/* * Copyright © 2019 Google LLC * SPDX-License-Identifier: MIT */ #include "tu_shader.h" #include "spirv/nir_spirv.h" #include "util/mesa-sha1.h" #include "nir/nir_xfb_info.h" #include "nir/nir_vulkan.h" #include "vk_pipeline.h" #include "vk_util.h" #include "ir3/ir3_nir.h" #include "tu_device.h" #include "tu_descriptor_set.h" #include "tu_pipeline.h" nir_shader * tu_spirv_to_nir(struct tu_device *dev, void *mem_ctx, const VkPipelineShaderStageCreateInfo *stage_info, gl_shader_stage stage) { /* TODO these are made-up */ const struct spirv_to_nir_options spirv_options = { /* ViewID is a sysval in geometry stages and an input in the FS */ .view_index_is_input = stage == MESA_SHADER_FRAGMENT, /* Use 16-bit math for RelaxedPrecision ALU ops */ .mediump_16bit_alu = true, .caps = { .demote_to_helper_invocation = true, .descriptor_array_dynamic_indexing = true, .descriptor_array_non_uniform_indexing = true, .descriptor_indexing = true, .device_group = true, .draw_parameters = true, .float_controls = true, .float16 = true, .fragment_density = true, .geometry_streams = true, .image_read_without_format = true, .image_write_without_format = true, .int16 = true, .multiview = true, .physical_storage_buffer_address = true, .post_depth_coverage = true, .runtime_descriptor_array = true, .shader_viewport_index_layer = true, .stencil_export = true, .storage_16bit = dev->physical_device->info->a6xx.storage_16bit, .subgroup_arithmetic = true, .subgroup_ballot = true, .subgroup_basic = true, .subgroup_quad = true, .subgroup_shuffle = true, .subgroup_vote = true, .tessellation = true, .transform_feedback = true, .variable_pointers = true, .vk_memory_model_device_scope = true, .vk_memory_model = true, }, .ubo_addr_format = nir_address_format_vec2_index_32bit_offset, .ssbo_addr_format = nir_address_format_vec2_index_32bit_offset, /* Accessed via stg/ldg */ .phys_ssbo_addr_format = nir_address_format_64bit_global, /* Accessed via the const register file */ .push_const_addr_format = nir_address_format_logical, /* Accessed via ldl/stl */ .shared_addr_format = nir_address_format_32bit_offset, /* Accessed via stg/ldg (not used with Vulkan?) */ .global_addr_format = nir_address_format_64bit_global, }; const nir_shader_compiler_options *nir_options = ir3_get_compiler_options(dev->compiler); nir_shader *nir; VkResult result = vk_pipeline_shader_stage_to_nir(&dev->vk, stage_info, &spirv_options, nir_options, mem_ctx, &nir); if (result != VK_SUCCESS) return NULL; if (TU_DEBUG(NIR)) { fprintf(stderr, "translated nir:\n"); nir_print_shader(nir, stderr); } const struct nir_lower_sysvals_to_varyings_options sysvals_to_varyings = { .point_coord = true, }; NIR_PASS_V(nir, nir_lower_sysvals_to_varyings, &sysvals_to_varyings); NIR_PASS_V(nir, nir_lower_global_vars_to_local); /* Older glslang missing bf6efd0316d8 ("SPV: Fix #2293: keep relaxed * precision on arg passed to relaxed param") will pass function args through * a highp temporary, so we need the nir_opt_find_array_copies() and a copy * prop before we lower mediump vars, or you'll be unable to optimize out * array copies after lowering. We do this before splitting copies, since * that works against nir_opt_find_array_copies(). * */ NIR_PASS_V(nir, nir_opt_find_array_copies); NIR_PASS_V(nir, nir_opt_copy_prop_vars); NIR_PASS_V(nir, nir_opt_dce); NIR_PASS_V(nir, nir_split_var_copies); NIR_PASS_V(nir, nir_lower_var_copies); NIR_PASS_V(nir, nir_lower_mediump_vars, nir_var_function_temp | nir_var_shader_temp | nir_var_mem_shared); NIR_PASS_V(nir, nir_opt_copy_prop_vars); NIR_PASS_V(nir, nir_opt_combine_stores, nir_var_all); NIR_PASS_V(nir, nir_lower_system_values); NIR_PASS_V(nir, nir_lower_is_helper_invocation); ir3_optimize_loop(dev->compiler, nir); NIR_PASS_V(nir, nir_opt_conditional_discard); return nir; } static void lower_load_push_constant(struct tu_device *dev, nir_builder *b, nir_intrinsic_instr *instr, struct tu_shader *shader, const struct tu_pipeline_layout *layout) { uint32_t base = nir_intrinsic_base(instr); assert(base % 4 == 0); if (tu6_shared_constants_enable(layout, dev->compiler)) { /* All stages share the same range. We could potentially add * push_constant_offset to layout and apply it, but this is good for * now. */ base += dev->compiler->shared_consts_base_offset * 4; } else { assert(base >= shader->const_state.push_consts.lo * 4); base -= shader->const_state.push_consts.lo * 4; } nir_ssa_def *load = nir_load_uniform(b, instr->num_components, instr->dest.ssa.bit_size, nir_ushr(b, instr->src[0].ssa, nir_imm_int(b, 2)), .base = base); nir_ssa_def_rewrite_uses(&instr->dest.ssa, load); nir_instr_remove(&instr->instr); } