_TYPELESS: + return true; + default: + return false; + } + } + + void RecordCopyBufferToTextureFromTextureCopySplit(ID3D12GraphicsCommandList* commandList, + const Texture2DCopySplit& baseCopySplit, + ID3D12Resource* bufferResource, + uint64_t baseOffsetBytes, + uint64_t bufferBytesPerRow, + Texture* texture, + uint32_t textureMiplevel, + uint32_t textureSlice, + Aspect aspect) { + ASSERT(HasOneBit(aspect)); + const D3D12_TEXTURE_COPY_LOCATION textureLocation = + ComputeTextureCopyLocationForTexture(texture, textureMiplevel, textureSlice, aspect); + + const uint64_t offsetBytes = baseCopySplit.offset + baseOffsetBytes; + + for (uint32_t i = 0; i < baseCopySplit.count; ++i) { + const Texture2DCopySplit::CopyInfo& info = baseCopySplit.copies[i]; + + // TODO(jiawei.shao@intel.com): pre-compute bufferLocation and sourceRegion as + // members in Texture2DCopySplit::CopyInfo. + const D3D12_TEXTURE_COPY_LOCATION bufferLocation = + ComputeBufferLocationForCopyTextureRegion(texture, bufferResource, info.bufferSize, + offsetBytes, bufferBytesPerRow, aspect); + const D3D12_BOX sourceRegion = + ComputeD3D12BoxFromOffsetAndSize(info.bufferOffset, info.copySize); + + commandList->CopyTextureRegion(&textureLocation, info.textureOffset.x, + info.textureOffset.y, info.textureOffset.z, + &bufferLocation, &sourceRegion); + } + } + + void CopyBufferToTextureWithCopySplit(CommandRecordingContext* commandContext, + const TextureCopy& textureCopy, + const Extent3D& copySize, + Texture* texture, + ID3D12Resource* bufferResource, + const uint64_t offsetBytes, + const uint32_t bytesPerRow, + const uint32_t rowsPerImage, + Aspect aspect) { + ASSERT(HasOneBit(aspect)); + // See comments in ComputeTextureCopySplits() for more details. + const TexelBlockInfo& blockInfo = texture->GetFormat().GetTexelBlockInfo(aspect); + const TextureCopySplits copySplits = ComputeTextureCopySplits( + textureCopy.origin, copySize, blockInfo, offsetBytes, bytesPerRow, rowsPerImage); + + const uint64_t bytesPerSlice = bytesPerRow * (rowsPerImage / blockInfo.blockHeight); + + // copySplits.copies2D[1] is always calculated for the second copy slice with + // extra "bytesPerSlice" copy offset compared with the first copy slice. So + // here we use an array bufferOffsetsForNextSlice to record the extra offsets + // for each copy slice: bufferOffsetsForNextSlice[0] is the extra offset for + // the next copy slice that uses copySplits.copies2D[0], and + // bufferOffsetsForNextSlice[1] is the extra offset for the next copy slice + // that uses copySplits.copies2D[1]. + std::array<uint64_t, TextureCopySplits::kMaxTextureCopySplits> bufferOffsetsForNextSlice = { + {0u, 0u}}; + + for (uint32_t copySlice = 0; copySlice < copySize.depth; ++copySlice) { + const uint32_t splitIndex = copySlice % copySplits.copies2D.size(); + + const Texture2DCopySplit& copySplitPerLayerBase = copySplits.copies2D[splitIndex]; + const uint64_t bufferOffsetForNextSlice = bufferOffsetsForNextSlice[splitIndex]; + const uint32_t copyTextureLayer = copySlice + textureCopy.origin.z; + + RecordCopyBufferToTextureFromTextureCopySplit( + commandContext->GetCommandList(), copySplitPerLayerBase, bufferResource, + bufferOffsetForNextSlice, bytesPerRow, texture, textureCopy.mipLevel, + copyTextureLayer, aspect); + + bufferOffsetsForNextSlice[splitIndex] += bytesPerSlice * copySplits.copies2D.size(); + } + } + }} // namespace dawn_native::d3d12 diff --git a/chromium/third_party/dawn/src/dawn_native/d3d12/UtilsD3D12.h b/chromium/third_party/dawn/src/dawn_native/d3d12/UtilsD3D12.h index d1559e72a14..6109c0f60b3 100644 --- a/chromium/third_party/dawn/src/dawn_native/d3d12/UtilsD3D12.h +++ b/chromium/third_party/dawn/src/dawn_native/d3d12/UtilsD3D12.h @@ -15,6 +15,7 @@ #ifndef DAWNNATIVE_D3D12_UTILSD3D12_H_ #define DAWNNATIVE_D3D12_UTILSD3D12_H_ +#include "dawn_native/Commands.h" #include "dawn_native/d3d12/BufferD3D12.h" #include "dawn_native/d3d12/TextureCopySplitter.h" #include "dawn_native/d3d12/TextureD3D12.h" @@ -29,16 +30,40 @@ namespace dawn_native { namespace d3d12 { D3D12_TEXTURE_COPY_LOCATION ComputeTextureCopyLocationForTexture(const Texture* texture, uint32_t level, - uint32_t slice); + uint32_t slice, + Aspect aspect); D3D12_TEXTURE_COPY_LOCATION ComputeBufferLocationForCopyTextureRegion( const Texture* texture, ID3D12Resource* bufferResource, const Extent3D& bufferSize, const uint64_t offset, - const uint32_t rowPitch); + const uint32_t rowPitch, + Aspect aspect); D3D12_BOX ComputeD3D12BoxFromOffsetAndSize(const Origin3D& offset, const Extent3D& copySize); + bool IsTypeless(DXGI_FORMAT format); + + void RecordCopyBufferToTextureFromTextureCopySplit(ID3D12GraphicsCommandList* commandList, + const Texture2DCopySplit& baseCopySplit, + Buffer* buffer, + uint64_t baseOffset, + uint64_t bufferBytesPerRow, + Texture* texture, + uint32_t textureMiplevel, + uint32_t textureSlice, + Aspect aspect); + + void CopyBufferToTextureWithCopySplit(CommandRecordingContext* commandContext, + const TextureCopy& textureCopy, + const Extent3D& copySize, + Texture* texture, + ID3D12Resource* bufferResource, + const uint64_t offset, + const uint32_t bytesPerRow, + const uint32_t rowsPerImage, + Aspect aspect); + }} // namespace dawn_native::d3d12 #endif // DAWNNATIVE_D3D12_UTILSD3D12_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/metal/BufferMTL.h b/chromium/third_party/dawn/src/dawn_native/metal/BufferMTL.h index 98bab96244c..ad6ed5faac9 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/BufferMTL.h +++ b/chromium/third_party/dawn/src/dawn_native/metal/BufferMTL.h @@ -22,28 +22,39 @@ namespace dawn_native { namespace metal { + class CommandRecordingContext; class Device; class Buffer : public BufferBase { public: - static ResultOrError<Buffer*> Create(Device* device, const BufferDescriptor* descriptor); + static ResultOrError<Ref<Buffer>> Create(Device* device, + const BufferDescriptor* descriptor); id<MTLBuffer> GetMTLBuffer() const; + void EnsureDataInitialized(CommandRecordingContext* commandContext); + void EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext, + uint64_t offset, + uint64_t size); + void EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext, + const CopyTextureToBufferCmd* copy); + private: using BufferBase::BufferBase; MaybeError Initialize(); ~Buffer() override; // Dawn API - MaybeError MapReadAsyncImpl(uint32_t serial) override; - MaybeError MapWriteAsyncImpl(uint32_t serial) override; + MaybeError MapReadAsyncImpl() override; + MaybeError MapWriteAsyncImpl() override; + MaybeError MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) override; void UnmapImpl() override; void DestroyImpl() override; void* GetMappedPointerImpl() override; - bool IsMapWritable() const override; - MaybeError MapAtCreationImpl(uint8_t** mappedPointer) override; + bool IsMappableAtCreation() const override; + MaybeError MapAtCreationImpl() override; - void ClearBuffer(BufferBase::ClearValue clearValue); + void InitializeToZero(CommandRecordingContext* commandContext); + void ClearBuffer(CommandRecordingContext* commandContext, uint8_t clearValue); id<MTLBuffer> mMtlBuffer = nil; }; diff --git a/chromium/third_party/dawn/src/dawn_native/metal/BufferMTL.mm b/chromium/third_party/dawn/src/dawn_native/metal/BufferMTL.mm index ccfd3b39ecb..2c82a7cf721 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/BufferMTL.mm +++ b/chromium/third_party/dawn/src/dawn_native/metal/BufferMTL.mm @@ -15,6 +15,8 @@ #include "dawn_native/metal/BufferMTL.h" #include "common/Math.h" +#include "dawn_native/CommandBuffer.h" +#include "dawn_native/metal/CommandRecordingContext.h" #include "dawn_native/metal/DeviceMTL.h" #include <limits> @@ -29,10 +31,10 @@ namespace dawn_native { namespace metal { static constexpr uint32_t kMaxBufferSizeFallback = 1024u * 1024u * 1024u; // static - ResultOrError<Buffer*> Buffer::Create(Device* device, const BufferDescriptor* descriptor) { + ResultOrError<Ref<Buffer>> Buffer::Create(Device* device, const BufferDescriptor* descriptor) { Ref<Buffer> buffer = AcquireRef(new Buffer(device, descriptor)); DAWN_TRY(buffer->Initialize()); - return buffer.Detach(); + return std::move(buffer); } MaybeError Buffer::Initialize() { @@ -67,6 +69,7 @@ namespace dawn_native { namespace metal { if (currentSize > maxBufferSize) { return DAWN_OUT_OF_MEMORY_ERROR("Buffer allocation is too large"); } +#if defined(DAWN_PLATFORM_MACOS) } else if (@available(macOS 10.12, *)) { // |maxBufferLength| isn't always available on older systems. If available, use // |recommendedMaxWorkingSetSize| instead. We can probably allocate more than this, @@ -76,6 +79,7 @@ namespace dawn_native { namespace metal { if (currentSize > maxWorkingSetSize) { return DAWN_OUT_OF_MEMORY_ERROR("Buffer allocation is too large"); } +#endif } else if (currentSize > kMaxBufferSizeFallback) { return DAWN_OUT_OF_MEMORY_ERROR("Buffer allocation is too large"); } @@ -87,7 +91,9 @@ namespace dawn_native { namespace metal { } if (GetDevice()->IsToggleEnabled(Toggle::NonzeroClearResourcesOnCreationForTesting)) { - ClearBuffer(BufferBase::ClearValue::NonZero); + CommandRecordingContext* commandContext = + ToBackend(GetDevice())->GetPendingCommandContext(); + ClearBuffer(commandContext, uint8_t(1u)); } return {}; @@ -101,26 +107,37 @@ namespace dawn_native { namespace metal { return mMtlBuffer; } - bool Buffer::IsMapWritable() const { + bool Buffer::IsMappableAtCreation() const { // TODO(enga): Handle CPU-visible memory on UMA return (GetUsage() & (wgpu::BufferUsage::MapRead | wgpu::BufferUsage::MapWrite)) != 0; } - MaybeError Buffer::MapAtCreationImpl(uint8_t** mappedPointer) { - *mappedPointer = reinterpret_cast<uint8_t*>([mMtlBuffer contents]); + MaybeError Buffer::MapAtCreationImpl() { + CommandRecordingContext* commandContext = + ToBackend(GetDevice())->GetPendingCommandContext(); + EnsureDataInitialized(commandContext); + + return {}; + } + + MaybeError Buffer::MapReadAsyncImpl() { return {}; } - MaybeError Buffer::MapReadAsyncImpl(uint32_t serial) { + MaybeError Buffer::MapWriteAsyncImpl() { return {}; } - MaybeError Buffer::MapWriteAsyncImpl(uint32_t serial) { + MaybeError Buffer::MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) { + CommandRecordingContext* commandContext = + ToBackend(GetDevice())->GetPendingCommandContext(); + EnsureDataInitialized(commandContext); + return {}; } void* Buffer::GetMappedPointerImpl() { - return reinterpret_cast<uint8_t*>([mMtlBuffer contents]); + return [mMtlBuffer contents]; } void Buffer::UnmapImpl() { @@ -132,16 +149,65 @@ namespace dawn_native { namespace metal { mMtlBuffer = nil; } - void Buffer::ClearBuffer(BufferBase::ClearValue clearValue) { - // TODO(jiawei.shao@intel.com): support buffer lazy-initialization to 0. - ASSERT(clearValue == BufferBase::ClearValue::NonZero); - const uint8_t clearBufferValue = 1; + void Buffer::EnsureDataInitialized(CommandRecordingContext* commandContext) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return; + } + + InitializeToZero(commandContext); + } + + void Buffer::EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext, + uint64_t offset, + uint64_t size) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return; + } + + if (IsFullBufferRange(offset, size)) { + SetIsDataInitialized(); + } else { + InitializeToZero(commandContext); + } + } + + void Buffer::EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext, + const CopyTextureToBufferCmd* copy) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return; + } + + if (IsFullBufferOverwrittenInTextureToBufferCopy(copy)) { + SetIsDataInitialized(); + } else { + InitializeToZero(commandContext); + } + } + + void Buffer::InitializeToZero(CommandRecordingContext* commandContext) { + ASSERT(GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)); + ASSERT(!IsDataInitialized()); + + ClearBuffer(commandContext, uint8_t(0u)); + + SetIsDataInitialized(); + GetDevice()->IncrementLazyClearCountForTesting(); + } - Device* device = ToBackend(GetDevice()); - CommandRecordingContext* commandContext = device->GetPendingCommandContext(); + void Buffer::ClearBuffer(CommandRecordingContext* commandContext, uint8_t clearValue) { + ASSERT(commandContext != nullptr); [commandContext->EnsureBlit() fillBuffer:mMtlBuffer range:NSMakeRange(0, GetSize()) - value:clearBufferValue]; + value:clearValue]; } }} // namespace dawn_native::metal diff --git a/chromium/third_party/dawn/src/dawn_native/metal/CommandBufferMTL.h b/chromium/third_party/dawn/src/dawn_native/metal/CommandBufferMTL.h index e8c81b7dc28..3410e57954d 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/CommandBufferMTL.h +++ b/chromium/third_party/dawn/src/dawn_native/metal/CommandBufferMTL.h @@ -15,8 +15,8 @@ #ifndef DAWNNATIVE_METAL_COMMANDBUFFERMTL_H_ #define DAWNNATIVE_METAL_COMMANDBUFFERMTL_H_ -#include "dawn_native/CommandAllocator.h" #include "dawn_native/CommandBuffer.h" +#include "dawn_native/Error.h" #import <Metal/Metal.h> @@ -33,22 +33,19 @@ namespace dawn_native { namespace metal { public: CommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor); - void FillCommands(CommandRecordingContext* commandContext); + MaybeError FillCommands(CommandRecordingContext* commandContext); private: - ~CommandBuffer() override; - void EncodeComputePass(CommandRecordingContext* commandContext); - void EncodeRenderPass(CommandRecordingContext* commandContext, - MTLRenderPassDescriptor* mtlRenderPass, - uint32_t width, - uint32_t height); - - void EncodeRenderPassInternal(CommandRecordingContext* commandContext, - MTLRenderPassDescriptor* mtlRenderPass, - uint32_t width, - uint32_t height); - - CommandIterator mCommands; + MaybeError EncodeComputePass(CommandRecordingContext* commandContext); + MaybeError EncodeRenderPass(CommandRecordingContext* commandContext, + MTLRenderPassDescriptor* mtlRenderPass, + uint32_t width, + uint32_t height); + + MaybeError EncodeRenderPassInternal(CommandRecordingContext* commandContext, + MTLRenderPassDescriptor* mtlRenderPass, + uint32_t width, + uint32_t height); }; }} // namespace dawn_native::metal diff --git a/chromium/third_party/dawn/src/dawn_native/metal/CommandBufferMTL.mm b/chromium/third_party/dawn/src/dawn_native/metal/CommandBufferMTL.mm index 64c098dce51..351090aaec4 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/CommandBufferMTL.mm +++ b/chromium/third_party/dawn/src/dawn_native/metal/CommandBufferMTL.mm @@ -26,6 +26,7 @@ #include "dawn_native/metal/RenderPipelineMTL.h" #include "dawn_native/metal/SamplerMTL.h" #include "dawn_native/metal/TextureMTL.h" +#include "dawn_native/metal/UtilsMetal.h" namespace dawn_native { namespace metal { @@ -262,7 +263,9 @@ namespace dawn_native { namespace metal { // MSL code generated by SPIRV-Cross expects. PerStage<std::array<uint32_t, kGenericMetalBufferSlots>> data; - void Apply(id<MTLRenderCommandEncoder> render, RenderPipeline* pipeline) { + void Apply(id<MTLRenderCommandEncoder> render, + RenderPipeline* pipeline, + bool enableVertexPulling) { wgpu::ShaderStage stagesToApply = dirtyStages & pipeline->GetStagesRequiringStorageBufferLength(); @@ -273,6 +276,11 @@ namespace dawn_native { namespace metal { if (stagesToApply & wgpu::ShaderStage::Vertex) { uint32_t bufferCount = ToBackend(pipeline->GetLayout()) ->GetBufferBindingCount(SingleShaderStage::Vertex); + + if (enableVertexPulling) { + bufferCount += pipeline->GetVertexStateDescriptor()->vertexBufferCount; + } + [render setVertexBytes:data[SingleShaderStage::Vertex].data() length:sizeof(uint32_t) * bufferCount atIndex:kBufferLengthBufferSlot]; @@ -309,159 +317,6 @@ namespace dawn_native { namespace metal { } }; - struct TextureBufferCopySplit { - static constexpr uint32_t kMaxTextureBufferCopyRegions = 3; - - struct CopyInfo { - NSUInteger bufferOffset; - NSUInteger bytesPerRow; - NSUInteger bytesPerImage; - Origin3D textureOrigin; - Extent3D copyExtent; - }; - - uint32_t count = 0; - std::array<CopyInfo, kMaxTextureBufferCopyRegions> copies; - }; - - MTLOrigin MakeMTLOrigin(Origin3D origin) { - return MTLOriginMake(origin.x, origin.y, origin.z); - } - - TextureBufferCopySplit ComputeTextureBufferCopySplit(wgpu::TextureDimension dimension, - Origin3D origin, - Extent3D copyExtent, - Format textureFormat, - Extent3D virtualSizeAtLevel, - uint64_t bufferSize, - uint64_t bufferOffset, - uint32_t bytesPerRow, - uint32_t rowsPerImage) { - TextureBufferCopySplit copy; - - // When copying textures from/to an unpacked buffer, the Metal validation layer doesn't - // compute the correct range when checking if the buffer is big enough to contain the - // data for the whole copy. Instead of looking at the position of the last texel in the - // buffer, it computes the volume of the 3D box with bytesPerRow * (rowsPerImage / - // format.blockHeight) * copySize.depth. For example considering the pixel buffer below - // where in memory, each row data (D) of the texture is followed by some padding data - // (P): - // |DDDDDDD|PP| - // |DDDDDDD|PP| - // |DDDDDDD|PP| - // |DDDDDDD|PP| - // |DDDDDDA|PP| - // The last pixel read will be A, but the driver will think it is the whole last padding - // row, causing it to generate an error when the pixel buffer is just big enough. - - // We work around this limitation by detecting when Metal would complain and copy the - // last image and row separately using tight sourceBytesPerRow or sourceBytesPerImage. - uint32_t dataRowsPerImage = rowsPerImage / textureFormat.blockHeight; - uint32_t bytesPerImage = bytesPerRow * dataRowsPerImage; - - // Metal validation layer requires that if the texture's pixel format is a compressed - // format, the sourceSize must be a multiple of the pixel format's block size or be - // clamped to the edge of the texture if the block extends outside the bounds of a - // texture. - uint32_t clampedCopyExtentWidth = - (origin.x + copyExtent.width > virtualSizeAtLevel.width) - ? (virtualSizeAtLevel.width - origin.x) - : copyExtent.width; - uint32_t clampedCopyExtentHeight = - (origin.y + copyExtent.height > virtualSizeAtLevel.height) - ? (virtualSizeAtLevel.height - origin.y) - : copyExtent.height; - - ASSERT(dimension == wgpu::TextureDimension::e2D); - - // Check whether buffer size is big enough. - bool needWorkaround = bufferSize - bufferOffset < bytesPerImage * copyExtent.depth; - if (!needWorkaround) { - copy.count = 1; - copy.copies[0].bufferOffset = bufferOffset; - copy.copies[0].bytesPerRow = bytesPerRow; - copy.copies[0].bytesPerImage = bytesPerImage; - copy.copies[0].textureOrigin = origin; - copy.copies[0].copyExtent = {clampedCopyExtentWidth, clampedCopyExtentHeight, - copyExtent.depth}; - return copy; - } - - uint64_t currentOffset = bufferOffset; - - // Doing all the copy except the last image. - if (copyExtent.depth > 1) { - copy.copies[copy.count].bufferOffset = currentOffset; - copy.copies[copy.count].bytesPerRow = bytesPerRow; - copy.copies[copy.count].bytesPerImage = bytesPerImage; - copy.copies[copy.count].textureOrigin = origin; - copy.copies[copy.count].copyExtent = { - clampedCopyExtentWidth, clampedCopyExtentHeight, copyExtent.depth - 1}; - - ++copy.count; - - // Update offset to copy to the last image. - currentOffset += (copyExtent.depth - 1) * bytesPerImage; - } - - // Doing all the copy in last image except the last row. - uint32_t copyBlockRowCount = copyExtent.height / textureFormat.blockHeight; - if (copyBlockRowCount > 1) { - copy.copies[copy.count].bufferOffset = currentOffset; - copy.copies[copy.count].bytesPerRow = bytesPerRow; - copy.copies[copy.count].bytesPerImage = bytesPerRow * (copyBlockRowCount - 1); - copy.copies[copy.count].textureOrigin = {origin.x, origin.y, - origin.z + copyExtent.depth - 1}; - - ASSERT(copyExtent.height - textureFormat.blockHeight < virtualSizeAtLevel.height); - copy.copies[copy.count].copyExtent = { - clampedCopyExtentWidth, copyExtent.height - textureFormat.blockHeight, 1}; - - ++copy.count; - - // Update offset to copy to the last row. - currentOffset += (copyBlockRowCount - 1) * bytesPerRow; - } - - // Doing the last row copy with the exact number of bytes in last row. - // Workaround this issue in a way just like the copy to a 1D texture. - uint32_t lastRowDataSize = - (copyExtent.width / textureFormat.blockWidth) * textureFormat.blockByteSize; - uint32_t lastRowCopyExtentHeight = - textureFormat.blockHeight + clampedCopyExtentHeight - copyExtent.height; - ASSERT(lastRowCopyExtentHeight <= textureFormat.blockHeight); - - copy.copies[copy.count].bufferOffset = currentOffset; - copy.copies[copy.count].bytesPerRow = lastRowDataSize; - copy.copies[copy.count].bytesPerImage = lastRowDataSize; - copy.copies[copy.count].textureOrigin = { - origin.x, origin.y + copyExtent.height - textureFormat.blockHeight, - origin.z + copyExtent.depth - 1}; - copy.copies[copy.count].copyExtent = {clampedCopyExtentWidth, lastRowCopyExtentHeight, - 1}; - ++copy.count; - - return copy; - } - - void EnsureSourceTextureInitialized(Texture* texture, - const Extent3D& size, - const TextureCopy& src) { - texture->EnsureSubresourceContentInitialized( - {src.mipLevel, 1, src.arrayLayer, size.depth}); - } - - void EnsureDestinationTextureInitialized(Texture* texture, - const Extent3D& size, - const TextureCopy& dst) { - SubresourceRange range = {dst.mipLevel, 1, dst.arrayLayer, size.depth}; - if (IsCompleteSubresourceCopiedTo(texture, size, dst.mipLevel)) { - texture->SetIsSubresourceContentInitialized(true, range); - } else { - texture->EnsureSubresourceContentInitialized(range); - } - } - // Keeps track of the dirty bind groups so they can be lazily applied when we know the // pipeline state. // Bind groups may be inherited because bind groups are packed in the buffer / @@ -635,10 +490,17 @@ namespace dawn_native { namespace metal { // all the relevant state. class VertexBufferTracker { public: + explicit VertexBufferTracker(StorageBufferLengthTracker* lengthTracker) + : mLengthTracker(lengthTracker) { + } + void OnSetVertexBuffer(uint32_t slot, Buffer* buffer, uint64_t offset) { mVertexBuffers[slot] = buffer->GetMTLBuffer(); mVertexBufferOffsets[slot] = offset; + ASSERT(buffer->GetSize() < std::numeric_limits<uint32_t>::max()); + mVertexBufferBindingSizes[slot] = static_cast<uint32_t>(buffer->GetSize() - offset); + // Use 64 bit masks and make sure there are no shift UB static_assert(kMaxVertexBuffers <= 8 * sizeof(unsigned long long) - 1, ""); mDirtyVertexBuffers |= 1ull << slot; @@ -651,13 +513,22 @@ namespace dawn_native { namespace metal { mDirtyVertexBuffers |= pipeline->GetVertexBufferSlotsUsed(); } - void Apply(id<MTLRenderCommandEncoder> encoder, RenderPipeline* pipeline) { + void Apply(id<MTLRenderCommandEncoder> encoder, + RenderPipeline* pipeline, + bool enableVertexPulling) { std::bitset<kMaxVertexBuffers> vertexBuffersToApply = mDirtyVertexBuffers & pipeline->GetVertexBufferSlotsUsed(); for (uint32_t dawnIndex : IterateBitSet(vertexBuffersToApply)) { uint32_t metalIndex = pipeline->GetMtlVertexBufferIndex(dawnIndex); + if (enableVertexPulling) { + // Insert lengths for vertex buffers bound as storage buffers + mLengthTracker->data[SingleShaderStage::Vertex][metalIndex] = + mVertexBufferBindingSizes[dawnIndex]; + mLengthTracker->dirtyStages |= wgpu::ShaderStage::Vertex; + } + [encoder setVertexBuffers:&mVertexBuffers[dawnIndex] offsets:&mVertexBufferOffsets[dawnIndex] withRange:NSMakeRange(metalIndex, 1)]; @@ -671,23 +542,23 @@ namespace dawn_native { namespace metal { std::bitset<kMaxVertexBuffers> mDirtyVertexBuffers; std::array<id<MTLBuffer>, kMaxVertexBuffers> mVertexBuffers; std::array<NSUInteger, kMaxVertexBuffers> mVertexBufferOffsets; + std::array<uint32_t, kMaxVertexBuffers> mVertexBufferBindingSizes; + + StorageBufferLengthTracker* mLengthTracker; }; } // anonymous namespace CommandBuffer::CommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor) - : CommandBufferBase(encoder, descriptor), mCommands(encoder->AcquireCommands()) { - } - - CommandBuffer::~CommandBuffer() { - FreeCommands(&mCommands); + : CommandBufferBase(encoder, descriptor) { } - void CommandBuffer::FillCommands(CommandRecordingContext* commandContext) { + MaybeError CommandBuffer::FillCommands(CommandRecordingContext* commandContext) { const std::vector<PassResourceUsage>& passResourceUsages = GetResourceUsages().perPass; size_t nextPassNumber = 0; - auto LazyClearForPass = [](const PassResourceUsage& usages) { + auto LazyClearForPass = [](const PassResourceUsage& usages, + CommandRecordingContext* commandContext) { for (size_t i = 0; i < usages.textures.size(); ++i) { Texture* texture = ToBackend(usages.textures[i]); // Clear textures that are not output attachments. Output attachments will be @@ -697,6 +568,9 @@ namespace dawn_native { namespace metal { texture->EnsureSubresourceContentInitialized(texture->GetAllSubresources()); } } + for (BufferBase* bufferBase : usages.buffers) { + ToBackend(bufferBase)->EnsureDataInitialized(commandContext); + } }; Command type; @@ -705,10 +579,10 @@ namespace dawn_native { namespace metal { case Command::BeginComputePass: { mCommands.NextCommand<BeginComputePassCmd>(); - LazyClearForPass(passResourceUsages[nextPassNumber]); + LazyClearForPass(passResourceUsages[nextPassNumber], commandContext); commandContext->EndBlit(); - EncodeComputePass(commandContext); + DAWN_TRY(EncodeComputePass(commandContext)); nextPassNumber++; break; @@ -717,12 +591,12 @@ namespace dawn_native { namespace metal { case Command::BeginRenderPass: { BeginRenderPassCmd* cmd = mCommands.NextCommand<BeginRenderPassCmd>(); - LazyClearForPass(passResourceUsages[nextPassNumber]); + LazyClearForPass(passResourceUsages[nextPassNumber], commandContext); commandContext->EndBlit(); LazyClearRenderPassAttachments(cmd); MTLRenderPassDescriptor* descriptor = CreateMTLRenderPassDescriptor(cmd); - EncodeRenderPass(commandContext, descriptor, cmd->width, cmd->height); + DAWN_TRY(EncodeRenderPass(commandContext, descriptor, cmd->width, cmd->height)); nextPassNumber++; break; @@ -731,6 +605,11 @@ namespace dawn_native { namespace metal { case Command::CopyBufferToBuffer: { CopyBufferToBufferCmd* copy = mCommands.NextCommand<CopyBufferToBufferCmd>(); + ToBackend(copy->source)->EnsureDataInitialized(commandContext); + ToBackend(copy->destination) + ->EnsureDataInitializedAsDestination(commandContext, + copy->destinationOffset, copy->size); + [commandContext->EnsureBlit() copyFromBuffer:ToBackend(copy->source)->GetMTLBuffer() sourceOffset:copy->sourceOffset @@ -748,17 +627,12 @@ namespace dawn_native { namespace metal { Buffer* buffer = ToBackend(src.buffer.Get()); Texture* texture = ToBackend(dst.texture.Get()); - EnsureDestinationTextureInitialized(texture, copy->copySize, copy->destination); - - const Extent3D virtualSizeAtLevel = - texture->GetMipLevelVirtualSize(dst.mipLevel); + buffer->EnsureDataInitialized(commandContext); + EnsureDestinationTextureInitialized(texture, copy->destination, copy->copySize); - Origin3D copyOrigin = dst.origin; - copyOrigin.z = dst.arrayLayer; TextureBufferCopySplit splitCopies = ComputeTextureBufferCopySplit( - texture->GetDimension(), copyOrigin, copySize, texture->GetFormat(), - virtualSizeAtLevel, buffer->GetSize(), src.offset, src.bytesPerRow, - src.rowsPerImage); + texture, dst.mipLevel, dst.origin, copySize, buffer->GetSize(), src.offset, + src.bytesPerRow, src.rowsPerImage); for (uint32_t i = 0; i < splitCopies.count; ++i) { const TextureBufferCopySplit::CopyInfo& copyInfo = splitCopies.copies[i]; @@ -770,6 +644,9 @@ namespace dawn_native { namespace metal { const MTLSize copyExtent = MTLSizeMake(copyInfo.copyExtent.width, copyInfo.copyExtent.height, 1); + MTLBlitOption blitOption = + ComputeMTLBlitOption(texture->GetFormat(), dst.aspect); + uint64_t bufferOffset = copyInfo.bufferOffset; for (uint32_t copyLayer = copyBaseLayer; copyLayer < copyBaseLayer + copyLayerCount; ++copyLayer) { @@ -781,7 +658,8 @@ namespace dawn_native { namespace metal { toTexture:texture->GetMTLTexture() destinationSlice:copyLayer destinationLevel:dst.mipLevel - destinationOrigin:textureOrigin]; + destinationOrigin:textureOrigin + options:blitOption]; bufferOffset += copyInfo.bytesPerImage; } } @@ -797,15 +675,14 @@ namespace dawn_native { namespace metal { Texture* texture = ToBackend(src.texture.Get()); Buffer* buffer = ToBackend(dst.buffer.Get()); - EnsureSourceTextureInitialized(texture, copy->copySize, copy->source); + buffer->EnsureDataInitializedAsDestination(commandContext, copy); + + texture->EnsureSubresourceContentInitialized( + GetSubresourcesAffectedByCopy(src, copySize)); - Extent3D virtualSizeAtLevel = texture->GetMipLevelVirtualSize(src.mipLevel); - Origin3D copyOrigin = src.origin; - copyOrigin.z = src.arrayLayer; TextureBufferCopySplit splitCopies = ComputeTextureBufferCopySplit( - texture->GetDimension(), copyOrigin, copySize, texture->GetFormat(), - virtualSizeAtLevel, buffer->GetSize(), dst.offset, dst.bytesPerRow, - dst.rowsPerImage); + texture, src.mipLevel, src.origin, copySize, buffer->GetSize(), dst.offset, + dst.bytesPerRow, dst.rowsPerImage); for (uint32_t i = 0; i < splitCopies.count; ++i) { const TextureBufferCopySplit::CopyInfo& copyInfo = splitCopies.copies[i]; @@ -817,6 +694,9 @@ namespace dawn_native { namespace metal { const MTLSize copyExtent = MTLSizeMake(copyInfo.copyExtent.width, copyInfo.copyExtent.height, 1); + MTLBlitOption blitOption = + ComputeMTLBlitOption(texture->GetFormat(), src.aspect); + uint64_t bufferOffset = copyInfo.bufferOffset; for (uint32_t copyLayer = copyBaseLayer; copyLayer < copyBaseLayer + copyLayerCount; ++copyLayer) { @@ -828,7 +708,8 @@ namespace dawn_native { namespace metal { toBuffer:buffer->GetMTLBuffer() destinationOffset:bufferOffset destinationBytesPerRow:copyInfo.bytesPerRow - destinationBytesPerImage:copyInfo.bytesPerImage]; + destinationBytesPerImage:copyInfo.bytesPerImage + options:blitOption]; bufferOffset += copyInfo.bytesPerImage; } } @@ -842,31 +723,44 @@ namespace dawn_native { namespace metal { Texture* srcTexture = ToBackend(copy->source.texture.Get()); Texture* dstTexture = ToBackend(copy->destination.texture.Get()); - EnsureSourceTextureInitialized(srcTexture, copy->copySize, copy->source); - EnsureDestinationTextureInitialized(dstTexture, copy->copySize, - copy->destination); + srcTexture->EnsureSubresourceContentInitialized( + GetSubresourcesAffectedByCopy(copy->source, copy->copySize)); + EnsureDestinationTextureInitialized(dstTexture, copy->destination, + copy->copySize); // TODO(jiawei.shao@intel.com): support copies with 1D and 3D textures. ASSERT(srcTexture->GetDimension() == wgpu::TextureDimension::e2D && dstTexture->GetDimension() == wgpu::TextureDimension::e2D); - const MTLSize mtlSizeOneLayer = + const MTLSize sizeOneLayer = MTLSizeMake(copy->copySize.width, copy->copySize.height, 1); + const MTLOrigin sourceOriginNoLayer = + MTLOriginMake(copy->source.origin.x, copy->source.origin.y, 0); + const MTLOrigin destinationOriginNoLayer = + MTLOriginMake(copy->destination.origin.x, copy->destination.origin.y, 0); + for (uint32_t slice = 0; slice < copy->copySize.depth; ++slice) { [commandContext->EnsureBlit() copyFromTexture:srcTexture->GetMTLTexture() - sourceSlice:copy->source.arrayLayer + slice + sourceSlice:copy->source.origin.z + slice sourceLevel:copy->source.mipLevel - sourceOrigin:MakeMTLOrigin(copy->source.origin) - sourceSize:mtlSizeOneLayer + sourceOrigin:sourceOriginNoLayer + sourceSize:sizeOneLayer toTexture:dstTexture->GetMTLTexture() - destinationSlice:copy->destination.arrayLayer + slice + destinationSlice:copy->destination.origin.z + slice destinationLevel:copy->destination.mipLevel - destinationOrigin:MakeMTLOrigin(copy->destination.origin)]; + destinationOrigin:destinationOriginNoLayer]; } - break; } + case Command::ResolveQuerySet: { + return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation."); + } + + case Command::WriteTimestamp: { + return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation."); + } + default: { UNREACHABLE(); break; @@ -875,9 +769,10 @@ namespace dawn_native { namespace metal { } commandContext->EndBlit(); + return {}; } - void CommandBuffer::EncodeComputePass(CommandRecordingContext* commandContext) { + MaybeError CommandBuffer::EncodeComputePass(CommandRecordingContext* commandContext) { ComputePipeline* lastPipeline = nullptr; StorageBufferLengthTracker storageBufferLengths = {}; BindGroupTracker bindGroups(&storageBufferLengths); @@ -890,7 +785,7 @@ namespace dawn_native { namespace metal { case Command::EndComputePass: { mCommands.NextCommand<EndComputePassCmd>(); commandContext->EndCompute(); - return; + return {}; } case Command::Dispatch: { @@ -968,6 +863,10 @@ namespace dawn_native { namespace metal { break; } + case Command::WriteTimestamp: { + return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation."); + } + default: { UNREACHABLE(); break; @@ -979,10 +878,10 @@ namespace dawn_native { namespace metal { UNREACHABLE(); } - void CommandBuffer::EncodeRenderPass(CommandRecordingContext* commandContext, - MTLRenderPassDescriptor* mtlRenderPass, - uint32_t width, - uint32_t height) { + MaybeError CommandBuffer::EncodeRenderPass(CommandRecordingContext* commandContext, + MTLRenderPassDescriptor* mtlRenderPass, + uint32_t width, + uint32_t height) { ASSERT(mtlRenderPass); Device* device = ToBackend(GetDevice()); @@ -1026,7 +925,7 @@ namespace dawn_native { namespace metal { // If we need to use a temporary resolve texture we need to copy the result of MSAA // resolve back to the true resolve targets. if (useTemporaryResolveTexture) { - EncodeRenderPass(commandContext, mtlRenderPass, width, height); + DAWN_TRY(EncodeRenderPass(commandContext, mtlRenderPass, width, height)); for (uint32_t i = 0; i < kMaxColorAttachments; ++i) { if (trueResolveTextures[i] == nil) { continue; @@ -1039,7 +938,7 @@ namespace dawn_native { namespace metal { [temporaryResolveTextures[i] release]; temporaryResolveTextures[i] = nil; } - return; + return {}; } } @@ -1062,24 +961,26 @@ namespace dawn_native { namespace metal { // If we found a store + MSAA resolve we need to resolve in a different render pass. if (hasStoreAndMSAAResolve) { - EncodeRenderPass(commandContext, mtlRenderPass, width, height); + DAWN_TRY(EncodeRenderPass(commandContext, mtlRenderPass, width, height)); ResolveInAnotherRenderPass(commandContext, mtlRenderPass, resolveTextures); - return; + return {}; } } - EncodeRenderPassInternal(commandContext, mtlRenderPass, width, height); + DAWN_TRY(EncodeRenderPassInternal(commandContext, mtlRenderPass, width, height)); + return {}; } - void CommandBuffer::EncodeRenderPassInternal(CommandRecordingContext* commandContext, - MTLRenderPassDescriptor* mtlRenderPass, - uint32_t width, - uint32_t height) { + MaybeError CommandBuffer::EncodeRenderPassInternal(CommandRecordingContext* commandContext, + MTLRenderPassDescriptor* mtlRenderPass, + uint32_t width, + uint32_t height) { + bool enableVertexPulling = GetDevice()->IsToggleEnabled(Toggle::MetalEnableVertexPulling); RenderPipeline* lastPipeline = nullptr; id<MTLBuffer> indexBuffer = nil; uint32_t indexBufferBaseOffset = 0; - VertexBufferTracker vertexBuffers; StorageBufferLengthTracker storageBufferLengths = {}; + VertexBufferTracker vertexBuffers(&storageBufferLengths); BindGroupTracker bindGroups(&storageBufferLengths); id<MTLRenderCommandEncoder> encoder = commandContext->BeginRender(mtlRenderPass); @@ -1089,9 +990,9 @@ namespace dawn_native { namespace metal { case Command::Draw: { DrawCmd* draw = iter->NextCommand<DrawCmd>(); - vertexBuffers.Apply(encoder, lastPipeline); + vertexBuffers.Apply(encoder, lastPipeline, enableVertexPulling); bindGroups.Apply(encoder); - storageBufferLengths.Apply(encoder, lastPipeline); + storageBufferLengths.Apply(encoder, lastPipeline, enableVertexPulling); // The instance count must be non-zero, otherwise no-op if (draw->instanceCount != 0) { @@ -1117,9 +1018,9 @@ namespace dawn_native { namespace metal { size_t formatSize = IndexFormatSize(lastPipeline->GetVertexStateDescriptor()->indexFormat); - vertexBuffers.Apply(encoder, lastPipeline); + vertexBuffers.Apply(encoder, lastPipeline, enableVertexPulling); bindGroups.Apply(encoder); - storageBufferLengths.Apply(encoder, lastPipeline); + storageBufferLengths.Apply(encoder, lastPipeline, enableVertexPulling); // The index and instance count must be non-zero, otherwise no-op if (draw->indexCount != 0 && draw->instanceCount != 0) { @@ -1151,9 +1052,9 @@ namespace dawn_native { namespace metal { case Command::DrawIndirect: { DrawIndirectCmd* draw = iter->NextCommand<DrawIndirectCmd>(); - vertexBuffers.Apply(encoder, lastPipeline); + vertexBuffers.Apply(encoder, lastPipeline, enableVertexPulling); bindGroups.Apply(encoder); - storageBufferLengths.Apply(encoder, lastPipeline); + storageBufferLengths.Apply(encoder, lastPipeline, enableVertexPulling); Buffer* buffer = ToBackend(draw->indirectBuffer.Get()); id<MTLBuffer> indirectBuffer = buffer->GetMTLBuffer(); @@ -1166,9 +1067,9 @@ namespace dawn_native { namespace metal { case Command::DrawIndexedIndirect: { DrawIndirectCmd* draw = iter->NextCommand<DrawIndirectCmd>(); - vertexBuffers.Apply(encoder, lastPipeline); + vertexBuffers.Apply(encoder, lastPipeline, enableVertexPulling); bindGroups.Apply(encoder); - storageBufferLengths.Apply(encoder, lastPipeline); + storageBufferLengths.Apply(encoder, lastPipeline, enableVertexPulling); Buffer* buffer = ToBackend(draw->indirectBuffer.Get()); id<MTLBuffer> indirectBuffer = buffer->GetMTLBuffer(); @@ -1264,7 +1165,7 @@ namespace dawn_native { namespace metal { case Command::EndRenderPass: { mCommands.NextCommand<EndRenderPassCmd>(); commandContext->EndRender(); - return; + return {}; } case Command::SetStencilReference: { @@ -1331,6 +1232,10 @@ namespace dawn_native { namespace metal { break; } + case Command::WriteTimestamp: { + return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation."); + } + default: { EncodeRenderBundleCommand(&mCommands, type); break; diff --git a/chromium/third_party/dawn/src/dawn_native/metal/DeviceMTL.h b/chromium/third_party/dawn/src/dawn_native/metal/DeviceMTL.h index 76e41518918..8c4bc493278 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/DeviceMTL.h +++ b/chromium/third_party/dawn/src/dawn_native/metal/DeviceMTL.h @@ -18,6 +18,7 @@ #include "dawn_native/dawn_platform.h" #include "common/Serial.h" +#include "dawn_native/Commands.h" #include "dawn_native/Device.h" #include "dawn_native/metal/CommandRecordingContext.h" #include "dawn_native/metal/Forward.h" @@ -63,6 +64,10 @@ namespace dawn_native { namespace metal { BufferBase* destination, uint64_t destinationOffset, uint64_t size) override; + MaybeError CopyFromStagingToTexture(StagingBufferBase* source, + const TextureDataLayout& dataLayout, + TextureCopy* dst, + const Extent3D& copySizePixels); private: Device(AdapterBase* adapter, id<MTLDevice> mtlDevice, const DeviceDescriptor* descriptor); @@ -71,7 +76,8 @@ namespace dawn_native { namespace metal { const BindGroupDescriptor* descriptor) override; ResultOrError<BindGroupLayoutBase*> CreateBindGroupLayoutImpl( const BindGroupLayoutDescriptor* descriptor) override; - ResultOrError<BufferBase*> CreateBufferImpl(const BufferDescriptor* descriptor) override; + ResultOrError<Ref<BufferBase>> CreateBufferImpl( + const BufferDescriptor* descriptor) override; ResultOrError<ComputePipelineBase*> CreateComputePipelineImpl( const ComputePipelineDescriptor* descriptor) override; ResultOrError<PipelineLayoutBase*> CreatePipelineLayoutImpl( diff --git a/chromium/third_party/dawn/src/dawn_native/metal/DeviceMTL.mm b/chromium/third_party/dawn/src/dawn_native/metal/DeviceMTL.mm index d8d0feb6a16..4004d788b3e 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/DeviceMTL.mm +++ b/chromium/third_party/dawn/src/dawn_native/metal/DeviceMTL.mm @@ -14,8 +14,10 @@ #include "dawn_native/metal/DeviceMTL.h" +#include "common/Platform.h" #include "dawn_native/BackendConnection.h" #include "dawn_native/BindGroupLayout.h" +#include "dawn_native/Commands.h" #include "dawn_native/ErrorData.h" #include "dawn_native/metal/BindGroupLayoutMTL.h" #include "dawn_native/metal/BindGroupMTL.h" @@ -30,6 +32,7 @@ #include "dawn_native/metal/StagingBufferMTL.h" #include "dawn_native/metal/SwapChainMTL.h" #include "dawn_native/metal/TextureMTL.h" +#include "dawn_native/metal/UtilsMetal.h" #include "dawn_platform/DawnPlatform.h" #include "dawn_platform/tracing/TraceEvent.h" @@ -49,9 +52,7 @@ namespace dawn_native { namespace metal { Device::Device(AdapterBase* adapter, id<MTLDevice> mtlDevice, const DeviceDescriptor* descriptor) - : DeviceBase(adapter, descriptor), - mMtlDevice([mtlDevice retain]), - mCompletedSerial(0) { + : DeviceBase(adapter, descriptor), mMtlDevice([mtlDevice retain]), mCompletedSerial(0) { [mMtlDevice retain]; } @@ -61,6 +62,11 @@ namespace dawn_native { namespace metal { MaybeError Device::Initialize() { InitTogglesFromDriver(); + + if (!IsRobustnessEnabled() || !IsToggleEnabled(Toggle::UseSpvc)) { + ForceSetToggle(Toggle::MetalEnableVertexPulling, false); + } + mCommandQueue = [mMtlDevice newCommandQueue]; return DeviceBase::Initialize(new Queue(this)); @@ -70,8 +76,10 @@ namespace dawn_native { namespace metal { { bool haveStoreAndMSAAResolve = false; #if defined(DAWN_PLATFORM_MACOS) - haveStoreAndMSAAResolve = - [mMtlDevice supportsFeatureSet:MTLFeatureSet_macOS_GPUFamily1_v2]; + if (@available(macOS 10.12, *)) { + haveStoreAndMSAAResolve = + [mMtlDevice supportsFeatureSet:MTLFeatureSet_macOS_GPUFamily1_v2]; + } #elif defined(DAWN_PLATFORM_IOS) haveStoreAndMSAAResolve = [mMtlDevice supportsFeatureSet:MTLFeatureSet_iOS_GPUFamily3_v2]; @@ -108,7 +116,7 @@ namespace dawn_native { namespace metal { const BindGroupLayoutDescriptor* descriptor) { return new BindGroupLayout(this, descriptor); } - ResultOrError<BufferBase*> Device::CreateBufferImpl(const BufferDescriptor* descriptor) { + ResultOrError<Ref<BufferBase>> Device::CreateBufferImpl(const BufferDescriptor* descriptor) { return Buffer::Create(this, descriptor); } CommandBufferBase* Device::CreateCommandBuffer(CommandEncoder* encoder, @@ -254,6 +262,10 @@ namespace dawn_native { namespace metal { // this function. ASSERT(size != 0); + ToBackend(destination) + ->EnsureDataInitializedAsDestination(GetPendingCommandContext(), destinationOffset, + size); + id<MTLBuffer> uploadBuffer = ToBackend(source)->GetBufferHandle(); id<MTLBuffer> buffer = ToBackend(destination)->GetMTLBuffer(); [GetPendingCommandContext()->EnsureBlit() copyFromBuffer:uploadBuffer @@ -264,20 +276,63 @@ namespace dawn_native { namespace metal { return {}; } + MaybeError Device::CopyFromStagingToTexture(StagingBufferBase* source, + const TextureDataLayout& dataLayout, + TextureCopy* dst, + const Extent3D& copySizePixels) { + Texture* texture = ToBackend(dst->texture.Get()); + + // This function assumes data is perfectly aligned. Otherwise, it might be necessary + // to split copying to several stages: see ComputeTextureBufferCopySplit. + const TexelBlockInfo& blockInfo = texture->GetFormat().GetTexelBlockInfo(dst->aspect); + uint32_t blockSize = blockInfo.blockByteSize; + uint32_t blockWidth = blockInfo.blockWidth; + uint32_t blockHeight = blockInfo.blockHeight; + ASSERT(dataLayout.rowsPerImage == (copySizePixels.height)); + ASSERT(dataLayout.bytesPerRow == (copySizePixels.width) / blockWidth * blockSize); + + EnsureDestinationTextureInitialized(texture, *dst, copySizePixels); + + // Metal validation layer requires that if the texture's pixel format is a compressed + // format, the sourceSize must be a multiple of the pixel format's block size or be + // clamped to the edge of the texture if the block extends outside the bounds of a + // texture. + const Extent3D clampedSize = + texture->ClampToMipLevelVirtualSize(dst->mipLevel, dst->origin, copySizePixels); + const uint32_t copyBaseLayer = dst->origin.z; + const uint32_t copyLayerCount = copySizePixels.depth; + const uint64_t bytesPerImage = + dataLayout.rowsPerImage * dataLayout.bytesPerRow / blockHeight; + + MTLBlitOption blitOption = ComputeMTLBlitOption(texture->GetFormat(), dst->aspect); + + uint64_t bufferOffset = dataLayout.offset; + for (uint32_t copyLayer = copyBaseLayer; copyLayer < copyBaseLayer + copyLayerCount; + ++copyLayer) { + [GetPendingCommandContext()->EnsureBlit() + copyFromBuffer:ToBackend(source)->GetBufferHandle() + sourceOffset:bufferOffset + sourceBytesPerRow:dataLayout.bytesPerRow + sourceBytesPerImage:bytesPerImage + sourceSize:MTLSizeMake(clampedSize.width, clampedSize.height, 1) + toTexture:texture->GetMTLTexture() + destinationSlice:copyLayer + destinationLevel:dst->mipLevel + destinationOrigin:MTLOriginMake(dst->origin.x, dst->origin.y, 0) + options:blitOption]; + + bufferOffset += bytesPerImage; + } + + return {}; + } + TextureBase* Device::CreateTextureWrappingIOSurface(const ExternalImageDescriptor* descriptor, IOSurfaceRef ioSurface, uint32_t plane) { const TextureDescriptor* textureDescriptor = reinterpret_cast<const TextureDescriptor*>(descriptor->cTextureDescriptor); - // TODO(dawn:22): Remove once migration from GPUTextureDescriptor.arrayLayerCount to - // GPUTextureDescriptor.size.depth is done. - TextureDescriptor fixedDescriptor; - if (ConsumedError(FixTextureDescriptor(this, textureDescriptor), &fixedDescriptor)) { - return nullptr; - } - textureDescriptor = &fixedDescriptor; - if (ConsumedError(ValidateTextureDescriptor(this, textureDescriptor))) { return nullptr; } diff --git a/chromium/third_party/dawn/src/dawn_native/metal/PipelineLayoutMTL.h b/chromium/third_party/dawn/src/dawn_native/metal/PipelineLayoutMTL.h index 7a3ad8084b3..b492e3b5827 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/PipelineLayoutMTL.h +++ b/chromium/third_party/dawn/src/dawn_native/metal/PipelineLayoutMTL.h @@ -15,7 +15,7 @@ #ifndef DAWNNATIVE_METAL_PIPELINELAYOUTMTL_H_ #define DAWNNATIVE_METAL_PIPELINELAYOUTMTL_H_ -#include "common/ityp_array.h" +#include "common/ityp_stack_vec.h" #include "dawn_native/BindingInfo.h" #include "dawn_native/PipelineLayout.h" @@ -44,7 +44,7 @@ namespace dawn_native { namespace metal { using BindingIndexInfo = ityp::array<BindGroupIndex, - ityp::array<BindingIndex, uint32_t, kMaxBindingsPerGroup>, + ityp::stack_vec<BindingIndex, uint32_t, kMaxOptimalBindingsPerGroup>, kMaxBindGroups>; const BindingIndexInfo& GetBindingIndexInfo(SingleShaderStage stage) const; diff --git a/chromium/third_party/dawn/src/dawn_native/metal/PipelineLayoutMTL.mm b/chromium/third_party/dawn/src/dawn_native/metal/PipelineLayoutMTL.mm index 3ee7d92bb1b..fa5d9262edf 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/PipelineLayoutMTL.mm +++ b/chromium/third_party/dawn/src/dawn_native/metal/PipelineLayoutMTL.mm @@ -29,6 +29,8 @@ namespace dawn_native { namespace metal { uint32_t textureIndex = 0; for (BindGroupIndex group : IterateBitSet(GetBindGroupLayoutsMask())) { + mIndexInfo[stage][group].resize(GetBindGroupLayout(group)->GetBindingCount()); + for (BindingIndex bindingIndex{0}; bindingIndex < GetBindGroupLayout(group)->GetBindingCount(); ++bindingIndex) { const BindingInfo& bindingInfo = diff --git a/chromium/third_party/dawn/src/dawn_native/metal/QueueMTL.h b/chromium/third_party/dawn/src/dawn_native/metal/QueueMTL.h index 2dd718e5c06..55915c25b52 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/QueueMTL.h +++ b/chromium/third_party/dawn/src/dawn_native/metal/QueueMTL.h @@ -28,6 +28,10 @@ namespace dawn_native { namespace metal { private: MaybeError SubmitImpl(uint32_t commandCount, CommandBufferBase* const* commands) override; + MaybeError WriteTextureImpl(const TextureCopyView& destination, + const void* data, + const TextureDataLayout& dataLayout, + const Extent3D& writeSizePixel) override; }; }} // namespace dawn_native::metal diff --git a/chromium/third_party/dawn/src/dawn_native/metal/QueueMTL.mm b/chromium/third_party/dawn/src/dawn_native/metal/QueueMTL.mm index 7c5967ad81d..5202ccfef10 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/QueueMTL.mm +++ b/chromium/third_party/dawn/src/dawn_native/metal/QueueMTL.mm @@ -14,12 +14,58 @@ #include "dawn_native/metal/QueueMTL.h" +#include "common/Math.h" +#include "dawn_native/Buffer.h" +#include "dawn_native/CommandValidation.h" +#include "dawn_native/Commands.h" +#include "dawn_native/DynamicUploader.h" #include "dawn_native/metal/CommandBufferMTL.h" #include "dawn_native/metal/DeviceMTL.h" #include "dawn_platform/DawnPlatform.h" #include "dawn_platform/tracing/TraceEvent.h" namespace dawn_native { namespace metal { + namespace { + ResultOrError<UploadHandle> UploadTextureDataAligningBytesPerRow( + DeviceBase* device, + const void* data, + uint32_t alignedBytesPerRow, + uint32_t alignedRowsPerImage, + const TextureDataLayout& dataLayout, + const TexelBlockInfo& blockInfo, + const Extent3D& writeSizePixel) { + uint64_t newDataSizeBytes; + DAWN_TRY_ASSIGN(newDataSizeBytes, + ComputeRequiredBytesInCopy(blockInfo, writeSizePixel, + alignedBytesPerRow, alignedRowsPerImage)); + + UploadHandle uploadHandle; + DAWN_TRY_ASSIGN(uploadHandle, device->GetDynamicUploader()->Allocate( + newDataSizeBytes, device->GetPendingCommandSerial(), + blockInfo.blockByteSize)); + ASSERT(uploadHandle.mappedBuffer != nullptr); + + uint8_t* dstPointer = static_cast<uint8_t*>(uploadHandle.mappedBuffer); + const uint8_t* srcPointer = static_cast<const uint8_t*>(data); + srcPointer += dataLayout.offset; + + uint32_t alignedRowsPerImageInBlock = alignedRowsPerImage / blockInfo.blockHeight; + uint32_t dataRowsPerImageInBlock = dataLayout.rowsPerImage / blockInfo.blockHeight; + if (dataRowsPerImageInBlock == 0) { + dataRowsPerImageInBlock = writeSizePixel.height / blockInfo.blockHeight; + } + + ASSERT(dataRowsPerImageInBlock >= alignedRowsPerImageInBlock); + uint64_t imageAdditionalStride = + dataLayout.bytesPerRow * (dataRowsPerImageInBlock - alignedRowsPerImageInBlock); + + CopyTextureData(dstPointer, srcPointer, writeSizePixel.depth, + alignedRowsPerImageInBlock, imageAdditionalStride, alignedBytesPerRow, + alignedBytesPerRow, dataLayout.bytesPerRow); + + return uploadHandle; + } + } Queue::Queue(Device* device) : QueueBase(device) { } @@ -31,7 +77,7 @@ namespace dawn_native { namespace metal { TRACE_EVENT_BEGIN0(GetDevice()->GetPlatform(), Recording, "CommandBufferMTL::FillCommands"); for (uint32_t i = 0; i < commandCount; ++i) { - ToBackend(commands[i])->FillCommands(commandContext); + DAWN_TRY(ToBackend(commands[i])->FillCommands(commandContext)); } TRACE_EVENT_END0(GetDevice()->GetPlatform(), Recording, "CommandBufferMTL::FillCommands"); @@ -39,4 +85,43 @@ namespace dawn_native { namespace metal { return {}; } + // We don't write from the CPU to the texture directly which can be done in Metal using the + // replaceRegion function, because the function requires a non-private storage mode and Dawn + // sets the private storage mode by default for all textures except IOSurfaces on macOS. + MaybeError Queue::WriteTextureImpl(const TextureCopyView& destination, + const void* data, + const TextureDataLayout& dataLayout, + const Extent3D& writeSizePixel) { + const TexelBlockInfo& blockInfo = + destination.texture->GetFormat().GetTexelBlockInfo(destination.aspect); + + // We are only copying the part of the data that will appear in the texture. + // Note that validating texture copy range ensures that writeSizePixel->width and + // writeSizePixel->height are multiples of blockWidth and blockHeight respectively. + uint32_t alignedBytesPerRow = + (writeSizePixel.width) / blockInfo.blockWidth * blockInfo.blockByteSize; + uint32_t alignedRowsPerImage = writeSizePixel.height; + + UploadHandle uploadHandle; + DAWN_TRY_ASSIGN(uploadHandle, + UploadTextureDataAligningBytesPerRow(GetDevice(), data, alignedBytesPerRow, + alignedRowsPerImage, dataLayout, + blockInfo, writeSizePixel)); + + TextureDataLayout passDataLayout = dataLayout; + passDataLayout.offset = uploadHandle.startOffset; + passDataLayout.bytesPerRow = alignedBytesPerRow; + passDataLayout.rowsPerImage = alignedRowsPerImage; + + TextureCopy textureCopy; + textureCopy.texture = destination.texture; + textureCopy.mipLevel = destination.mipLevel; + textureCopy.origin = destination.origin; + textureCopy.aspect = ConvertAspect(destination.texture->GetFormat(), destination.aspect); + + return ToBackend(GetDevice()) + ->CopyFromStagingToTexture(uploadHandle.stagingBuffer, passDataLayout, &textureCopy, + writeSizePixel); + } + }} // namespace dawn_native::metal diff --git a/chromium/third_party/dawn/src/dawn_native/metal/RenderPipelineMTL.mm b/chromium/third_party/dawn/src/dawn_native/metal/RenderPipelineMTL.mm index 9deba399bda..1e9efe1b30c 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/RenderPipelineMTL.mm +++ b/chromium/third_party/dawn/src/dawn_native/metal/RenderPipelineMTL.mm @@ -329,11 +329,24 @@ namespace dawn_native { namespace metal { MTLRenderPipelineDescriptor* descriptorMTL = [MTLRenderPipelineDescriptor new]; + // TODO: MakeVertexDesc should be const in the future, so we don't need to call it here when + // vertex pulling is enabled + MTLVertexDescriptor* vertexDesc = MakeVertexDesc(); + descriptorMTL.vertexDescriptor = vertexDesc; + [vertexDesc release]; + + if (GetDevice()->IsToggleEnabled(Toggle::MetalEnableVertexPulling)) { + // Calling MakeVertexDesc first is important since it sets indices for packed bindings + MTLVertexDescriptor* emptyVertexDesc = [MTLVertexDescriptor new]; + descriptorMTL.vertexDescriptor = emptyVertexDesc; + [emptyVertexDesc release]; + } + ShaderModule* vertexModule = ToBackend(descriptor->vertexStage.module); const char* vertexEntryPoint = descriptor->vertexStage.entryPoint; ShaderModule::MetalFunctionData vertexData; DAWN_TRY(vertexModule->GetFunction(vertexEntryPoint, SingleShaderStage::Vertex, - ToBackend(GetLayout()), &vertexData)); + ToBackend(GetLayout()), &vertexData, 0xFFFFFFFF, this)); descriptorMTL.vertexFunction = vertexData.function; if (vertexData.needsStorageBufferLength) { @@ -344,7 +357,8 @@ namespace dawn_native { namespace metal { const char* fragmentEntryPoint = descriptor->fragmentStage->entryPoint; ShaderModule::MetalFunctionData fragmentData; DAWN_TRY(fragmentModule->GetFunction(fragmentEntryPoint, SingleShaderStage::Fragment, - ToBackend(GetLayout()), &fragmentData)); + ToBackend(GetLayout()), &fragmentData, + descriptor->sampleMask)); descriptorMTL.fragmentFunction = fragmentData.function; if (fragmentData.needsStorageBufferLength) { @@ -370,18 +384,14 @@ namespace dawn_native { namespace metal { descriptorMTL.colorAttachments[i].pixelFormat = MetalPixelFormat(GetColorAttachmentFormat(i)); const ColorStateDescriptor* descriptor = GetColorStateDescriptor(i); - bool isDeclaredInFragmentShader = fragmentOutputBaseTypes[i] != Format::Other; + bool isDeclaredInFragmentShader = fragmentOutputBaseTypes[i] != Format::Type::Other; ComputeBlendDesc(descriptorMTL.colorAttachments[i], descriptor, isDeclaredInFragmentShader); } descriptorMTL.inputPrimitiveTopology = MTLInputPrimitiveTopology(GetPrimitiveTopology()); - - MTLVertexDescriptor* vertexDesc = MakeVertexDesc(); - descriptorMTL.vertexDescriptor = vertexDesc; - [vertexDesc release]; - descriptorMTL.sampleCount = GetSampleCount(); + descriptorMTL.alphaToCoverageEnabled = descriptor->alphaToCoverageEnabled; { NSError* error = nil; diff --git a/chromium/third_party/dawn/src/dawn_native/metal/ShaderModuleMTL.h b/chromium/third_party/dawn/src/dawn_native/metal/ShaderModuleMTL.h index d4d41abc687..9193b25029a 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/ShaderModuleMTL.h +++ b/chromium/third_party/dawn/src/dawn_native/metal/ShaderModuleMTL.h @@ -29,6 +29,7 @@ namespace dawn_native { namespace metal { class Device; class PipelineLayout; + class RenderPipeline; class ShaderModule final : public ShaderModuleBase { public: @@ -46,14 +47,16 @@ namespace dawn_native { namespace metal { MaybeError GetFunction(const char* functionName, SingleShaderStage functionStage, const PipelineLayout* layout, - MetalFunctionData* out); + MetalFunctionData* out, + uint32_t sampleMask = 0xFFFFFFFF, + const RenderPipeline* renderPipeline = nullptr); private: ShaderModule(Device* device, const ShaderModuleDescriptor* descriptor); ~ShaderModule() override = default; MaybeError Initialize(); - shaderc_spvc::CompileOptions GetMSLCompileOptions(); + shaderc_spvc::CompileOptions GetMSLCompileOptions(uint32_t sampleMask = 0xFFFFFFFF); }; }} // namespace dawn_native::metal diff --git a/chromium/third_party/dawn/src/dawn_native/metal/ShaderModuleMTL.mm b/chromium/third_party/dawn/src/dawn_native/metal/ShaderModuleMTL.mm index 5983e9ea82d..208612e6318 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/ShaderModuleMTL.mm +++ b/chromium/third_party/dawn/src/dawn_native/metal/ShaderModuleMTL.mm @@ -17,6 +17,7 @@ #include "dawn_native/BindGroupLayout.h" #include "dawn_native/metal/DeviceMTL.h" #include "dawn_native/metal/PipelineLayoutMTL.h" +#include "dawn_native/metal/RenderPipelineMTL.h" #include <spirv_msl.hpp> @@ -91,19 +92,35 @@ namespace dawn_native { namespace metal { MaybeError ShaderModule::GetFunction(const char* functionName, SingleShaderStage functionStage, const PipelineLayout* layout, - ShaderModule::MetalFunctionData* out) { + ShaderModule::MetalFunctionData* out, + uint32_t sampleMask, + const RenderPipeline* renderPipeline) { ASSERT(!IsError()); ASSERT(out); - const std::vector<uint32_t>& spirv = GetSpirv(); + const std::vector<uint32_t>* spirv = &GetSpirv(); + +#ifdef DAWN_ENABLE_WGSL + // Use set 4 since it is bigger than what users can access currently + static const uint32_t kPullingBufferBindingSet = 4; + std::vector<uint32_t> pullingSpirv; + if (GetDevice()->IsToggleEnabled(Toggle::MetalEnableVertexPulling) && + functionStage == SingleShaderStage::Vertex) { + DAWN_TRY_ASSIGN(pullingSpirv, + GeneratePullingSpirv(*renderPipeline->GetVertexStateDescriptor(), + functionName, kPullingBufferBindingSet)); + spirv = &pullingSpirv; + } +#endif std::unique_ptr<spirv_cross::CompilerMSL> compilerImpl; spirv_cross::CompilerMSL* compiler; if (GetDevice()->IsToggleEnabled(Toggle::UseSpvc)) { // Initializing the compiler is needed every call, because this method uses reflection // to mutate the compiler's IR. - DAWN_TRY(CheckSpvcSuccess( - mSpvcContext.InitializeForMsl(spirv.data(), spirv.size(), GetMSLCompileOptions()), - "Unable to initialize instance of spvc")); + DAWN_TRY( + CheckSpvcSuccess(mSpvcContext.InitializeForMsl(spirv->data(), spirv->size(), + GetMSLCompileOptions(sampleMask)), + "Unable to initialize instance of spvc")); DAWN_TRY(CheckSpvcSuccess(mSpvcContext.GetCompiler(reinterpret_cast<void**>(&compiler)), "Unable to get cross compiler")); } else { @@ -122,7 +139,9 @@ namespace dawn_native { namespace metal { // the shader storage buffer lengths. options_msl.buffer_size_buffer_index = kBufferLengthBufferSlot; - compilerImpl = std::make_unique<spirv_cross::CompilerMSL>(spirv); + options_msl.additional_fixed_sample_mask = sampleMask; + + compilerImpl = std::make_unique<spirv_cross::CompilerMSL>(*spirv); compiler = compilerImpl.get(); compiler->set_msl_options(options_msl); } @@ -168,6 +187,22 @@ namespace dawn_native { namespace metal { } } + // Add vertex buffers bound as storage buffers + if (GetDevice()->IsToggleEnabled(Toggle::MetalEnableVertexPulling) && + functionStage == SingleShaderStage::Vertex) { + for (uint32_t dawnIndex : IterateBitSet(renderPipeline->GetVertexBufferSlotsUsed())) { + uint32_t metalIndex = renderPipeline->GetMtlVertexBufferIndex(dawnIndex); + + shaderc_spvc_msl_resource_binding mslBinding; + mslBinding.stage = ToSpvcExecutionModel(SingleShaderStage::Vertex); + mslBinding.desc_set = kPullingBufferBindingSet; + mslBinding.binding = dawnIndex; + mslBinding.msl_buffer = metalIndex; + DAWN_TRY(CheckSpvcSuccess(mSpvcContext.AddMSLResourceBinding(mslBinding), + "Unable to add MSL Resource Binding")); + } + } + { if (GetDevice()->IsToggleEnabled(Toggle::UseSpvc)) { shaderc_spvc_execution_model executionModel = ToSpvcExecutionModel(functionStage); @@ -187,29 +222,37 @@ namespace dawn_native { namespace metal { // SPIRV-Cross also supports re-ordering attributes but it seems to do the correct thing // by default. NSString* mslSource; + std::string msl; if (GetDevice()->IsToggleEnabled(Toggle::UseSpvc)) { shaderc_spvc::CompilationResult result; DAWN_TRY(CheckSpvcSuccess(mSpvcContext.CompileShader(&result), "Unable to compile MSL shader")); - std::string result_str; - DAWN_TRY(CheckSpvcSuccess(result.GetStringOutput(&result_str), + DAWN_TRY(CheckSpvcSuccess(result.GetStringOutput(&msl), "Unable to get MSL shader text")); - mslSource = [[NSString alloc] initWithUTF8String:result_str.c_str()]; } else { - std::string msl = compiler->compile(); - mslSource = [[NSString alloc] initWithUTF8String:msl.c_str()]; + msl = compiler->compile(); } + // Metal uses Clang to compile the shader as C++14. Disable everything in the -Wall + // category. -Wunused-variable in particular comes up a lot in generated code, and some + // (old?) Metal drivers accidentally treat it as a MTLLibraryErrorCompileError instead + // of a warning. + msl = R"(\ +#ifdef __clang__ +#pragma clang diagnostic ignored "-Wall" +#endif +)" + msl; + mslSource = [[NSString alloc] initWithUTF8String:msl.c_str()]; + auto mtlDevice = ToBackend(GetDevice())->GetMTLDevice(); NSError* error = nil; id<MTLLibrary> library = [mtlDevice newLibraryWithSource:mslSource options:nil error:&error]; if (error != nil) { - // TODO(cwallez@chromium.org): Switch that NSLog to use dawn::InfoLog or even be - // folded in the DAWN_VALIDATION_ERROR - NSLog(@"MTLDevice newLibraryWithSource => %@", error); if (error.code != MTLLibraryErrorCompileWarning) { - return DAWN_VALIDATION_ERROR("Unable to create library object"); + const char* errorString = [error.localizedDescription UTF8String]; + return DAWN_VALIDATION_ERROR(std::string("Unable to create library object: ") + + errorString); } } @@ -233,10 +276,15 @@ namespace dawn_native { namespace metal { out->needsStorageBufferLength = compiler->needs_buffer_size_buffer(); } + if (GetDevice()->IsToggleEnabled(Toggle::MetalEnableVertexPulling) && + functionStage == SingleShaderStage::Vertex && GetUsedVertexAttributes().any()) { + out->needsStorageBufferLength = true; + } + return {}; } - shaderc_spvc::CompileOptions ShaderModule::GetMSLCompileOptions() { + shaderc_spvc::CompileOptions ShaderModule::GetMSLCompileOptions(uint32_t sampleMask) { // If these options are changed, the values in DawnSPIRVCrossGLSLFastFuzzer.cpp need to // be updated. shaderc_spvc::CompileOptions options = GetCompileOptions(); @@ -252,6 +300,8 @@ namespace dawn_native { namespace metal { // the shader storage buffer lengths. options.SetMSLBufferSizeBufferIndex(kBufferLengthBufferSlot); + options.SetMSLAdditionalFixedSampleMask(sampleMask); + return options; } diff --git a/chromium/third_party/dawn/src/dawn_native/metal/TextureMTL.mm b/chromium/third_party/dawn/src/dawn_native/metal/TextureMTL.mm index 454a14b2739..449f3389516 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/TextureMTL.mm +++ b/chromium/third_party/dawn/src/dawn_native/metal/TextureMTL.mm @@ -18,8 +18,10 @@ #include "common/Math.h" #include "common/Platform.h" #include "dawn_native/DynamicUploader.h" +#include "dawn_native/EnumMaskIterator.h" #include "dawn_native/metal/DeviceMTL.h" #include "dawn_native/metal/StagingBufferMTL.h" +#include "dawn_native/metal/UtilsMetal.h" #include <CoreVideo/CVPixelBuffer.h> @@ -47,10 +49,6 @@ namespace dawn_native { namespace metal { result |= MTLTextureUsageRenderTarget; } - if (UsageNeedsTextureView(usage)) { - result |= MTLTextureUsagePixelFormatView; - } - return result; } @@ -174,8 +172,10 @@ namespace dawn_native { namespace metal { return MTLPixelFormatBGRA8Unorm_sRGB; case wgpu::TextureFormat::RGB10A2Unorm: return MTLPixelFormatRGB10A2Unorm; - case wgpu::TextureFormat::RG11B10Float: + case wgpu::TextureFormat::RG11B10Ufloat: return MTLPixelFormatRG11B10Float; + case wgpu::TextureFormat::RGB9E5Ufloat: + return MTLPixelFormatRGB9E5Float; case wgpu::TextureFormat::RG32Uint: return MTLPixelFormatRG32Uint; @@ -289,6 +289,8 @@ namespace dawn_native { namespace metal { mtlDesc.width = descriptor->size.width; mtlDesc.height = descriptor->size.height; mtlDesc.sampleCount = descriptor->sampleCount; + // TODO: add MTLTextureUsagePixelFormatView when needed when we support format + // reinterpretation. mtlDesc.usage = MetalTextureUsage(descriptor->usage); mtlDesc.pixelFormat = MetalPixelFormat(descriptor->format); mtlDesc.mipmapLevelCount = descriptor->mipLevelCount; @@ -353,7 +355,7 @@ namespace dawn_native { namespace metal { plane:plane]; [mtlDesc release]; - SetIsSubresourceContentInitialized(descriptor->isCleared, {0, 1, 0, 1}); + SetIsSubresourceContentInitialized(descriptor->isCleared, GetAllSubresources()); } Texture::~Texture() { @@ -393,8 +395,8 @@ namespace dawn_native { namespace metal { for (uint32_t arrayLayer = range.baseArrayLayer; arrayLayer < range.baseArrayLayer + range.layerCount; arrayLayer++) { if (clearValue == TextureBase::ClearValue::Zero && - IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, arrayLayer))) { + IsSubresourceContentInitialized(SubresourceRange::SingleMipAndLayer( + level, arrayLayer, range.aspects))) { // Skip lazy clears if already initialized. continue; } @@ -402,18 +404,34 @@ namespace dawn_native { namespace metal { MTLRenderPassDescriptor* descriptor = [MTLRenderPassDescriptor renderPassDescriptor]; - if (GetFormat().HasDepth()) { - descriptor.depthAttachment.texture = GetMTLTexture(); - descriptor.depthAttachment.loadAction = MTLLoadActionClear; - descriptor.depthAttachment.storeAction = MTLStoreActionStore; - descriptor.depthAttachment.clearDepth = dClearColor; - } - if (GetFormat().HasStencil()) { - descriptor.stencilAttachment.texture = GetMTLTexture(); - descriptor.stencilAttachment.loadAction = MTLLoadActionClear; - descriptor.stencilAttachment.storeAction = MTLStoreActionStore; - descriptor.stencilAttachment.clearStencil = - static_cast<uint32_t>(clearColor); + // At least one aspect needs clearing. Iterate the aspects individually to + // determine which to clear. + for (Aspect aspect : IterateEnumMask(range.aspects)) { + if (clearValue == TextureBase::ClearValue::Zero && + IsSubresourceContentInitialized(SubresourceRange::SingleMipAndLayer( + level, arrayLayer, aspect))) { + // Skip lazy clears if already initialized. + continue; + } + + switch (aspect) { + case Aspect::Depth: + descriptor.depthAttachment.texture = GetMTLTexture(); + descriptor.depthAttachment.loadAction = MTLLoadActionClear; + descriptor.depthAttachment.storeAction = MTLStoreActionStore; + descriptor.depthAttachment.clearDepth = dClearColor; + break; + case Aspect::Stencil: + descriptor.stencilAttachment.texture = GetMTLTexture(); + descriptor.stencilAttachment.loadAction = MTLLoadActionClear; + descriptor.stencilAttachment.storeAction = MTLStoreActionStore; + descriptor.stencilAttachment.clearStencil = + static_cast<uint32_t>(clearColor); + break; + default: + UNREACHABLE(); + break; + } } commandContext->BeginRender(descriptor); @@ -433,8 +451,8 @@ namespace dawn_native { namespace metal { for (uint32_t arrayLayer = range.baseArrayLayer; arrayLayer < range.baseArrayLayer + range.layerCount; arrayLayer++) { if (clearValue == TextureBase::ClearValue::Zero && - IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, arrayLayer))) { + IsSubresourceContentInitialized(SubresourceRange::SingleMipAndLayer( + level, arrayLayer, Aspect::Color))) { // Skip lazy clears if already initialized. continue; } @@ -492,41 +510,29 @@ namespace dawn_native { namespace metal { DynamicUploader* uploader = device->GetDynamicUploader(); UploadHandle uploadHandle; DAWN_TRY_ASSIGN(uploadHandle, - uploader->Allocate(bufferSize, device->GetPendingCommandSerial())); + uploader->Allocate(bufferSize, device->GetPendingCommandSerial(), + GetFormat().blockByteSize)); memset(uploadHandle.mappedBuffer, clearColor, bufferSize); id<MTLBlitCommandEncoder> encoder = commandContext->EnsureBlit(); id<MTLBuffer> uploadBuffer = ToBackend(uploadHandle.stagingBuffer)->GetBufferHandle(); // Encode a buffer to texture copy to clear each subresource. - for (uint32_t level = range.baseMipLevel; level < range.baseMipLevel + range.levelCount; - ++level) { - Extent3D virtualSize = GetMipLevelVirtualSize(level); - - for (uint32_t arrayLayer = range.baseArrayLayer; - arrayLayer < range.baseArrayLayer + range.layerCount; ++arrayLayer) { - if (clearValue == TextureBase::ClearValue::Zero && - IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, arrayLayer))) { - // Skip lazy clears if already initialized. - continue; - } + for (Aspect aspect : IterateEnumMask(range.aspects)) { + for (uint32_t level = range.baseMipLevel; + level < range.baseMipLevel + range.levelCount; ++level) { + Extent3D virtualSize = GetMipLevelVirtualSize(level); - // If the texture’s pixel format is a combined depth/stencil format, then - // options must be set to either blit the depth attachment portion or blit the - // stencil attachment portion. - std::array<MTLBlitOption, 3> blitOptions = { - MTLBlitOptionNone, MTLBlitOptionDepthFromDepthStencil, - MTLBlitOptionStencilFromDepthStencil}; - - auto blitOptionStart = blitOptions.begin(); - auto blitOptionEnd = blitOptionStart + 1; - if (GetFormat().format == wgpu::TextureFormat::Depth24PlusStencil8) { - blitOptionStart = blitOptions.begin() + 1; - blitOptionEnd = blitOptionStart + 2; - } + for (uint32_t arrayLayer = range.baseArrayLayer; + arrayLayer < range.baseArrayLayer + range.layerCount; ++arrayLayer) { + if (clearValue == TextureBase::ClearValue::Zero && + IsSubresourceContentInitialized( + SubresourceRange::SingleMipAndLayer(level, arrayLayer, aspect))) { + // Skip lazy clears if already initialized. + continue; + } - for (auto it = blitOptionStart; it != blitOptionEnd; ++it) { + MTLBlitOption blitOption = ComputeMTLBlitOption(GetFormat(), aspect); [encoder copyFromBuffer:uploadBuffer sourceOffset:uploadHandle.startOffset sourceBytesPerRow:largestMipBytesPerRow @@ -537,7 +543,7 @@ namespace dawn_native { namespace metal { destinationSlice:arrayLayer destinationLevel:level destinationOrigin:MTLOriginMake(0, 0, 0) - options:(*it)]; + options:blitOption]; } } } diff --git a/chromium/third_party/dawn/src/dawn_native/metal/UtilsMetal.h b/chromium/third_party/dawn/src/dawn_native/metal/UtilsMetal.h index 091d8284f0d..d7c0a70e52c 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/UtilsMetal.h +++ b/chromium/third_party/dawn/src/dawn_native/metal/UtilsMetal.h @@ -16,6 +16,8 @@ #define DAWNNATIVE_METAL_UTILSMETAL_H_ #include "dawn_native/dawn_platform.h" +#include "dawn_native/metal/DeviceMTL.h" +#include "dawn_native/metal/TextureMTL.h" #import <Metal/Metal.h> @@ -23,6 +25,36 @@ namespace dawn_native { namespace metal { MTLCompareFunction ToMetalCompareFunction(wgpu::CompareFunction compareFunction); + struct TextureBufferCopySplit { + static constexpr uint32_t kMaxTextureBufferCopyRegions = 3; + + struct CopyInfo { + NSUInteger bufferOffset; + NSUInteger bytesPerRow; + NSUInteger bytesPerImage; + Origin3D textureOrigin; + Extent3D copyExtent; + }; + + uint32_t count = 0; + std::array<CopyInfo, kMaxTextureBufferCopyRegions> copies; + }; + + TextureBufferCopySplit ComputeTextureBufferCopySplit(const Texture* texture, + uint32_t mipLevel, + Origin3D origin, + Extent3D copyExtent, + uint64_t bufferSize, + uint64_t bufferOffset, + uint32_t bytesPerRow, + uint32_t rowsPerImage); + + void EnsureDestinationTextureInitialized(Texture* texture, + const TextureCopy& dst, + const Extent3D& size); + + MTLBlitOption ComputeMTLBlitOption(const Format& format, Aspect aspect); + }} // namespace dawn_native::metal #endif // DAWNNATIVE_METAL_UTILSMETAL_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/metal/UtilsMetal.mm b/chromium/third_party/dawn/src/dawn_native/metal/UtilsMetal.mm index aeb4f7aae8c..3b8f64c0d8c 100644 --- a/chromium/third_party/dawn/src/dawn_native/metal/UtilsMetal.mm +++ b/chromium/third_party/dawn/src/dawn_native/metal/UtilsMetal.mm @@ -13,6 +13,7 @@ // limitations under the License. #include "dawn_native/metal/UtilsMetal.h" +#include "dawn_native/CommandBuffer.h" #include "common/Assert.h" @@ -41,4 +42,146 @@ namespace dawn_native { namespace metal { } } + TextureBufferCopySplit ComputeTextureBufferCopySplit(const Texture* texture, + uint32_t mipLevel, + Origin3D origin, + Extent3D copyExtent, + uint64_t bufferSize, + uint64_t bufferOffset, + uint32_t bytesPerRow, + uint32_t rowsPerImage) { + TextureBufferCopySplit copy; + const Format textureFormat = texture->GetFormat(); + + // When copying textures from/to an unpacked buffer, the Metal validation layer doesn't + // compute the correct range when checking if the buffer is big enough to contain the + // data for the whole copy. Instead of looking at the position of the last texel in the + // buffer, it computes the volume of the 3D box with bytesPerRow * (rowsPerImage / + // format.blockHeight) * copySize.depth. For example considering the pixel buffer below + // where in memory, each row data (D) of the texture is followed by some padding data + // (P): + // |DDDDDDD|PP| + // |DDDDDDD|PP| + // |DDDDDDD|PP| + // |DDDDDDD|PP| + // |DDDDDDA|PP| + // The last pixel read will be A, but the driver will think it is the whole last padding + // row, causing it to generate an error when the pixel buffer is just big enough. + + // We work around this limitation by detecting when Metal would complain and copy the + // last image and row separately using tight sourceBytesPerRow or sourceBytesPerImage. + uint32_t dataRowsPerImage = rowsPerImage / textureFormat.blockHeight; + uint32_t bytesPerImage = bytesPerRow * dataRowsPerImage; + + // Metal validation layer requires that if the texture's pixel format is a compressed + // format, the sourceSize must be a multiple of the pixel format's block size or be + // clamped to the edge of the texture if the block extends outside the bounds of a + // texture. + const Extent3D clampedCopyExtent = + texture->ClampToMipLevelVirtualSize(mipLevel, origin, copyExtent); + + ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D); + + // Check whether buffer size is big enough. + bool needWorkaround = bufferSize - bufferOffset < bytesPerImage * copyExtent.depth; + if (!needWorkaround) { + copy.count = 1; + copy.copies[0].bufferOffset = bufferOffset; + copy.copies[0].bytesPerRow = bytesPerRow; + copy.copies[0].bytesPerImage = bytesPerImage; + copy.copies[0].textureOrigin = origin; + copy.copies[0].copyExtent = {clampedCopyExtent.width, clampedCopyExtent.height, + copyExtent.depth}; + return copy; + } + + uint64_t currentOffset = bufferOffset; + + // Doing all the copy except the last image. + if (copyExtent.depth > 1) { + copy.copies[copy.count].bufferOffset = currentOffset; + copy.copies[copy.count].bytesPerRow = bytesPerRow; + copy.copies[copy.count].bytesPerImage = bytesPerImage; + copy.copies[copy.count].textureOrigin = origin; + copy.copies[copy.count].copyExtent = {clampedCopyExtent.width, clampedCopyExtent.height, + copyExtent.depth - 1}; + + ++copy.count; + + // Update offset to copy to the last image. + currentOffset += (copyExtent.depth - 1) * bytesPerImage; + } + + // Doing all the copy in last image except the last row. + uint32_t copyBlockRowCount = copyExtent.height / textureFormat.blockHeight; + if (copyBlockRowCount > 1) { + copy.copies[copy.count].bufferOffset = currentOffset; + copy.copies[copy.count].bytesPerRow = bytesPerRow; + copy.copies[copy.count].bytesPerImage = bytesPerRow * (copyBlockRowCount - 1); + copy.copies[copy.count].textureOrigin = {origin.x, origin.y, + origin.z + copyExtent.depth - 1}; + + ASSERT(copyExtent.height - textureFormat.blockHeight < + texture->GetMipLevelVirtualSize(mipLevel).height); + copy.copies[copy.count].copyExtent = {clampedCopyExtent.width, + copyExtent.height - textureFormat.blockHeight, 1}; + + ++copy.count; + + // Update offset to copy to the last row. + currentOffset += (copyBlockRowCount - 1) * bytesPerRow; + } + + // Doing the last row copy with the exact number of bytes in last row. + // Workaround this issue in a way just like the copy to a 1D texture. + uint32_t lastRowDataSize = + (copyExtent.width / textureFormat.blockWidth) * textureFormat.blockByteSize; + uint32_t lastRowCopyExtentHeight = + textureFormat.blockHeight + clampedCopyExtent.height - copyExtent.height; + ASSERT(lastRowCopyExtentHeight <= textureFormat.blockHeight); + + copy.copies[copy.count].bufferOffset = currentOffset; + copy.copies[copy.count].bytesPerRow = lastRowDataSize; + copy.copies[copy.count].bytesPerImage = lastRowDataSize; + copy.copies[copy.count].textureOrigin = { + origin.x, origin.y + copyExtent.height - textureFormat.blockHeight, + origin.z + copyExtent.depth - 1}; + copy.copies[copy.count].copyExtent = {clampedCopyExtent.width, lastRowCopyExtentHeight, 1}; + ++copy.count; + + return copy; + } + + void EnsureDestinationTextureInitialized(Texture* texture, + const TextureCopy& dst, + const Extent3D& size) { + ASSERT(texture == dst.texture.Get()); + SubresourceRange range = GetSubresourcesAffectedByCopy(dst, size); + if (IsCompleteSubresourceCopiedTo(dst.texture.Get(), size, dst.mipLevel)) { + texture->SetIsSubresourceContentInitialized(true, range); + } else { + texture->EnsureSubresourceContentInitialized(range); + } + } + + MTLBlitOption ComputeMTLBlitOption(const Format& format, Aspect aspect) { + ASSERT(HasOneBit(aspect)); + ASSERT(format.aspects & aspect); + + constexpr Aspect kDepthStencil = Aspect::Depth | Aspect::Stencil; + if ((format.aspects & kDepthStencil) == kDepthStencil) { + // We only provide a blit option if the format has both depth and stencil. + // It is invalid to provide a blit option otherwise. + switch (aspect) { + case Aspect::Depth: + return MTLBlitOptionDepthFromDepthStencil; + case Aspect::Stencil: + return MTLBlitOptionStencilFromDepthStencil; + default: + UNREACHABLE(); + } + } + return MTLBlitOptionNone; + } + }} // namespace dawn_native::metal diff --git a/chromium/third_party/dawn/src/dawn_native/null/DeviceNull.cpp b/chromium/third_party/dawn/src/dawn_native/null/DeviceNull.cpp index adc44f0974c..d382bb8874a 100644 --- a/chromium/third_party/dawn/src/dawn_native/null/DeviceNull.cpp +++ b/chromium/third_party/dawn/src/dawn_native/null/DeviceNull.cpp @@ -100,9 +100,9 @@ namespace dawn_native { namespace null { const BindGroupLayoutDescriptor* descriptor) { return new BindGroupLayout(this, descriptor); } - ResultOrError<BufferBase*> Device::CreateBufferImpl(const BufferDescriptor* descriptor) { + ResultOrError<Ref<BufferBase>> Device::CreateBufferImpl(const BufferDescriptor* descriptor) { DAWN_TRY(IncrementMemoryUsage(descriptor->size)); - return new Buffer(this, descriptor); + return AcquireRef(new Buffer(this, descriptor)); } CommandBufferBase* Device::CreateCommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor) { @@ -197,6 +197,10 @@ namespace dawn_native { namespace null { BufferBase* destination, uint64_t destinationOffset, uint64_t size) { + if (IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + destination->SetIsDataInitialized(); + } + auto operation = std::make_unique<CopyFromStagingToBufferOperation>(); operation->staging = source; operation->destination = ToBackend(destination); @@ -209,8 +213,8 @@ namespace dawn_native { namespace null { return {}; } - MaybeError Device::IncrementMemoryUsage(size_t bytes) { - static_assert(kMaxMemoryUsage <= std::numeric_limits<size_t>::max() / 2, ""); + MaybeError Device::IncrementMemoryUsage(uint64_t bytes) { + static_assert(kMaxMemoryUsage <= std::numeric_limits<size_t>::max(), ""); if (bytes > kMaxMemoryUsage || mMemoryUsage + bytes > kMaxMemoryUsage) { return DAWN_OUT_OF_MEMORY_ERROR("Out of memory."); } @@ -218,7 +222,7 @@ namespace dawn_native { namespace null { return {}; } - void Device::DecrementMemoryUsage(size_t bytes) { + void Device::DecrementMemoryUsage(uint64_t bytes) { ASSERT(mMemoryUsage >= bytes); mMemoryUsage -= bytes; } @@ -266,17 +270,6 @@ namespace dawn_native { namespace null { // Buffer - struct BufferMapOperation : PendingOperation { - virtual void Execute() { - buffer->OnMapCommandSerialFinished(serial, isWrite); - } - - Ref<Buffer> buffer; - void* ptr; - uint32_t serial; - bool isWrite; - }; - Buffer::Buffer(Device* device, const BufferDescriptor* descriptor) : BufferBase(device, descriptor) { mBackingData = std::unique_ptr<uint8_t[]>(new uint8_t[GetSize()]); @@ -287,14 +280,13 @@ namespace dawn_native { namespace null { ToBackend(GetDevice())->DecrementMemoryUsage(GetSize()); } - bool Buffer::IsMapWritable() const { + bool Buffer::IsMappableAtCreation() const { // Only return true for mappable buffers so we can test cases that need / don't need a // staging buffer. return (GetUsage() & (wgpu::BufferUsage::MapRead | wgpu::BufferUsage::MapWrite)) != 0; } - MaybeError Buffer::MapAtCreationImpl(uint8_t** mappedPointer) { - *mappedPointer = mBackingData.get(); + MaybeError Buffer::MapAtCreationImpl() { return {}; } @@ -312,26 +304,16 @@ namespace dawn_native { namespace null { memcpy(mBackingData.get() + bufferOffset, data, size); } - MaybeError Buffer::MapReadAsyncImpl(uint32_t serial) { - MapAsyncImplCommon(serial, false); + MaybeError Buffer::MapReadAsyncImpl() { return {}; } - MaybeError Buffer::MapWriteAsyncImpl(uint32_t serial) { - MapAsyncImplCommon(serial, true); + MaybeError Buffer::MapWriteAsyncImpl() { return {}; } - void Buffer::MapAsyncImplCommon(uint32_t serial, bool isWrite) { - ASSERT(mBackingData); - - auto operation = std::make_unique<BufferMapOperation>(); - operation->buffer = this; - operation->ptr = mBackingData.get(); - operation->serial = serial; - operation->isWrite = isWrite; - - ToBackend(GetDevice())->AddPendingOperation(std::move(operation)); + MaybeError Buffer::MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) { + return {}; } void* Buffer::GetMappedPointerImpl() { @@ -347,11 +329,7 @@ namespace dawn_native { namespace null { // CommandBuffer CommandBuffer::CommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor) - : CommandBufferBase(encoder, descriptor), mCommands(encoder->AcquireCommands()) { - } - - CommandBuffer::~CommandBuffer() { - FreeCommands(&mCommands); + : CommandBufferBase(encoder, descriptor) { } // QuerySet diff --git a/chromium/third_party/dawn/src/dawn_native/null/DeviceNull.h b/chromium/third_party/dawn/src/dawn_native/null/DeviceNull.h index f6f361185c8..b2dccfe918e 100644 --- a/chromium/third_party/dawn/src/dawn_native/null/DeviceNull.h +++ b/chromium/third_party/dawn/src/dawn_native/null/DeviceNull.h @@ -106,8 +106,8 @@ namespace dawn_native { namespace null { uint64_t destinationOffset, uint64_t size) override; - MaybeError IncrementMemoryUsage(size_t bytes); - void DecrementMemoryUsage(size_t bytes); + MaybeError IncrementMemoryUsage(uint64_t bytes); + void DecrementMemoryUsage(uint64_t bytes); private: using DeviceBase::DeviceBase; @@ -116,7 +116,8 @@ namespace dawn_native { namespace null { const BindGroupDescriptor* descriptor) override; ResultOrError<BindGroupLayoutBase*> CreateBindGroupLayoutImpl( const BindGroupLayoutDescriptor* descriptor) override; - ResultOrError<BufferBase*> CreateBufferImpl(const BufferDescriptor* descriptor) override; + ResultOrError<Ref<BufferBase>> CreateBufferImpl( + const BufferDescriptor* descriptor) override; ResultOrError<ComputePipelineBase*> CreateComputePipelineImpl( const ComputePipelineDescriptor* descriptor) override; ResultOrError<PipelineLayoutBase*> CreatePipelineLayoutImpl( @@ -147,7 +148,7 @@ namespace dawn_native { namespace null { std::vector<std::unique_ptr<PendingOperation>> mPendingOperations; - static constexpr size_t kMaxMemoryUsage = 256 * 1024 * 1024; + static constexpr uint64_t kMaxMemoryUsage = 256 * 1024 * 1024; size_t mMemoryUsage = 0; }; @@ -198,14 +199,14 @@ namespace dawn_native { namespace null { ~Buffer() override; // Dawn API - MaybeError MapReadAsyncImpl(uint32_t serial) override; - MaybeError MapWriteAsyncImpl(uint32_t serial) override; + MaybeError MapReadAsyncImpl() override; + MaybeError MapWriteAsyncImpl() override; + MaybeError MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) override; void UnmapImpl() override; void DestroyImpl() override; - bool IsMapWritable() const override; - MaybeError MapAtCreationImpl(uint8_t** mappedPointer) override; - void MapAsyncImplCommon(uint32_t serial, bool isWrite); + bool IsMappableAtCreation() const override; + MaybeError MapAtCreationImpl() override; void* GetMappedPointerImpl() override; std::unique_ptr<uint8_t[]> mBackingData; @@ -215,10 +216,6 @@ namespace dawn_native { namespace null { public: CommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor); - private: - ~CommandBuffer() override; - - CommandIterator mCommands; }; class QuerySet final : public QuerySetBase { diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/BackendGL.cpp b/chromium/third_party/dawn/src/dawn_native/opengl/BackendGL.cpp index 5ce73d88b54..2d3685980ec 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/BackendGL.cpp +++ b/chromium/third_party/dawn/src/dawn_native/opengl/BackendGL.cpp @@ -162,6 +162,7 @@ namespace dawn_native { namespace opengl { mFunctions.Enable(GL_PRIMITIVE_RESTART_FIXED_INDEX); mFunctions.Enable(GL_MULTISAMPLE); mFunctions.Enable(GL_FRAMEBUFFER_SRGB); + mFunctions.Enable(GL_SAMPLE_MASK); mPCIInfo.name = reinterpret_cast<const char*>(mFunctions.GetString(GL_RENDERER)); diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/BufferGL.cpp b/chromium/third_party/dawn/src/dawn_native/opengl/BufferGL.cpp index 7e91a4940cd..6289c6edb05 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/BufferGL.cpp +++ b/chromium/third_party/dawn/src/dawn_native/opengl/BufferGL.cpp @@ -14,6 +14,7 @@ #include "dawn_native/opengl/BufferGL.h" +#include "dawn_native/CommandBuffer.h" #include "dawn_native/opengl/DeviceGL.h" namespace dawn_native { namespace opengl { @@ -24,7 +25,7 @@ namespace dawn_native { namespace opengl { : BufferBase(device, descriptor) { // TODO(cwallez@chromium.org): Have a global "zero" buffer instead of creating a new 4-byte // buffer? - uint64_t size = std::max(GetSize(), uint64_t(4u)); + uint64_t size = GetAppliedSize(); device->gl.GenBuffers(1, &mBuffer); device->gl.BindBuffer(GL_ARRAY_BUFFER, mBuffer); @@ -45,22 +46,84 @@ namespace dawn_native { namespace opengl { return mBuffer; } - bool Buffer::IsMapWritable() const { + uint64_t Buffer::GetAppliedSize() const { + // TODO(cwallez@chromium.org): Have a global "zero" buffer instead of creating a new 4-byte + // buffer? + return std::max(GetSize(), uint64_t(4u)); + } + + void Buffer::EnsureDataInitialized() { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return; + } + + InitializeToZero(); + } + + void Buffer::EnsureDataInitializedAsDestination(uint64_t offset, uint64_t size) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return; + } + + if (IsFullBufferRange(offset, size)) { + SetIsDataInitialized(); + } else { + InitializeToZero(); + } + } + + void Buffer::EnsureDataInitializedAsDestination(const CopyTextureToBufferCmd* copy) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return; + } + + if (IsFullBufferOverwrittenInTextureToBufferCopy(copy)) { + SetIsDataInitialized(); + } else { + InitializeToZero(); + } + } + + void Buffer::InitializeToZero() { + ASSERT(GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)); + ASSERT(!IsDataInitialized()); + + const uint64_t size = GetAppliedSize(); + Device* device = ToBackend(GetDevice()); + + const std::vector<uint8_t> clearValues(size, 0u); + device->gl.BindBuffer(GL_ARRAY_BUFFER, mBuffer); + device->gl.BufferSubData(GL_ARRAY_BUFFER, 0, size, clearValues.data()); + device->IncrementLazyClearCountForTesting(); + + SetIsDataInitialized(); + } + + bool Buffer::IsMappableAtCreation() const { // TODO(enga): All buffers in GL can be mapped. Investigate if mapping them will cause the // driver to migrate it to shared memory. return true; } - MaybeError Buffer::MapAtCreationImpl(uint8_t** mappedPointer) { - const OpenGLFunctions& gl = ToBackend(GetDevice())->gl; + MaybeError Buffer::MapAtCreationImpl() { + EnsureDataInitialized(); + const OpenGLFunctions& gl = ToBackend(GetDevice())->gl; gl.BindBuffer(GL_ARRAY_BUFFER, mBuffer); mMappedData = gl.MapBuffer(GL_ARRAY_BUFFER, GL_WRITE_ONLY); - *mappedPointer = reinterpret_cast<uint8_t*>(mMappedData); return {}; } - MaybeError Buffer::MapReadAsyncImpl(uint32_t serial) { + MaybeError Buffer::MapReadAsyncImpl() { const OpenGLFunctions& gl = ToBackend(GetDevice())->gl; // TODO(cwallez@chromium.org): this does GPU->CPU synchronization, we could require a high @@ -70,7 +133,7 @@ namespace dawn_native { namespace opengl { return {}; } - MaybeError Buffer::MapWriteAsyncImpl(uint32_t serial) { + MaybeError Buffer::MapWriteAsyncImpl() { const OpenGLFunctions& gl = ToBackend(GetDevice())->gl; // TODO(cwallez@chromium.org): this does GPU->CPU synchronization, we could require a high @@ -80,7 +143,39 @@ namespace dawn_native { namespace opengl { return {}; } + MaybeError Buffer::MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) { + const OpenGLFunctions& gl = ToBackend(GetDevice())->gl; + + // It is an error to map an empty range in OpenGL. We always have at least a 4-byte buffer + // so we extend the range to be 4 bytes. + if (size == 0) { + if (offset != 0) { + offset -= 4; + } + size = 4; + } + + EnsureDataInitialized(); + + // TODO(cwallez@chromium.org): this does GPU->CPU synchronization, we could require a high + // version of OpenGL that would let us map the buffer unsynchronized. + gl.BindBuffer(GL_ARRAY_BUFFER, mBuffer); + void* mappedData = nullptr; + if (mode & wgpu::MapMode::Read) { + mappedData = gl.MapBufferRange(GL_ARRAY_BUFFER, offset, size, GL_MAP_READ_BIT); + } else { + ASSERT(mode & wgpu::MapMode::Write); + mappedData = gl.MapBufferRange(GL_ARRAY_BUFFER, offset, size, GL_MAP_WRITE_BIT); + } + + // The frontend asks that the pointer returned by GetMappedPointerImpl is from the start of + // the resource but OpenGL gives us the pointer at offset. Remove the offset. + mMappedData = static_cast<uint8_t*>(mappedData) - offset; + return {}; + } + void* Buffer::GetMappedPointerImpl() { + // The mapping offset has already been removed. return mMappedData; } diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/BufferGL.h b/chromium/third_party/dawn/src/dawn_native/opengl/BufferGL.h index 9949829a4be..8fcc7fcc346 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/BufferGL.h +++ b/chromium/third_party/dawn/src/dawn_native/opengl/BufferGL.h @@ -29,17 +29,25 @@ namespace dawn_native { namespace opengl { GLuint GetHandle() const; + void EnsureDataInitialized(); + void EnsureDataInitializedAsDestination(uint64_t offset, uint64_t size); + void EnsureDataInitializedAsDestination(const CopyTextureToBufferCmd* copy); + private: ~Buffer() override; // Dawn API - MaybeError MapReadAsyncImpl(uint32_t serial) override; - MaybeError MapWriteAsyncImpl(uint32_t serial) override; + MaybeError MapReadAsyncImpl() override; + MaybeError MapWriteAsyncImpl() override; + MaybeError MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) override; void UnmapImpl() override; void DestroyImpl() override; - bool IsMapWritable() const override; - MaybeError MapAtCreationImpl(uint8_t** mappedPointer) override; + bool IsMappableAtCreation() const override; + MaybeError MapAtCreationImpl() override; void* GetMappedPointerImpl() override; + uint64_t GetAppliedSize() const; + + void InitializeToZero(); GLuint mBuffer = 0; void* mMappedData = nullptr; diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/CommandBufferGL.cpp b/chromium/third_party/dawn/src/dawn_native/opengl/CommandBufferGL.cpp index 28c606adeac..be6b58d9d8e 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/CommandBufferGL.cpp +++ b/chromium/third_party/dawn/src/dawn_native/opengl/CommandBufferGL.cpp @@ -442,14 +442,10 @@ namespace dawn_native { namespace opengl { } // namespace CommandBuffer::CommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor) - : CommandBufferBase(encoder, descriptor), mCommands(encoder->AcquireCommands()) { + : CommandBufferBase(encoder, descriptor) { } - CommandBuffer::~CommandBuffer() { - FreeCommands(&mCommands); - } - - void CommandBuffer::Execute() { + MaybeError CommandBuffer::Execute() { const OpenGLFunctions& gl = ToBackend(GetDevice())->gl; auto TransitionForPass = [](const PassResourceUsage& usages) { @@ -462,6 +458,10 @@ namespace dawn_native { namespace opengl { texture->EnsureSubresourceContentInitialized(texture->GetAllSubresources()); } } + + for (BufferBase* bufferBase : usages.buffers) { + ToBackend(bufferBase)->EnsureDataInitialized(); + } }; const std::vector<PassResourceUsage>& passResourceUsages = GetResourceUsages().perPass; @@ -493,6 +493,10 @@ namespace dawn_native { namespace opengl { case Command::CopyBufferToBuffer: { CopyBufferToBufferCmd* copy = mCommands.NextCommand<CopyBufferToBufferCmd>(); + ToBackend(copy->source)->EnsureDataInitialized(); + ToBackend(copy->destination) + ->EnsureDataInitializedAsDestination(copy->destinationOffset, copy->size); + gl.BindBuffer(GL_PIXEL_PACK_BUFFER, ToBackend(copy->source)->GetHandle()); gl.BindBuffer(GL_PIXEL_UNPACK_BUFFER, ToBackend(copy->destination)->GetHandle()); @@ -514,12 +518,17 @@ namespace dawn_native { namespace opengl { GLenum target = texture->GetGLTarget(); const GLFormat& format = texture->GetGLFormat(); + if (dst.aspect == Aspect::Stencil) { + return DAWN_VALIDATION_ERROR( + "Copies to stencil textures unsupported on OpenGL"); + } + ASSERT(dst.aspect == Aspect::Color); + + buffer->EnsureDataInitialized(); + ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D); - // TODO(jiawei.shao@intel.com): use copy->destination.origin.z instead of - // copy->destination.arrayLayer once GPUTextureCopyView.arrayLayer to - // GPUTextureCopyView.origin.z is done. - SubresourceRange subresources = {dst.mipLevel, 1, dst.arrayLayer, - copy->copySize.depth}; + SubresourceRange subresources = + GetSubresourcesAffectedByCopy(dst, copy->copySize); if (IsCompleteSubresourceCopiedTo(texture, copySize, dst.mipLevel)) { texture->SetIsSubresourceContentInitialized(true, subresources); } else { @@ -545,12 +554,12 @@ namespace dawn_native { namespace opengl { ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D); uint64_t copyDataSize = (copySize.width / formatInfo.blockWidth) * (copySize.height / formatInfo.blockHeight) * - formatInfo.blockByteSize; + formatInfo.blockByteSize * copySize.depth; Extent3D copyExtent = ComputeTextureCopyExtent(dst, copySize); if (texture->GetArrayLayers() > 1) { gl.CompressedTexSubImage3D( - target, dst.mipLevel, dst.origin.x, dst.origin.y, dst.arrayLayer, + target, dst.mipLevel, dst.origin.x, dst.origin.y, dst.origin.z, copyExtent.width, copyExtent.height, copyExtent.depth, format.internalFormat, copyDataSize, reinterpret_cast<void*>(static_cast<uintptr_t>(src.offset))); @@ -565,7 +574,7 @@ namespace dawn_native { namespace opengl { case wgpu::TextureDimension::e2D: if (texture->GetArrayLayers() > 1) { gl.TexSubImage3D(target, dst.mipLevel, dst.origin.x, - dst.origin.y, dst.arrayLayer, copySize.width, + dst.origin.y, dst.origin.z, copySize.width, copySize.height, copySize.depth, format.format, format.type, reinterpret_cast<void*>( @@ -598,19 +607,21 @@ namespace dawn_native { namespace opengl { auto& copySize = copy->copySize; Texture* texture = ToBackend(src.texture.Get()); Buffer* buffer = ToBackend(dst.buffer.Get()); - const Format& format = texture->GetFormat(); - const GLFormat& glFormat = texture->GetGLFormat(); + const Format& formatInfo = texture->GetFormat(); + const GLFormat& format = texture->GetGLFormat(); GLenum target = texture->GetGLTarget(); // TODO(jiawei.shao@intel.com): support texture-to-buffer copy with compressed // texture formats. - if (format.isCompressed) { + if (formatInfo.isCompressed) { UNREACHABLE(); } + buffer->EnsureDataInitializedAsDestination(copy); + ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D); - SubresourceRange subresources = {src.mipLevel, 1, src.arrayLayer, - copy->copySize.depth}; + SubresourceRange subresources = + GetSubresourcesAffectedByCopy(src, copy->copySize); texture->EnsureSubresourceContentInitialized(subresources); // The only way to move data from a texture to a buffer in GL is via // glReadPixels with a pack buffer. Create a temporary FBO for the copy. @@ -620,29 +631,36 @@ namespace dawn_native { namespace opengl { gl.GenFramebuffers(1, &readFBO); gl.BindFramebuffer(GL_READ_FRAMEBUFFER, readFBO); - GLenum glAttachment = 0; - switch (format.aspect) { - case Format::Aspect::Color: + const TexelBlockInfo& blockInfo = formatInfo.GetTexelBlockInfo(src.aspect); + + gl.BindBuffer(GL_PIXEL_PACK_BUFFER, buffer->GetHandle()); + gl.PixelStorei(GL_PACK_IMAGE_HEIGHT, dst.rowsPerImage); + gl.PixelStorei(GL_PACK_ROW_LENGTH, dst.bytesPerRow / blockInfo.blockByteSize); + + GLenum glAttachment; + GLenum glFormat; + GLenum glType; + switch (src.aspect) { + case Aspect::Color: glAttachment = GL_COLOR_ATTACHMENT0; + glFormat = format.format; + glType = format.type; break; - case Format::Aspect::Depth: + case Aspect::Depth: glAttachment = GL_DEPTH_ATTACHMENT; + glFormat = GL_DEPTH_COMPONENT; + glType = GL_FLOAT; break; - case Format::Aspect::Stencil: + case Aspect::Stencil: glAttachment = GL_STENCIL_ATTACHMENT; - break; - case Format::Aspect::DepthStencil: - glAttachment = GL_DEPTH_STENCIL_ATTACHMENT; + glFormat = GL_STENCIL_INDEX; + glType = GL_UNSIGNED_BYTE; break; default: UNREACHABLE(); break; } - gl.BindBuffer(GL_PIXEL_PACK_BUFFER, buffer->GetHandle()); - gl.PixelStorei(GL_PACK_ROW_LENGTH, dst.bytesPerRow / format.blockByteSize); - gl.PixelStorei(GL_PACK_IMAGE_HEIGHT, dst.rowsPerImage); - uint8_t* offset = reinterpret_cast<uint8_t*>(static_cast<uintptr_t>(dst.offset)); switch (texture->GetDimension()) { @@ -651,8 +669,7 @@ namespace dawn_native { namespace opengl { gl.FramebufferTexture2D(GL_READ_FRAMEBUFFER, glAttachment, target, texture->GetHandle(), src.mipLevel); gl.ReadPixels(src.origin.x, src.origin.y, copySize.width, - copySize.height, glFormat.format, glFormat.type, - offset); + copySize.height, glFormat, glType, offset); break; } @@ -660,10 +677,9 @@ namespace dawn_native { namespace opengl { for (uint32_t layer = 0; layer < copySize.depth; ++layer) { gl.FramebufferTextureLayer(GL_READ_FRAMEBUFFER, glAttachment, texture->GetHandle(), src.mipLevel, - src.arrayLayer + layer); + src.origin.z + layer); gl.ReadPixels(src.origin.x, src.origin.y, copySize.width, - copySize.height, glFormat.format, glFormat.type, - offset); + copySize.height, glFormat, glType, offset); offset += bytesPerImage; } @@ -696,10 +712,9 @@ namespace dawn_native { namespace opengl { Extent3D copySize = ComputeTextureCopyExtent(dst, copy->copySize); Texture* srcTexture = ToBackend(src.texture.Get()); Texture* dstTexture = ToBackend(dst.texture.Get()); - SubresourceRange srcRange = {src.mipLevel, 1, src.arrayLayer, - copy->copySize.depth}; - SubresourceRange dstRange = {dst.mipLevel, 1, dst.arrayLayer, - copy->copySize.depth}; + + SubresourceRange srcRange = GetSubresourcesAffectedByCopy(src, copy->copySize); + SubresourceRange dstRange = GetSubresourcesAffectedByCopy(dst, copy->copySize); srcTexture->EnsureSubresourceContentInitialized(srcRange); if (IsCompleteSubresourceCopiedTo(dstTexture, copySize, dst.mipLevel)) { @@ -708,19 +723,33 @@ namespace dawn_native { namespace opengl { dstTexture->EnsureSubresourceContentInitialized(dstRange); } gl.CopyImageSubData(srcTexture->GetHandle(), srcTexture->GetGLTarget(), - src.mipLevel, src.origin.x, src.origin.y, src.arrayLayer, + src.mipLevel, src.origin.x, src.origin.y, src.origin.z, dstTexture->GetHandle(), dstTexture->GetGLTarget(), - dst.mipLevel, dst.origin.x, dst.origin.y, dst.arrayLayer, + dst.mipLevel, dst.origin.x, dst.origin.y, dst.origin.z, copySize.width, copySize.height, copy->copySize.depth); break; } + case Command::ResolveQuerySet: { + // TODO(hao.x.li@intel.com): Resolve non-precise occlusion query. + SkipCommand(&mCommands, type); + break; + } + + case Command::WriteTimestamp: { + // WriteTimestamp is not supported on OpenGL + UNREACHABLE(); + break; + } + default: { UNREACHABLE(); break; } } } + + return {}; } void CommandBuffer::ExecuteComputePass() { @@ -789,6 +818,12 @@ namespace dawn_native { namespace opengl { break; } + case Command::WriteTimestamp: { + // WriteTimestamp is not supported on OpenGL + UNREACHABLE(); + break; + } + default: { UNREACHABLE(); break; @@ -853,19 +888,14 @@ namespace dawn_native { namespace opengl { GLenum glAttachment = 0; // TODO(kainino@chromium.org): it may be valid to just always use // GL_DEPTH_STENCIL_ATTACHMENT here. - switch (format.aspect) { - case Format::Aspect::Depth: - glAttachment = GL_DEPTH_ATTACHMENT; - break; - case Format::Aspect::Stencil: - glAttachment = GL_STENCIL_ATTACHMENT; - break; - case Format::Aspect::DepthStencil: - glAttachment = GL_DEPTH_STENCIL_ATTACHMENT; - break; - default: - UNREACHABLE(); - break; + if (format.aspects == (Aspect::Depth | Aspect::Stencil)) { + glAttachment = GL_DEPTH_STENCIL_ATTACHMENT; + } else if (format.aspects == Aspect::Depth) { + glAttachment = GL_DEPTH_ATTACHMENT; + } else if (format.aspects == Aspect::Stencil) { + glAttachment = GL_STENCIL_ATTACHMENT; + } else { + UNREACHABLE(); } if (textureView->GetTexture()->GetArrayLayers() == 1) { @@ -897,13 +927,23 @@ namespace dawn_native { namespace opengl { auto* attachmentInfo = &renderPass->colorAttachments[i]; // Load op - color - // TODO(cwallez@chromium.org): Choose the clear function depending on the - // componentType: things work for now because the clear color is always a float, but - // when that's fixed will lose precision on integer formats when converting to - // float. if (attachmentInfo->loadOp == wgpu::LoadOp::Clear) { gl.ColorMaski(i, true, true, true, true); - gl.ClearBufferfv(GL_COLOR, i, &attachmentInfo->clearColor.r); + + const Format& attachmentFormat = attachmentInfo->view->GetFormat(); + if (attachmentFormat.HasComponentType(Format::Type::Float)) { + gl.ClearBufferfv(GL_COLOR, i, &attachmentInfo->clearColor.r); + } else if (attachmentFormat.HasComponentType(Format::Type::Uint)) { + const std::array<uint32_t, 4> appliedClearColor = + ConvertToUnsignedIntegerColor(attachmentInfo->clearColor); + gl.ClearBufferuiv(GL_COLOR, i, appliedClearColor.data()); + } else if (attachmentFormat.HasComponentType(Format::Type::Sint)) { + const std::array<int32_t, 4> appliedClearColor = + ConvertToSignedIntegerColor(attachmentInfo->clearColor); + gl.ClearBufferiv(GL_COLOR, i, appliedClearColor.data()); + } else { + UNREACHABLE(); + } } if (attachmentInfo->storeOp == wgpu::StoreOp::Clear) { @@ -1140,6 +1180,12 @@ namespace dawn_native { namespace opengl { break; } + case Command::WriteTimestamp: { + // WriteTimestamp is not supported on OpenGL + UNREACHABLE(); + break; + } + default: { DoRenderBundleCommand(&mCommands, type); break; diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/CommandBufferGL.h b/chromium/third_party/dawn/src/dawn_native/opengl/CommandBufferGL.h index 482f10d09d4..b860be8c845 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/CommandBufferGL.h +++ b/chromium/third_party/dawn/src/dawn_native/opengl/CommandBufferGL.h @@ -15,7 +15,6 @@ #ifndef DAWNNATIVE_OPENGL_COMMANDBUFFERGL_H_ #define DAWNNATIVE_OPENGL_COMMANDBUFFERGL_H_ -#include "dawn_native/CommandAllocator.h" #include "dawn_native/CommandBuffer.h" namespace dawn_native { @@ -30,14 +29,11 @@ namespace dawn_native { namespace opengl { public: CommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor); - void Execute(); + MaybeError Execute(); private: - ~CommandBuffer() override; void ExecuteComputePass(); void ExecuteRenderPass(BeginRenderPassCmd* renderPass); - - CommandIterator mCommands; }; }} // namespace dawn_native::opengl diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/DeviceGL.cpp b/chromium/third_party/dawn/src/dawn_native/opengl/DeviceGL.cpp index f47474f2aef..2f06d85da9b 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/DeviceGL.cpp +++ b/chromium/third_party/dawn/src/dawn_native/opengl/DeviceGL.cpp @@ -102,8 +102,8 @@ namespace dawn_native { namespace opengl { const BindGroupLayoutDescriptor* descriptor) { return new BindGroupLayout(this, descriptor); } - ResultOrError<BufferBase*> Device::CreateBufferImpl(const BufferDescriptor* descriptor) { - return new Buffer(this, descriptor); + ResultOrError<Ref<BufferBase>> Device::CreateBufferImpl(const BufferDescriptor* descriptor) { + return AcquireRef(new Buffer(this, descriptor)); } CommandBufferBase* Device::CreateCommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor) { diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/DeviceGL.h b/chromium/third_party/dawn/src/dawn_native/opengl/DeviceGL.h index 4a03f4aeaac..a297081d83c 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/DeviceGL.h +++ b/chromium/third_party/dawn/src/dawn_native/opengl/DeviceGL.h @@ -71,7 +71,8 @@ namespace dawn_native { namespace opengl { const BindGroupDescriptor* descriptor) override; ResultOrError<BindGroupLayoutBase*> CreateBindGroupLayoutImpl( const BindGroupLayoutDescriptor* descriptor) override; - ResultOrError<BufferBase*> CreateBufferImpl(const BufferDescriptor* descriptor) override; + ResultOrError<Ref<BufferBase>> CreateBufferImpl( + const BufferDescriptor* descriptor) override; ResultOrError<ComputePipelineBase*> CreateComputePipelineImpl( const ComputePipelineDescriptor* descriptor) override; ResultOrError<PipelineLayoutBase*> CreatePipelineLayoutImpl( diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/GLFormat.cpp b/chromium/third_party/dawn/src/dawn_native/opengl/GLFormat.cpp index 9e1a20c09bb..763f5554ddb 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/GLFormat.cpp +++ b/chromium/third_party/dawn/src/dawn_native/opengl/GLFormat.cpp @@ -74,7 +74,8 @@ namespace dawn_native { namespace opengl { // This doesn't have an enum for the internal format in OpenGL, so use RGBA8. AddFormat(wgpu::TextureFormat::BGRA8Unorm, GL_RGBA8, GL_BGRA, GL_UNSIGNED_BYTE, Type::Float); AddFormat(wgpu::TextureFormat::RGB10A2Unorm, GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, Type::Float); - AddFormat(wgpu::TextureFormat::RG11B10Float, GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV, Type::Float); + AddFormat(wgpu::TextureFormat::RG11B10Ufloat, GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV, Type::Float); + AddFormat(wgpu::TextureFormat::RGB9E5Ufloat, GL_RGB9_E5, GL_RGB, GL_UNSIGNED_INT_5_9_9_9_REV, Type::Float); // 8 bytes color formats AddFormat(wgpu::TextureFormat::RG32Uint, GL_RG32UI, GL_RG_INTEGER, GL_UNSIGNED_INT, Type::Uint); diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/PipelineGL.h b/chromium/third_party/dawn/src/dawn_native/opengl/PipelineGL.h index 7f681f94b8a..a79909ab55c 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/PipelineGL.h +++ b/chromium/third_party/dawn/src/dawn_native/opengl/PipelineGL.h @@ -36,10 +36,6 @@ namespace dawn_native { namespace opengl { const PipelineLayout* layout, const PerStage<const ShaderModule*>& modules); - using BindingLocations = ityp::array<BindGroupIndex, - ityp::array<BindingIndex, GLint, kMaxBindingsPerGroup>, - kMaxBindGroups>; - // For each unit a sampler is bound to we need to know if we should use filtering or not // because int and uint texture are only complete without filtering. struct SamplerUnit { diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/PipelineLayoutGL.cpp b/chromium/third_party/dawn/src/dawn_native/opengl/PipelineLayoutGL.cpp index 0e98c617610..8faab78b0d9 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/PipelineLayoutGL.cpp +++ b/chromium/third_party/dawn/src/dawn_native/opengl/PipelineLayoutGL.cpp @@ -30,6 +30,7 @@ namespace dawn_native { namespace opengl { for (BindGroupIndex group : IterateBitSet(GetBindGroupLayoutsMask())) { const BindGroupLayoutBase* bgl = GetBindGroupLayout(group); + mIndexInfo[group].resize(bgl->GetBindingCount()); for (BindingIndex bindingIndex{0}; bindingIndex < bgl->GetBindingCount(); ++bindingIndex) { diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/PipelineLayoutGL.h b/chromium/third_party/dawn/src/dawn_native/opengl/PipelineLayoutGL.h index 3d511d6d513..eeff7182dda 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/PipelineLayoutGL.h +++ b/chromium/third_party/dawn/src/dawn_native/opengl/PipelineLayoutGL.h @@ -18,6 +18,7 @@ #include "dawn_native/PipelineLayout.h" #include "common/ityp_array.h" +#include "common/ityp_vector.h" #include "dawn_native/BindingInfo.h" #include "dawn_native/opengl/opengl_platform.h" @@ -30,9 +31,7 @@ namespace dawn_native { namespace opengl { PipelineLayout(Device* device, const PipelineLayoutDescriptor* descriptor); using BindingIndexInfo = - ityp::array<BindGroupIndex, - ityp::array<BindingIndex, GLuint, kMaxBindingsPerGroup>, - kMaxBindGroups>; + ityp::array<BindGroupIndex, ityp::vector<BindingIndex, GLuint>, kMaxBindGroups>; const BindingIndexInfo& GetBindingIndexInfo() const; GLuint GetTextureUnitsUsed() const; diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/QueueGL.cpp b/chromium/third_party/dawn/src/dawn_native/opengl/QueueGL.cpp index a33cbd0da5f..340ec250c78 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/QueueGL.cpp +++ b/chromium/third_party/dawn/src/dawn_native/opengl/QueueGL.cpp @@ -30,7 +30,7 @@ namespace dawn_native { namespace opengl { TRACE_EVENT_BEGIN0(GetDevice()->GetPlatform(), Recording, "CommandBufferGL::Execute"); for (uint32_t i = 0; i < commandCount; ++i) { - ToBackend(commands[i])->Execute(); + DAWN_TRY(ToBackend(commands[i])->Execute()); } TRACE_EVENT_END0(GetDevice()->GetPlatform(), Recording, "CommandBufferGL::Execute"); @@ -44,6 +44,8 @@ namespace dawn_native { namespace opengl { size_t size) { const OpenGLFunctions& gl = ToBackend(GetDevice())->gl; + ToBackend(buffer)->EnsureDataInitializedAsDestination(bufferOffset, size); + gl.BindBuffer(GL_ARRAY_BUFFER, ToBackend(buffer)->GetHandle()); gl.BufferSubData(GL_ARRAY_BUFFER, bufferOffset, size, data); return {}; diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/RenderPipelineGL.cpp b/chromium/third_party/dawn/src/dawn_native/opengl/RenderPipelineGL.cpp index b16783ce904..7c70396f21f 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/RenderPipelineGL.cpp +++ b/chromium/third_party/dawn/src/dawn_native/opengl/RenderPipelineGL.cpp @@ -43,14 +43,15 @@ namespace dawn_native { namespace opengl { void ApplyFrontFaceAndCulling(const OpenGLFunctions& gl, wgpu::FrontFace face, wgpu::CullMode mode) { + // Note that we invert winding direction in OpenGL. Because Y axis is up in OpenGL, + // which is different from WebGPU and other backends (Y axis is down). + GLenum direction = (face == wgpu::FrontFace::CCW) ? GL_CW : GL_CCW; + gl.FrontFace(direction); + if (mode == wgpu::CullMode::None) { gl.Disable(GL_CULL_FACE); } else { gl.Enable(GL_CULL_FACE); - // Note that we invert winding direction in OpenGL. Because Y axis is up in OpenGL, - // which is different from WebGPU and other backends (Y axis is down). - GLenum direction = (face == wgpu::FrontFace::CCW) ? GL_CW : GL_CCW; - gl.FrontFace(direction); GLenum cullMode = (mode == wgpu::CullMode::Front) ? GL_FRONT : GL_BACK; gl.CullFace(cullMode); @@ -257,6 +258,13 @@ namespace dawn_native { namespace opengl { ApplyDepthStencilState(gl, GetDepthStencilStateDescriptor(), &persistentPipelineState); + gl.SampleMaski(0, GetSampleMask()); + if (IsAlphaToCoverageEnabled()) { + gl.Enable(GL_SAMPLE_ALPHA_TO_COVERAGE); + } else { + gl.Disable(GL_SAMPLE_ALPHA_TO_COVERAGE); + } + for (uint32_t attachmentSlot : IterateBitSet(GetColorAttachmentsMask())) { ApplyColorState(gl, attachmentSlot, GetColorStateDescriptor(attachmentSlot)); } diff --git a/chromium/third_party/dawn/src/dawn_native/opengl/TextureGL.cpp b/chromium/third_party/dawn/src/dawn_native/opengl/TextureGL.cpp index ee64bcda572..e0930d7c02a 100644 --- a/chromium/third_party/dawn/src/dawn_native/opengl/TextureGL.cpp +++ b/chromium/third_party/dawn/src/dawn_native/opengl/TextureGL.cpp @@ -17,6 +17,7 @@ #include "common/Assert.h" #include "common/Constants.h" #include "common/Math.h" +#include "dawn_native/EnumMaskIterator.h" #include "dawn_native/opengl/BufferGL.h" #include "dawn_native/opengl/DeviceGL.h" #include "dawn_native/opengl/UtilsGL.h" @@ -199,25 +200,25 @@ namespace dawn_native { namespace opengl { float fClearColor = (clearValue == TextureBase::ClearValue::Zero) ? 0.f : 1.f; if (GetFormat().isRenderable) { - if (GetFormat().HasDepthOrStencil()) { - bool doDepthClear = GetFormat().HasDepth(); - bool doStencilClear = GetFormat().HasStencil(); + if ((range.aspects & (Aspect::Depth | Aspect::Stencil)) != 0) { GLfloat depth = fClearColor; GLint stencil = clearColor; - if (doDepthClear) { + if (range.aspects & Aspect::Depth) { gl.DepthMask(GL_TRUE); } - if (doStencilClear) { + if (range.aspects & Aspect::Stencil) { gl.StencilMask(GetStencilMaskFromStencilFormat(GetFormat().format)); } - auto DoClear = [&]() { - if (doDepthClear && doStencilClear) { + auto DoClear = [&](Aspect aspects) { + if (aspects == (Aspect::Depth | Aspect::Stencil)) { gl.ClearBufferfi(GL_DEPTH_STENCIL, 0, depth, stencil); - } else if (doDepthClear) { + } else if (aspects == Aspect::Depth) { gl.ClearBufferfv(GL_DEPTH, 0, &depth); - } else if (doStencilClear) { + } else if (aspects == Aspect::Stencil) { gl.ClearBufferiv(GL_STENCIL, 0, &stencil); + } else { + UNREACHABLE(); } }; @@ -230,23 +231,42 @@ namespace dawn_native { namespace opengl { switch (GetDimension()) { case wgpu::TextureDimension::e2D: if (GetArrayLayers() == 1) { - if (clearValue == TextureBase::ClearValue::Zero && - IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, 0))) { - // Skip lazy clears if already initialized. + Aspect aspectsToClear = Aspect::None; + for (Aspect aspect : IterateEnumMask(range.aspects)) { + if (clearValue == TextureBase::ClearValue::Zero && + IsSubresourceContentInitialized( + SubresourceRange::SingleMipAndLayer(level, 0, + aspect))) { + // Skip lazy clears if already initialized. + continue; + } + aspectsToClear |= aspect; + } + + if (aspectsToClear == Aspect::None) { continue; } + gl.FramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GetGLTarget(), GetHandle(), static_cast<GLint>(level)); - DoClear(); + DoClear(aspectsToClear); } else { for (uint32_t layer = range.baseArrayLayer; layer < range.baseArrayLayer + range.layerCount; ++layer) { - if (clearValue == TextureBase::ClearValue::Zero && - IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, layer))) { - // Skip lazy clears if already initialized. + Aspect aspectsToClear = Aspect::None; + for (Aspect aspect : IterateEnumMask(range.aspects)) { + if (clearValue == TextureBase::ClearValue::Zero && + IsSubresourceContentInitialized( + SubresourceRange::SingleMipAndLayer(level, layer, + aspect))) { + // Skip lazy clears if already initialized. + continue; + } + aspectsToClear |= aspect; + } + + if (aspectsToClear == Aspect::None) { continue; } @@ -254,7 +274,7 @@ namespace dawn_native { namespace opengl { GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GetHandle(), static_cast<GLint>(level), static_cast<GLint>(layer)); - DoClear(); + DoClear(aspectsToClear); } } break; @@ -266,6 +286,8 @@ namespace dawn_native { namespace opengl { gl.DeleteFramebuffers(1, &framebuffer); } else { + ASSERT(range.aspects == Aspect::Color); + static constexpr uint32_t MAX_TEXEL_SIZE = 16; ASSERT(GetFormat().blockByteSize <= MAX_TEXEL_SIZE); std::array<GLbyte, MAX_TEXEL_SIZE> clearColorData; @@ -280,7 +302,7 @@ namespace dawn_native { namespace opengl { layer < range.baseArrayLayer + range.layerCount; ++layer) { if (clearValue == TextureBase::ClearValue::Zero && IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, layer))) { + SubresourceRange::SingleMipAndLayer(level, layer, Aspect::Color))) { // Skip lazy clears if already initialized. continue; } @@ -292,7 +314,8 @@ namespace dawn_native { namespace opengl { } } } else { - // TODO(natlee@microsoft.com): test compressed textures are cleared + ASSERT(range.aspects == Aspect::Color); + // create temp buffer with clear color to copy to the texture image ASSERT(kTextureBytesPerRowAlignment % GetFormat().blockByteSize == 0); uint32_t bytesPerRow = @@ -303,24 +326,19 @@ namespace dawn_native { namespace opengl { ASSERT(bytesPerRow % GetFormat().blockByteSize == 0); ASSERT(GetHeight() % GetFormat().blockHeight == 0); - dawn_native::BufferDescriptor descriptor; + dawn_native::BufferDescriptor descriptor = {}; + descriptor.mappedAtCreation = true; + descriptor.usage = wgpu::BufferUsage::CopySrc; descriptor.size = bytesPerRow * (GetHeight() / GetFormat().blockHeight); if (descriptor.size > std::numeric_limits<uint32_t>::max()) { return DAWN_OUT_OF_MEMORY_ERROR("Unable to allocate buffer."); } - descriptor.nextInChain = nullptr; - descriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::MapWrite; + // TODO(natlee@microsoft.com): use Dynamic Uplaoder here for temp buffer - Ref<Buffer> srcBuffer = ToBackend(device->CreateBuffer(&descriptor)); - // Call release here to prevent memory leak since CreateBuffer will up the ref count to - // 1, then assigning to Ref<Buffer> ups the ref count to 2. Release will reduce the ref - // count and ensure it to reach 0 when out of use. - srcBuffer->Release(); + Ref<Buffer> srcBuffer = AcquireRef(ToBackend(device->CreateBuffer(&descriptor))); // Fill the buffer with clear color - uint8_t* clearBuffer = nullptr; - DAWN_TRY(srcBuffer->MapAtCreation(&clearBuffer)); - memset(clearBuffer, clearColor, descriptor.size); + memset(srcBuffer->GetMappedRange(0, descriptor.size), clearColor, descriptor.size); srcBuffer->Unmap(); // Bind buffer and texture, and make the buffer to texture copy @@ -339,7 +357,7 @@ namespace dawn_native { namespace opengl { if (GetArrayLayers() == 1) { if (clearValue == TextureBase::ClearValue::Zero && IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, 0))) { + SubresourceRange::SingleMipAndLayer(level, 0, Aspect::Color))) { // Skip lazy clears if already initialized. continue; } @@ -351,7 +369,8 @@ namespace dawn_native { namespace opengl { layer < range.baseArrayLayer + range.layerCount; ++layer) { if (clearValue == TextureBase::ClearValue::Zero && IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, layer))) { + SubresourceRange::SingleMipAndLayer(level, layer, + Aspect::Color))) { // Skip lazy clears if already initialized. continue; } diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/BackendVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/BackendVk.cpp index 0a68231e7b2..613f0a01260 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/BackendVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/BackendVk.cpp @@ -231,7 +231,17 @@ namespace dawn_native { namespace vulkan { appInfo.applicationVersion = 0; appInfo.pEngineName = nullptr; appInfo.engineVersion = 0; - appInfo.apiVersion = mGlobalInfo.apiVersion; + // Vulkan 1.0 implementations were required to return VK_ERROR_INCOMPATIBLE_DRIVER if + // apiVersion was larger than 1.0. Meanwhile, as long as the instance supports at least + // Vulkan 1.1, an application can use different versions of Vulkan with an instance than + // it does with a device or physical device. So we should set apiVersion to Vulkan 1.0 + // if the instance only supports Vulkan 1.0. Otherwise we set apiVersion to Vulkan 1.2, + // treat 1.2 as the highest API version dawn targets. + if (mGlobalInfo.apiVersion == VK_MAKE_VERSION(1, 0, 0)) { + appInfo.apiVersion = mGlobalInfo.apiVersion; + } else { + appInfo.apiVersion = VK_MAKE_VERSION(1, 2, 0); + } VkInstanceCreateInfo createInfo; createInfo.sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO; diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/BindGroupLayoutVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/BindGroupLayoutVk.cpp index 1b325bee5b8..54c3dcfc40f 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/BindGroupLayoutVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/BindGroupLayoutVk.cpp @@ -15,6 +15,7 @@ #include "dawn_native/vulkan/BindGroupLayoutVk.h" #include "common/BitSetIterator.h" +#include "common/ityp_vector.h" #include "dawn_native/vulkan/BindGroupVk.h" #include "dawn_native/vulkan/DescriptorSetAllocator.h" #include "dawn_native/vulkan/DeviceVk.h" @@ -85,29 +86,30 @@ namespace dawn_native { namespace vulkan { // Compute the bindings that will be chained in the DescriptorSetLayout create info. We add // one entry per binding set. This might be optimized by computing continuous ranges of // bindings of the same type. - uint32_t numBindings = 0; - std::array<VkDescriptorSetLayoutBinding, kMaxBindingsPerGroup> bindings; + ityp::vector<BindingIndex, VkDescriptorSetLayoutBinding> bindings; + bindings.reserve(GetBindingCount()); + for (const auto& it : GetBindingMap()) { BindingNumber bindingNumber = it.first; BindingIndex bindingIndex = it.second; const BindingInfo& bindingInfo = GetBindingInfo(bindingIndex); - VkDescriptorSetLayoutBinding* vkBinding = &bindings[numBindings]; - vkBinding->binding = static_cast<uint32_t>(bindingNumber); - vkBinding->descriptorType = + VkDescriptorSetLayoutBinding vkBinding; + vkBinding.binding = static_cast<uint32_t>(bindingNumber); + vkBinding.descriptorType = VulkanDescriptorType(bindingInfo.type, bindingInfo.hasDynamicOffset); - vkBinding->descriptorCount = 1; - vkBinding->stageFlags = VulkanShaderStageFlags(bindingInfo.visibility); - vkBinding->pImmutableSamplers = nullptr; + vkBinding.descriptorCount = 1; + vkBinding.stageFlags = VulkanShaderStageFlags(bindingInfo.visibility); + vkBinding.pImmutableSamplers = nullptr; - numBindings++; + bindings.emplace_back(vkBinding); } VkDescriptorSetLayoutCreateInfo createInfo; createInfo.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO; createInfo.pNext = nullptr; createInfo.flags = 0; - createInfo.bindingCount = numBindings; + createInfo.bindingCount = static_cast<uint32_t>(bindings.size()); createInfo.pBindings = bindings.data(); Device* device = ToBackend(GetDevice()); diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/BindGroupVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/BindGroupVk.cpp index eb31182c0a4..1124e15d8fb 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/BindGroupVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/BindGroupVk.cpp @@ -15,6 +15,7 @@ #include "dawn_native/vulkan/BindGroupVk.h" #include "common/BitSetIterator.h" +#include "common/ityp_stack_vec.h" #include "dawn_native/vulkan/BindGroupLayoutVk.h" #include "dawn_native/vulkan/BufferVk.h" #include "dawn_native/vulkan/DeviceVk.h" @@ -38,11 +39,15 @@ namespace dawn_native { namespace vulkan { mDescriptorSetAllocation(descriptorSetAllocation) { // Now do a write of a single descriptor set with all possible chained data allocated on the // stack. - uint32_t numWrites = 0; - std::array<VkWriteDescriptorSet, kMaxBindingsPerGroup> writes; - std::array<VkDescriptorBufferInfo, kMaxBindingsPerGroup> writeBufferInfo; - std::array<VkDescriptorImageInfo, kMaxBindingsPerGroup> writeImageInfo; + const uint32_t bindingCount = static_cast<uint32_t>((GetLayout()->GetBindingCount())); + ityp::stack_vec<uint32_t, VkWriteDescriptorSet, kMaxOptimalBindingsPerGroup> writes( + bindingCount); + ityp::stack_vec<uint32_t, VkDescriptorBufferInfo, kMaxOptimalBindingsPerGroup> + writeBufferInfo(bindingCount); + ityp::stack_vec<uint32_t, VkDescriptorImageInfo, kMaxOptimalBindingsPerGroup> + writeImageInfo(bindingCount); + uint32_t numWrites = 0; for (const auto& it : GetLayout()->GetBindingMap()) { BindingNumber bindingNumber = it.first; BindingIndex bindingIndex = it.second; diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/BufferVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/BufferVk.cpp index 0b9385e9393..e011455348b 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/BufferVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/BufferVk.cpp @@ -14,6 +14,7 @@ #include "dawn_native/vulkan/BufferVk.h" +#include "dawn_native/CommandBuffer.h" #include "dawn_native/vulkan/DeviceVk.h" #include "dawn_native/vulkan/FencedDeleter.h" #include "dawn_native/vulkan/ResourceHeapVk.h" @@ -116,10 +117,10 @@ namespace dawn_native { namespace vulkan { } // namespace // static - ResultOrError<Buffer*> Buffer::Create(Device* device, const BufferDescriptor* descriptor) { + ResultOrError<Ref<Buffer>> Buffer::Create(Device* device, const BufferDescriptor* descriptor) { Ref<Buffer> buffer = AcquireRef(new Buffer(device, descriptor)); DAWN_TRY(buffer->Initialize()); - return buffer.Detach(); + return std::move(buffer); } MaybeError Buffer::Initialize() { @@ -166,7 +167,7 @@ namespace dawn_native { namespace vulkan { "vkBindBufferMemory")); if (device->IsToggleEnabled(Toggle::NonzeroClearResourcesOnCreationForTesting)) { - ClearBuffer(device->GetPendingRecordingContext(), ClearValue::NonZero); + ClearBuffer(device->GetPendingRecordingContext(), 0x01010101); } return {}; @@ -176,14 +177,6 @@ namespace dawn_native { namespace vulkan { DestroyInternal(); } - void Buffer::OnMapReadCommandSerialFinished(uint32_t mapSerial, const void* data) { - CallMapReadCallback(mapSerial, WGPUBufferMapAsyncStatus_Success, data, GetSize()); - } - - void Buffer::OnMapWriteCommandSerialFinished(uint32_t mapSerial, void* data) { - CallMapWriteCallback(mapSerial, WGPUBufferMapAsyncStatus_Success, data, GetSize()); - } - VkBuffer Buffer::GetHandle() const { return mHandle; } @@ -242,17 +235,22 @@ namespace dawn_native { namespace vulkan { mLastUsage = usage; } - bool Buffer::IsMapWritable() const { + bool Buffer::IsMappableAtCreation() const { // TODO(enga): Handle CPU-visible memory on UMA return mMemoryAllocation.GetMappedPointer() != nullptr; } - MaybeError Buffer::MapAtCreationImpl(uint8_t** mappedPointer) { - *mappedPointer = mMemoryAllocation.GetMappedPointer(); + MaybeError Buffer::MapAtCreationImpl() { + CommandRecordingContext* recordingContext = + ToBackend(GetDevice())->GetPendingRecordingContext(); + + // TODO(jiawei.shao@intel.com): initialize mapped buffer in CPU side. + EnsureDataInitialized(recordingContext); + return {}; } - MaybeError Buffer::MapReadAsyncImpl(uint32_t serial) { + MaybeError Buffer::MapReadAsyncImpl() { Device* device = ToBackend(GetDevice()); CommandRecordingContext* recordingContext = device->GetPendingRecordingContext(); @@ -260,7 +258,7 @@ namespace dawn_native { namespace vulkan { return {}; } - MaybeError Buffer::MapWriteAsyncImpl(uint32_t serial) { + MaybeError Buffer::MapWriteAsyncImpl() { Device* device = ToBackend(GetDevice()); CommandRecordingContext* recordingContext = device->GetPendingRecordingContext(); @@ -268,6 +266,23 @@ namespace dawn_native { namespace vulkan { return {}; } + MaybeError Buffer::MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) { + Device* device = ToBackend(GetDevice()); + + CommandRecordingContext* recordingContext = device->GetPendingRecordingContext(); + + // TODO(jiawei.shao@intel.com): initialize mapped buffer in CPU side. + EnsureDataInitialized(recordingContext); + + if (mode & wgpu::MapMode::Read) { + TransitionUsageNow(recordingContext, wgpu::BufferUsage::MapRead); + } else { + ASSERT(mode & wgpu::MapMode::Write); + TransitionUsageNow(recordingContext, wgpu::BufferUsage::MapWrite); + } + return {}; + } + void Buffer::UnmapImpl() { // No need to do anything, we keep CPU-visible memory mapped at all time. } @@ -287,13 +302,61 @@ namespace dawn_native { namespace vulkan { } } - void Buffer::ClearBuffer(CommandRecordingContext* recordingContext, ClearValue clearValue) { - ASSERT(recordingContext != nullptr); + void Buffer::EnsureDataInitialized(CommandRecordingContext* recordingContext) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return; + } - // TODO(jiawei.shao@intel.com): support buffer lazy-initialization to 0. - ASSERT(clearValue == BufferBase::ClearValue::NonZero); + InitializeToZero(recordingContext); + } - constexpr uint32_t kClearBufferValue = 0x01010101; + void Buffer::EnsureDataInitializedAsDestination(CommandRecordingContext* recordingContext, + uint64_t offset, + uint64_t size) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return; + } + + if (IsFullBufferRange(offset, size)) { + SetIsDataInitialized(); + } else { + InitializeToZero(recordingContext); + } + } + + void Buffer::EnsureDataInitializedAsDestination(CommandRecordingContext* recordingContext, + const CopyTextureToBufferCmd* copy) { + // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse + // instead when buffer lazy initialization is completely supported. + if (IsDataInitialized() || + !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) { + return; + } + + if (IsFullBufferOverwrittenInTextureToBufferCopy(copy)) { + SetIsDataInitialized(); + } else { + InitializeToZero(recordingContext); + } + } + + void Buffer::InitializeToZero(CommandRecordingContext* recordingContext) { + ASSERT(GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)); + ASSERT(!IsDataInitialized()); + + ClearBuffer(recordingContext, 0u); + GetDevice()->IncrementLazyClearCountForTesting(); + SetIsDataInitialized(); + } + + void Buffer::ClearBuffer(CommandRecordingContext* recordingContext, uint32_t clearValue) { + ASSERT(recordingContext != nullptr); TransitionUsageNow(recordingContext, wgpu::BufferUsage::CopyDst); @@ -301,6 +364,6 @@ namespace dawn_native { namespace vulkan { // TODO(jiawei.shao@intel.com): find out why VK_WHOLE_SIZE doesn't work on old Windows Intel // Vulkan drivers. device->fn.CmdFillBuffer(recordingContext->commandBuffer, mHandle, 0, GetSize(), - kClearBufferValue); + clearValue); } }} // namespace dawn_native::vulkan diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/BufferVk.h b/chromium/third_party/dawn/src/dawn_native/vulkan/BufferVk.h index 1c04870376a..14495d19d46 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/BufferVk.h +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/BufferVk.h @@ -28,10 +28,8 @@ namespace dawn_native { namespace vulkan { class Buffer final : public BufferBase { public: - static ResultOrError<Buffer*> Create(Device* device, const BufferDescriptor* descriptor); - - void OnMapReadCommandSerialFinished(uint32_t mapSerial, const void* data); - void OnMapWriteCommandSerialFinished(uint32_t mapSerial, void* data); + static ResultOrError<Ref<Buffer>> Create(Device* device, + const BufferDescriptor* descriptor); VkBuffer GetHandle() const; @@ -45,20 +43,29 @@ namespace dawn_native { namespace vulkan { VkPipelineStageFlags* srcStages, VkPipelineStageFlags* dstStages); + void EnsureDataInitialized(CommandRecordingContext* recordingContext); + void EnsureDataInitializedAsDestination(CommandRecordingContext* recordingContext, + uint64_t offset, + uint64_t size); + void EnsureDataInitializedAsDestination(CommandRecordingContext* recordingContext, + const CopyTextureToBufferCmd* copy); + private: ~Buffer() override; using BufferBase::BufferBase; MaybeError Initialize(); - void ClearBuffer(CommandRecordingContext* recordingContext, ClearValue clearValue); + void InitializeToZero(CommandRecordingContext* recordingContext); + void ClearBuffer(CommandRecordingContext* recordingContext, uint32_t clearValue); // Dawn API - MaybeError MapReadAsyncImpl(uint32_t serial) override; - MaybeError MapWriteAsyncImpl(uint32_t serial) override; + MaybeError MapReadAsyncImpl() override; + MaybeError MapWriteAsyncImpl() override; + MaybeError MapAsyncImpl(wgpu::MapMode mode, size_t offset, size_t size) override; void UnmapImpl() override; void DestroyImpl() override; - bool IsMapWritable() const override; - MaybeError MapAtCreationImpl(uint8_t** mappedPointer) override; + bool IsMappableAtCreation() const override; + MaybeError MapAtCreationImpl() override; void* GetMappedPointerImpl() override; VkBuffer mHandle = VK_NULL_HANDLE; diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/CommandBufferVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/CommandBufferVk.cpp index 575f66f8704..ce86612b7e9 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/CommandBufferVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/CommandBufferVk.cpp @@ -68,23 +68,23 @@ namespace dawn_native { namespace vulkan { // TODO(jiawei.shao@intel.com): support 1D and 3D textures ASSERT(srcTexture->GetDimension() == wgpu::TextureDimension::e2D && dstTexture->GetDimension() == wgpu::TextureDimension::e2D); - region.srcSubresource.aspectMask = srcTexture->GetVkAspectMask(); + region.srcSubresource.aspectMask = VulkanAspectMask(srcCopy.aspect); region.srcSubresource.mipLevel = srcCopy.mipLevel; - region.srcSubresource.baseArrayLayer = srcCopy.arrayLayer; + region.srcSubresource.baseArrayLayer = srcCopy.origin.z; region.srcSubresource.layerCount = copySize.depth; region.srcOffset.x = srcCopy.origin.x; region.srcOffset.y = srcCopy.origin.y; - region.srcOffset.z = srcCopy.origin.z; + region.srcOffset.z = 0; - region.dstSubresource.aspectMask = dstTexture->GetVkAspectMask(); + region.dstSubresource.aspectMask = VulkanAspectMask(dstCopy.aspect); region.dstSubresource.mipLevel = dstCopy.mipLevel; - region.dstSubresource.baseArrayLayer = dstCopy.arrayLayer; + region.dstSubresource.baseArrayLayer = dstCopy.origin.z; region.dstSubresource.layerCount = copySize.depth; region.dstOffset.x = dstCopy.origin.x; region.dstOffset.y = dstCopy.origin.y; - region.dstOffset.z = dstCopy.origin.z; + region.dstOffset.z = 0; ASSERT(HasSameTextureCopyExtent(srcCopy, dstCopy, copySize)); Extent3D imageExtent = ComputeTextureCopyExtent(dstCopy, copySize); @@ -104,7 +104,7 @@ namespace dawn_native { namespace vulkan { const ityp::array<BindGroupIndex, BindGroupBase*, kMaxBindGroups>& bindGroups, const ityp::array<BindGroupIndex, uint32_t, kMaxBindGroups>& dynamicOffsetCounts, const ityp::array<BindGroupIndex, - std::array<uint32_t, kMaxBindingsPerGroup>, + std::array<uint32_t, kMaxDynamicBuffersPerPipelineLayout>, kMaxBindGroups>& dynamicOffsets) { for (BindGroupIndex dirtyIndex : IterateBitSet(bindGroupsToApply)) { VkDescriptorSet set = ToBackend(bindGroups[dirtyIndex])->GetHandle(); @@ -234,10 +234,27 @@ namespace dawn_native { namespace vulkan { attachments[attachmentCount] = view->GetHandle(); - clearValues[attachmentCount].color.float32[0] = attachmentInfo.clearColor.r; - clearValues[attachmentCount].color.float32[1] = attachmentInfo.clearColor.g; - clearValues[attachmentCount].color.float32[2] = attachmentInfo.clearColor.b; - clearValues[attachmentCount].color.float32[3] = attachmentInfo.clearColor.a; + const Format& attachmentFormat = view->GetFormat(); + if (attachmentFormat.HasComponentType(Format::Type::Float)) { + clearValues[attachmentCount].color.float32[0] = attachmentInfo.clearColor.r; + clearValues[attachmentCount].color.float32[1] = attachmentInfo.clearColor.g; + clearValues[attachmentCount].color.float32[2] = attachmentInfo.clearColor.b; + clearValues[attachmentCount].color.float32[3] = attachmentInfo.clearColor.a; + } else if (attachmentFormat.HasComponentType(Format::Type::Uint)) { + const std::array<uint32_t, 4> appliedClearColor = + ConvertToUnsignedIntegerColor(attachmentInfo.clearColor); + for (uint32_t i = 0; i < 4; ++i) { + clearValues[attachmentCount].color.uint32[i] = appliedClearColor[i]; + } + } else if (attachmentFormat.HasComponentType(Format::Type::Sint)) { + const std::array<int32_t, 4> appliedClearColor = + ConvertToSignedIntegerColor(attachmentInfo.clearColor); + for (uint32_t i = 0; i < 4; ++i) { + clearValues[attachmentCount].color.int32[i] = appliedClearColor[i]; + } + } else { + UNREACHABLE(); + } attachmentCount++; } @@ -313,11 +330,7 @@ namespace dawn_native { namespace vulkan { } CommandBuffer::CommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor) - : CommandBufferBase(encoder, descriptor), mCommands(encoder->AcquireCommands()) { - } - - CommandBuffer::~CommandBuffer() { - FreeCommands(&mCommands); + : CommandBufferBase(encoder, descriptor) { } void CommandBuffer::RecordCopyImageWithTemporaryBuffer( @@ -387,6 +400,7 @@ namespace dawn_native { namespace vulkan { for (size_t i = 0; i < usages.buffers.size(); ++i) { Buffer* buffer = ToBackend(usages.buffers[i]); + buffer->EnsureDataInitialized(recordingContext); buffer->TransitionUsageNow(recordingContext, usages.bufferUsages[i], &bufferBarriers, &srcStages, &dstStages); } @@ -424,6 +438,10 @@ namespace dawn_native { namespace vulkan { Buffer* srcBuffer = ToBackend(copy->source.Get()); Buffer* dstBuffer = ToBackend(copy->destination.Get()); + srcBuffer->EnsureDataInitialized(recordingContext); + dstBuffer->EnsureDataInitializedAsDestination( + recordingContext, copy->destinationOffset, copy->size); + srcBuffer->TransitionUsageNow(recordingContext, wgpu::BufferUsage::CopySrc); dstBuffer->TransitionUsageNow(recordingContext, wgpu::BufferUsage::CopyDst); @@ -443,13 +461,16 @@ namespace dawn_native { namespace vulkan { auto& src = copy->source; auto& dst = copy->destination; + ToBackend(src.buffer)->EnsureDataInitialized(recordingContext); + VkBufferImageCopy region = ComputeBufferImageCopyRegion(src, dst, copy->copySize); VkImageSubresourceLayers subresource = region.imageSubresource; ASSERT(dst.texture->GetDimension() == wgpu::TextureDimension::e2D); - SubresourceRange range = {subresource.mipLevel, 1, subresource.baseArrayLayer, - subresource.layerCount}; + SubresourceRange range = + GetSubresourcesAffectedByCopy(copy->destination, copy->copySize); + if (IsCompleteSubresourceCopiedTo(dst.texture.Get(), copy->copySize, subresource.mipLevel)) { // Since texture has been overwritten, it has been "initialized" @@ -478,14 +499,16 @@ namespace dawn_native { namespace vulkan { auto& src = copy->source; auto& dst = copy->destination; + ToBackend(dst.buffer) + ->EnsureDataInitializedAsDestination(recordingContext, copy); + VkBufferImageCopy region = ComputeBufferImageCopyRegion(dst, src, copy->copySize); - VkImageSubresourceLayers subresource = region.imageSubresource; ASSERT(src.texture->GetDimension() == wgpu::TextureDimension::e2D); - const SubresourceRange range = {subresource.mipLevel, 1, - subresource.baseArrayLayer, - subresource.layerCount}; + SubresourceRange range = + GetSubresourcesAffectedByCopy(copy->source, copy->copySize); + ToBackend(src.texture) ->EnsureSubresourceContentInitialized(recordingContext, range); @@ -507,10 +530,8 @@ namespace dawn_native { namespace vulkan { mCommands.NextCommand<CopyTextureToTextureCmd>(); TextureCopy& src = copy->source; TextureCopy& dst = copy->destination; - SubresourceRange srcRange = {src.mipLevel, 1, src.arrayLayer, - copy->copySize.depth}; - SubresourceRange dstRange = {dst.mipLevel, 1, dst.arrayLayer, - copy->copySize.depth}; + SubresourceRange srcRange = GetSubresourcesAffectedByCopy(src, copy->copySize); + SubresourceRange dstRange = GetSubresourcesAffectedByCopy(dst, copy->copySize); ToBackend(src.texture) ->EnsureSubresourceContentInitialized(recordingContext, srcRange); @@ -528,10 +549,11 @@ namespace dawn_native { namespace vulkan { // subresources should all be GENERAL instead of what we set now. Currently // it is not allowed to copy with overlapped subresources, but we still // add the ASSERT here as a reminder for this possible misuse. - ASSERT(!IsRangeOverlapped(src.arrayLayer, dst.arrayLayer, - copy->copySize.depth)); + ASSERT( + !IsRangeOverlapped(src.origin.z, dst.origin.z, copy->copySize.depth)); } + // TODO after Yunchao's CL ToBackend(src.texture) ->TransitionUsageNow(recordingContext, wgpu::TextureUsage::CopySrc, srcRange); @@ -590,12 +612,20 @@ namespace dawn_native { namespace vulkan { mCommands.NextCommand<BeginComputePassCmd>(); TransitionForPass(device, recordingContext, passResourceUsages[nextPassNumber]); - RecordComputePass(recordingContext); + DAWN_TRY(RecordComputePass(recordingContext)); nextPassNumber++; break; } + case Command::ResolveQuerySet: { + return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation."); + } + + case Command::WriteTimestamp: { + return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation."); + } + default: { UNREACHABLE(); break; @@ -606,7 +636,7 @@ namespace dawn_native { namespace vulkan { return {}; } - void CommandBuffer::RecordComputePass(CommandRecordingContext* recordingContext) { + MaybeError CommandBuffer::RecordComputePass(CommandRecordingContext* recordingContext) { Device* device = ToBackend(GetDevice()); VkCommandBuffer commands = recordingContext->commandBuffer; @@ -617,7 +647,7 @@ namespace dawn_native { namespace vulkan { switch (type) { case Command::EndComputePass: { mCommands.NextCommand<EndComputePassCmd>(); - return; + return {}; } case Command::Dispatch: { @@ -713,6 +743,10 @@ namespace dawn_native { namespace vulkan { break; } + case Command::WriteTimestamp: { + return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation."); + } + default: { UNREACHABLE(); break; @@ -982,6 +1016,10 @@ namespace dawn_native { namespace vulkan { break; } + case Command::WriteTimestamp: { + return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation."); + } + default: { EncodeRenderBundleCommand(&mCommands, type); break; diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/CommandBufferVk.h b/chromium/third_party/dawn/src/dawn_native/vulkan/CommandBufferVk.h index c60fa9278f2..c5476d38ad0 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/CommandBufferVk.h +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/CommandBufferVk.h @@ -15,7 +15,6 @@ #ifndef DAWNNATIVE_VULKAN_COMMANDBUFFERVK_H_ #define DAWNNATIVE_VULKAN_COMMANDBUFFERVK_H_ -#include "dawn_native/CommandAllocator.h" #include "dawn_native/CommandBuffer.h" #include "dawn_native/Error.h" @@ -40,17 +39,14 @@ namespace dawn_native { namespace vulkan { private: CommandBuffer(CommandEncoder* encoder, const CommandBufferDescriptor* descriptor); - ~CommandBuffer() override; - void RecordComputePass(CommandRecordingContext* recordingContext); + MaybeError RecordComputePass(CommandRecordingContext* recordingContext); MaybeError RecordRenderPass(CommandRecordingContext* recordingContext, BeginRenderPassCmd* renderPass); void RecordCopyImageWithTemporaryBuffer(CommandRecordingContext* recordingContext, const TextureCopy& srcCopy, const TextureCopy& dstCopy, const Extent3D& copySize); - - CommandIterator mCommands; }; }} // namespace dawn_native::vulkan diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/DescriptorSetAllocator.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/DescriptorSetAllocator.cpp index 9f1999c3a5d..a7b794ff424 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/DescriptorSetAllocator.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/DescriptorSetAllocator.cpp @@ -38,7 +38,6 @@ namespace dawn_native { namespace vulkan { totalDescriptorCount += it.second; mPoolSizes.push_back(VkDescriptorPoolSize{it.first, it.second}); } - ASSERT(totalDescriptorCount <= kMaxBindingsPerGroup); if (totalDescriptorCount == 0) { // Vulkan requires that valid usage of vkCreateDescriptorPool must have a non-zero @@ -49,6 +48,9 @@ namespace dawn_native { namespace vulkan { mPoolSizes.push_back(VkDescriptorPoolSize{VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, 1}); mMaxSets = kMaxDescriptorsPerPool; } else { + ASSERT(totalDescriptorCount <= kMaxBindingsPerPipelineLayout); + static_assert(kMaxBindingsPerPipelineLayout <= kMaxDescriptorsPerPool, ""); + // Compute the total number of descriptors sets that fits given the max. mMaxSets = kMaxDescriptorsPerPool / totalDescriptorCount; ASSERT(mMaxSets > 0); diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/DeviceVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/DeviceVk.cpp index 921dd7341c3..01ba55ce461 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/DeviceVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/DeviceVk.cpp @@ -16,7 +16,6 @@ #include "common/Platform.h" #include "dawn_native/BackendConnection.h" -#include "dawn_native/Commands.h" #include "dawn_native/Error.h" #include "dawn_native/ErrorData.h" #include "dawn_native/VulkanBackend.h" @@ -29,6 +28,7 @@ #include "dawn_native/vulkan/ComputePipelineVk.h" #include "dawn_native/vulkan/FencedDeleter.h" #include "dawn_native/vulkan/PipelineLayoutVk.h" +#include "dawn_native/vulkan/QuerySetVk.h" #include "dawn_native/vulkan/QueueVk.h" #include "dawn_native/vulkan/RenderPassCache.h" #include "dawn_native/vulkan/RenderPipelineVk.h" @@ -110,7 +110,7 @@ namespace dawn_native { namespace vulkan { const BindGroupLayoutDescriptor* descriptor) { return BindGroupLayout::Create(this, descriptor); } - ResultOrError<BufferBase*> Device::CreateBufferImpl(const BufferDescriptor* descriptor) { + ResultOrError<Ref<BufferBase>> Device::CreateBufferImpl(const BufferDescriptor* descriptor) { return Buffer::Create(this, descriptor); } CommandBufferBase* Device::CreateCommandBuffer(CommandEncoder* encoder, @@ -126,7 +126,7 @@ namespace dawn_native { namespace vulkan { return PipelineLayout::Create(this, descriptor); } ResultOrError<QuerySetBase*> Device::CreateQuerySetImpl(const QuerySetDescriptor* descriptor) { - return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation"); + return QuerySet::Create(this, descriptor); } ResultOrError<RenderPipelineBase*> Device::CreateRenderPipelineImpl( const RenderPipelineDescriptor* descriptor) { @@ -299,6 +299,10 @@ namespace dawn_native { namespace vulkan { // Always require fragmentStoresAndAtomics because it is required by end2end tests. usedKnobs.features.fragmentStoresAndAtomics = VK_TRUE; + if (IsRobustnessEnabled()) { + usedKnobs.features.robustBufferAccess = VK_TRUE; + } + if (mDeviceInfo.HasExt(DeviceExt::SubgroupSizeControl)) { ASSERT(usedKnobs.HasExt(DeviceExt::SubgroupSizeControl)); @@ -315,6 +319,12 @@ namespace dawn_native { namespace vulkan { usedKnobs.features.textureCompressionBC = VK_TRUE; } + if (IsExtensionEnabled(Extension::PipelineStatisticsQuery)) { + ASSERT(ToBackend(GetAdapter())->GetDeviceInfo().features.pipelineStatisticsQuery == + VK_TRUE); + usedKnobs.features.pipelineStatisticsQuery = VK_TRUE; + } + if (IsExtensionEnabled(Extension::ShaderFloat16)) { const VulkanDeviceInfo& deviceInfo = ToBackend(GetAdapter())->GetDeviceInfo(); ASSERT(deviceInfo.HasExt(DeviceExt::ShaderFloat16Int8) && @@ -586,16 +596,17 @@ namespace dawn_native { namespace vulkan { // calling this function. ASSERT(size != 0); - // Insert memory barrier to ensure host write operations are made visible before - // copying from the staging buffer. However, this barrier can be removed (see note below). - // - // Note: Depending on the spec understanding, an explicit barrier may not be required when - // used with HOST_COHERENT as vkQueueSubmit does an implicit barrier between host and - // device. See "Availability, Visibility, and Domain Operations" in Vulkan spec for details. + CommandRecordingContext* recordingContext = GetPendingRecordingContext(); + + ToBackend(destination) + ->EnsureDataInitializedAsDestination(recordingContext, destinationOffset, size); + + // There is no need of a barrier to make host writes available and visible to the copy + // operation for HOST_COHERENT memory. The Vulkan spec for vkQueueSubmit describes that it + // does an implicit availability, visibility and domain operation. // Insert pipeline barrier to ensure correct ordering with previous memory operations on the // buffer. - CommandRecordingContext* recordingContext = GetPendingRecordingContext(); ToBackend(destination)->TransitionUsageNow(recordingContext, wgpu::BufferUsage::CopyDst); VkBufferCopy copy; @@ -610,6 +621,43 @@ namespace dawn_native { namespace vulkan { return {}; } + MaybeError Device::CopyFromStagingToTexture(StagingBufferBase* source, + const TextureDataLayout& src, + TextureCopy* dst, + const Extent3D& copySizePixels) { + // There is no need of a barrier to make host writes available and visible to the copy + // operation for HOST_COHERENT memory. The Vulkan spec for vkQueueSubmit describes that it + // does an implicit availability, visibility and domain operation. + + CommandRecordingContext* recordingContext = GetPendingRecordingContext(); + + VkBufferImageCopy region = ComputeBufferImageCopyRegion(src, *dst, copySizePixels); + VkImageSubresourceLayers subresource = region.imageSubresource; + + ASSERT(dst->texture->GetDimension() == wgpu::TextureDimension::e2D); + SubresourceRange range = GetSubresourcesAffectedByCopy(*dst, copySizePixels); + + if (IsCompleteSubresourceCopiedTo(dst->texture.Get(), copySizePixels, + subresource.mipLevel)) { + // Since texture has been overwritten, it has been "initialized" + dst->texture->SetIsSubresourceContentInitialized(true, range); + } else { + ToBackend(dst->texture)->EnsureSubresourceContentInitialized(recordingContext, range); + } + // Insert pipeline barrier to ensure correct ordering with previous memory operations on the + // texture. + ToBackend(dst->texture) + ->TransitionUsageNow(recordingContext, wgpu::TextureUsage::CopyDst, range); + VkImage dstImage = ToBackend(dst->texture)->GetHandle(); + + // Dawn guarantees dstImage be in the TRANSFER_DST_OPTIMAL layout after the + // copy command. + this->fn.CmdCopyBufferToImage(recordingContext->commandBuffer, + ToBackend(source)->GetBufferHandle(), dstImage, + VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, ®ion); + return {}; + } + MaybeError Device::ImportExternalImage(const ExternalImageDescriptor* descriptor, ExternalMemoryHandle memoryHandle, VkImage image, @@ -620,12 +668,6 @@ namespace dawn_native { namespace vulkan { const TextureDescriptor* textureDescriptor = reinterpret_cast<const TextureDescriptor*>(descriptor->cTextureDescriptor); - // TODO(dawn:22): Remove once migration from GPUTextureDescriptor.arrayLayerCount to - // GPUTextureDescriptor.size.depth is done. - TextureDescriptor fixedDescriptor; - DAWN_TRY_ASSIGN(fixedDescriptor, FixTextureDescriptor(this, textureDescriptor)); - textureDescriptor = &fixedDescriptor; - // Check services support this combination of handle type / image info if (!mExternalSemaphoreService->Supported()) { return DAWN_VALIDATION_ERROR("External semaphore usage not supported"); @@ -812,7 +854,15 @@ namespace dawn_native { namespace vulkan { // Immediately tag the recording context as unused so we don't try to submit it in Tick. mRecordingContext.used = false; - fn.DestroyCommandPool(mVkDevice, mRecordingContext.commandPool, nullptr); + if (mRecordingContext.commandPool != VK_NULL_HANDLE) { + // The VkCommandBuffer memory should be wholly owned by the pool and freed when it is + // destroyed, but that's not the case in some drivers and the leak memory. + // So we call FreeCommandBuffers before DestroyCommandPool to be safe. + // TODO(enga): Only do this on a known list of bad drivers. + fn.FreeCommandBuffers(mVkDevice, mRecordingContext.commandPool, 1, + &mRecordingContext.commandBuffer); + fn.DestroyCommandPool(mVkDevice, mRecordingContext.commandPool, nullptr); + } for (VkSemaphore semaphore : mRecordingContext.waitSemaphores) { fn.DestroySemaphore(mVkDevice, semaphore, nullptr); @@ -826,6 +876,11 @@ namespace dawn_native { namespace vulkan { ASSERT(mCommandsInFlight.Empty()); for (const CommandPoolAndBuffer& commands : mUnusedCommands) { + // The VkCommandBuffer memory should be wholly owned by the pool and freed when it is + // destroyed, but that's not the case in some drivers and the leak memory. + // So we call FreeCommandBuffers before DestroyCommandPool to be safe. + // TODO(enga): Only do this on a known list of bad drivers. + fn.FreeCommandBuffers(mVkDevice, commands.pool, 1, &commands.commandBuffer); fn.DestroyCommandPool(mVkDevice, commands.pool, nullptr); } mUnusedCommands.clear(); @@ -837,8 +892,12 @@ namespace dawn_native { namespace vulkan { // Releasing the uploader enqueues buffers to be released. // Call Tick() again to clear them before releasing the deleter. + mResourceMemoryAllocator->Tick(GetCompletedCommandSerial()); mDeleter->Tick(GetCompletedCommandSerial()); + // Allow recycled memory to be deleted. + mResourceMemoryAllocator->DestroyPool(); + // The VkRenderPasses in the cache can be destroyed immediately since all commands referring // to them are guaranteed to be finished executing. mRenderPassCache = nullptr; diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/DeviceVk.h b/chromium/third_party/dawn/src/dawn_native/vulkan/DeviceVk.h index 9e9ded9727c..89ec1ea8e8d 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/DeviceVk.h +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/DeviceVk.h @@ -19,6 +19,7 @@ #include "common/Serial.h" #include "common/SerialQueue.h" +#include "dawn_native/Commands.h" #include "dawn_native/Device.h" #include "dawn_native/vulkan/CommandRecordingContext.h" #include "dawn_native/vulkan/Forward.h" @@ -87,6 +88,10 @@ namespace dawn_native { namespace vulkan { BufferBase* destination, uint64_t destinationOffset, uint64_t size) override; + MaybeError CopyFromStagingToTexture(StagingBufferBase* source, + const TextureDataLayout& src, + TextureCopy* dst, + const Extent3D& copySizePixels); ResultOrError<ResourceMemoryAllocation> AllocateMemory(VkMemoryRequirements requirements, bool mappable); @@ -107,7 +112,8 @@ namespace dawn_native { namespace vulkan { const BindGroupDescriptor* descriptor) override; ResultOrError<BindGroupLayoutBase*> CreateBindGroupLayoutImpl( const BindGroupLayoutDescriptor* descriptor) override; - ResultOrError<BufferBase*> CreateBufferImpl(const BufferDescriptor* descriptor) override; + ResultOrError<Ref<BufferBase>> CreateBufferImpl( + const BufferDescriptor* descriptor) override; ResultOrError<ComputePipelineBase*> CreateComputePipelineImpl( const ComputePipelineDescriptor* descriptor) override; ResultOrError<PipelineLayoutBase*> CreatePipelineLayoutImpl( @@ -194,4 +200,4 @@ namespace dawn_native { namespace vulkan { }} // namespace dawn_native::vulkan -#endif // DAWNNATIVE_VULKAN_DEVICEVK_H_ +#endif // DAWNNATIVE_VULKAN_DEVICEVK_H_
\ No newline at end of file diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/FencedDeleter.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/FencedDeleter.cpp index 388eb93b2e5..42a42373e25 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/FencedDeleter.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/FencedDeleter.cpp @@ -30,6 +30,7 @@ namespace dawn_native { namespace vulkan { ASSERT(mMemoriesToDelete.Empty()); ASSERT(mPipelinesToDelete.Empty()); ASSERT(mPipelineLayoutsToDelete.Empty()); + ASSERT(mQueryPoolsToDelete.Empty()); ASSERT(mRenderPassesToDelete.Empty()); ASSERT(mSamplersToDelete.Empty()); ASSERT(mSemaphoresToDelete.Empty()); @@ -70,6 +71,10 @@ namespace dawn_native { namespace vulkan { mPipelineLayoutsToDelete.Enqueue(layout, mDevice->GetPendingCommandSerial()); } + void FencedDeleter::DeleteWhenUnused(VkQueryPool querypool) { + mQueryPoolsToDelete.Enqueue(querypool, mDevice->GetPendingCommandSerial()); + } + void FencedDeleter::DeleteWhenUnused(VkRenderPass renderPass) { mRenderPassesToDelete.Enqueue(renderPass, mDevice->GetPendingCommandSerial()); } @@ -164,6 +169,11 @@ namespace dawn_native { namespace vulkan { } mDescriptorPoolsToDelete.ClearUpTo(completedSerial); + for (VkQueryPool pool : mQueryPoolsToDelete.IterateUpTo(completedSerial)) { + mDevice->fn.DestroyQueryPool(vkDevice, pool, nullptr); + } + mQueryPoolsToDelete.ClearUpTo(completedSerial); + for (VkSampler sampler : mSamplersToDelete.IterateUpTo(completedSerial)) { mDevice->fn.DestroySampler(vkDevice, sampler, nullptr); } diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/FencedDeleter.h b/chromium/third_party/dawn/src/dawn_native/vulkan/FencedDeleter.h index 7200eda94d6..9e516b1006c 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/FencedDeleter.h +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/FencedDeleter.h @@ -36,6 +36,7 @@ namespace dawn_native { namespace vulkan { void DeleteWhenUnused(VkPipelineLayout layout); void DeleteWhenUnused(VkRenderPass renderPass); void DeleteWhenUnused(VkPipeline pipeline); + void DeleteWhenUnused(VkQueryPool querypool); void DeleteWhenUnused(VkSampler sampler); void DeleteWhenUnused(VkSemaphore semaphore); void DeleteWhenUnused(VkShaderModule module); @@ -54,6 +55,7 @@ namespace dawn_native { namespace vulkan { SerialQueue<VkImageView> mImageViewsToDelete; SerialQueue<VkPipeline> mPipelinesToDelete; SerialQueue<VkPipelineLayout> mPipelineLayoutsToDelete; + SerialQueue<VkQueryPool> mQueryPoolsToDelete; SerialQueue<VkRenderPass> mRenderPassesToDelete; SerialQueue<VkSampler> mSamplersToDelete; SerialQueue<VkSemaphore> mSemaphoresToDelete; diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/QuerySetVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/QuerySetVk.cpp new file mode 100644 index 00000000000..0d740a63ffe --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/QuerySetVk.cpp @@ -0,0 +1,112 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "dawn_native/vulkan/QuerySetVk.h" + +#include "dawn_native/vulkan/DeviceVk.h" +#include "dawn_native/vulkan/FencedDeleter.h" +#include "dawn_native/vulkan/VulkanError.h" +#include "dawn_platform/DawnPlatform.h" + +namespace dawn_native { namespace vulkan { + + namespace { + VkQueryType VulkanQueryType(wgpu::QueryType type) { + switch (type) { + case wgpu::QueryType::Occlusion: + return VK_QUERY_TYPE_OCCLUSION; + case wgpu::QueryType::PipelineStatistics: + return VK_QUERY_TYPE_PIPELINE_STATISTICS; + case wgpu::QueryType::Timestamp: + return VK_QUERY_TYPE_TIMESTAMP; + default: + UNREACHABLE(); + } + } + + VkQueryPipelineStatisticFlags VulkanQueryPipelineStatisticFlags( + std::vector<wgpu::PipelineStatisticName> pipelineStatisticsSet) { + VkQueryPipelineStatisticFlags pipelineStatistics = 0; + for (size_t i = 0; i < pipelineStatisticsSet.size(); ++i) { + switch (pipelineStatisticsSet[i]) { + case wgpu::PipelineStatisticName::ClipperInvocations: + pipelineStatistics |= VK_QUERY_PIPELINE_STATISTIC_CLIPPING_INVOCATIONS_BIT; + break; + case wgpu::PipelineStatisticName::ClipperPrimitivesOut: + pipelineStatistics |= VK_QUERY_PIPELINE_STATISTIC_CLIPPING_PRIMITIVES_BIT; + break; + case wgpu::PipelineStatisticName::ComputeShaderInvocations: + pipelineStatistics |= + VK_QUERY_PIPELINE_STATISTIC_COMPUTE_SHADER_INVOCATIONS_BIT; + break; + case wgpu::PipelineStatisticName::FragmentShaderInvocations: + pipelineStatistics |= + VK_QUERY_PIPELINE_STATISTIC_FRAGMENT_SHADER_INVOCATIONS_BIT; + break; + case wgpu::PipelineStatisticName::VertexShaderInvocations: + pipelineStatistics |= + VK_QUERY_PIPELINE_STATISTIC_VERTEX_SHADER_INVOCATIONS_BIT; + break; + default: + UNREACHABLE(); + break; + } + } + + return pipelineStatistics; + } + } // anonymous namespace + + // static + ResultOrError<QuerySet*> QuerySet::Create(Device* device, + const QuerySetDescriptor* descriptor) { + Ref<QuerySet> queryset = AcquireRef(new QuerySet(device, descriptor)); + DAWN_TRY(queryset->Initialize()); + return queryset.Detach(); + } + + MaybeError QuerySet::Initialize() { + VkQueryPoolCreateInfo createInfo; + createInfo.sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO; + createInfo.pNext = NULL; + createInfo.flags = 0; + createInfo.queryType = VulkanQueryType(GetQueryType()); + createInfo.queryCount = GetQueryCount(); + if (GetQueryType() == wgpu::QueryType::PipelineStatistics) { + createInfo.pipelineStatistics = + VulkanQueryPipelineStatisticFlags(GetPipelineStatistics()); + } + + Device* device = ToBackend(GetDevice()); + return CheckVkOOMThenSuccess( + device->fn.CreateQueryPool(device->GetVkDevice(), &createInfo, nullptr, &*mHandle), + "vkCreateQueryPool"); + } + + VkQueryPool QuerySet::GetHandle() const { + return mHandle; + } + + QuerySet::~QuerySet() { + DestroyInternal(); + } + + void QuerySet::DestroyImpl() { + if (mHandle != VK_NULL_HANDLE) { + ToBackend(GetDevice())->GetFencedDeleter()->DeleteWhenUnused(mHandle); + mHandle = VK_NULL_HANDLE; + } + } + +}} // namespace dawn_native::vulkan diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/QuerySetVk.h b/chromium/third_party/dawn/src/dawn_native/vulkan/QuerySetVk.h new file mode 100644 index 00000000000..18cd0012908 --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/QuerySetVk.h @@ -0,0 +1,45 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef DAWNNATIVE_VULKAN_QUERYSETVK_H_ +#define DAWNNATIVE_VULKAN_QUERYSETVK_H_ + +#include "dawn_native/QuerySet.h" + +#include "common/vulkan_platform.h" + +namespace dawn_native { namespace vulkan { + + class Device; + + class QuerySet final : public QuerySetBase { + public: + static ResultOrError<QuerySet*> Create(Device* device, + const QuerySetDescriptor* descriptor); + + VkQueryPool GetHandle() const; + + private: + ~QuerySet() override; + using QuerySetBase::QuerySetBase; + MaybeError Initialize(); + + void DestroyImpl() override; + + VkQueryPool mHandle = VK_NULL_HANDLE; + }; + +}} // namespace dawn_native::vulkan + +#endif // DAWNNATIVE_VULKAN_QUERYSETVK_H_ diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/QueueVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/QueueVk.cpp index 19ab88c7e3f..fc11c10bb13 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/QueueVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/QueueVk.cpp @@ -14,6 +14,11 @@ #include "dawn_native/vulkan/QueueVk.h" +#include "common/Math.h" +#include "dawn_native/Buffer.h" +#include "dawn_native/CommandValidation.h" +#include "dawn_native/Commands.h" +#include "dawn_native/DynamicUploader.h" #include "dawn_native/vulkan/CommandBufferVk.h" #include "dawn_native/vulkan/CommandRecordingContext.h" #include "dawn_native/vulkan/DeviceVk.h" @@ -22,6 +27,61 @@ namespace dawn_native { namespace vulkan { + namespace { + ResultOrError<UploadHandle> UploadTextureDataAligningBytesPerRow( + DeviceBase* device, + const void* data, + uint32_t alignedBytesPerRow, + uint32_t optimallyAlignedBytesPerRow, + uint32_t alignedRowsPerImage, + const TextureDataLayout& dataLayout, + const TexelBlockInfo& blockInfo, + const Extent3D& writeSizePixel) { + uint64_t newDataSizeBytes; + DAWN_TRY_ASSIGN( + newDataSizeBytes, + ComputeRequiredBytesInCopy(blockInfo, writeSizePixel, optimallyAlignedBytesPerRow, + alignedRowsPerImage)); + + uint64_t optimalOffsetAlignment = + ToBackend(device) + ->GetDeviceInfo() + .properties.limits.optimalBufferCopyOffsetAlignment; + ASSERT(IsPowerOfTwo(optimalOffsetAlignment)); + ASSERT(IsPowerOfTwo(blockInfo.blockByteSize)); + // We need the offset to be aligned to both optimalOffsetAlignment and blockByteSize, + // since both of them are powers of two, we only need to align to the max value. + uint64_t offsetAlignment = + std::max(optimalOffsetAlignment, uint64_t(blockInfo.blockByteSize)); + + UploadHandle uploadHandle; + DAWN_TRY_ASSIGN(uploadHandle, device->GetDynamicUploader()->Allocate( + newDataSizeBytes, device->GetPendingCommandSerial(), + offsetAlignment)); + ASSERT(uploadHandle.mappedBuffer != nullptr); + + uint8_t* dstPointer = static_cast<uint8_t*>(uploadHandle.mappedBuffer); + const uint8_t* srcPointer = static_cast<const uint8_t*>(data); + srcPointer += dataLayout.offset; + + uint32_t alignedRowsPerImageInBlock = alignedRowsPerImage / blockInfo.blockHeight; + uint32_t dataRowsPerImageInBlock = dataLayout.rowsPerImage / blockInfo.blockHeight; + if (dataRowsPerImageInBlock == 0) { + dataRowsPerImageInBlock = writeSizePixel.height / blockInfo.blockHeight; + } + + ASSERT(dataRowsPerImageInBlock >= alignedRowsPerImageInBlock); + uint64_t imageAdditionalStride = + dataLayout.bytesPerRow * (dataRowsPerImageInBlock - alignedRowsPerImageInBlock); + + CopyTextureData(dstPointer, srcPointer, writeSizePixel.depth, + alignedRowsPerImageInBlock, imageAdditionalStride, alignedBytesPerRow, + optimallyAlignedBytesPerRow, dataLayout.bytesPerRow); + + return uploadHandle; + } + } // namespace + // static Queue* Queue::Create(Device* device) { return new Queue(device); @@ -48,4 +108,46 @@ namespace dawn_native { namespace vulkan { return {}; } + MaybeError Queue::WriteTextureImpl(const TextureCopyView& destination, + const void* data, + const TextureDataLayout& dataLayout, + const Extent3D& writeSizePixel) { + const TexelBlockInfo& blockInfo = + destination.texture->GetFormat().GetTexelBlockInfo(destination.aspect); + + // We are only copying the part of the data that will appear in the texture. + // Note that validating texture copy range ensures that writeSizePixel->width and + // writeSizePixel->height are multiples of blockWidth and blockHeight respectively. + uint32_t alignedBytesPerRow = + (writeSizePixel.width) / blockInfo.blockWidth * blockInfo.blockByteSize; + uint32_t alignedRowsPerImage = writeSizePixel.height; + + uint32_t optimalBytesPerRowAlignment = + ToBackend(GetDevice()) + ->GetDeviceInfo() + .properties.limits.optimalBufferCopyRowPitchAlignment; + uint32_t optimallyAlignedBytesPerRow = + Align(alignedBytesPerRow, optimalBytesPerRowAlignment); + + UploadHandle uploadHandle; + DAWN_TRY_ASSIGN(uploadHandle, + UploadTextureDataAligningBytesPerRow( + GetDevice(), data, alignedBytesPerRow, optimallyAlignedBytesPerRow, + alignedRowsPerImage, dataLayout, blockInfo, writeSizePixel)); + + TextureDataLayout passDataLayout = dataLayout; + passDataLayout.offset = uploadHandle.startOffset; + passDataLayout.bytesPerRow = optimallyAlignedBytesPerRow; + passDataLayout.rowsPerImage = alignedRowsPerImage; + + TextureCopy textureCopy; + textureCopy.texture = destination.texture; + textureCopy.mipLevel = destination.mipLevel; + textureCopy.origin = destination.origin; + textureCopy.aspect = ConvertAspect(destination.texture->GetFormat(), destination.aspect); + + return ToBackend(GetDevice()) + ->CopyFromStagingToTexture(uploadHandle.stagingBuffer, passDataLayout, &textureCopy, + writeSizePixel); + } }} // namespace dawn_native::vulkan diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/QueueVk.h b/chromium/third_party/dawn/src/dawn_native/vulkan/QueueVk.h index 715f5eb90d9..34a0e6a5795 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/QueueVk.h +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/QueueVk.h @@ -31,8 +31,12 @@ namespace dawn_native { namespace vulkan { using QueueBase::QueueBase; MaybeError SubmitImpl(uint32_t commandCount, CommandBufferBase* const* commands) override; + MaybeError WriteTextureImpl(const TextureCopyView& destination, + const void* data, + const TextureDataLayout& dataLayout, + const Extent3D& writeSizePixel) override; }; }} // namespace dawn_native::vulkan -#endif // DAWNNATIVE_VULKAN_QUEUEVK_H_ +#endif // DAWNNATIVE_VULKAN_QUEUEVK_H_
\ No newline at end of file diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/RenderPipelineVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/RenderPipelineVk.cpp index 5053e01e792..271aaaad6eb 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/RenderPipelineVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/RenderPipelineVk.cpp @@ -410,8 +410,13 @@ namespace dawn_native { namespace vulkan { multisample.rasterizationSamples = VulkanSampleCount(GetSampleCount()); multisample.sampleShadingEnable = VK_FALSE; multisample.minSampleShading = 0.0f; - multisample.pSampleMask = nullptr; - multisample.alphaToCoverageEnable = VK_FALSE; + // VkPipelineMultisampleStateCreateInfo.pSampleMask is an array of length + // ceil(rasterizationSamples / 32) and since we're passing a single uint32_t + // we have to assert that this length is indeed 1. + ASSERT(multisample.rasterizationSamples <= 32); + VkSampleMask sampleMask = GetSampleMask(); + multisample.pSampleMask = &sampleMask; + multisample.alphaToCoverageEnable = descriptor->alphaToCoverageEnabled; multisample.alphaToOneEnable = VK_FALSE; VkPipelineDepthStencilStateCreateInfo depthStencilState = @@ -424,7 +429,7 @@ namespace dawn_native { namespace vulkan { descriptor->fragmentStage->module->GetFragmentOutputBaseTypes(); for (uint32_t i : IterateBitSet(GetColorAttachmentsMask())) { const ColorStateDescriptor* colorStateDescriptor = GetColorStateDescriptor(i); - bool isDeclaredInFragmentShader = fragmentOutputBaseTypes[i] != Format::Other; + bool isDeclaredInFragmentShader = fragmentOutputBaseTypes[i] != Format::Type::Other; colorBlendAttachments[i] = ComputeColorDesc(colorStateDescriptor, isDeclaredInFragmentShader); } diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/ResourceMemoryAllocatorVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/ResourceMemoryAllocatorVk.cpp index 22523a36874..924a47b9fed 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/ResourceMemoryAllocatorVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/ResourceMemoryAllocatorVk.cpp @@ -46,17 +46,22 @@ namespace dawn_native { namespace vulkan { : mDevice(device), mMemoryTypeIndex(memoryTypeIndex), mMemoryHeapSize(memoryHeapSize), + mPooledMemoryAllocator(this), mBuddySystem( // Round down to a power of 2 that's <= mMemoryHeapSize. This will always // be a multiple of kBuddyHeapsSize because kBuddyHeapsSize is a power of 2. uint64_t(1) << Log2(mMemoryHeapSize), // Take the min in the very unlikely case the memory heap is tiny. std::min(uint64_t(1) << Log2(mMemoryHeapSize), kBuddyHeapsSize), - this) { + &mPooledMemoryAllocator) { ASSERT(IsPowerOfTwo(kBuddyHeapsSize)); } ~SingleTypeAllocator() override = default; + void DestroyPool() { + mPooledMemoryAllocator.DestroyPool(); + } + ResultOrError<ResourceMemoryAllocation> AllocateMemory( const VkMemoryRequirements& requirements) { return mBuddySystem.Allocate(requirements.size, requirements.alignment); @@ -100,6 +105,7 @@ namespace dawn_native { namespace vulkan { Device* mDevice; size_t mMemoryTypeIndex; VkDeviceSize mMemoryHeapSize; + PooledResourceMemoryAllocator mPooledMemoryAllocator; BuddyMemoryAllocator mBuddySystem; }; @@ -258,4 +264,10 @@ namespace dawn_native { namespace vulkan { return bestType; } + void ResourceMemoryAllocator::DestroyPool() { + for (auto& alloc : mAllocatorsPerType) { + alloc->DestroyPool(); + } + } + }} // namespace dawn_native::vulkan diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/ResourceMemoryAllocatorVk.h b/chromium/third_party/dawn/src/dawn_native/vulkan/ResourceMemoryAllocatorVk.h index 88f6d4e0c66..04176a33000 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/ResourceMemoryAllocatorVk.h +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/ResourceMemoryAllocatorVk.h @@ -18,6 +18,7 @@ #include "common/SerialQueue.h" #include "common/vulkan_platform.h" #include "dawn_native/Error.h" +#include "dawn_native/PooledResourceMemoryAllocator.h" #include "dawn_native/ResourceMemoryAllocation.h" #include <memory> @@ -36,6 +37,8 @@ namespace dawn_native { namespace vulkan { bool mappable); void Deallocate(ResourceMemoryAllocation* allocation); + void DestroyPool(); + void Tick(Serial completedSerial); int FindBestTypeIndex(VkMemoryRequirements requirements, bool mappable); diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/SwapChainVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/SwapChainVk.cpp index 660128776d6..30d774d195e 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/SwapChainVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/SwapChainVk.cpp @@ -47,8 +47,8 @@ namespace dawn_native { namespace vulkan { return nullptr; } - VkImage nativeTexture = - VkImage::CreateFromHandle(reinterpret_cast<::VkImage>(next.texture.u64)); + ::VkImage image = NativeNonDispatachableHandleFromU64<::VkImage>(next.texture.u64); + VkImage nativeTexture = VkImage::CreateFromHandle(image); return Texture::CreateForSwapChain(ToBackend(GetDevice()), descriptor, nativeTexture) .Detach(); } diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/TextureVk.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/TextureVk.cpp index eb502f8c1f3..e61facd09fd 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/TextureVk.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/TextureVk.cpp @@ -17,6 +17,7 @@ #include "common/Assert.h" #include "common/Math.h" #include "dawn_native/DynamicUploader.h" +#include "dawn_native/EnumMaskIterator.h" #include "dawn_native/Error.h" #include "dawn_native/VulkanBackend.h" #include "dawn_native/vulkan/AdapterVk.h" @@ -192,23 +193,6 @@ namespace dawn_native { namespace vulkan { return flags; } - // Computes which Vulkan texture aspects are relevant for the given Dawn format - VkImageAspectFlags VulkanAspectMask(const Format& format) { - switch (format.aspect) { - case Format::Aspect::Color: - return VK_IMAGE_ASPECT_COLOR_BIT; - case Format::Aspect::Depth: - return VK_IMAGE_ASPECT_DEPTH_BIT; - case Format::Aspect::Stencil: - return VK_IMAGE_ASPECT_STENCIL_BIT; - case Format::Aspect::DepthStencil: - return VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT; - default: - UNREACHABLE(); - return 0; - } - } - VkImageMemoryBarrier BuildMemoryBarrier(const Format& format, const VkImage& image, wgpu::TextureUsage lastUsage, @@ -222,7 +206,7 @@ namespace dawn_native { namespace vulkan { barrier.oldLayout = VulkanImageLayout(lastUsage, format); barrier.newLayout = VulkanImageLayout(usage, format); barrier.image = image; - barrier.subresourceRange.aspectMask = VulkanAspectMask(format); + barrier.subresourceRange.aspectMask = VulkanAspectMask(format.aspects); barrier.subresourceRange.baseMipLevel = range.baseMipLevel; barrier.subresourceRange.levelCount = range.levelCount; barrier.subresourceRange.baseArrayLayer = range.baseArrayLayer; @@ -311,8 +295,10 @@ namespace dawn_native { namespace vulkan { return VK_FORMAT_B8G8R8A8_SRGB; case wgpu::TextureFormat::RGB10A2Unorm: return VK_FORMAT_A2B10G10R10_UNORM_PACK32; - case wgpu::TextureFormat::RG11B10Float: + case wgpu::TextureFormat::RG11B10Ufloat: return VK_FORMAT_B10G11R11_UFLOAT_PACK32; + case wgpu::TextureFormat::RGB9E5Ufloat: + return VK_FORMAT_E5B9G9R9_UFLOAT_PACK32; case wgpu::TextureFormat::RG32Uint: return VK_FORMAT_R32G32_UINT; @@ -591,7 +577,7 @@ namespace dawn_native { namespace vulkan { // Don't clear imported texture if already cleared if (descriptor->isCleared) { - SetIsSubresourceContentInitialized(true, {0, 1, 0, 1}); + SetIsSubresourceContentInitialized(true, GetAllSubresources()); } // Success, acquire all the external objects. @@ -663,8 +649,21 @@ namespace dawn_native { namespace vulkan { return mHandle; } - VkImageAspectFlags Texture::GetVkAspectMask() const { - return VulkanAspectMask(GetFormat()); + VkImageAspectFlags Texture::GetVkAspectMask(wgpu::TextureAspect aspect) const { + // TODO(enga): These masks could be precomputed. + switch (aspect) { + case wgpu::TextureAspect::All: + return VulkanAspectMask(GetFormat().aspects); + case wgpu::TextureAspect::DepthOnly: + ASSERT(GetFormat().aspects & Aspect::Depth); + return VulkanAspectMask(Aspect::Depth); + case wgpu::TextureAspect::StencilOnly: + ASSERT(GetFormat().aspects & Aspect::Stencil); + return VulkanAspectMask(Aspect::Stencil); + default: + UNREACHABLE(); + return 0; + } } void Texture::TweakTransitionForExternalUsage(CommandRecordingContext* recordingContext, @@ -681,7 +680,7 @@ namespace dawn_native { namespace vulkan { if (barriers->size() == transitionBarrierStart) { barriers->push_back(BuildMemoryBarrier( GetFormat(), mHandle, wgpu::TextureUsage::None, wgpu::TextureUsage::None, - SubresourceRange::SingleSubresource(0, 0))); + SubresourceRange::SingleMipAndLayer(0, 0, GetFormat().aspects))); } // Transfer texture from external queue to graphics queue @@ -695,7 +694,7 @@ namespace dawn_native { namespace vulkan { if (barriers->size() == transitionBarrierStart) { barriers->push_back(BuildMemoryBarrier( GetFormat(), mHandle, wgpu::TextureUsage::None, wgpu::TextureUsage::None, - SubresourceRange::SingleSubresource(0, 0))); + SubresourceRange::SingleMipAndLayer(0, 0, GetFormat().aspects))); } // Transfer texture from graphics queue to external queue @@ -762,24 +761,40 @@ namespace dawn_native { namespace vulkan { } else { for (uint32_t arrayLayer = 0; arrayLayer < GetArrayLayers(); ++arrayLayer) { for (uint32_t mipLevel = 0; mipLevel < GetNumMipLevels(); ++mipLevel) { - uint32_t index = GetSubresourceIndex(mipLevel, arrayLayer); + wgpu::TextureUsage lastUsage = wgpu::TextureUsage::None; + wgpu::TextureUsage usage = wgpu::TextureUsage::None; + + // Accumulate usage for all format aspects because we cannot transition + // separately. + // TODO(enga): Use VK_KHR_separate_depth_stencil_layouts. + for (Aspect aspect : IterateEnumMask(GetFormat().aspects)) { + uint32_t index = GetSubresourceIndex(mipLevel, arrayLayer, aspect); + + usage |= textureUsages.subresourceUsages[index]; + lastUsage |= mSubresourceLastUsages[index]; + } // Avoid encoding barriers when it isn't needed. - if (textureUsages.subresourceUsages[index] == wgpu::TextureUsage::None) { + if (usage == wgpu::TextureUsage::None) { continue; } - if (CanReuseWithoutBarrier(mSubresourceLastUsages[index], - textureUsages.subresourceUsages[index])) { + if (CanReuseWithoutBarrier(lastUsage, usage)) { continue; } - imageBarriers->push_back(BuildMemoryBarrier( - format, mHandle, mSubresourceLastUsages[index], - textureUsages.subresourceUsages[index], - SubresourceRange::SingleSubresource(mipLevel, arrayLayer))); - allLastUsages |= mSubresourceLastUsages[index]; - allUsages |= textureUsages.subresourceUsages[index]; - mSubresourceLastUsages[index] = textureUsages.subresourceUsages[index]; + + allLastUsages |= lastUsage; + allUsages |= usage; + + for (Aspect aspect : IterateEnumMask(GetFormat().aspects)) { + uint32_t index = GetSubresourceIndex(mipLevel, arrayLayer, aspect); + mSubresourceLastUsages[index] = usage; + } + + imageBarriers->push_back( + BuildMemoryBarrier(format, mHandle, lastUsage, usage, + SubresourceRange::SingleMipAndLayer( + mipLevel, arrayLayer, GetFormat().aspects))); } } } @@ -801,7 +816,6 @@ namespace dawn_native { namespace vulkan { const Format& format = GetFormat(); wgpu::TextureUsage allLastUsages = wgpu::TextureUsage::None; - uint32_t subresourceCount = GetSubresourceCount(); // This transitions assume it is a 2D texture ASSERT(GetDimension() == wgpu::TextureDimension::e2D); @@ -810,7 +824,9 @@ namespace dawn_native { namespace vulkan { // are the same, then we can use one barrier to do state transition for all subresources. // Note that if the texture has only one mip level and one array slice, it will fall into // this category. - bool areAllSubresourcesCovered = range.levelCount * range.layerCount == subresourceCount; + bool areAllSubresourcesCovered = (range.levelCount == GetNumMipLevels() && // + range.layerCount == GetArrayLayers() && // + range.aspects == format.aspects); if (mSameLastUsagesAcrossSubresources && areAllSubresourcesCovered) { ASSERT(range.baseMipLevel == 0 && range.baseArrayLayer == 0); if (CanReuseWithoutBarrier(mSubresourceLastUsages[0], usage)) { @@ -819,7 +835,7 @@ namespace dawn_native { namespace vulkan { barriers.push_back( BuildMemoryBarrier(format, mHandle, mSubresourceLastUsages[0], usage, range)); allLastUsages = mSubresourceLastUsages[0]; - for (uint32_t i = 0; i < subresourceCount; ++i) { + for (uint32_t i = 0; i < GetSubresourceCount(); ++i) { mSubresourceLastUsages[i] = usage; } } else { @@ -827,17 +843,29 @@ namespace dawn_native { namespace vulkan { layer < range.baseArrayLayer + range.layerCount; ++layer) { for (uint32_t level = range.baseMipLevel; level < range.baseMipLevel + range.levelCount; ++level) { - uint32_t index = GetSubresourceIndex(level, layer); + // Accumulate usage for all format aspects because we cannot transition + // separately. + // TODO(enga): Use VK_KHR_separate_depth_stencil_layouts. + wgpu::TextureUsage lastUsage = wgpu::TextureUsage::None; + for (Aspect aspect : IterateEnumMask(format.aspects)) { + uint32_t index = GetSubresourceIndex(level, layer, aspect); + lastUsage |= mSubresourceLastUsages[index]; + } - if (CanReuseWithoutBarrier(mSubresourceLastUsages[index], usage)) { + if (CanReuseWithoutBarrier(lastUsage, usage)) { continue; } - barriers.push_back( - BuildMemoryBarrier(format, mHandle, mSubresourceLastUsages[index], usage, - SubresourceRange::SingleSubresource(level, layer))); - allLastUsages |= mSubresourceLastUsages[index]; - mSubresourceLastUsages[index] = usage; + allLastUsages |= lastUsage; + + for (Aspect aspect : IterateEnumMask(format.aspects)) { + uint32_t index = GetSubresourceIndex(level, layer, aspect); + mSubresourceLastUsages[index] = usage; + } + + barriers.push_back(BuildMemoryBarrier( + format, mHandle, lastUsage, usage, + SubresourceRange::SingleMipAndLayer(level, layer, format.aspects))); } } } @@ -866,7 +894,6 @@ namespace dawn_native { namespace vulkan { TransitionUsageNow(recordingContext, wgpu::TextureUsage::CopyDst, range); if (GetFormat().isRenderable) { VkImageSubresourceRange imageRange = {}; - imageRange.aspectMask = GetVkAspectMask(); imageRange.levelCount = 1; imageRange.layerCount = 1; @@ -875,16 +902,25 @@ namespace dawn_native { namespace vulkan { imageRange.baseMipLevel = level; for (uint32_t layer = range.baseArrayLayer; layer < range.baseArrayLayer + range.layerCount; ++layer) { - if (clearValue == TextureBase::ClearValue::Zero && - IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, layer))) { - // Skip lazy clears if already initialized. + Aspect aspects = Aspect::None; + for (Aspect aspect : IterateEnumMask(range.aspects)) { + if (clearValue == TextureBase::ClearValue::Zero && + IsSubresourceContentInitialized( + SubresourceRange::SingleMipAndLayer(level, layer, aspect))) { + // Skip lazy clears if already initialized. + continue; + } + aspects |= aspect; + } + + if (aspects == Aspect::None) { continue; } + imageRange.aspectMask = VulkanAspectMask(aspects); imageRange.baseArrayLayer = layer; - if (GetFormat().HasDepthOrStencil()) { + if (aspects & (Aspect::Depth | Aspect::Stencil)) { VkClearDepthStencilValue clearDepthStencilValue[1]; clearDepthStencilValue[0].depth = fClearColor; clearDepthStencilValue[0].stencil = clearColor; @@ -893,6 +929,7 @@ namespace dawn_native { namespace vulkan { VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, clearDepthStencilValue, 1, &imageRange); } else { + ASSERT(aspects == Aspect::Color); VkClearColorValue clearColorValue = { {fClearColor, fClearColor, fClearColor, fClearColor}}; device->fn.CmdClearColorImage(recordingContext->commandBuffer, GetHandle(), @@ -902,7 +939,6 @@ namespace dawn_native { namespace vulkan { } } } else { - // TODO(natlee@microsoft.com): test compressed textures are cleared // create temp buffer with clear color to copy to the texture image uint32_t bytesPerRow = Align((GetWidth() / GetFormat().blockWidth) * GetFormat().blockByteSize, @@ -915,7 +951,8 @@ namespace dawn_native { namespace vulkan { DynamicUploader* uploader = device->GetDynamicUploader(); UploadHandle uploadHandle; DAWN_TRY_ASSIGN(uploadHandle, - uploader->Allocate(bufferSize, device->GetPendingCommandSerial())); + uploader->Allocate(bufferSize, device->GetPendingCommandSerial(), + GetFormat().blockByteSize)); memset(uploadHandle.mappedBuffer, clearColor, bufferSize); // compute the buffer image copy to set the clear region of entire texture @@ -924,6 +961,7 @@ namespace dawn_native { namespace vulkan { bufferCopy.offset = uploadHandle.startOffset; bufferCopy.bytesPerRow = bytesPerRow; + ASSERT(range.aspects == Aspect::Color); for (uint32_t level = range.baseMipLevel; level < range.baseMipLevel + range.levelCount; ++level) { Extent3D copySize = GetMipLevelVirtualSize(level); @@ -932,16 +970,17 @@ namespace dawn_native { namespace vulkan { layer < range.baseArrayLayer + range.layerCount; ++layer) { if (clearValue == TextureBase::ClearValue::Zero && IsSubresourceContentInitialized( - SubresourceRange::SingleSubresource(level, layer))) { + SubresourceRange::SingleMipAndLayer(level, layer, Aspect::Color))) { // Skip lazy clears if already initialized. continue; } + ASSERT(GetFormat().aspects == Aspect::Color); dawn_native::TextureCopy textureCopy; textureCopy.texture = this; - textureCopy.origin = {0, 0, 0}; + textureCopy.origin = {0, 0, layer}; textureCopy.mipLevel = level; - textureCopy.arrayLayer = layer; + textureCopy.aspect = GetFormat().aspects; VkBufferImageCopy region = ComputeBufferImageCopyRegion(bufferCopy, textureCopy, copySize); @@ -1008,11 +1047,13 @@ namespace dawn_native { namespace vulkan { createInfo.format = VulkanImageFormat(device, descriptor->format); createInfo.components = VkComponentMapping{VK_COMPONENT_SWIZZLE_R, VK_COMPONENT_SWIZZLE_G, VK_COMPONENT_SWIZZLE_B, VK_COMPONENT_SWIZZLE_A}; - createInfo.subresourceRange.aspectMask = VulkanAspectMask(GetFormat()); - createInfo.subresourceRange.baseMipLevel = descriptor->baseMipLevel; - createInfo.subresourceRange.levelCount = descriptor->mipLevelCount; - createInfo.subresourceRange.baseArrayLayer = descriptor->baseArrayLayer; - createInfo.subresourceRange.layerCount = descriptor->arrayLayerCount; + + const SubresourceRange& subresources = GetSubresourceRange(); + createInfo.subresourceRange.baseMipLevel = subresources.baseMipLevel; + createInfo.subresourceRange.levelCount = subresources.levelCount; + createInfo.subresourceRange.baseArrayLayer = subresources.baseArrayLayer; + createInfo.subresourceRange.layerCount = subresources.layerCount; + createInfo.subresourceRange.aspectMask = VulkanAspectMask(subresources.aspects); return CheckVkSuccess( device->fn.CreateImageView(device->GetVkDevice(), &createInfo, nullptr, &*mHandle), diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/TextureVk.h b/chromium/third_party/dawn/src/dawn_native/vulkan/TextureVk.h index 8a1564af72b..6748ebd0d4f 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/TextureVk.h +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/TextureVk.h @@ -59,7 +59,7 @@ namespace dawn_native { namespace vulkan { VkImage nativeImage); VkImage GetHandle() const; - VkImageAspectFlags GetVkAspectMask() const; + VkImageAspectFlags GetVkAspectMask(wgpu::TextureAspect aspect) const; // Transitions the texture to be used as `usage`, recording any necessary barrier in // `commands`. diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/UtilsVulkan.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/UtilsVulkan.cpp index 4ce513f5840..1b6b2d66d31 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/UtilsVulkan.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/UtilsVulkan.cpp @@ -15,6 +15,7 @@ #include "dawn_native/vulkan/UtilsVulkan.h" #include "common/Assert.h" +#include "dawn_native/EnumMaskIterator.h" #include "dawn_native/Format.h" #include "dawn_native/vulkan/Forward.h" #include "dawn_native/vulkan/TextureVk.h" @@ -44,6 +45,28 @@ namespace dawn_native { namespace vulkan { } } + // Convert Dawn texture aspects to Vulkan texture aspect flags + VkImageAspectFlags VulkanAspectMask(const Aspect& aspects) { + VkImageAspectFlags flags = 0; + for (Aspect aspect : IterateEnumMask(aspects)) { + switch (aspect) { + case Aspect::Color: + flags |= VK_IMAGE_ASPECT_COLOR_BIT; + break; + case Aspect::Depth: + flags |= VK_IMAGE_ASPECT_DEPTH_BIT; + break; + case Aspect::Stencil: + flags |= VK_IMAGE_ASPECT_STENCIL_BIT; + break; + default: + UNREACHABLE(); + break; + } + } + return flags; + } + // Vulkan SPEC requires the source/destination region specified by each element of // pRegions must be a region that is contained within srcImage/dstImage. Here the size of // the image refers to the virtual size, while Dawn validates texture copy extent with the @@ -68,32 +91,52 @@ namespace dawn_native { namespace vulkan { VkBufferImageCopy ComputeBufferImageCopyRegion(const BufferCopy& bufferCopy, const TextureCopy& textureCopy, const Extent3D& copySize) { + TextureDataLayout passDataLayout; + passDataLayout.offset = bufferCopy.offset; + passDataLayout.rowsPerImage = bufferCopy.rowsPerImage; + passDataLayout.bytesPerRow = bufferCopy.bytesPerRow; + return ComputeBufferImageCopyRegion(passDataLayout, textureCopy, copySize); + } + + VkBufferImageCopy ComputeBufferImageCopyRegion(const TextureDataLayout& dataLayout, + const TextureCopy& textureCopy, + const Extent3D& copySize) { const Texture* texture = ToBackend(textureCopy.texture.Get()); VkBufferImageCopy region; - region.bufferOffset = bufferCopy.offset; + region.bufferOffset = dataLayout.offset; // In Vulkan the row length is in texels while it is in bytes for Dawn - const Format& format = texture->GetFormat(); - ASSERT(bufferCopy.bytesPerRow % format.blockByteSize == 0); - region.bufferRowLength = bufferCopy.bytesPerRow / format.blockByteSize * format.blockWidth; - region.bufferImageHeight = bufferCopy.rowsPerImage; + const TexelBlockInfo& blockInfo = + texture->GetFormat().GetTexelBlockInfo(textureCopy.aspect); + ASSERT(dataLayout.bytesPerRow % blockInfo.blockByteSize == 0); + region.bufferRowLength = + dataLayout.bytesPerRow / blockInfo.blockByteSize * blockInfo.blockWidth; + region.bufferImageHeight = dataLayout.rowsPerImage; - region.imageSubresource.aspectMask = texture->GetVkAspectMask(); + region.imageSubresource.aspectMask = VulkanAspectMask(textureCopy.aspect); region.imageSubresource.mipLevel = textureCopy.mipLevel; - region.imageSubresource.baseArrayLayer = textureCopy.arrayLayer; - region.imageOffset.x = textureCopy.origin.x; - region.imageOffset.y = textureCopy.origin.y; - region.imageOffset.z = textureCopy.origin.z; + switch (textureCopy.texture->GetDimension()) { + case wgpu::TextureDimension::e2D: { + region.imageOffset.x = textureCopy.origin.x; + region.imageOffset.y = textureCopy.origin.y; + region.imageOffset.z = 0; + + region.imageSubresource.baseArrayLayer = textureCopy.origin.z; + region.imageSubresource.layerCount = copySize.depth; - Extent3D imageExtent = ComputeTextureCopyExtent(textureCopy, copySize); - region.imageExtent.width = imageExtent.width; - region.imageExtent.height = imageExtent.height; + Extent3D imageExtent = ComputeTextureCopyExtent(textureCopy, copySize); + region.imageExtent.width = imageExtent.width; + region.imageExtent.height = imageExtent.height; + region.imageExtent.depth = 1; + break; + } - ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D); - region.imageSubresource.layerCount = copySize.depth; - region.imageExtent.depth = 1; + default: + UNREACHABLE(); + break; + } return region; } diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/UtilsVulkan.h b/chromium/third_party/dawn/src/dawn_native/vulkan/UtilsVulkan.h index 36ebd34fe67..e57e3f4c768 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/UtilsVulkan.h +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/UtilsVulkan.h @@ -88,10 +88,16 @@ namespace dawn_native { namespace vulkan { VkCompareOp ToVulkanCompareOp(wgpu::CompareFunction op); + VkImageAspectFlags VulkanAspectMask(const Aspect& aspects); + Extent3D ComputeTextureCopyExtent(const TextureCopy& textureCopy, const Extent3D& copySize); + VkBufferImageCopy ComputeBufferImageCopyRegion(const BufferCopy& bufferCopy, const TextureCopy& textureCopy, const Extent3D& copySize); + VkBufferImageCopy ComputeBufferImageCopyRegion(const TextureDataLayout& dataLayout, + const TextureCopy& textureCopy, + const Extent3D& copySize); }} // namespace dawn_native::vulkan diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/external_memory/MemoryServiceOpaqueFD.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/external_memory/MemoryServiceOpaqueFD.cpp index 99057771f63..d43a10f7031 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/external_memory/MemoryServiceOpaqueFD.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/external_memory/MemoryServiceOpaqueFD.cpp @@ -111,7 +111,7 @@ namespace dawn_native { namespace vulkan { namespace external_memory { VkImportMemoryFdInfoKHR importMemoryFdInfo; importMemoryFdInfo.sType = VK_STRUCTURE_TYPE_IMPORT_MEMORY_FD_INFO_KHR; importMemoryFdInfo.pNext = nullptr; - importMemoryFdInfo.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR; + importMemoryFdInfo.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT; importMemoryFdInfo.fd = handle; VkMemoryAllocateInfo allocateInfo; @@ -129,7 +129,13 @@ namespace dawn_native { namespace vulkan { namespace external_memory { ResultOrError<VkImage> Service::CreateImage(const ExternalImageDescriptor* descriptor, const VkImageCreateInfo& baseCreateInfo) { + VkExternalMemoryImageCreateInfo externalMemoryImageCreateInfo; + externalMemoryImageCreateInfo.sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO; + externalMemoryImageCreateInfo.pNext = nullptr; + externalMemoryImageCreateInfo.handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT; + VkImageCreateInfo createInfo = baseCreateInfo; + createInfo.pNext = &externalMemoryImageCreateInfo; createInfo.flags = VK_IMAGE_CREATE_ALIAS_BIT_KHR; createInfo.tiling = VK_IMAGE_TILING_OPTIMAL; createInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; diff --git a/chromium/third_party/dawn/src/dawn_native/vulkan/external_memory/MemoryServiceZirconHandle.cpp b/chromium/third_party/dawn/src/dawn_native/vulkan/external_memory/MemoryServiceZirconHandle.cpp index 08d8d630045..85c4e4a8d3e 100644 --- a/chromium/third_party/dawn/src/dawn_native/vulkan/external_memory/MemoryServiceZirconHandle.cpp +++ b/chromium/third_party/dawn/src/dawn_native/vulkan/external_memory/MemoryServiceZirconHandle.cpp @@ -131,7 +131,14 @@ namespace dawn_native { namespace vulkan { namespace external_memory { ResultOrError<VkImage> Service::CreateImage(const ExternalImageDescriptor* descriptor, const VkImageCreateInfo& baseCreateInfo) { + VkExternalMemoryImageCreateInfo externalMemoryImageCreateInfo; + externalMemoryImageCreateInfo.sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO; + externalMemoryImageCreateInfo.pNext = nullptr; + externalMemoryImageCreateInfo.handleTypes = + VK_EXTERNAL_MEMORY_HANDLE_TYPE_TEMP_ZIRCON_VMO_BIT_FUCHSIA; + VkImageCreateInfo createInfo = baseCreateInfo; + createInfo.pNext = &externalMemoryImageCreateInfo; createInfo.flags = VK_IMAGE_CREATE_ALIAS_BIT_KHR; createInfo.tiling = VK_IMAGE_TILING_OPTIMAL; createInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; diff --git a/chromium/third_party/dawn/src/dawn_platform/BUILD.gn b/chromium/third_party/dawn/src/dawn_platform/BUILD.gn index a152b9ecdea..c2b10933a32 100644 --- a/chromium/third_party/dawn/src/dawn_platform/BUILD.gn +++ b/chromium/third_party/dawn/src/dawn_platform/BUILD.gn @@ -24,7 +24,5 @@ source_set("dawn_platform") { "tracing/TraceEvent.h", ] - deps = [ - "${dawn_root}/src/common", - ] + deps = [ "${dawn_root}/src/common" ] } diff --git a/chromium/third_party/dawn/src/dawn_wire/BUILD.gn b/chromium/third_party/dawn/src/dawn_wire/BUILD.gn index 5d1b2d7f2df..9e463423242 100644 --- a/chromium/third_party/dawn/src/dawn_wire/BUILD.gn +++ b/chromium/third_party/dawn/src/dawn_wire/BUILD.gn @@ -20,9 +20,7 @@ import("${dawn_root}/scripts/dawn_component.gni") # Public dawn_wire headers so they can be publically visible for # dependencies of dawn_wire source_set("dawn_wire_headers") { - public_deps = [ - "${dawn_root}/src/dawn:dawn_headers", - ] + public_deps = [ "${dawn_root}/src/dawn:dawn_headers" ] all_dependent_configs = [ "${dawn_root}/src/common:dawn_public_include_dirs" ] sources = [ "${dawn_root}/src/include/dawn_wire/Wire.h", @@ -39,7 +37,6 @@ dawn_json_generator("dawn_wire_gen") { "src/dawn_wire/WireCmd_autogen.cpp", "src/dawn_wire/client/ApiObjects_autogen.h", "src/dawn_wire/client/ApiProcs_autogen.cpp", - "src/dawn_wire/client/ApiProcs_autogen.h", "src/dawn_wire/client/ClientBase_autogen.h", "src/dawn_wire/client/ClientHandlers_autogen.cpp", "src/dawn_wire/client/ClientPrototypes_autogen.inc", @@ -66,7 +63,6 @@ dawn_component("dawn_wire") { "WireDeserializeAllocator.h", "WireServer.cpp", "client/ApiObjects.h", - "client/ApiProcs.cpp", "client/Buffer.cpp", "client/Buffer.h", "client/Client.cpp", @@ -78,6 +74,8 @@ dawn_component("dawn_wire") { "client/Fence.cpp", "client/Fence.h", "client/ObjectAllocator.h", + "client/Queue.cpp", + "client/Queue.h", "server/ObjectStorage.h", "server/Server.cpp", "server/Server.h", @@ -89,7 +87,5 @@ dawn_component("dawn_wire") { ] # Make headers publicly visible - public_deps = [ - ":dawn_wire_headers", - ] + public_deps = [ ":dawn_wire_headers" ] } diff --git a/chromium/third_party/dawn/src/dawn_wire/CMakeLists.txt b/chromium/third_party/dawn/src/dawn_wire/CMakeLists.txt index 8ec7bff9ab9..e7927fa6b56 100644 --- a/chromium/third_party/dawn/src/dawn_wire/CMakeLists.txt +++ b/chromium/third_party/dawn/src/dawn_wire/CMakeLists.txt @@ -30,7 +30,6 @@ target_sources(dawn_wire PRIVATE "WireDeserializeAllocator.h" "WireServer.cpp" "client/ApiObjects.h" - "client/ApiProcs.cpp" "client/Buffer.cpp" "client/Buffer.h" "client/Client.cpp" @@ -42,6 +41,8 @@ target_sources(dawn_wire PRIVATE "client/Fence.cpp" "client/Fence.h" "client/ObjectAllocator.h" + "client/Queue.cpp" + "client/Queue.h" "server/ObjectStorage.h" "server/Server.cpp" "server/Server.h" diff --git a/chromium/third_party/dawn/src/dawn_wire/client/ApiObjects.h b/chromium/third_party/dawn/src/dawn_wire/client/ApiObjects.h index b74eefe6153..f842d53f469 100644 --- a/chromium/third_party/dawn/src/dawn_wire/client/ApiObjects.h +++ b/chromium/third_party/dawn/src/dawn_wire/client/ApiObjects.h @@ -20,6 +20,7 @@ #include "dawn_wire/client/Buffer.h" #include "dawn_wire/client/Device.h" #include "dawn_wire/client/Fence.h" +#include "dawn_wire/client/Queue.h" #include "dawn_wire/client/ApiObjects_autogen.h" diff --git a/chromium/third_party/dawn/src/dawn_wire/client/ApiProcs.cpp b/chromium/third_party/dawn/src/dawn_wire/client/ApiProcs.cpp deleted file mode 100644 index 358e7175252..00000000000 --- a/chromium/third_party/dawn/src/dawn_wire/client/ApiProcs.cpp +++ /dev/null @@ -1,433 +0,0 @@ -// Copyright 2019 The Dawn Authors -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -#include "dawn_wire/client/ApiObjects.h" -#include "dawn_wire/client/ApiProcs_autogen.h" -#include "dawn_wire/client/Client.h" - -namespace dawn_wire { namespace client { - - namespace { - template <typename Handle> - void SerializeBufferMapAsync(const Buffer* buffer, uint32_t serial, Handle* handle) { - // TODO(enga): Remove the template when Read/Write handles are combined in a tagged - // pointer. - constexpr bool isWrite = - std::is_same<Handle, MemoryTransferService::WriteHandle>::value; - - // Get the serialization size of the handle. - size_t handleCreateInfoLength = handle->SerializeCreateSize(); - - BufferMapAsyncCmd cmd; - cmd.bufferId = buffer->id; - cmd.requestSerial = serial; - cmd.isWrite = isWrite; - cmd.handleCreateInfoLength = handleCreateInfoLength; - cmd.handleCreateInfo = nullptr; - - char* writeHandleSpace = - buffer->device->GetClient()->SerializeCommand(cmd, handleCreateInfoLength); - - // Serialize the handle into the space after the command. - handle->SerializeCreate(writeHandleSpace); - } - } // namespace - - void ClientHandwrittenBufferMapReadAsync(WGPUBuffer cBuffer, - WGPUBufferMapReadCallback callback, - void* userdata) { - Buffer* buffer = reinterpret_cast<Buffer*>(cBuffer); - - uint32_t serial = buffer->requestSerial++; - ASSERT(buffer->requests.find(serial) == buffer->requests.end()); - - if (buffer->size > std::numeric_limits<size_t>::max()) { - // On buffer creation, we check that mappable buffers do not exceed this size. - // So this buffer must not have mappable usage. Inject a validation error. - ClientDeviceInjectError(reinterpret_cast<WGPUDevice>(buffer->device), - WGPUErrorType_Validation, - "Buffer needs the correct map usage bit"); - callback(WGPUBufferMapAsyncStatus_Error, nullptr, 0, userdata); - return; - } - - // Create a ReadHandle for the map request. This is the client's intent to read GPU - // memory. - MemoryTransferService::ReadHandle* readHandle = - buffer->device->GetClient()->GetMemoryTransferService()->CreateReadHandle( - static_cast<size_t>(buffer->size)); - if (readHandle == nullptr) { - ClientDeviceInjectError(reinterpret_cast<WGPUDevice>(buffer->device), - WGPUErrorType_OutOfMemory, "Failed to create buffer mapping"); - callback(WGPUBufferMapAsyncStatus_Error, nullptr, 0, userdata); - return; - } - - Buffer::MapRequestData request = {}; - request.readCallback = callback; - request.userdata = userdata; - // The handle is owned by the MapRequest until the callback returns. - request.readHandle = std::unique_ptr<MemoryTransferService::ReadHandle>(readHandle); - - // Store a mapping from serial -> MapRequest. The client can map/unmap before the map - // operations are returned by the server so multiple requests may be in flight. - buffer->requests[serial] = std::move(request); - - SerializeBufferMapAsync(buffer, serial, readHandle); - } - - void ClientHandwrittenBufferMapWriteAsync(WGPUBuffer cBuffer, - WGPUBufferMapWriteCallback callback, - void* userdata) { - Buffer* buffer = reinterpret_cast<Buffer*>(cBuffer); - - uint32_t serial = buffer->requestSerial++; - ASSERT(buffer->requests.find(serial) == buffer->requests.end()); - - if (buffer->size > std::numeric_limits<size_t>::max()) { - // On buffer creation, we check that mappable buffers do not exceed this size. - // So this buffer must not have mappable usage. Inject a validation error. - ClientDeviceInjectError(reinterpret_cast<WGPUDevice>(buffer->device), - WGPUErrorType_Validation, - "Buffer needs the correct map usage bit"); - callback(WGPUBufferMapAsyncStatus_Error, nullptr, 0, userdata); - return; - } - - // Create a WriteHandle for the map request. This is the client's intent to write GPU - // memory. - MemoryTransferService::WriteHandle* writeHandle = - buffer->device->GetClient()->GetMemoryTransferService()->CreateWriteHandle( - static_cast<size_t>(buffer->size)); - if (writeHandle == nullptr) { - ClientDeviceInjectError(reinterpret_cast<WGPUDevice>(buffer->device), - WGPUErrorType_OutOfMemory, "Failed to create buffer mapping"); - callback(WGPUBufferMapAsyncStatus_Error, nullptr, 0, userdata); - return; - } - - Buffer::MapRequestData request = {}; - request.writeCallback = callback; - request.userdata = userdata; - // The handle is owned by the MapRequest until the callback returns. - request.writeHandle = std::unique_ptr<MemoryTransferService::WriteHandle>(writeHandle); - - // Store a mapping from serial -> MapRequest. The client can map/unmap before the map - // operations are returned by the server so multiple requests may be in flight. - buffer->requests[serial] = std::move(request); - - SerializeBufferMapAsync(buffer, serial, writeHandle); - } - - WGPUBuffer ClientHandwrittenDeviceCreateBuffer(WGPUDevice cDevice, - const WGPUBufferDescriptor* descriptor) { - Device* device = reinterpret_cast<Device*>(cDevice); - Client* wireClient = device->GetClient(); - - if ((descriptor->usage & (WGPUBufferUsage_MapRead | WGPUBufferUsage_MapWrite)) != 0 && - descriptor->size > std::numeric_limits<size_t>::max()) { - ClientDeviceInjectError(cDevice, WGPUErrorType_OutOfMemory, - "Buffer is too large for map usage"); - return ClientDeviceCreateErrorBuffer(cDevice); - } - - auto* bufferObjectAndSerial = wireClient->BufferAllocator().New(device); - Buffer* buffer = bufferObjectAndSerial->object.get(); - // Store the size of the buffer so that mapping operations can allocate a - // MemoryTransfer handle of the proper size. - buffer->size = descriptor->size; - - DeviceCreateBufferCmd cmd; - cmd.self = cDevice; - cmd.descriptor = descriptor; - cmd.result = ObjectHandle{buffer->id, bufferObjectAndSerial->generation}; - - wireClient->SerializeCommand(cmd); - - return reinterpret_cast<WGPUBuffer>(buffer); - } - - WGPUCreateBufferMappedResult ClientHandwrittenDeviceCreateBufferMapped( - WGPUDevice cDevice, - const WGPUBufferDescriptor* descriptor) { - Device* device = reinterpret_cast<Device*>(cDevice); - Client* wireClient = device->GetClient(); - - WGPUCreateBufferMappedResult result; - result.data = nullptr; - result.dataLength = 0; - - // This buffer is too large to be mapped and to make a WriteHandle for. - if (descriptor->size > std::numeric_limits<size_t>::max()) { - ClientDeviceInjectError(cDevice, WGPUErrorType_OutOfMemory, - "Buffer is too large for mapping"); - result.buffer = ClientDeviceCreateErrorBuffer(cDevice); - return result; - } - - // Create a WriteHandle for the map request. This is the client's intent to write GPU - // memory. - std::unique_ptr<MemoryTransferService::WriteHandle> writeHandle = - std::unique_ptr<MemoryTransferService::WriteHandle>( - wireClient->GetMemoryTransferService()->CreateWriteHandle(descriptor->size)); - - if (writeHandle == nullptr) { - ClientDeviceInjectError(cDevice, WGPUErrorType_OutOfMemory, - "Buffer mapping allocation failed"); - result.buffer = ClientDeviceCreateErrorBuffer(cDevice); - return result; - } - - // CreateBufferMapped is synchronous and the staging buffer for upload should be immediately - // available. - // Open the WriteHandle. This returns a pointer and size of mapped memory. - // |result.data| may be null on error. - std::tie(result.data, result.dataLength) = writeHandle->Open(); - if (result.data == nullptr) { - ClientDeviceInjectError(cDevice, WGPUErrorType_OutOfMemory, - "Buffer mapping allocation failed"); - result.buffer = ClientDeviceCreateErrorBuffer(cDevice); - return result; - } - - auto* bufferObjectAndSerial = wireClient->BufferAllocator().New(device); - Buffer* buffer = bufferObjectAndSerial->object.get(); - buffer->size = descriptor->size; - // Successfully created staging memory. The buffer now owns the WriteHandle. - buffer->writeHandle = std::move(writeHandle); - - result.buffer = reinterpret_cast<WGPUBuffer>(buffer); - - // Get the serialization size of the WriteHandle. - size_t handleCreateInfoLength = buffer->writeHandle->SerializeCreateSize(); - - DeviceCreateBufferMappedCmd cmd; - cmd.device = cDevice; - cmd.descriptor = descriptor; - cmd.result = ObjectHandle{buffer->id, bufferObjectAndSerial->generation}; - cmd.handleCreateInfoLength = handleCreateInfoLength; - cmd.handleCreateInfo = nullptr; - - char* writeHandleSpace = - buffer->device->GetClient()->SerializeCommand(cmd, handleCreateInfoLength); - - // Serialize the WriteHandle into the space after the command. - buffer->writeHandle->SerializeCreate(writeHandleSpace); - - return result; - } - - void ClientHandwrittenDevicePushErrorScope(WGPUDevice cDevice, WGPUErrorFilter filter) { - Device* device = reinterpret_cast<Device*>(cDevice); - device->PushErrorScope(filter); - } - - bool ClientHandwrittenDevicePopErrorScope(WGPUDevice cDevice, - WGPUErrorCallback callback, - void* userdata) { - Device* device = reinterpret_cast<Device*>(cDevice); - return device->RequestPopErrorScope(callback, userdata); - } - - uint64_t ClientHandwrittenFenceGetCompletedValue(WGPUFence cSelf) { - auto fence = reinterpret_cast<Fence*>(cSelf); - return fence->completedValue; - } - - void ClientHandwrittenFenceOnCompletion(WGPUFence cFence, - uint64_t value, - WGPUFenceOnCompletionCallback callback, - void* userdata) { - Fence* fence = reinterpret_cast<Fence*>(cFence); - if (value > fence->signaledValue) { - ClientDeviceInjectError(reinterpret_cast<WGPUDevice>(fence->device), - WGPUErrorType_Validation, - "Value greater than fence signaled value"); - callback(WGPUFenceCompletionStatus_Error, userdata); - return; - } - - if (value <= fence->completedValue) { - callback(WGPUFenceCompletionStatus_Success, userdata); - return; - } - - Fence::OnCompletionData request; - request.completionCallback = callback; - request.userdata = userdata; - fence->requests.Enqueue(std::move(request), value); - } - - void ClientHandwrittenBufferSetSubData(WGPUBuffer cBuffer, - uint64_t start, - uint64_t count, - const void* data) { - Buffer* buffer = reinterpret_cast<Buffer*>(cBuffer); - - BufferSetSubDataInternalCmd cmd; - cmd.bufferId = buffer->id; - cmd.start = start; - cmd.count = count; - cmd.data = static_cast<const uint8_t*>(data); - - buffer->device->GetClient()->SerializeCommand(cmd); - } - - void ClientHandwrittenBufferUnmap(WGPUBuffer cBuffer) { - Buffer* buffer = reinterpret_cast<Buffer*>(cBuffer); - - // Invalidate the local pointer, and cancel all other in-flight requests that would - // turn into errors anyway (you can't double map). This prevents race when the following - // happens, where the application code would have unmapped a buffer but still receive a - // callback: - // - Client -> Server: MapRequest1, Unmap, MapRequest2 - // - Server -> Client: Result of MapRequest1 - // - Unmap locally on the client - // - Server -> Client: Result of MapRequest2 - if (buffer->writeHandle) { - // Writes need to be flushed before Unmap is sent. Unmap calls all associated - // in-flight callbacks which may read the updated data. - ASSERT(buffer->readHandle == nullptr); - - // Get the serialization size of metadata to flush writes. - size_t writeFlushInfoLength = buffer->writeHandle->SerializeFlushSize(); - - BufferUpdateMappedDataCmd cmd; - cmd.bufferId = buffer->id; - cmd.writeFlushInfoLength = writeFlushInfoLength; - cmd.writeFlushInfo = nullptr; - - char* writeHandleSpace = - buffer->device->GetClient()->SerializeCommand(cmd, writeFlushInfoLength); - - // Serialize flush metadata into the space after the command. - // This closes the handle for writing. - buffer->writeHandle->SerializeFlush(writeHandleSpace); - buffer->writeHandle = nullptr; - - } else if (buffer->readHandle) { - buffer->readHandle = nullptr; - } - buffer->ClearMapRequests(WGPUBufferMapAsyncStatus_Unknown); - - BufferUnmapCmd cmd; - cmd.self = cBuffer; - buffer->device->GetClient()->SerializeCommand(cmd); - } - - void ClientHandwrittenBufferDestroy(WGPUBuffer cBuffer) { - Buffer* buffer = reinterpret_cast<Buffer*>(cBuffer); - - // Cancel or remove all mappings - buffer->writeHandle = nullptr; - buffer->readHandle = nullptr; - buffer->ClearMapRequests(WGPUBufferMapAsyncStatus_Unknown); - - BufferDestroyCmd cmd; - cmd.self = cBuffer; - buffer->device->GetClient()->SerializeCommand(cmd); - } - - WGPUFence ClientHandwrittenQueueCreateFence(WGPUQueue cSelf, - WGPUFenceDescriptor const* descriptor) { - Queue* queue = reinterpret_cast<Queue*>(cSelf); - Device* device = queue->device; - - QueueCreateFenceCmd cmd; - cmd.self = cSelf; - auto* allocation = device->GetClient()->FenceAllocator().New(device); - cmd.result = ObjectHandle{allocation->object->id, allocation->generation}; - cmd.descriptor = descriptor; - - device->GetClient()->SerializeCommand(cmd); - - WGPUFence cFence = reinterpret_cast<WGPUFence>(allocation->object.get()); - - Fence* fence = reinterpret_cast<Fence*>(cFence); - fence->queue = queue; - - uint64_t initialValue = descriptor != nullptr ? descriptor->initialValue : 0u; - fence->signaledValue = initialValue; - fence->completedValue = initialValue; - return cFence; - } - - void ClientHandwrittenQueueSignal(WGPUQueue cQueue, WGPUFence cFence, uint64_t signalValue) { - Fence* fence = reinterpret_cast<Fence*>(cFence); - Queue* queue = reinterpret_cast<Queue*>(cQueue); - if (fence->queue != queue) { - ClientDeviceInjectError(reinterpret_cast<WGPUDevice>(fence->device), - WGPUErrorType_Validation, - "Fence must be signaled on the queue on which it was created."); - return; - } - if (signalValue <= fence->signaledValue) { - ClientDeviceInjectError(reinterpret_cast<WGPUDevice>(fence->device), - WGPUErrorType_Validation, - "Fence value less than or equal to signaled value"); - return; - } - fence->signaledValue = signalValue; - - QueueSignalCmd cmd; - cmd.self = cQueue; - cmd.fence = cFence; - cmd.signalValue = signalValue; - - queue->device->GetClient()->SerializeCommand(cmd); - } - - void ClientHandwrittenQueueWriteBuffer(WGPUQueue cQueue, - WGPUBuffer cBuffer, - uint64_t bufferOffset, - const void* data, - size_t size) { - Queue* queue = reinterpret_cast<Queue*>(cQueue); - Buffer* buffer = reinterpret_cast<Buffer*>(cBuffer); - - QueueWriteBufferInternalCmd cmd; - cmd.queueId = queue->id; - cmd.bufferId = buffer->id; - cmd.bufferOffset = bufferOffset; - cmd.data = static_cast<const uint8_t*>(data); - cmd.size = size; - - queue->device->GetClient()->SerializeCommand(cmd); - } - - void ClientDeviceReference(WGPUDevice) { - } - - void ClientDeviceRelease(WGPUDevice) { - } - - WGPUQueue ClientHandwrittenDeviceGetDefaultQueue(WGPUDevice cSelf) { - Device* device = reinterpret_cast<Device*>(cSelf); - return device->GetDefaultQueue(); - } - - void ClientHandwrittenDeviceSetUncapturedErrorCallback(WGPUDevice cSelf, - WGPUErrorCallback callback, - void* userdata) { - Device* device = reinterpret_cast<Device*>(cSelf); - device->SetUncapturedErrorCallback(callback, userdata); - } - void ClientHandwrittenDeviceSetDeviceLostCallback(WGPUDevice cSelf, - WGPUDeviceLostCallback callback, - void* userdata) { - Device* device = reinterpret_cast<Device*>(cSelf); - device->SetDeviceLostCallback(callback, userdata); - } - -}} // namespace dawn_wire::client diff --git a/chromium/third_party/dawn/src/dawn_wire/client/Buffer.cpp b/chromium/third_party/dawn/src/dawn_wire/client/Buffer.cpp index 3548f5133fb..86f5944cc4a 100644 --- a/chromium/third_party/dawn/src/dawn_wire/client/Buffer.cpp +++ b/chromium/third_party/dawn/src/dawn_wire/client/Buffer.cpp @@ -14,8 +14,92 @@ #include "dawn_wire/client/Buffer.h" +#include "dawn_wire/client/Client.h" +#include "dawn_wire/client/Device.h" + namespace dawn_wire { namespace client { + // static + WGPUBuffer Buffer::Create(Device* device_, const WGPUBufferDescriptor* descriptor) { + Client* wireClient = device_->GetClient(); + + bool mappable = + (descriptor->usage & (WGPUBufferUsage_MapRead | WGPUBufferUsage_MapWrite)) != 0 || + descriptor->mappedAtCreation; + if (mappable && descriptor->size > std::numeric_limits<size_t>::max()) { + device_->InjectError(WGPUErrorType_OutOfMemory, "Buffer is too large for map usage"); + return device_->CreateErrorBuffer(); + } + + std::unique_ptr<MemoryTransferService::WriteHandle> writeHandle = nullptr; + void* writeData = nullptr; + size_t writeHandleCreateInfoLength = 0; + + // If the buffer is mapped at creation, create a write handle that will represent the + // mapping of the whole buffer. + if (descriptor->mappedAtCreation) { + // Create the handle. + writeHandle.reset( + wireClient->GetMemoryTransferService()->CreateWriteHandle(descriptor->size)); + if (writeHandle == nullptr) { + device_->InjectError(WGPUErrorType_OutOfMemory, "Buffer mapping allocation failed"); + return device_->CreateErrorBuffer(); + } + + // Open the handle, it may fail by returning a nullptr in writeData. + size_t writeDataLength = 0; + std::tie(writeData, writeDataLength) = writeHandle->Open(); + if (writeData == nullptr) { + device_->InjectError(WGPUErrorType_OutOfMemory, "Buffer mapping allocation failed"); + return device_->CreateErrorBuffer(); + } + ASSERT(writeDataLength == descriptor->size); + + // Get the serialization size of the write handle. + writeHandleCreateInfoLength = writeHandle->SerializeCreateSize(); + } + + // Create the buffer and send the creation command. + auto* bufferObjectAndSerial = wireClient->BufferAllocator().New(device_); + Buffer* buffer = bufferObjectAndSerial->object.get(); + buffer->mSize = descriptor->size; + + DeviceCreateBufferCmd cmd; + cmd.device = ToAPI(device_); + cmd.descriptor = descriptor; + cmd.result = ObjectHandle{buffer->id, bufferObjectAndSerial->generation}; + cmd.handleCreateInfoLength = writeHandleCreateInfoLength; + cmd.handleCreateInfo = nullptr; + + char* writeHandleSpace = wireClient->SerializeCommand(cmd, writeHandleCreateInfoLength); + + if (descriptor->mappedAtCreation) { + // Serialize the WriteHandle into the space after the command. + writeHandle->SerializeCreate(writeHandleSpace); + + // Set the buffer state for the mapping at creation. The buffer now owns the write + // handle.. + buffer->mWriteHandle = std::move(writeHandle); + buffer->mMappedData = writeData; + buffer->mMapOffset = 0; + buffer->mMapSize = buffer->mSize; + } + + return ToAPI(buffer); + } + + // static + WGPUBuffer Buffer::CreateError(Device* device_) { + auto* allocation = device_->GetClient()->BufferAllocator().New(device_); + + DeviceCreateErrorBufferCmd cmd; + cmd.self = ToAPI(device_); + cmd.result = ObjectHandle{allocation->object->id, allocation->generation}; + device_->GetClient()->SerializeCommand(cmd); + + return ToAPI(allocation->object.get()); + } + Buffer::~Buffer() { // Callbacks need to be fired in all cases, as they can handle freeing resources // so we call them with "Unknown" status. @@ -23,14 +107,330 @@ namespace dawn_wire { namespace client { } void Buffer::ClearMapRequests(WGPUBufferMapAsyncStatus status) { - for (auto& it : requests) { - if (it.second.writeHandle) { - it.second.writeCallback(status, nullptr, 0, it.second.userdata); + for (auto& it : mRequests) { + if (it.second.callback) { + it.second.callback(status, it.second.userdata); + } + } + mRequests.clear(); + } + + void Buffer::MapReadAsync(WGPUBufferMapReadCallback callback, void* userdata) { + struct ProxyData { + WGPUBufferMapReadCallback callback; + void* userdata; + Buffer* self; + }; + ProxyData* proxy = new ProxyData; + proxy->callback = callback; + proxy->userdata = userdata; + proxy->self = this; + + MapAsync( + WGPUMapMode_Read, 0, mSize, + [](WGPUBufferMapAsyncStatus status, void* userdata) { + ProxyData* proxy = static_cast<ProxyData*>(userdata); + Buffer* self = proxy->self; + + if (status == WGPUBufferMapAsyncStatus_Success) { + // On buffer creation we assert that a mappable buffer cannot be bigger than + // MAX_SIZE_T so we should never have a successful mapping in this case. + ASSERT(self->mSize <= std::numeric_limits<size_t>::max()); + self->mMapOffset = 0; + self->mMapSize = self->mSize; + } + + if (proxy->callback) { + const void* data = self->GetConstMappedRange(0, self->mSize); + uint64_t dataLength = data == nullptr ? 0 : self->mSize; + proxy->callback(status, data, dataLength, proxy->userdata); + } + + delete proxy; + }, + proxy); + } + + void Buffer::MapWriteAsync(WGPUBufferMapWriteCallback callback, void* userdata) { + struct ProxyData { + WGPUBufferMapWriteCallback callback; + void* userdata; + Buffer* self; + }; + ProxyData* proxy = new ProxyData; + proxy->callback = callback; + proxy->userdata = userdata; + proxy->self = this; + + MapAsync( + WGPUMapMode_Write, 0, mSize, + [](WGPUBufferMapAsyncStatus status, void* userdata) { + ProxyData* proxy = static_cast<ProxyData*>(userdata); + Buffer* self = proxy->self; + + if (status == WGPUBufferMapAsyncStatus_Success) { + // On buffer creation we assert that a mappable buffer cannot be bigger than + // MAX_SIZE_T so we should never have a successful mapping in this case. + ASSERT(self->mSize <= std::numeric_limits<size_t>::max()); + self->mMapOffset = 0; + self->mMapSize = self->mSize; + } + + if (proxy->callback) { + void* data = self->GetMappedRange(0, self->mSize); + uint64_t dataLength = data == nullptr ? 0 : self->mSize; + proxy->callback(status, data, dataLength, proxy->userdata); + } + + delete proxy; + }, + proxy); + } + + void Buffer::MapAsync(WGPUMapModeFlags mode, + size_t offset, + size_t size, + WGPUBufferMapCallback callback, + void* userdata) { + // Handle the defaulting of size required by WebGPU. + if (size == 0 && offset < mSize) { + size = mSize - offset; + } + + bool isReadMode = mode & WGPUMapMode_Read; + bool isWriteMode = mode & WGPUMapMode_Write; + + // Step 1. Do early validation of READ ^ WRITE because the server rejects mode = 0. + if (!(isReadMode ^ isWriteMode)) { + device->InjectError(WGPUErrorType_Validation, "MapAsync error (you figure out :P)"); + if (callback != nullptr) { + callback(WGPUBufferMapAsyncStatus_Error, userdata); + } + return; + } + + // Step 2. Create the request structure that will hold information while this mapping is + // in flight. + uint32_t serial = mRequestSerial++; + ASSERT(mRequests.find(serial) == mRequests.end()); + + Buffer::MapRequestData request = {}; + request.callback = callback; + request.userdata = userdata; + request.size = size; + request.offset = offset; + + // Step 2a: Create the read / write handles for this request. + if (isReadMode) { + request.readHandle.reset( + device->GetClient()->GetMemoryTransferService()->CreateReadHandle(size)); + if (request.readHandle == nullptr) { + device->InjectError(WGPUErrorType_OutOfMemory, "Failed to create buffer mapping"); + callback(WGPUBufferMapAsyncStatus_Error, userdata); + return; + } + } else { + ASSERT(isWriteMode); + request.writeHandle.reset( + device->GetClient()->GetMemoryTransferService()->CreateWriteHandle(size)); + if (request.writeHandle == nullptr) { + device->InjectError(WGPUErrorType_OutOfMemory, "Failed to create buffer mapping"); + callback(WGPUBufferMapAsyncStatus_Error, userdata); + return; + } + } + + // Step 3. Serialize the command to send to the server. + BufferMapAsyncCmd cmd; + cmd.bufferId = this->id; + cmd.requestSerial = serial; + cmd.mode = mode; + cmd.offset = offset; + cmd.size = size; + cmd.handleCreateInfo = nullptr; + + // Step 3a. Fill the handle create info in the command. + if (isReadMode) { + cmd.handleCreateInfoLength = request.readHandle->SerializeCreateSize(); + char* handleCreateInfoSpace = + device->GetClient()->SerializeCommand(cmd, cmd.handleCreateInfoLength); + request.readHandle->SerializeCreate(handleCreateInfoSpace); + } else { + ASSERT(isWriteMode); + cmd.handleCreateInfoLength = request.writeHandle->SerializeCreateSize(); + char* handleCreateInfoSpace = + device->GetClient()->SerializeCommand(cmd, cmd.handleCreateInfoLength); + request.writeHandle->SerializeCreate(handleCreateInfoSpace); + } + + // Step 4. Register this request so that we can retrieve it from its serial when the server + // sends the callback. + mRequests[serial] = std::move(request); + } + + bool Buffer::OnMapAsyncCallback(uint32_t requestSerial, + uint32_t status, + uint64_t readInitialDataInfoLength, + const uint8_t* readInitialDataInfo) { + // The requests can have been deleted via an Unmap so this isn't an error. + auto requestIt = mRequests.find(requestSerial); + if (requestIt == mRequests.end()) { + return true; + } + + auto request = std::move(requestIt->second); + // Delete the request before calling the callback otherwise the callback could be fired a + // second time. If, for example, buffer.Unmap() is called inside the callback. + mRequests.erase(requestIt); + + auto FailRequest = [&request]() -> bool { + if (request.callback != nullptr) { + request.callback(WGPUBufferMapAsyncStatus_DeviceLost, request.userdata); + } + return false; + }; + + bool isRead = request.readHandle != nullptr; + bool isWrite = request.writeHandle != nullptr; + ASSERT(isRead != isWrite); + + size_t mappedDataLength = 0; + const void* mappedData = nullptr; + if (status == WGPUBufferMapAsyncStatus_Success) { + if (mReadHandle || mWriteHandle) { + // Buffer is already mapped. + return FailRequest(); + } + + if (isRead) { + if (readInitialDataInfoLength > std::numeric_limits<size_t>::max()) { + // This is the size of data deserialized from the command stream, which must be + // CPU-addressable. + return FailRequest(); + } + + // The server serializes metadata to initialize the contents of the ReadHandle. + // Deserialize the message and return a pointer and size of the mapped data for + // reading. + if (!request.readHandle->DeserializeInitialData( + readInitialDataInfo, static_cast<size_t>(readInitialDataInfoLength), + &mappedData, &mappedDataLength)) { + // Deserialization shouldn't fail. This is a fatal error. + return FailRequest(); + } + ASSERT(mappedData != nullptr); + } else { - it.second.readCallback(status, nullptr, 0, it.second.userdata); + // Open the WriteHandle. This returns a pointer and size of mapped memory. + // On failure, |mappedData| may be null. + std::tie(mappedData, mappedDataLength) = request.writeHandle->Open(); + + if (mappedData == nullptr) { + return FailRequest(); + } } + + // The MapAsync request was successful. The buffer now owns the Read/Write handles + // until Unmap(). + mReadHandle = std::move(request.readHandle); + mWriteHandle = std::move(request.writeHandle); + } + + mMapOffset = request.offset; + mMapSize = request.size; + mMappedData = const_cast<void*>(mappedData); + if (request.callback) { + request.callback(static_cast<WGPUBufferMapAsyncStatus>(status), request.userdata); + } + + return true; + } + + void* Buffer::GetMappedRange(size_t offset, size_t size) { + if (!IsMappedForWriting() || !CheckGetMappedRangeOffsetSize(offset, size)) { + return nullptr; } - requests.clear(); + return static_cast<uint8_t*>(mMappedData) + (offset - mMapOffset); } + const void* Buffer::GetConstMappedRange(size_t offset, size_t size) { + if (!(IsMappedForWriting() || IsMappedForReading()) || + !CheckGetMappedRangeOffsetSize(offset, size)) { + return nullptr; + } + return static_cast<uint8_t*>(mMappedData) + (offset - mMapOffset); + } + + void Buffer::Unmap() { + // Invalidate the local pointer, and cancel all other in-flight requests that would + // turn into errors anyway (you can't double map). This prevents race when the following + // happens, where the application code would have unmapped a buffer but still receive a + // callback: + // - Client -> Server: MapRequest1, Unmap, MapRequest2 + // - Server -> Client: Result of MapRequest1 + // - Unmap locally on the client + // - Server -> Client: Result of MapRequest2 + if (mWriteHandle) { + // Writes need to be flushed before Unmap is sent. Unmap calls all associated + // in-flight callbacks which may read the updated data. + ASSERT(mReadHandle == nullptr); + + // Get the serialization size of metadata to flush writes. + size_t writeFlushInfoLength = mWriteHandle->SerializeFlushSize(); + + BufferUpdateMappedDataCmd cmd; + cmd.bufferId = id; + cmd.writeFlushInfoLength = writeFlushInfoLength; + cmd.writeFlushInfo = nullptr; + + char* writeHandleSpace = + device->GetClient()->SerializeCommand(cmd, writeFlushInfoLength); + + // Serialize flush metadata into the space after the command. + // This closes the handle for writing. + mWriteHandle->SerializeFlush(writeHandleSpace); + mWriteHandle = nullptr; + + } else if (mReadHandle) { + mReadHandle = nullptr; + } + + mMappedData = nullptr; + mMapOffset = 0; + mMapSize = 0; + ClearMapRequests(WGPUBufferMapAsyncStatus_Unknown); + + BufferUnmapCmd cmd; + cmd.self = ToAPI(this); + device->GetClient()->SerializeCommand(cmd); + } + + void Buffer::Destroy() { + // Cancel or remove all mappings + mWriteHandle = nullptr; + mReadHandle = nullptr; + mMappedData = nullptr; + ClearMapRequests(WGPUBufferMapAsyncStatus_Unknown); + + BufferDestroyCmd cmd; + cmd.self = ToAPI(this); + device->GetClient()->SerializeCommand(cmd); + } + + bool Buffer::IsMappedForReading() const { + return mReadHandle != nullptr; + } + + bool Buffer::IsMappedForWriting() const { + return mWriteHandle != nullptr; + } + + bool Buffer::CheckGetMappedRangeOffsetSize(size_t offset, size_t size) const { + if (size > mMapSize || offset < mMapOffset) { + return false; + } + + size_t offsetInMappedRange = offset - mMapOffset; + return offsetInMappedRange <= mMapSize - size; + } }} // namespace dawn_wire::client diff --git a/chromium/third_party/dawn/src/dawn_wire/client/Buffer.h b/chromium/third_party/dawn/src/dawn_wire/client/Buffer.h index 09da8e34977..5c32c751fc7 100644 --- a/chromium/third_party/dawn/src/dawn_wire/client/Buffer.h +++ b/chromium/third_party/dawn/src/dawn_wire/client/Buffer.h @@ -24,33 +24,62 @@ namespace dawn_wire { namespace client { - struct Buffer : ObjectBase { + class Buffer : public ObjectBase { + public: using ObjectBase::ObjectBase; + static WGPUBuffer Create(Device* device, const WGPUBufferDescriptor* descriptor); + static WGPUBuffer CreateError(Device* device); + ~Buffer(); void ClearMapRequests(WGPUBufferMapAsyncStatus status); + void MapReadAsync(WGPUBufferMapReadCallback callback, void* userdata); + void MapWriteAsync(WGPUBufferMapWriteCallback callback, void* userdata); + bool OnMapAsyncCallback(uint32_t requestSerial, + uint32_t status, + uint64_t readInitialDataInfoLength, + const uint8_t* readInitialDataInfo); + void MapAsync(WGPUMapModeFlags mode, + size_t offset, + size_t size, + WGPUBufferMapCallback callback, + void* userdata); + void* GetMappedRange(size_t offset, size_t size); + const void* GetConstMappedRange(size_t offset, size_t size); + void Unmap(); + + void Destroy(); + + private: + bool IsMappedForReading() const; + bool IsMappedForWriting() const; + bool CheckGetMappedRangeOffsetSize(size_t offset, size_t size) const; + // We want to defer all the validation to the server, which means we could have multiple // map request in flight at a single time and need to track them separately. // On well-behaved applications, only one request should exist at a single time. struct MapRequestData { - // TODO(enga): Use a tagged pointer to save space. - WGPUBufferMapReadCallback readCallback = nullptr; - WGPUBufferMapWriteCallback writeCallback = nullptr; + WGPUBufferMapCallback callback = nullptr; void* userdata = nullptr; + size_t offset = 0; + size_t size = 0; // TODO(enga): Use a tagged pointer to save space. std::unique_ptr<MemoryTransferService::ReadHandle> readHandle = nullptr; std::unique_ptr<MemoryTransferService::WriteHandle> writeHandle = nullptr; }; - std::map<uint32_t, MapRequestData> requests; - uint32_t requestSerial = 0; - uint64_t size = 0; + std::map<uint32_t, MapRequestData> mRequests; + uint32_t mRequestSerial = 0; + uint64_t mSize = 0; // Only one mapped pointer can be active at a time because Unmap clears all the in-flight // requests. // TODO(enga): Use a tagged pointer to save space. - std::unique_ptr<MemoryTransferService::ReadHandle> readHandle = nullptr; - std::unique_ptr<MemoryTransferService::WriteHandle> writeHandle = nullptr; + std::unique_ptr<MemoryTransferService::ReadHandle> mReadHandle = nullptr; + std::unique_ptr<MemoryTransferService::WriteHandle> mWriteHandle = nullptr; + void* mMappedData = nullptr; + size_t mMapOffset = 0; + size_t mMapSize = 0; }; }} // namespace dawn_wire::client diff --git a/chromium/third_party/dawn/src/dawn_wire/client/Client.cpp b/chromium/third_party/dawn/src/dawn_wire/client/Client.cpp index 1953347f44c..6b5b7be7ea2 100644 --- a/chromium/third_party/dawn/src/dawn_wire/client/Client.cpp +++ b/chromium/third_party/dawn/src/dawn_wire/client/Client.cpp @@ -20,9 +20,7 @@ namespace dawn_wire { namespace client { Client::Client(CommandSerializer* serializer, MemoryTransferService* memoryTransferService) - : ClientBase(), - mSerializer(serializer), - mMemoryTransferService(memoryTransferService) { + : ClientBase(), mSerializer(serializer), mMemoryTransferService(memoryTransferService) { if (mMemoryTransferService == nullptr) { // If a MemoryTransferService is not provided, fall back to inline memory. mOwnedMemoryTransferService = CreateInlineMemoryTransferService(); @@ -44,11 +42,11 @@ namespace dawn_wire { namespace client { } ReservedTexture Client::ReserveTexture(WGPUDevice cDevice) { - Device* device = reinterpret_cast<Device*>(cDevice); + Device* device = FromAPI(cDevice); ObjectAllocator<Texture>::ObjectAndSerial* allocation = TextureAllocator().New(device); ReservedTexture result; - result.texture = reinterpret_cast<WGPUTexture>(allocation->object.get()); + result.texture = ToAPI(allocation->object.get()); result.id = allocation->object->id; result.generation = allocation->generation; return result; diff --git a/chromium/third_party/dawn/src/dawn_wire/client/ClientDoers.cpp b/chromium/third_party/dawn/src/dawn_wire/client/ClientDoers.cpp index dd904068692..7fdabc08052 100644 --- a/chromium/third_party/dawn/src/dawn_wire/client/ClientDoers.cpp +++ b/chromium/third_party/dawn/src/dawn_wire/client/ClientDoers.cpp @@ -43,144 +43,21 @@ namespace dawn_wire { namespace client { bool Client::DoDevicePopErrorScopeCallback(uint64_t requestSerial, WGPUErrorType errorType, const char* message) { - return mDevice->PopErrorScope(requestSerial, errorType, message); + return mDevice->OnPopErrorScopeCallback(requestSerial, errorType, message); } - bool Client::DoBufferMapReadAsyncCallback(Buffer* buffer, - uint32_t requestSerial, - uint32_t status, - uint64_t initialDataInfoLength, - const uint8_t* initialDataInfo) { + bool Client::DoBufferMapAsyncCallback(Buffer* buffer, + uint32_t requestSerial, + uint32_t status, + uint64_t readInitialDataInfoLength, + const uint8_t* readInitialDataInfo) { // The buffer might have been deleted or recreated so this isn't an error. if (buffer == nullptr) { return true; } - // The requests can have been deleted via an Unmap so this isn't an error. - auto requestIt = buffer->requests.find(requestSerial); - if (requestIt == buffer->requests.end()) { - return true; - } - - auto request = std::move(requestIt->second); - // Delete the request before calling the callback otherwise the callback could be fired a - // second time. If, for example, buffer.Unmap() is called inside the callback. - buffer->requests.erase(requestIt); - - const void* mappedData = nullptr; - size_t mappedDataLength = 0; - - auto GetMappedData = [&]() -> bool { - // It is an error for the server to call the read callback when we asked for a map write - if (request.writeHandle) { - return false; - } - - if (status == WGPUBufferMapAsyncStatus_Success) { - if (buffer->readHandle || buffer->writeHandle) { - // Buffer is already mapped. - return false; - } - if (initialDataInfoLength > std::numeric_limits<size_t>::max()) { - // This is the size of data deserialized from the command stream, which must be - // CPU-addressable. - return false; - } - ASSERT(request.readHandle != nullptr); - - // The server serializes metadata to initialize the contents of the ReadHandle. - // Deserialize the message and return a pointer and size of the mapped data for - // reading. - if (!request.readHandle->DeserializeInitialData( - initialDataInfo, static_cast<size_t>(initialDataInfoLength), &mappedData, - &mappedDataLength)) { - // Deserialization shouldn't fail. This is a fatal error. - return false; - } - ASSERT(mappedData != nullptr); - - // The MapRead request was successful. The buffer now owns the ReadHandle until - // Unmap(). - buffer->readHandle = std::move(request.readHandle); - } - - return true; - }; - - if (!GetMappedData()) { - // Dawn promises that all callbacks are called in finite time. Even if a fatal error - // occurs, the callback is called. - request.readCallback(WGPUBufferMapAsyncStatus_DeviceLost, nullptr, 0, request.userdata); - return false; - } else { - request.readCallback(static_cast<WGPUBufferMapAsyncStatus>(status), mappedData, - static_cast<uint64_t>(mappedDataLength), request.userdata); - return true; - } - } - - bool Client::DoBufferMapWriteAsyncCallback(Buffer* buffer, - uint32_t requestSerial, - uint32_t status) { - // The buffer might have been deleted or recreated so this isn't an error. - if (buffer == nullptr) { - return true; - } - - // The requests can have been deleted via an Unmap so this isn't an error. - auto requestIt = buffer->requests.find(requestSerial); - if (requestIt == buffer->requests.end()) { - return true; - } - - auto request = std::move(requestIt->second); - // Delete the request before calling the callback otherwise the callback could be fired a - // second time. If, for example, buffer.Unmap() is called inside the callback. - buffer->requests.erase(requestIt); - - void* mappedData = nullptr; - size_t mappedDataLength = 0; - - auto GetMappedData = [&]() -> bool { - // It is an error for the server to call the write callback when we asked for a map read - if (request.readHandle) { - return false; - } - - if (status == WGPUBufferMapAsyncStatus_Success) { - if (buffer->readHandle || buffer->writeHandle) { - // Buffer is already mapped. - return false; - } - ASSERT(request.writeHandle != nullptr); - - // Open the WriteHandle. This returns a pointer and size of mapped memory. - // On failure, |mappedData| may be null. - std::tie(mappedData, mappedDataLength) = request.writeHandle->Open(); - - if (mappedData == nullptr) { - return false; - } - - // The MapWrite request was successful. The buffer now owns the WriteHandle until - // Unmap(). - buffer->writeHandle = std::move(request.writeHandle); - } - - return true; - }; - - if (!GetMappedData()) { - // Dawn promises that all callbacks are called in finite time. Even if a fatal error - // occurs, the callback is called. - request.writeCallback(WGPUBufferMapAsyncStatus_DeviceLost, nullptr, 0, - request.userdata); - return false; - } else { - request.writeCallback(static_cast<WGPUBufferMapAsyncStatus>(status), mappedData, - static_cast<uint64_t>(mappedDataLength), request.userdata); - return true; - } + return buffer->OnMapAsyncCallback(requestSerial, status, readInitialDataInfoLength, + readInitialDataInfo); } bool Client::DoFenceUpdateCompletedValue(Fence* fence, uint64_t value) { @@ -189,8 +66,7 @@ namespace dawn_wire { namespace client { return true; } - fence->completedValue = value; - fence->CheckPassedFences(); + fence->OnUpdateCompletedValueCallback(value); return true; } diff --git a/chromium/third_party/dawn/src/dawn_wire/client/Device.cpp b/chromium/third_party/dawn/src/dawn_wire/client/Device.cpp index 43361b3e625..addc9773599 100644 --- a/chromium/third_party/dawn/src/dawn_wire/client/Device.cpp +++ b/chromium/third_party/dawn/src/dawn_wire/client/Device.cpp @@ -31,7 +31,7 @@ namespace dawn_wire { namespace client { mDefaultQueue = allocation->object.get(); DeviceGetDefaultQueueCmd cmd; - cmd.self = reinterpret_cast<WGPUDevice>(this); + cmd.self = ToAPI(this); cmd.result = ObjectHandle{allocation->object->id, allocation->generation}; mClient->SerializeCommand(cmd); @@ -85,13 +85,13 @@ namespace dawn_wire { namespace client { mErrorScopeStackSize++; DevicePushErrorScopeCmd cmd; - cmd.self = reinterpret_cast<WGPUDevice>(this); + cmd.self = ToAPI(this); cmd.filter = filter; mClient->SerializeCommand(cmd); } - bool Device::RequestPopErrorScope(WGPUErrorCallback callback, void* userdata) { + bool Device::PopErrorScope(WGPUErrorCallback callback, void* userdata) { if (mErrorScopeStackSize == 0) { return false; } @@ -103,7 +103,7 @@ namespace dawn_wire { namespace client { mErrorScopes[serial] = {callback, userdata}; DevicePopErrorScopeCmd cmd; - cmd.device = reinterpret_cast<WGPUDevice>(this); + cmd.device = ToAPI(this); cmd.requestSerial = serial; mClient->SerializeCommand(cmd); @@ -111,7 +111,9 @@ namespace dawn_wire { namespace client { return true; } - bool Device::PopErrorScope(uint64_t requestSerial, WGPUErrorType type, const char* message) { + bool Device::OnPopErrorScopeCallback(uint64_t requestSerial, + WGPUErrorType type, + const char* message) { switch (type) { case WGPUErrorType_NoError: case WGPUErrorType_Validation: @@ -135,9 +137,38 @@ namespace dawn_wire { namespace client { return true; } + void Device::InjectError(WGPUErrorType type, const char* message) { + DeviceInjectErrorCmd cmd; + cmd.self = ToAPI(this); + cmd.type = type; + cmd.message = message; + mClient->SerializeCommand(cmd); + } + + WGPUBuffer Device::CreateBuffer(const WGPUBufferDescriptor* descriptor) { + return Buffer::Create(this, descriptor); + } + + WGPUCreateBufferMappedResult Device::CreateBufferMapped( + const WGPUBufferDescriptor* descriptor) { + WGPUBufferDescriptor descMappedAtCreation = *descriptor; + descMappedAtCreation.mappedAtCreation = true; + + WGPUCreateBufferMappedResult result; + result.buffer = CreateBuffer(&descMappedAtCreation); + result.data = FromAPI(result.buffer)->GetMappedRange(0, descriptor->size); + result.dataLength = result.data == nullptr ? 0 : descriptor->size; + + return result; + } + + WGPUBuffer Device::CreateErrorBuffer() { + return Buffer::CreateError(this); + } + WGPUQueue Device::GetDefaultQueue() { mDefaultQueue->refcount++; - return reinterpret_cast<WGPUQueue>(mDefaultQueue); + return ToAPI(mDefaultQueue); } }} // namespace dawn_wire::client diff --git a/chromium/third_party/dawn/src/dawn_wire/client/Device.h b/chromium/third_party/dawn/src/dawn_wire/client/Device.h index f32259a76f7..2b554e34e89 100644 --- a/chromium/third_party/dawn/src/dawn_wire/client/Device.h +++ b/chromium/third_party/dawn/src/dawn_wire/client/Device.h @@ -24,7 +24,7 @@ namespace dawn_wire { namespace client { class Client; - struct Queue; + class Queue; class Device : public ObjectBase { public: @@ -32,14 +32,20 @@ namespace dawn_wire { namespace client { ~Device(); Client* GetClient(); - void HandleError(WGPUErrorType errorType, const char* message); - void HandleDeviceLost(const char* message); void SetUncapturedErrorCallback(WGPUErrorCallback errorCallback, void* errorUserdata); void SetDeviceLostCallback(WGPUDeviceLostCallback errorCallback, void* errorUserdata); - + void InjectError(WGPUErrorType type, const char* message); void PushErrorScope(WGPUErrorFilter filter); - bool RequestPopErrorScope(WGPUErrorCallback callback, void* userdata); - bool PopErrorScope(uint64_t requestSerial, WGPUErrorType type, const char* message); + bool PopErrorScope(WGPUErrorCallback callback, void* userdata); + WGPUBuffer CreateBuffer(const WGPUBufferDescriptor* descriptor); + WGPUCreateBufferMappedResult CreateBufferMapped(const WGPUBufferDescriptor* descriptor); + WGPUBuffer CreateErrorBuffer(); + + void HandleError(WGPUErrorType errorType, const char* message); + void HandleDeviceLost(const char* message); + bool OnPopErrorScopeCallback(uint64_t requestSerial, + WGPUErrorType type, + const char* message); WGPUQueue GetDefaultQueue(); diff --git a/chromium/third_party/dawn/src/dawn_wire/client/Fence.cpp b/chromium/third_party/dawn/src/dawn_wire/client/Fence.cpp index 607483e520f..d9e1fa7ed32 100644 --- a/chromium/third_party/dawn/src/dawn_wire/client/Fence.cpp +++ b/chromium/third_party/dawn/src/dawn_wire/client/Fence.cpp @@ -14,22 +14,74 @@ #include "dawn_wire/client/Fence.h" +#include "dawn_wire/client/Device.h" + namespace dawn_wire { namespace client { Fence::~Fence() { // Callbacks need to be fired in all cases, as they can handle freeing resources // so we call them with "Unknown" status. - for (auto& request : requests.IterateAll()) { + for (auto& request : mRequests.IterateAll()) { request.completionCallback(WGPUFenceCompletionStatus_Unknown, request.userdata); } - requests.Clear(); + mRequests.Clear(); + } + + void Fence::Initialize(Queue* queue, const WGPUFenceDescriptor* descriptor) { + mQueue = queue; + + uint64_t initialValue = descriptor != nullptr ? descriptor->initialValue : 0u; + mSignaledValue = initialValue; + mCompletedValue = initialValue; } void Fence::CheckPassedFences() { - for (auto& request : requests.IterateUpTo(completedValue)) { + for (auto& request : mRequests.IterateUpTo(mCompletedValue)) { request.completionCallback(WGPUFenceCompletionStatus_Success, request.userdata); } - requests.ClearUpTo(completedValue); + mRequests.ClearUpTo(mCompletedValue); + } + + void Fence::OnCompletion(uint64_t value, + WGPUFenceOnCompletionCallback callback, + void* userdata) { + if (value > mSignaledValue) { + device->InjectError(WGPUErrorType_Validation, + "Value greater than fence signaled value"); + callback(WGPUFenceCompletionStatus_Error, userdata); + return; + } + + if (value <= mCompletedValue) { + callback(WGPUFenceCompletionStatus_Success, userdata); + return; + } + + Fence::OnCompletionData request; + request.completionCallback = callback; + request.userdata = userdata; + mRequests.Enqueue(std::move(request), value); + } + + void Fence::OnUpdateCompletedValueCallback(uint64_t value) { + mCompletedValue = value; + CheckPassedFences(); + } + + uint64_t Fence::GetCompletedValue() const { + return mCompletedValue; + } + + uint64_t Fence::GetSignaledValue() const { + return mSignaledValue; + } + + Queue* Fence::GetQueue() const { + return mQueue; + } + + void Fence::SetSignaledValue(uint64_t value) { + mSignaledValue = value; } }} // namespace dawn_wire::client diff --git a/chromium/third_party/dawn/src/dawn_wire/client/Fence.h b/chromium/third_party/dawn/src/dawn_wire/client/Fence.h index 4acde6d6837..107b4e7f440 100644 --- a/chromium/third_party/dawn/src/dawn_wire/client/Fence.h +++ b/chromium/third_party/dawn/src/dawn_wire/client/Fence.h @@ -22,21 +22,33 @@ namespace dawn_wire { namespace client { - struct Queue; - struct Fence : ObjectBase { + class Queue; + class Fence : public ObjectBase { + public: using ObjectBase::ObjectBase; ~Fence(); + void Initialize(Queue* queue, const WGPUFenceDescriptor* descriptor); + void CheckPassedFences(); + void OnCompletion(uint64_t value, WGPUFenceOnCompletionCallback callback, void* userdata); + void OnUpdateCompletedValueCallback(uint64_t value); + + uint64_t GetCompletedValue() const; + uint64_t GetSignaledValue() const; + Queue* GetQueue() const; + + void SetSignaledValue(uint64_t value); + private: struct OnCompletionData { WGPUFenceOnCompletionCallback completionCallback = nullptr; void* userdata = nullptr; }; - Queue* queue = nullptr; - uint64_t signaledValue = 0; - uint64_t completedValue = 0; - SerialMap<OnCompletionData> requests; + Queue* mQueue = nullptr; + uint64_t mSignaledValue = 0; + uint64_t mCompletedValue = 0; + SerialMap<OnCompletionData> mRequests; }; }} // namespace dawn_wire::client diff --git a/chromium/third_party/dawn/src/dawn_wire/client/Queue.cpp b/chromium/third_party/dawn/src/dawn_wire/client/Queue.cpp new file mode 100644 index 00000000000..ad116732ff3 --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_wire/client/Queue.cpp @@ -0,0 +1,91 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "dawn_wire/client/Queue.h" + +#include "dawn_wire/client/Client.h" +#include "dawn_wire/client/Device.h" + +namespace dawn_wire { namespace client { + + WGPUFence Queue::CreateFence(WGPUFenceDescriptor const* descriptor) { + auto* allocation = device->GetClient()->FenceAllocator().New(device); + + QueueCreateFenceCmd cmd; + cmd.self = ToAPI(this); + cmd.result = ObjectHandle{allocation->object->id, allocation->generation}; + cmd.descriptor = descriptor; + device->GetClient()->SerializeCommand(cmd); + + Fence* fence = allocation->object.get(); + fence->Initialize(this, descriptor); + return ToAPI(fence); + } + + void Queue::Signal(WGPUFence cFence, uint64_t signalValue) { + Fence* fence = FromAPI(cFence); + if (fence->GetQueue() != this) { + device->InjectError(WGPUErrorType_Validation, + "Fence must be signaled on the queue on which it was created."); + return; + } + if (signalValue <= fence->GetSignaledValue()) { + device->InjectError(WGPUErrorType_Validation, + "Fence value less than or equal to signaled value"); + return; + } + + fence->SetSignaledValue(signalValue); + + QueueSignalCmd cmd; + cmd.self = ToAPI(this); + cmd.fence = cFence; + cmd.signalValue = signalValue; + + device->GetClient()->SerializeCommand(cmd); + } + + void Queue::WriteBuffer(WGPUBuffer cBuffer, + uint64_t bufferOffset, + const void* data, + size_t size) { + Buffer* buffer = FromAPI(cBuffer); + + QueueWriteBufferInternalCmd cmd; + cmd.queueId = id; + cmd.bufferId = buffer->id; + cmd.bufferOffset = bufferOffset; + cmd.data = static_cast<const uint8_t*>(data); + cmd.size = size; + + device->GetClient()->SerializeCommand(cmd); + } + + void Queue::WriteTexture(const WGPUTextureCopyView* destination, + const void* data, + size_t dataSize, + const WGPUTextureDataLayout* dataLayout, + const WGPUExtent3D* writeSize) { + QueueWriteTextureInternalCmd cmd; + cmd.queueId = id; + cmd.destination = destination; + cmd.data = static_cast<const uint8_t*>(data); + cmd.dataSize = dataSize; + cmd.dataLayout = dataLayout; + cmd.writeSize = writeSize; + + device->GetClient()->SerializeCommand(cmd); + } + +}} // namespace dawn_wire::client diff --git a/chromium/third_party/dawn/src/dawn_wire/client/Queue.h b/chromium/third_party/dawn/src/dawn_wire/client/Queue.h new file mode 100644 index 00000000000..866bccde066 --- /dev/null +++ b/chromium/third_party/dawn/src/dawn_wire/client/Queue.h @@ -0,0 +1,43 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef DAWNWIRE_CLIENT_QUEUE_H_ +#define DAWNWIRE_CLIENT_QUEUE_H_ + +#include <dawn/webgpu.h> + +#include "dawn_wire/WireClient.h" +#include "dawn_wire/client/ObjectBase.h" + +#include <map> + +namespace dawn_wire { namespace client { + + class Queue : public ObjectBase { + public: + using ObjectBase::ObjectBase; + + WGPUFence CreateFence(const WGPUFenceDescriptor* descriptor); + void Signal(WGPUFence fence, uint64_t signalValue); + void WriteBuffer(WGPUBuffer cBuffer, uint64_t bufferOffset, const void* data, size_t size); + void WriteTexture(const WGPUTextureCopyView* destination, + const void* data, + size_t dataSize, + const WGPUTextureDataLayout* dataLayout, + const WGPUExtent3D* writeSize); + }; + +}} // namespace dawn_wire::client + +#endif // DAWNWIRE_CLIENT_QUEUE_H_ diff --git a/chromium/third_party/dawn/src/dawn_wire/server/Server.h b/chromium/third_party/dawn/src/dawn_wire/server/Server.h index 7dd0303df7c..8049e80ada1 100644 --- a/chromium/third_party/dawn/src/dawn_wire/server/Server.h +++ b/chromium/third_party/dawn/src/dawn_wire/server/Server.h @@ -25,8 +25,11 @@ namespace dawn_wire { namespace server { struct MapUserdata { Server* server; ObjectHandle buffer; + WGPUBuffer bufferObj; uint32_t requestSerial; + uint64_t offset; uint64_t size; + WGPUMapModeFlags mode; // TODO(enga): Use a tagged pointer to save space. std::unique_ptr<MemoryTransferService::ReadHandle> readHandle = nullptr; std::unique_ptr<MemoryTransferService::WriteHandle> writeHandle = nullptr; @@ -72,14 +75,7 @@ namespace dawn_wire { namespace server { static void ForwardUncapturedError(WGPUErrorType type, const char* message, void* userdata); static void ForwardDeviceLost(const char* message, void* userdata); static void ForwardPopErrorScope(WGPUErrorType type, const char* message, void* userdata); - static void ForwardBufferMapReadAsync(WGPUBufferMapAsyncStatus status, - const void* ptr, - uint64_t dataLength, - void* userdata); - static void ForwardBufferMapWriteAsync(WGPUBufferMapAsyncStatus status, - void* ptr, - uint64_t dataLength, - void* userdata); + static void ForwardBufferMapAsync(WGPUBufferMapAsyncStatus status, void* userdata); static void ForwardFenceCompletedValue(WGPUFenceCompletionStatus status, void* userdata); // Error callbacks @@ -88,14 +84,7 @@ namespace dawn_wire { namespace server { void OnDevicePopErrorScope(WGPUErrorType type, const char* message, ErrorScopeUserdata* userdata); - void OnBufferMapReadAsyncCallback(WGPUBufferMapAsyncStatus status, - const void* ptr, - uint64_t dataLength, - MapUserdata* userdata); - void OnBufferMapWriteAsyncCallback(WGPUBufferMapAsyncStatus status, - void* ptr, - uint64_t dataLength, - MapUserdata* userdata); + void OnBufferMapAsyncCallback(WGPUBufferMapAsyncStatus status, MapUserdata* userdata); void OnFenceCompletedValueUpdated(WGPUFenceCompletionStatus status, FenceCompletionUserdata* userdata); diff --git a/chromium/third_party/dawn/src/dawn_wire/server/ServerBuffer.cpp b/chromium/third_party/dawn/src/dawn_wire/server/ServerBuffer.cpp index 1b516f96be5..f85ff154147 100644 --- a/chromium/third_party/dawn/src/dawn_wire/server/ServerBuffer.cpp +++ b/chromium/third_party/dawn/src/dawn_wire/server/ServerBuffer.cpp @@ -46,7 +46,9 @@ namespace dawn_wire { namespace server { bool Server::DoBufferMapAsync(ObjectId bufferId, uint32_t requestSerial, - bool isWrite, + WGPUMapModeFlags mode, + size_t offset, + size_t size, uint64_t handleCreateInfoLength, const uint8_t* handleCreateInfo) { // These requests are just forwarded to the buffer, with userdata containing what the @@ -62,6 +64,13 @@ namespace dawn_wire { namespace server { return false; } + // The server only knows how to deal with write XOR read. Validate that. + bool isReadMode = mode & WGPUMapMode_Read; + bool isWriteMode = mode & WGPUMapMode_Write; + if (!(isReadMode ^ isWriteMode)) { + return false; + } + if (handleCreateInfoLength > std::numeric_limits<size_t>::max()) { // This is the size of data deserialized from the command stream, which must be // CPU-addressable. @@ -71,11 +80,15 @@ namespace dawn_wire { namespace server { std::unique_ptr<MapUserdata> userdata = std::make_unique<MapUserdata>(); userdata->server = this; userdata->buffer = ObjectHandle{bufferId, buffer->generation}; + userdata->bufferObj = buffer->handle; userdata->requestSerial = requestSerial; + userdata->offset = offset; + userdata->size = size; + userdata->mode = mode; // The handle will point to the mapped memory or staging memory for the mapping. // Store it on the map request. - if (isWrite) { + if (isWriteMode) { // Deserialize metadata produced from the client to create a companion server handle. MemoryTransferService::WriteHandle* writeHandle = nullptr; if (!mMemoryTransferService->DeserializeWriteHandle( @@ -86,9 +99,8 @@ namespace dawn_wire { namespace server { userdata->writeHandle = std::unique_ptr<MemoryTransferService::WriteHandle>(writeHandle); - mProcs.bufferMapWriteAsync(buffer->handle, ForwardBufferMapWriteAsync, - userdata.release()); } else { + ASSERT(isReadMode); // Deserialize metadata produced from the client to create a companion server handle. MemoryTransferService::ReadHandle* readHandle = nullptr; if (!mMemoryTransferService->DeserializeReadHandle( @@ -98,76 +110,61 @@ namespace dawn_wire { namespace server { ASSERT(readHandle != nullptr); userdata->readHandle = std::unique_ptr<MemoryTransferService::ReadHandle>(readHandle); - mProcs.bufferMapReadAsync(buffer->handle, ForwardBufferMapReadAsync, - userdata.release()); } + mProcs.bufferMapAsync(buffer->handle, mode, offset, size, ForwardBufferMapAsync, + userdata.release()); + return true; } - bool Server::DoDeviceCreateBufferMapped(WGPUDevice device, - const WGPUBufferDescriptor* descriptor, - ObjectHandle bufferResult, - uint64_t handleCreateInfoLength, - const uint8_t* handleCreateInfo) { - if (handleCreateInfoLength > std::numeric_limits<size_t>::max()) { - // This is the size of data deserialized from the command stream, which must be - // CPU-addressable. - return false; - } - + bool Server::DoDeviceCreateBuffer(WGPUDevice device, + const WGPUBufferDescriptor* descriptor, + ObjectHandle bufferResult, + uint64_t handleCreateInfoLength, + const uint8_t* handleCreateInfo) { + // Create and register the buffer object. auto* resultData = BufferObjects().Allocate(bufferResult.id); if (resultData == nullptr) { return false; } resultData->generation = bufferResult.generation; + resultData->handle = mProcs.deviceCreateBuffer(device, descriptor); - WGPUCreateBufferMappedResult result = mProcs.deviceCreateBufferMapped(device, descriptor); - ASSERT(result.buffer != nullptr); - if (result.data == nullptr && result.dataLength != 0) { - // Non-zero dataLength but null data is used to indicate an allocation error. - // Don't return false because this is not fatal. result.buffer is an ErrorBuffer - // and subsequent operations will be errors. - // This should only happen when fuzzing with the Null backend. - resultData->mapWriteState = BufferMapWriteState::MapError; - } else { - // Deserialize metadata produced from the client to create a companion server handle. - MemoryTransferService::WriteHandle* writeHandle = nullptr; - if (!mMemoryTransferService->DeserializeWriteHandle( - handleCreateInfo, static_cast<size_t>(handleCreateInfoLength), &writeHandle)) { - return false; - } - ASSERT(writeHandle != nullptr); - - // Set the target of the WriteHandle to the mapped GPU memory. - writeHandle->SetTarget(result.data, result.dataLength); - - // The buffer is mapped and has a valid mappedData pointer. - // The buffer may still be an error with fake staging data. - resultData->mapWriteState = BufferMapWriteState::Mapped; - resultData->writeHandle = - std::unique_ptr<MemoryTransferService::WriteHandle>(writeHandle); + // If the buffer isn't mapped at creation, we are done. + if (!descriptor->mappedAtCreation) { + return handleCreateInfoLength == 0; } - resultData->handle = result.buffer; - return true; - } - - bool Server::DoBufferSetSubDataInternal(ObjectId bufferId, - uint64_t start, - uint64_t offset, - const uint8_t* data) { - // The null object isn't valid as `self` - if (bufferId == 0) { + // This is the size of data deserialized from the command stream to create the write handle, + // which must be CPU-addressable. + if (handleCreateInfoLength > std::numeric_limits<size_t>::max()) { return false; } - auto* buffer = BufferObjects().Get(bufferId); - if (buffer == nullptr) { + void* mapping = mProcs.bufferGetMappedRange(resultData->handle, 0, descriptor->size); + if (mapping == nullptr) { + // A zero mapping is used to indicate an allocation error of an error buffer. This is a + // valid case and isn't fatal. Remember the buffer is an error so as to skip subsequent + // mapping operations. + resultData->mapWriteState = BufferMapWriteState::MapError; + return true; + } + + // Deserialize metadata produced from the client to create a companion server handle. + MemoryTransferService::WriteHandle* writeHandle = nullptr; + if (!mMemoryTransferService->DeserializeWriteHandle( + handleCreateInfo, static_cast<size_t>(handleCreateInfoLength), &writeHandle)) { return false; } - mProcs.bufferSetSubData(buffer->handle, start, offset, data); + // Set the target of the WriteHandle to the mapped GPU memory. + ASSERT(writeHandle != nullptr); + writeHandle->SetTarget(mapping, descriptor->size); + + resultData->mapWriteState = BufferMapWriteState::Mapped; + resultData->writeHandle.reset(writeHandle); + return true; } @@ -209,26 +206,12 @@ namespace dawn_wire { namespace server { static_cast<size_t>(writeFlushInfoLength)); } - void Server::ForwardBufferMapReadAsync(WGPUBufferMapAsyncStatus status, - const void* ptr, - uint64_t dataLength, - void* userdata) { - auto data = static_cast<MapUserdata*>(userdata); - data->server->OnBufferMapReadAsyncCallback(status, ptr, dataLength, data); - } - - void Server::ForwardBufferMapWriteAsync(WGPUBufferMapAsyncStatus status, - void* ptr, - uint64_t dataLength, - void* userdata) { + void Server::ForwardBufferMapAsync(WGPUBufferMapAsyncStatus status, void* userdata) { auto data = static_cast<MapUserdata*>(userdata); - data->server->OnBufferMapWriteAsyncCallback(status, ptr, dataLength, data); + data->server->OnBufferMapAsyncCallback(status, data); } - void Server::OnBufferMapReadAsyncCallback(WGPUBufferMapAsyncStatus status, - const void* ptr, - uint64_t dataLength, - MapUserdata* userdata) { + void Server::OnBufferMapAsyncCallback(WGPUBufferMapAsyncStatus status, MapUserdata* userdata) { std::unique_ptr<MapUserdata> data(userdata); // Skip sending the callback if the buffer has already been destroyed. @@ -237,59 +220,43 @@ namespace dawn_wire { namespace server { return; } - size_t initialDataInfoLength = 0; - if (status == WGPUBufferMapAsyncStatus_Success) { - // Get the serialization size of the message to initialize ReadHandle data. - initialDataInfoLength = data->readHandle->SerializeInitialDataSize(ptr, dataLength); - } else { - dataLength = 0; - } + bool isRead = data->mode & WGPUMapMode_Read; + bool isSuccess = status == WGPUBufferMapAsyncStatus_Success; - ReturnBufferMapReadAsyncCallbackCmd cmd; + ReturnBufferMapAsyncCallbackCmd cmd; cmd.buffer = data->buffer; cmd.requestSerial = data->requestSerial; cmd.status = status; - cmd.initialDataInfoLength = initialDataInfoLength; - cmd.initialDataInfo = nullptr; - - char* readHandleSpace = SerializeCommand(cmd, initialDataInfoLength); + cmd.readInitialDataInfoLength = 0; + cmd.readInitialDataInfo = nullptr; - if (status == WGPUBufferMapAsyncStatus_Success) { - // Serialize the initialization message into the space after the command. - data->readHandle->SerializeInitialData(ptr, dataLength, readHandleSpace); - - // The in-flight map request returned successfully. - // Move the ReadHandle so it is owned by the buffer. - bufferData->readHandle = std::move(data->readHandle); - } - } - - void Server::OnBufferMapWriteAsyncCallback(WGPUBufferMapAsyncStatus status, - void* ptr, - uint64_t dataLength, - MapUserdata* userdata) { - std::unique_ptr<MapUserdata> data(userdata); - - // Skip sending the callback if the buffer has already been destroyed. - auto* bufferData = BufferObjects().Get(data->buffer.id); - if (bufferData == nullptr || bufferData->generation != data->buffer.generation) { - return; + const void* readData = nullptr; + if (isSuccess && isRead) { + // Get the serialization size of the message to initialize ReadHandle data. + readData = mProcs.bufferGetConstMappedRange(data->bufferObj, data->offset, data->size); + cmd.readInitialDataInfoLength = + data->readHandle->SerializeInitialDataSize(readData, data->size); } - ReturnBufferMapWriteAsyncCallbackCmd cmd; - cmd.buffer = data->buffer; - cmd.requestSerial = data->requestSerial; - cmd.status = status; - - SerializeCommand(cmd); - - if (status == WGPUBufferMapAsyncStatus_Success) { - // The in-flight map request returned successfully. - // Move the WriteHandle so it is owned by the buffer. - bufferData->writeHandle = std::move(data->writeHandle); - bufferData->mapWriteState = BufferMapWriteState::Mapped; - // Set the target of the WriteHandle to the mapped buffer data. - bufferData->writeHandle->SetTarget(ptr, dataLength); + char* readHandleSpace = SerializeCommand(cmd, cmd.readInitialDataInfoLength); + + if (isSuccess) { + if (isRead) { + // Serialize the initialization message into the space after the command. + data->readHandle->SerializeInitialData(readData, data->size, readHandleSpace); + // The in-flight map request returned successfully. + // Move the ReadHandle so it is owned by the buffer. + bufferData->readHandle = std::move(data->readHandle); + } else { + // The in-flight map request returned successfully. + // Move the WriteHandle so it is owned by the buffer. + bufferData->writeHandle = std::move(data->writeHandle); + bufferData->mapWriteState = BufferMapWriteState::Mapped; + // Set the target of the WriteHandle to the mapped buffer data. + bufferData->writeHandle->SetTarget( + mProcs.bufferGetMappedRange(data->bufferObj, data->offset, data->size), + data->size); + } } } diff --git a/chromium/third_party/dawn/src/dawn_wire/server/ServerQueue.cpp b/chromium/third_party/dawn/src/dawn_wire/server/ServerQueue.cpp index 6e47492d3b0..0c5a6b880f8 100644 --- a/chromium/third_party/dawn/src/dawn_wire/server/ServerQueue.cpp +++ b/chromium/third_party/dawn/src/dawn_wire/server/ServerQueue.cpp @@ -55,4 +55,21 @@ namespace dawn_wire { namespace server { return true; } + bool Server::DoQueueWriteTextureInternal(ObjectId queueId, + const WGPUTextureCopyView* destination, + const uint8_t* data, + size_t dataSize, + const WGPUTextureDataLayout* dataLayout, + const WGPUExtent3D* writeSize) { + // The null object isn't valid as `self` so we can combine the check with the + // check that the ID is valid. + auto* queue = QueueObjects().Get(queueId); + if (queue == nullptr) { + return false; + } + + mProcs.queueWriteTexture(queue->handle, destination, data, dataSize, dataLayout, writeSize); + return true; + } + }} // namespace dawn_wire::server diff --git a/chromium/third_party/dawn/src/include/dawn_native/D3D12Backend.h b/chromium/third_party/dawn/src/include/dawn_native/D3D12Backend.h index bdefdf37be8..035ad96188e 100644 --- a/chromium/third_party/dawn/src/include/dawn_native/D3D12Backend.h +++ b/chromium/third_party/dawn/src/include/dawn_native/D3D12Backend.h @@ -18,6 +18,7 @@ #include <dawn/dawn_wsi.h> #include <dawn_native/DawnNative.h> +#include <DXGI1_4.h> #include <windows.h> #include <wrl/client.h> @@ -52,6 +53,12 @@ namespace dawn_native { namespace d3d12 { DAWN_NATIVE_EXPORT WGPUTexture WrapSharedHandle(WGPUDevice device, const ExternalImageDescriptorDXGISharedHandle* descriptor); + struct DAWN_NATIVE_EXPORT AdapterDiscoveryOptions : public AdapterDiscoveryOptionsBase { + AdapterDiscoveryOptions(Microsoft::WRL::ComPtr<IDXGIAdapter> adapter); + + Microsoft::WRL::ComPtr<IDXGIAdapter> dxgiAdapter; + }; + }} // namespace dawn_native::d3d12 #endif // DAWNNATIVE_D3D12BACKEND_H_ diff --git a/chromium/third_party/dawn/src/include/dawn_native/DawnNative.h b/chromium/third_party/dawn/src/include/dawn_native/DawnNative.h index 219e53fa94b..a57baeb1193 100644 --- a/chromium/third_party/dawn/src/include/dawn_native/DawnNative.h +++ b/chromium/third_party/dawn/src/include/dawn_native/DawnNative.h @@ -157,6 +157,9 @@ namespace dawn_native { // Enable debug capture on Dawn startup void EnableBeginCaptureOnStartup(bool beginCaptureOnStartup); + // Enable GPU based backend validation if it has. + void EnableGPUBasedBackendValidation(bool enableGPUBasedBackendValidation); + void SetPlatform(dawn_platform::Platform* platform); // Returns the underlying WGPUInstance object. @@ -179,11 +182,13 @@ namespace dawn_native { DAWN_NATIVE_EXPORT size_t GetDeprecationWarningCountForTesting(WGPUDevice device); // Query if texture has been initialized - DAWN_NATIVE_EXPORT bool IsTextureSubresourceInitialized(WGPUTexture texture, - uint32_t baseMipLevel, - uint32_t levelCount, - uint32_t baseArrayLayer, - uint32_t layerCount); + DAWN_NATIVE_EXPORT bool IsTextureSubresourceInitialized( + WGPUTexture texture, + uint32_t baseMipLevel, + uint32_t levelCount, + uint32_t baseArrayLayer, + uint32_t layerCount, + WGPUTextureAspect aspect = WGPUTextureAspect_All); // Backdoor to get the order of the ProcMap for testing DAWN_NATIVE_EXPORT std::vector<const char*> GetProcMapNamesForTesting(); diff --git a/chromium/third_party/dawn/src/include/dawn_native/VulkanBackend.h b/chromium/third_party/dawn/src/include/dawn_native/VulkanBackend.h index 0965871941c..4e5aee995e8 100644 --- a/chromium/third_party/dawn/src/include/dawn_native/VulkanBackend.h +++ b/chromium/third_party/dawn/src/include/dawn_native/VulkanBackend.h @@ -23,6 +23,7 @@ #include <vector> namespace dawn_native { namespace vulkan { + DAWN_NATIVE_EXPORT VkInstance GetInstance(WGPUDevice device); DAWN_NATIVE_EXPORT PFN_vkVoidFunction GetInstanceProcAddr(WGPUDevice device, const char* pName); @@ -75,6 +76,7 @@ namespace dawn_native { namespace vulkan { DAWN_NATIVE_EXPORT WGPUTexture WrapVulkanImage(WGPUDevice cDevice, const ExternalImageDescriptor* descriptor); #endif // __linux__ + }} // namespace dawn_native::vulkan #endif // DAWNNATIVE_VULKANBACKEND_H_ diff --git a/chromium/third_party/dawn/src/include/dawn_wire/WireClient.h b/chromium/third_party/dawn/src/include/dawn_wire/WireClient.h index 5b5f33c2ace..815b66b877a 100644 --- a/chromium/third_party/dawn/src/include/dawn_wire/WireClient.h +++ b/chromium/third_party/dawn/src/include/dawn_wire/WireClient.h @@ -26,7 +26,7 @@ namespace dawn_wire { namespace client { class Client; class MemoryTransferService; - } + } // namespace client struct ReservedTexture { WGPUTexture texture; diff --git a/chromium/third_party/dawn/src/include/dawn_wire/WireServer.h b/chromium/third_party/dawn/src/include/dawn_wire/WireServer.h index b59572180a8..b3d1e62a1ff 100644 --- a/chromium/third_party/dawn/src/include/dawn_wire/WireServer.h +++ b/chromium/third_party/dawn/src/include/dawn_wire/WireServer.h @@ -26,7 +26,7 @@ namespace dawn_wire { namespace server { class Server; class MemoryTransferService; - } + } // namespace server struct DAWN_WIRE_EXPORT WireServerDescriptor { WGPUDevice device; diff --git a/chromium/third_party/dawn/src/include/webgpu/webgpu.h b/chromium/third_party/dawn/src/include/webgpu/webgpu.h new file mode 100644 index 00000000000..4a29d373468 --- /dev/null +++ b/chromium/third_party/dawn/src/include/webgpu/webgpu.h @@ -0,0 +1 @@ +#include "dawn/webgpu.h" diff --git a/chromium/third_party/dawn/src/include/webgpu/webgpu_cpp.h b/chromium/third_party/dawn/src/include/webgpu/webgpu_cpp.h new file mode 100644 index 00000000000..5bbd869b944 --- /dev/null +++ b/chromium/third_party/dawn/src/include/webgpu/webgpu_cpp.h @@ -0,0 +1 @@ +#include <dawn/webgpu_cpp.h> diff --git a/chromium/third_party/dawn/src/tests/BUILD.gn b/chromium/third_party/dawn/src/tests/BUILD.gn index a1e77415ccc..7817d8ac70c 100644 --- a/chromium/third_party/dawn/src/tests/BUILD.gn +++ b/chromium/third_party/dawn/src/tests/BUILD.gn @@ -156,12 +156,14 @@ test("dawn_unittests") { "unittests/BuddyMemoryAllocatorTests.cpp", "unittests/CommandAllocatorTests.cpp", "unittests/EnumClassBitmasksTests.cpp", + "unittests/EnumMaskIteratorTests.cpp", "unittests/ErrorTests.cpp", "unittests/ExtensionTests.cpp", "unittests/GetProcAddressTests.cpp", "unittests/ITypArrayTests.cpp", "unittests/ITypBitsetTests.cpp", "unittests/ITypSpanTests.cpp", + "unittests/ITypVectorTests.cpp", "unittests/LinkedListTests.cpp", "unittests/MathTests.cpp", "unittests/ObjectBaseTests.cpp", @@ -173,6 +175,7 @@ test("dawn_unittests") { "unittests/SerialMapTests.cpp", "unittests/SerialQueueTests.cpp", "unittests/SlabAllocatorTests.cpp", + "unittests/StackContainerTests.cpp", "unittests/SystemUtilsTests.cpp", "unittests/ToBackendTests.cpp", "unittests/TypedIntegerTests.cpp", @@ -192,6 +195,7 @@ test("dawn_unittests") { "unittests/validation/MinimumBufferSizeValidationTests.cpp", "unittests/validation/QuerySetValidationTests.cpp", "unittests/validation/QueueSubmitValidationTests.cpp", + "unittests/validation/QueueWriteTextureValidationTests.cpp", "unittests/validation/RenderBundleValidationTests.cpp", "unittests/validation/RenderPassDescriptorValidationTests.cpp", "unittests/validation/RenderPipelineValidationTests.cpp", @@ -259,6 +263,7 @@ source_set("dawn_end2end_tests_sources") { "end2end/BasicTests.cpp", "end2end/BindGroupTests.cpp", "end2end/BufferTests.cpp", + "end2end/BufferZeroInitTests.cpp", "end2end/ClipSpaceTests.cpp", "end2end/ColorStateTests.cpp", "end2end/CompressedTextureFormatTests.cpp", @@ -271,6 +276,7 @@ source_set("dawn_end2end_tests_sources") { "end2end/DebugMarkerTests.cpp", "end2end/DeprecatedAPITests.cpp", "end2end/DepthSamplingTests.cpp", + "end2end/DepthStencilCopyTests.cpp", "end2end/DepthStencilStateTests.cpp", "end2end/DestroyTests.cpp", "end2end/DeviceLostTests.cpp", @@ -290,6 +296,7 @@ source_set("dawn_end2end_tests_sources") { "end2end/OpArrayLengthTests.cpp", "end2end/PipelineLayoutTests.cpp", "end2end/PrimitiveTopologyTests.cpp", + "end2end/QueryTests.cpp", "end2end/QueueTests.cpp", "end2end/RenderBundleTests.cpp", "end2end/RenderPassLoadOpTests.cpp", @@ -327,7 +334,11 @@ source_set("dawn_end2end_tests_sources") { if (dawn_enable_metal) { sources += [ "end2end/IOSurfaceWrappingTests.cpp" ] - libs += [ "IOSurface.framework" ] + frameworks = [ "IOSurface.framework" ] + } + + if (dawn_enable_wgsl) { + sources += [ "end2end/VertexBufferRobustnessTests.cpp" ] } if (dawn_enable_opengl) { @@ -470,16 +481,17 @@ test("dawn_perf_tests") { libs = [] - # When building inside Chromium, use their gtest main function because it is - # needed to run in swarming correctly. + # When building inside Chromium, use their gtest main function and the + # other perf test scaffolding in order to run in swarming correctly. if (build_with_chromium) { deps += [ ":dawn_perf_tests_main" ] + data_deps = [ "//testing:run_perf_test" ] } else { sources += [ "PerfTestsMain.cpp" ] } if (dawn_enable_metal) { - libs += [ "IOSurface.framework" ] + frameworks = [ "IOSurface.framework" ] } if (dawn_enable_opengl) { diff --git a/chromium/third_party/dawn/src/utils/BUILD.gn b/chromium/third_party/dawn/src/utils/BUILD.gn index 9d5d69c7541..38fd58797be 100644 --- a/chromium/third_party/dawn/src/utils/BUILD.gn +++ b/chromium/third_party/dawn/src/utils/BUILD.gn @@ -25,9 +25,7 @@ import("${dawn_root}/scripts/dawn_features.gni") # use the real library from third_party/. if (dawn_supports_glfw_for_windowing) { group("dawn_glfw") { - public_deps = [ - "${dawn_root}/third_party/gn/glfw", - ] + public_deps = [ "${dawn_root}/third_party/gn/glfw" ] } } else if (is_fuchsia) { # The mock implementation of GLFW on Fuchsia @@ -52,9 +50,7 @@ if (dawn_supports_glfw_for_windowing) { "Glfw3Fuchsia.cpp", ] public_configs = [ ":dawn_glfw_public_config" ] - deps = [ - "${dawn_root}/src/common", - ] + deps = [ "${dawn_root}/src/common" ] } } else { # Just skip GLFW on other systems @@ -74,10 +70,13 @@ static_library("dawn_utils") { "ComboRenderBundleEncoderDescriptor.h", "ComboRenderPipelineDescriptor.cpp", "ComboRenderPipelineDescriptor.h", + "PlatformDebugLogger.h", "SystemUtils.cpp", "SystemUtils.h", "TerribleCommandBuffer.cpp", "TerribleCommandBuffer.h", + "TestUtils.cpp", + "TestUtils.h", "TextureFormatUtils.cpp", "TextureFormatUtils.h", "Timer.h", @@ -91,6 +90,13 @@ static_library("dawn_utils") { "${dawn_shaderc_dir}:libshaderc", ] libs = [] + frameworks = [] + + if (is_win) { + sources += [ "WindowsDebugLogger.cpp" ] + } else { + sources += [ "EmptyDebugLogger.cpp" ] + } if (is_win) { sources += [ "WindowsTimer.cpp" ] @@ -100,7 +106,7 @@ static_library("dawn_utils") { "ObjCUtils.h", "ObjCUtils.mm", ] - libs += [ "QuartzCore.framework" ] + frameworks += [ "QuartzCore.framework" ] } else { sources += [ "PosixTimer.cpp" ] } @@ -114,13 +120,11 @@ static_library("dawn_utils") { if (dawn_enable_metal) { sources += [ "GLFWUtils_metal.mm" ] - libs += [ "Metal.framework" ] + frameworks += [ "Metal.framework" ] } } - public_deps = [ - "${dawn_root}/src/dawn:dawncpp_headers", - ] + public_deps = [ "${dawn_root}/src/dawn:dawncpp_headers" ] } ############################################################################### @@ -137,9 +141,7 @@ if (dawn_standalone) { "BackendBinding.h", ] - public_deps = [ - "${dawn_root}/src/dawn:dawn_headers", - ] + public_deps = [ "${dawn_root}/src/dawn:dawn_headers" ] deps = [ ":dawn_glfw", @@ -147,6 +149,7 @@ if (dawn_standalone) { "${dawn_root}/src/dawn_native", ] libs = [] + frameworks = [] if (dawn_enable_d3d12) { sources += [ "D3D12Binding.cpp" ] @@ -154,7 +157,7 @@ if (dawn_standalone) { if (dawn_enable_metal) { sources += [ "MetalBinding.mm" ] - libs += [ + frameworks += [ "Metal.framework", "QuartzCore.framework", ] diff --git a/chromium/third_party/dawn/src/utils/CMakeLists.txt b/chromium/third_party/dawn/src/utils/CMakeLists.txt index 3a959401a2b..e215f516377 100644 --- a/chromium/third_party/dawn/src/utils/CMakeLists.txt +++ b/chromium/third_party/dawn/src/utils/CMakeLists.txt @@ -22,10 +22,13 @@ target_sources(dawn_utils PRIVATE "ComboRenderPipelineDescriptor.h" "GLFWUtils.cpp" "GLFWUtils.h" + "PlatformDebugLogger.h" "SystemUtils.cpp" "SystemUtils.h" "TerribleCommandBuffer.cpp" "TerribleCommandBuffer.h" + "TestUtils.cpp" + "TestUtils.h" "TextureFormatUtils.cpp" "TextureFormatUtils.h" "Timer.h" @@ -43,6 +46,12 @@ target_link_libraries(dawn_utils ) if(WIN32) + target_sources(dawn_utils PRIVATE "WindowsDebugLogger.cpp") +else() + target_sources(dawn_utils PRIVATE "EmptyDebugLogger.cpp") +endif() + +if(WIN32) target_sources(dawn_utils PRIVATE "WindowsTimer.cpp") elseif(APPLE) target_sources(dawn_utils PRIVATE diff --git a/chromium/third_party/dawn/src/utils/EmptyDebugLogger.cpp b/chromium/third_party/dawn/src/utils/EmptyDebugLogger.cpp new file mode 100644 index 00000000000..ed0ad7ff532 --- /dev/null +++ b/chromium/third_party/dawn/src/utils/EmptyDebugLogger.cpp @@ -0,0 +1,29 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "utils/PlatformDebugLogger.h" + +namespace utils { + + class EmptyDebugLogger : public PlatformDebugLogger { + public: + EmptyDebugLogger() = default; + ~EmptyDebugLogger() override = default; + }; + + PlatformDebugLogger* CreatePlatformDebugLogger() { + return new EmptyDebugLogger(); + } + +} // namespace utils diff --git a/chromium/third_party/dawn/src/utils/PlatformDebugLogger.h b/chromium/third_party/dawn/src/utils/PlatformDebugLogger.h new file mode 100644 index 00000000000..33c46dec44b --- /dev/null +++ b/chromium/third_party/dawn/src/utils/PlatformDebugLogger.h @@ -0,0 +1,29 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef UTILS_PLATFORMDEBUGLOGGER_H_ +#define UTILS_PLATFORMDEBUGLOGGER_H_ + +namespace utils { + + class PlatformDebugLogger { + public: + virtual ~PlatformDebugLogger() = default; + }; + + PlatformDebugLogger* CreatePlatformDebugLogger(); + +} // namespace utils + +#endif // UTILS_PLATFORMDEBUGLOGGER_H_ diff --git a/chromium/third_party/dawn/src/utils/SystemUtils.h b/chromium/third_party/dawn/src/utils/SystemUtils.h index 828eb58aabf..1f42cc539a6 100644 --- a/chromium/third_party/dawn/src/utils/SystemUtils.h +++ b/chromium/third_party/dawn/src/utils/SystemUtils.h @@ -20,4 +20,4 @@ namespace utils { void USleep(unsigned int usecs); } -#endif // UTILS_SYSTEMUTILS_H_
\ No newline at end of file +#endif // UTILS_SYSTEMUTILS_H_ diff --git a/chromium/third_party/dawn/src/utils/TestUtils.cpp b/chromium/third_party/dawn/src/utils/TestUtils.cpp new file mode 100644 index 00000000000..cdbe3b3fc85 --- /dev/null +++ b/chromium/third_party/dawn/src/utils/TestUtils.cpp @@ -0,0 +1,107 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "utils/TestUtils.h" + +#include "common/Assert.h" +#include "common/Constants.h" +#include "common/Math.h" +#include "utils/TextureFormatUtils.h" +#include "utils/WGPUHelpers.h" + +#include <vector> + +namespace utils { + + uint32_t GetMinimumBytesPerRow(wgpu::TextureFormat format, uint32_t width) { + const uint32_t bytesPerTexel = utils::GetTexelBlockSizeInBytes(format); + return Align(bytesPerTexel * width, kTextureBytesPerRowAlignment); + } + + uint32_t GetBytesInBufferTextureCopy(wgpu::TextureFormat format, + uint32_t width, + uint32_t bytesPerRow, + uint32_t rowsPerImage, + uint32_t copyArrayLayerCount) { + ASSERT(rowsPerImage > 0); + const uint32_t bytesPerTexel = utils::GetTexelBlockSizeInBytes(format); + const uint32_t bytesAtLastImage = bytesPerRow * (rowsPerImage - 1) + bytesPerTexel * width; + return bytesPerRow * rowsPerImage * (copyArrayLayerCount - 1) + bytesAtLastImage; + } + + // TODO(jiawei.shao@intel.com): support compressed texture formats + TextureDataCopyLayout GetTextureDataCopyLayoutForTexture2DAtLevel( + wgpu::TextureFormat format, + wgpu::Extent3D textureSizeAtLevel0, + uint32_t mipmapLevel, + uint32_t rowsPerImage) { + TextureDataCopyLayout layout; + + layout.mipSize = {textureSizeAtLevel0.width >> mipmapLevel, + textureSizeAtLevel0.height >> mipmapLevel, textureSizeAtLevel0.depth}; + + layout.bytesPerRow = GetMinimumBytesPerRow(format, layout.mipSize.width); + + uint32_t appliedRowsPerImage = rowsPerImage > 0 ? rowsPerImage : layout.mipSize.height; + layout.bytesPerImage = layout.bytesPerRow * appliedRowsPerImage; + + layout.byteLength = + GetBytesInBufferTextureCopy(format, layout.mipSize.width, layout.bytesPerRow, + appliedRowsPerImage, textureSizeAtLevel0.depth); + + const uint32_t bytesPerTexel = utils::GetTexelBlockSizeInBytes(format); + layout.texelBlocksPerRow = layout.bytesPerRow / bytesPerTexel; + layout.texelBlocksPerImage = layout.bytesPerImage / bytesPerTexel; + layout.texelBlockCount = layout.byteLength / bytesPerTexel; + + return layout; + } + + uint64_t RequiredBytesInCopy(uint64_t bytesPerRow, + uint64_t rowsPerImage, + wgpu::Extent3D copyExtent, + wgpu::TextureFormat textureFormat) { + if (copyExtent.width == 0 || copyExtent.height == 0 || copyExtent.depth == 0) { + return 0; + } else { + uint32_t blockSize = utils::GetTexelBlockSizeInBytes(textureFormat); + uint32_t blockWidth = utils::GetTextureFormatBlockWidth(textureFormat); + uint32_t blockHeight = utils::GetTextureFormatBlockHeight(textureFormat); + + uint64_t texelBlockRowsPerImage = rowsPerImage / blockHeight; + uint64_t bytesPerImage = bytesPerRow * texelBlockRowsPerImage; + uint64_t bytesInLastSlice = bytesPerRow * (copyExtent.height / blockHeight - 1) + + (copyExtent.width / blockWidth * blockSize); + return bytesPerImage * (copyExtent.depth - 1) + bytesInLastSlice; + } + } + + void UnalignDynamicUploader(wgpu::Device device) { + std::vector<uint8_t> data = {1}; + + wgpu::TextureDescriptor descriptor = {}; + descriptor.size = {1, 1, 1}; + descriptor.format = wgpu::TextureFormat::R8Unorm; + descriptor.usage = wgpu::TextureUsage::CopyDst | wgpu::TextureUsage::CopySrc; + wgpu::Texture texture = device.CreateTexture(&descriptor); + + wgpu::TextureCopyView textureCopyView = utils::CreateTextureCopyView(texture, 0, {0, 0, 0}); + wgpu::TextureDataLayout textureDataLayout = utils::CreateTextureDataLayout(0, 0, 0); + wgpu::Extent3D copyExtent = {1, 1, 1}; + + // WriteTexture with exactly 1 byte of data. + device.GetDefaultQueue().WriteTexture(&textureCopyView, data.data(), 1, &textureDataLayout, + ©Extent); + } +} // namespace utils
\ No newline at end of file diff --git a/chromium/third_party/dawn/src/utils/TestUtils.h b/chromium/third_party/dawn/src/utils/TestUtils.h new file mode 100644 index 00000000000..d1ba25fa410 --- /dev/null +++ b/chromium/third_party/dawn/src/utils/TestUtils.h @@ -0,0 +1,57 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef UTILS_TESTHELPERS_H_ +#define UTILS_TESTHELPERS_H_ + +#include <dawn/webgpu_cpp.h> + +namespace utils { + + struct TextureDataCopyLayout { + uint64_t byteLength; + uint64_t texelBlockCount; + uint32_t bytesPerRow; + uint32_t texelBlocksPerRow; + uint32_t bytesPerImage; + uint32_t texelBlocksPerImage; + wgpu::Extent3D mipSize; + }; + + uint32_t GetMinimumBytesPerRow(wgpu::TextureFormat format, uint32_t width); + uint32_t GetBytesInBufferTextureCopy(wgpu::TextureFormat format, + uint32_t width, + uint32_t bytesPerRow, + uint32_t rowsPerImage, + uint32_t copyArrayLayerCount); + TextureDataCopyLayout GetTextureDataCopyLayoutForTexture2DAtLevel( + wgpu::TextureFormat format, + wgpu::Extent3D textureSizeAtLevel0, + uint32_t mipmapLevel, + uint32_t rowsPerImage); + + uint64_t RequiredBytesInCopy(uint64_t bytesPerRow, + uint64_t rowsPerImage, + wgpu::Extent3D copyExtent, + wgpu::TextureFormat textureFormat); + + // A helper function used for testing DynamicUploader offset alignment. + // A call of this function will do a Queue::WriteTexture with 1 byte of data, + // so that assuming that WriteTexture uses DynamicUploader, the first RingBuffer + // in it will contain 1 byte of data. + void UnalignDynamicUploader(wgpu::Device device); + +} // namespace utils + +#endif // UTILS_TESTHELPERS_H_ diff --git a/chromium/third_party/dawn/src/utils/TextureFormatUtils.cpp b/chromium/third_party/dawn/src/utils/TextureFormatUtils.cpp index 795d1f948ac..89cd7d0ee2f 100644 --- a/chromium/third_party/dawn/src/utils/TextureFormatUtils.cpp +++ b/chromium/third_party/dawn/src/utils/TextureFormatUtils.cpp @@ -27,7 +27,8 @@ namespace utils { case wgpu::TextureFormat::RGBA8Unorm: case wgpu::TextureFormat::RGBA8Snorm: case wgpu::TextureFormat::RGB10A2Unorm: - case wgpu::TextureFormat::RG11B10Float: + case wgpu::TextureFormat::RG11B10Ufloat: + case wgpu::TextureFormat::RGB9E5Ufloat: case wgpu::TextureFormat::RG32Float: case wgpu::TextureFormat::RGBA16Float: case wgpu::TextureFormat::RGBA32Float: @@ -118,7 +119,9 @@ namespace utils { case wgpu::TextureFormat::BGRA8Unorm: case wgpu::TextureFormat::BGRA8UnormSrgb: case wgpu::TextureFormat::RGB10A2Unorm: - case wgpu::TextureFormat::RG11B10Float: + case wgpu::TextureFormat::RG11B10Ufloat: + case wgpu::TextureFormat::RGB9E5Ufloat: + case wgpu::TextureFormat::Depth32Float: return 4u; case wgpu::TextureFormat::RG32Float: @@ -152,9 +155,140 @@ namespace utils { case wgpu::TextureFormat::BC7RGBAUnormSrgb: return 16u; + case wgpu::TextureFormat::Depth24Plus: + case wgpu::TextureFormat::Depth24PlusStencil8: + case wgpu::TextureFormat::Undefined: + default: + UNREACHABLE(); + return 0u; + } + } + + uint32_t GetTextureFormatBlockWidth(wgpu::TextureFormat textureFormat) { + switch (textureFormat) { + case wgpu::TextureFormat::R8Unorm: + case wgpu::TextureFormat::R8Snorm: + case wgpu::TextureFormat::R8Uint: + case wgpu::TextureFormat::R8Sint: + case wgpu::TextureFormat::R16Uint: + case wgpu::TextureFormat::R16Sint: + case wgpu::TextureFormat::R16Float: + case wgpu::TextureFormat::RG8Unorm: + case wgpu::TextureFormat::RG8Snorm: + case wgpu::TextureFormat::RG8Uint: + case wgpu::TextureFormat::RG8Sint: + case wgpu::TextureFormat::R32Float: + case wgpu::TextureFormat::R32Uint: + case wgpu::TextureFormat::R32Sint: + case wgpu::TextureFormat::RG16Uint: + case wgpu::TextureFormat::RG16Sint: + case wgpu::TextureFormat::RG16Float: + case wgpu::TextureFormat::RGBA8Unorm: + case wgpu::TextureFormat::RGBA8UnormSrgb: + case wgpu::TextureFormat::RGBA8Snorm: + case wgpu::TextureFormat::RGBA8Uint: + case wgpu::TextureFormat::RGBA8Sint: + case wgpu::TextureFormat::BGRA8Unorm: + case wgpu::TextureFormat::BGRA8UnormSrgb: + case wgpu::TextureFormat::RGB10A2Unorm: + case wgpu::TextureFormat::RG11B10Ufloat: + case wgpu::TextureFormat::RGB9E5Ufloat: + case wgpu::TextureFormat::RG32Float: + case wgpu::TextureFormat::RG32Uint: + case wgpu::TextureFormat::RG32Sint: + case wgpu::TextureFormat::RGBA16Uint: + case wgpu::TextureFormat::RGBA16Sint: + case wgpu::TextureFormat::RGBA16Float: + case wgpu::TextureFormat::RGBA32Float: + case wgpu::TextureFormat::RGBA32Uint: + case wgpu::TextureFormat::RGBA32Sint: + case wgpu::TextureFormat::Depth32Float: + case wgpu::TextureFormat::Depth24Plus: + case wgpu::TextureFormat::Depth24PlusStencil8: + return 1u; + + case wgpu::TextureFormat::BC1RGBAUnorm: + case wgpu::TextureFormat::BC1RGBAUnormSrgb: + case wgpu::TextureFormat::BC4RUnorm: + case wgpu::TextureFormat::BC4RSnorm: + case wgpu::TextureFormat::BC2RGBAUnorm: + case wgpu::TextureFormat::BC2RGBAUnormSrgb: + case wgpu::TextureFormat::BC3RGBAUnorm: + case wgpu::TextureFormat::BC3RGBAUnormSrgb: + case wgpu::TextureFormat::BC5RGUnorm: + case wgpu::TextureFormat::BC5RGSnorm: + case wgpu::TextureFormat::BC6HRGBUfloat: + case wgpu::TextureFormat::BC6HRGBSfloat: + case wgpu::TextureFormat::BC7RGBAUnorm: + case wgpu::TextureFormat::BC7RGBAUnormSrgb: + return 4u; + + case wgpu::TextureFormat::Undefined: + default: + UNREACHABLE(); + return 0u; + } + } + + uint32_t GetTextureFormatBlockHeight(wgpu::TextureFormat textureFormat) { + switch (textureFormat) { + case wgpu::TextureFormat::R8Unorm: + case wgpu::TextureFormat::R8Snorm: + case wgpu::TextureFormat::R8Uint: + case wgpu::TextureFormat::R8Sint: + case wgpu::TextureFormat::R16Uint: + case wgpu::TextureFormat::R16Sint: + case wgpu::TextureFormat::R16Float: + case wgpu::TextureFormat::RG8Unorm: + case wgpu::TextureFormat::RG8Snorm: + case wgpu::TextureFormat::RG8Uint: + case wgpu::TextureFormat::RG8Sint: + case wgpu::TextureFormat::R32Float: + case wgpu::TextureFormat::R32Uint: + case wgpu::TextureFormat::R32Sint: + case wgpu::TextureFormat::RG16Uint: + case wgpu::TextureFormat::RG16Sint: + case wgpu::TextureFormat::RG16Float: + case wgpu::TextureFormat::RGBA8Unorm: + case wgpu::TextureFormat::RGBA8UnormSrgb: + case wgpu::TextureFormat::RGBA8Snorm: + case wgpu::TextureFormat::RGBA8Uint: + case wgpu::TextureFormat::RGBA8Sint: + case wgpu::TextureFormat::BGRA8Unorm: + case wgpu::TextureFormat::BGRA8UnormSrgb: + case wgpu::TextureFormat::RGB10A2Unorm: + case wgpu::TextureFormat::RG11B10Ufloat: + case wgpu::TextureFormat::RGB9E5Ufloat: + case wgpu::TextureFormat::RG32Float: + case wgpu::TextureFormat::RG32Uint: + case wgpu::TextureFormat::RG32Sint: + case wgpu::TextureFormat::RGBA16Uint: + case wgpu::TextureFormat::RGBA16Sint: + case wgpu::TextureFormat::RGBA16Float: + case wgpu::TextureFormat::RGBA32Float: + case wgpu::TextureFormat::RGBA32Uint: + case wgpu::TextureFormat::RGBA32Sint: case wgpu::TextureFormat::Depth32Float: case wgpu::TextureFormat::Depth24Plus: case wgpu::TextureFormat::Depth24PlusStencil8: + return 1u; + + case wgpu::TextureFormat::BC1RGBAUnorm: + case wgpu::TextureFormat::BC1RGBAUnormSrgb: + case wgpu::TextureFormat::BC4RUnorm: + case wgpu::TextureFormat::BC4RSnorm: + case wgpu::TextureFormat::BC2RGBAUnorm: + case wgpu::TextureFormat::BC2RGBAUnormSrgb: + case wgpu::TextureFormat::BC3RGBAUnorm: + case wgpu::TextureFormat::BC3RGBAUnormSrgb: + case wgpu::TextureFormat::BC5RGUnorm: + case wgpu::TextureFormat::BC5RGSnorm: + case wgpu::TextureFormat::BC6HRGBUfloat: + case wgpu::TextureFormat::BC6HRGBSfloat: + case wgpu::TextureFormat::BC7RGBAUnorm: + case wgpu::TextureFormat::BC7RGBAUnormSrgb: + return 4u; + case wgpu::TextureFormat::Undefined: default: UNREACHABLE(); @@ -208,7 +342,7 @@ namespace utils { return "rgba8i"; case wgpu::TextureFormat::RGB10A2Unorm: return "rgb10_a2"; - case wgpu::TextureFormat::RG11B10Float: + case wgpu::TextureFormat::RG11B10Ufloat: return "r11f_g11f_b10f"; case wgpu::TextureFormat::RG32Float: return "rg32f"; @@ -229,6 +363,7 @@ namespace utils { case wgpu::TextureFormat::RGBA32Sint: return "rgba32i"; + case wgpu::TextureFormat::RGB9E5Ufloat: case wgpu::TextureFormat::RGBA8UnormSrgb: case wgpu::TextureFormat::BGRA8Unorm: case wgpu::TextureFormat::BGRA8UnormSrgb: diff --git a/chromium/third_party/dawn/src/utils/TextureFormatUtils.h b/chromium/third_party/dawn/src/utils/TextureFormatUtils.h index 2976e94d3fa..7a7fa429bd6 100644 --- a/chromium/third_party/dawn/src/utils/TextureFormatUtils.h +++ b/chromium/third_party/dawn/src/utils/TextureFormatUtils.h @@ -22,39 +22,77 @@ #include "common/Assert.h" namespace utils { - static constexpr std::array<wgpu::TextureFormat, 52> kAllTextureFormats = { - wgpu::TextureFormat::R8Unorm, wgpu::TextureFormat::R8Snorm, - wgpu::TextureFormat::R8Uint, wgpu::TextureFormat::R8Sint, - wgpu::TextureFormat::R16Uint, wgpu::TextureFormat::R16Sint, - wgpu::TextureFormat::R16Float, wgpu::TextureFormat::RG8Unorm, - wgpu::TextureFormat::RG8Snorm, wgpu::TextureFormat::RG8Uint, - wgpu::TextureFormat::RG8Sint, wgpu::TextureFormat::R32Float, - wgpu::TextureFormat::R32Uint, wgpu::TextureFormat::R32Sint, - wgpu::TextureFormat::RG16Uint, wgpu::TextureFormat::RG16Sint, - wgpu::TextureFormat::RG16Float, wgpu::TextureFormat::RGBA8Unorm, - wgpu::TextureFormat::RGBA8UnormSrgb, wgpu::TextureFormat::RGBA8Snorm, - wgpu::TextureFormat::RGBA8Uint, wgpu::TextureFormat::RGBA8Sint, - wgpu::TextureFormat::BGRA8Unorm, wgpu::TextureFormat::BGRA8UnormSrgb, - wgpu::TextureFormat::RGB10A2Unorm, wgpu::TextureFormat::RG11B10Float, - wgpu::TextureFormat::RG32Float, wgpu::TextureFormat::RG32Uint, - wgpu::TextureFormat::RG32Sint, wgpu::TextureFormat::RGBA16Uint, - wgpu::TextureFormat::RGBA16Sint, wgpu::TextureFormat::RGBA16Float, - wgpu::TextureFormat::RGBA32Float, wgpu::TextureFormat::RGBA32Uint, - wgpu::TextureFormat::RGBA32Sint, wgpu::TextureFormat::Depth32Float, - wgpu::TextureFormat::Depth24Plus, wgpu::TextureFormat::Depth24PlusStencil8, - wgpu::TextureFormat::BC1RGBAUnorm, wgpu::TextureFormat::BC1RGBAUnormSrgb, - wgpu::TextureFormat::BC2RGBAUnorm, wgpu::TextureFormat::BC2RGBAUnormSrgb, - wgpu::TextureFormat::BC3RGBAUnorm, wgpu::TextureFormat::BC3RGBAUnormSrgb, - wgpu::TextureFormat::BC4RUnorm, wgpu::TextureFormat::BC4RSnorm, - wgpu::TextureFormat::BC5RGUnorm, wgpu::TextureFormat::BC5RGSnorm, - wgpu::TextureFormat::BC6HRGBUfloat, wgpu::TextureFormat::BC6HRGBSfloat, - wgpu::TextureFormat::BC7RGBAUnorm, wgpu::TextureFormat::BC7RGBAUnormSrgb, + static constexpr std::array<wgpu::TextureFormat, 53> kAllTextureFormats = { + wgpu::TextureFormat::R8Unorm, + wgpu::TextureFormat::R8Snorm, + wgpu::TextureFormat::R8Uint, + wgpu::TextureFormat::R8Sint, + wgpu::TextureFormat::R16Uint, + wgpu::TextureFormat::R16Sint, + wgpu::TextureFormat::R16Float, + wgpu::TextureFormat::RG8Unorm, + wgpu::TextureFormat::RG8Snorm, + wgpu::TextureFormat::RG8Uint, + wgpu::TextureFormat::RG8Sint, + wgpu::TextureFormat::R32Float, + wgpu::TextureFormat::R32Uint, + wgpu::TextureFormat::R32Sint, + wgpu::TextureFormat::RG16Uint, + wgpu::TextureFormat::RG16Sint, + wgpu::TextureFormat::RG16Float, + wgpu::TextureFormat::RGBA8Unorm, + wgpu::TextureFormat::RGBA8UnormSrgb, + wgpu::TextureFormat::RGBA8Snorm, + wgpu::TextureFormat::RGBA8Uint, + wgpu::TextureFormat::RGBA8Sint, + wgpu::TextureFormat::BGRA8Unorm, + wgpu::TextureFormat::BGRA8UnormSrgb, + wgpu::TextureFormat::RGB10A2Unorm, + wgpu::TextureFormat::RG11B10Ufloat, + wgpu::TextureFormat::RGB9E5Ufloat, + wgpu::TextureFormat::RG32Float, + wgpu::TextureFormat::RG32Uint, + wgpu::TextureFormat::RG32Sint, + wgpu::TextureFormat::RGBA16Uint, + wgpu::TextureFormat::RGBA16Sint, + wgpu::TextureFormat::RGBA16Float, + wgpu::TextureFormat::RGBA32Float, + wgpu::TextureFormat::RGBA32Uint, + wgpu::TextureFormat::RGBA32Sint, + wgpu::TextureFormat::Depth32Float, + wgpu::TextureFormat::Depth24Plus, + wgpu::TextureFormat::Depth24PlusStencil8, + wgpu::TextureFormat::BC1RGBAUnorm, + wgpu::TextureFormat::BC1RGBAUnormSrgb, + wgpu::TextureFormat::BC2RGBAUnorm, + wgpu::TextureFormat::BC2RGBAUnormSrgb, + wgpu::TextureFormat::BC3RGBAUnorm, + wgpu::TextureFormat::BC3RGBAUnormSrgb, + wgpu::TextureFormat::BC4RUnorm, + wgpu::TextureFormat::BC4RSnorm, + wgpu::TextureFormat::BC5RGUnorm, + wgpu::TextureFormat::BC5RGSnorm, + wgpu::TextureFormat::BC6HRGBUfloat, + wgpu::TextureFormat::BC6HRGBSfloat, + wgpu::TextureFormat::BC7RGBAUnorm, + wgpu::TextureFormat::BC7RGBAUnormSrgb, }; + static constexpr std::array<wgpu::TextureFormat, 14> kBCFormats = { + wgpu::TextureFormat::BC1RGBAUnorm, wgpu::TextureFormat::BC1RGBAUnormSrgb, + wgpu::TextureFormat::BC2RGBAUnorm, wgpu::TextureFormat::BC2RGBAUnormSrgb, + wgpu::TextureFormat::BC3RGBAUnorm, wgpu::TextureFormat::BC3RGBAUnormSrgb, + wgpu::TextureFormat::BC4RUnorm, wgpu::TextureFormat::BC4RSnorm, + wgpu::TextureFormat::BC5RGUnorm, wgpu::TextureFormat::BC5RGSnorm, + wgpu::TextureFormat::BC6HRGBUfloat, wgpu::TextureFormat::BC6HRGBSfloat, + wgpu::TextureFormat::BC7RGBAUnorm, wgpu::TextureFormat::BC7RGBAUnormSrgb}; + const char* GetColorTextureComponentTypePrefix(wgpu::TextureFormat textureFormat); bool TextureFormatSupportsStorageTexture(wgpu::TextureFormat format); uint32_t GetTexelBlockSizeInBytes(wgpu::TextureFormat textureFormat); + uint32_t GetTextureFormatBlockWidth(wgpu::TextureFormat textureFormat); + uint32_t GetTextureFormatBlockHeight(wgpu::TextureFormat textureFormat); const char* GetGLSLImageFormatQualifier(wgpu::TextureFormat textureFormat); } // namespace utils diff --git a/chromium/third_party/dawn/src/utils/WGPUHelpers.cpp b/chromium/third_party/dawn/src/utils/WGPUHelpers.cpp index 686b223f35a..4e2f6e87117 100644 --- a/chromium/third_party/dawn/src/utils/WGPUHelpers.cpp +++ b/chromium/third_party/dawn/src/utils/WGPUHelpers.cpp @@ -14,11 +14,8 @@ #include "utils/WGPUHelpers.h" -#include "common/Assert.h" #include "common/Constants.h" #include "common/Log.h" -#include "common/Math.h" -#include "utils/TextureFormatUtils.h" #include <shaderc/shaderc.hpp> @@ -144,6 +141,14 @@ namespace utils { return CreateShaderModuleFromResult(device, result); } + wgpu::ShaderModule CreateShaderModuleFromWGSL(const wgpu::Device& device, const char* source) { + wgpu::ShaderModuleWGSLDescriptor wgslDesc; + wgslDesc.source = source; + wgpu::ShaderModuleDescriptor descriptor; + descriptor.nextInChain = &wgslDesc; + return device.CreateShaderModule(&descriptor); + } + std::vector<uint32_t> CompileGLSLToSpirv(SingleShaderStage stage, const char* source) { shaderc_shader_kind kind = ShadercShaderKind(stage); @@ -267,26 +272,37 @@ namespace utils { uint64_t offset, uint32_t bytesPerRow, uint32_t rowsPerImage) { - wgpu::BufferCopyView bufferCopyView; + wgpu::BufferCopyView bufferCopyView = {}; bufferCopyView.buffer = buffer; - bufferCopyView.offset = offset; - bufferCopyView.bytesPerRow = bytesPerRow; - bufferCopyView.rowsPerImage = rowsPerImage; + bufferCopyView.layout = CreateTextureDataLayout(offset, bytesPerRow, rowsPerImage); return bufferCopyView; } wgpu::TextureCopyView CreateTextureCopyView(wgpu::Texture texture, uint32_t mipLevel, - wgpu::Origin3D origin) { + wgpu::Origin3D origin, + wgpu::TextureAspect aspect) { wgpu::TextureCopyView textureCopyView; textureCopyView.texture = texture; textureCopyView.mipLevel = mipLevel; textureCopyView.origin = origin; + textureCopyView.aspect = aspect; return textureCopyView; } + wgpu::TextureDataLayout CreateTextureDataLayout(uint64_t offset, + uint32_t bytesPerRow, + uint32_t rowsPerImage) { + wgpu::TextureDataLayout textureDataLayout; + textureDataLayout.offset = offset; + textureDataLayout.bytesPerRow = bytesPerRow; + textureDataLayout.rowsPerImage = rowsPerImage; + + return textureDataLayout; + } + wgpu::SamplerDescriptor GetDefaultSamplerDescriptor() { wgpu::SamplerDescriptor desc = {}; @@ -374,47 +390,4 @@ namespace utils { return device.CreateBindGroup(&descriptor); } - uint32_t GetMinimumBytesPerRow(wgpu::TextureFormat format, uint32_t width) { - const uint32_t bytesPerTexel = utils::GetTexelBlockSizeInBytes(format); - return Align(bytesPerTexel * width, kTextureBytesPerRowAlignment); - } - - uint32_t GetBytesInBufferTextureCopy(wgpu::TextureFormat format, - uint32_t width, - uint32_t bytesPerRow, - uint32_t rowsPerImage, - uint32_t copyArrayLayerCount) { - ASSERT(rowsPerImage > 0); - const uint32_t bytesPerTexel = utils::GetTexelBlockSizeInBytes(format); - const uint32_t bytesAtLastImage = bytesPerRow * (rowsPerImage - 1) + bytesPerTexel * width; - return bytesPerRow * rowsPerImage * (copyArrayLayerCount - 1) + bytesAtLastImage; - } - - // TODO(jiawei.shao@intel.com): support compressed texture formats - BufferTextureCopyLayout GetBufferTextureCopyLayoutForTexture2DAtLevel( - wgpu::TextureFormat format, - wgpu::Extent3D textureSizeAtLevel0, - uint32_t mipmapLevel, - uint32_t rowsPerImage) { - BufferTextureCopyLayout layout; - - layout.mipSize = {textureSizeAtLevel0.width >> mipmapLevel, - textureSizeAtLevel0.height >> mipmapLevel, textureSizeAtLevel0.depth}; - - layout.bytesPerRow = GetMinimumBytesPerRow(format, layout.mipSize.width); - - uint32_t appliedRowsPerImage = rowsPerImage > 0 ? rowsPerImage : layout.mipSize.height; - layout.bytesPerImage = layout.bytesPerRow * appliedRowsPerImage; - - layout.byteLength = - GetBytesInBufferTextureCopy(format, layout.mipSize.width, layout.bytesPerRow, - appliedRowsPerImage, textureSizeAtLevel0.depth); - - const uint32_t bytesPerTexel = utils::GetTexelBlockSizeInBytes(format); - layout.texelBlocksPerRow = layout.bytesPerRow / bytesPerTexel; - layout.texelBlocksPerImage = layout.bytesPerImage / bytesPerTexel; - layout.texelBlockCount = layout.byteLength / bytesPerTexel; - - return layout; - } } // namespace utils diff --git a/chromium/third_party/dawn/src/utils/WGPUHelpers.h b/chromium/third_party/dawn/src/utils/WGPUHelpers.h index fdbd8648f77..b77b6a438bc 100644 --- a/chromium/third_party/dawn/src/utils/WGPUHelpers.h +++ b/chromium/third_party/dawn/src/utils/WGPUHelpers.h @@ -22,6 +22,7 @@ #include <vector> #include "common/Constants.h" +#include "utils/TextureFormatUtils.h" namespace utils { @@ -33,6 +34,8 @@ namespace utils { SingleShaderStage stage, const char* source); wgpu::ShaderModule CreateShaderModuleFromASM(const wgpu::Device& device, const char* source); + wgpu::ShaderModule CreateShaderModuleFromWGSL(const wgpu::Device& device, const char* source); + std::vector<uint32_t> CompileGLSLToSpirv(SingleShaderStage stage, const char* source); wgpu::Buffer CreateBufferFromData(const wgpu::Device& device, @@ -51,9 +54,14 @@ namespace utils { uint64_t offset, uint32_t bytesPerRow, uint32_t rowsPerImage); - wgpu::TextureCopyView CreateTextureCopyView(wgpu::Texture texture, - uint32_t level, - wgpu::Origin3D origin); + wgpu::TextureCopyView CreateTextureCopyView( + wgpu::Texture texture, + uint32_t level, + wgpu::Origin3D origin, + wgpu::TextureAspect aspect = wgpu::TextureAspect::All); + wgpu::TextureDataLayout CreateTextureDataLayout(uint64_t offset, + uint32_t bytesPerRow, + uint32_t rowsPerImage); struct ComboRenderPassDescriptor : public wgpu::RenderPassDescriptor { public: @@ -66,7 +74,7 @@ namespace utils { std::array<wgpu::RenderPassColorAttachmentDescriptor, kMaxColorAttachments> cColorAttachments; - wgpu::RenderPassDepthStencilAttachmentDescriptor cDepthStencilAttachmentInfo; + wgpu::RenderPassDepthStencilAttachmentDescriptor cDepthStencilAttachmentInfo = {}; }; struct BasicRenderPass { @@ -129,28 +137,6 @@ namespace utils { const wgpu::BindGroupLayout& layout, std::initializer_list<BindingInitializationHelper> entriesInitializer); - struct BufferTextureCopyLayout { - uint64_t byteLength; - uint64_t texelBlockCount; - uint32_t bytesPerRow; - uint32_t texelBlocksPerRow; - uint32_t bytesPerImage; - uint32_t texelBlocksPerImage; - wgpu::Extent3D mipSize; - }; - - uint32_t GetMinimumBytesPerRow(wgpu::TextureFormat format, uint32_t width); - uint32_t GetBytesInBufferTextureCopy(wgpu::TextureFormat format, - uint32_t width, - uint32_t bytesPerRow, - uint32_t rowsPerImage, - uint32_t copyArrayLayerCount); - BufferTextureCopyLayout GetBufferTextureCopyLayoutForTexture2DAtLevel( - wgpu::TextureFormat format, - wgpu::Extent3D textureSizeAtLevel0, - uint32_t mipmapLevel, - uint32_t rowsPerImage); - } // namespace utils #endif // UTILS_DAWNHELPERS_H_ diff --git a/chromium/third_party/dawn/src/utils/WindowsDebugLogger.cpp b/chromium/third_party/dawn/src/utils/WindowsDebugLogger.cpp new file mode 100644 index 00000000000..76c9faa81a8 --- /dev/null +++ b/chromium/third_party/dawn/src/utils/WindowsDebugLogger.cpp @@ -0,0 +1,104 @@ +// Copyright 2020 The Dawn Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "utils/PlatformDebugLogger.h" + +#include "common/Assert.h" +#include "common/windows_with_undefs.h" + +#include <array> +#include <thread> + +namespace utils { + + class WindowsDebugLogger : public PlatformDebugLogger { + public: + WindowsDebugLogger() : PlatformDebugLogger() { + if (IsDebuggerPresent()) { + // This condition is true when running inside Visual Studio or some other debugger. + // Messages are already printed there so we don't need to do anything. + return; + } + + mShouldExitHandle = CreateEventA(nullptr, TRUE, FALSE, nullptr); + ASSERT(mShouldExitHandle != nullptr); + + mThread = std::thread( + [](HANDLE shouldExit) { + // https://blogs.msdn.microsoft.com/reiley/2011/07/29/a-debugging-approach-to-outputdebugstring/ + // for the layout of this struct. + struct { + DWORD process_id; + char data[4096 - sizeof(DWORD)]; + }* dbWinBuffer = nullptr; + + HANDLE file = CreateFileMappingA(INVALID_HANDLE_VALUE, nullptr, PAGE_READWRITE, + 0, sizeof(*dbWinBuffer), "DBWIN_BUFFER"); + ASSERT(file != nullptr); + ASSERT(file != INVALID_HANDLE_VALUE); + + dbWinBuffer = static_cast<decltype(dbWinBuffer)>( + MapViewOfFile(file, SECTION_MAP_READ, 0, 0, 0)); + ASSERT(dbWinBuffer != nullptr); + + HANDLE dbWinBufferReady = + CreateEventA(nullptr, FALSE, FALSE, "DBWIN_BUFFER_READY"); + ASSERT(dbWinBufferReady != nullptr); + + HANDLE dbWinDataReady = CreateEventA(nullptr, FALSE, FALSE, "DBWIN_DATA_READY"); + ASSERT(dbWinDataReady != nullptr); + + std::array<HANDLE, 2> waitHandles = {shouldExit, dbWinDataReady}; + while (true) { + SetEvent(dbWinBufferReady); + DWORD wait = WaitForMultipleObjects(waitHandles.size(), waitHandles.data(), + FALSE, INFINITE); + if (wait == WAIT_OBJECT_0) { + break; + } + ASSERT(wait == WAIT_OBJECT_0 + 1); + fprintf(stderr, "%.*s\n", static_cast<int>(sizeof(dbWinBuffer->data)), + dbWinBuffer->data); + fflush(stderr); + } + + CloseHandle(dbWinDataReady); + CloseHandle(dbWinBufferReady); + UnmapViewOfFile(dbWinBuffer); + CloseHandle(file); + }, + mShouldExitHandle); + } + + ~WindowsDebugLogger() override { + if (mShouldExitHandle != nullptr) { + ASSERT(SetEvent(mShouldExitHandle) != 0); + CloseHandle(mShouldExitHandle); + } + + if (mThread.joinable()) { + mThread.join(); + } + } + + private: + std::thread mThread; + HANDLE mShouldExitHandle = INVALID_HANDLE_VALUE; + }; + + PlatformDebugLogger* CreatePlatformDebugLogger() { + return new WindowsDebugLogger(); + } + +} // namespace utils diff --git a/chromium/third_party/dawn/third_party/.clang-format b/chromium/third_party/dawn/third_party/.clang-format new file mode 100644 index 00000000000..9d159247d51 --- /dev/null +++ b/chromium/third_party/dawn/third_party/.clang-format @@ -0,0 +1,2 @@ +DisableFormat: true +SortIncludes: false diff --git a/chromium/third_party/dawn/third_party/gn/glfw/BUILD.gn b/chromium/third_party/dawn/third_party/gn/glfw/BUILD.gn index a5a46412aca..ba2336fbd6a 100644 --- a/chromium/third_party/dawn/third_party/gn/glfw/BUILD.gn +++ b/chromium/third_party/dawn/third_party/gn/glfw/BUILD.gn @@ -134,7 +134,7 @@ static_library("glfw") { "${glfw_dir}/src/nsgl_context.h", "${glfw_dir}/src/nsgl_context.m", ] - libs += [ + frameworks = [ "Cocoa.framework", "IOKit.framework", "CoreFoundation.framework", diff --git a/chromium/third_party/dawn/third_party/khronos/BUILD.gn b/chromium/third_party/dawn/third_party/khronos/BUILD.gn index 52ce94bd53b..f9e6df25096 100644 --- a/chromium/third_party/dawn/third_party/khronos/BUILD.gn +++ b/chromium/third_party/dawn/third_party/khronos/BUILD.gn @@ -41,9 +41,7 @@ config("khronos_headers_public") { } source_set("khronos_platform") { - sources = [ - "KHR/khrplatform.h", - ] + sources = [ "KHR/khrplatform.h" ] public_configs = [ ":khronos_headers_public" ] } |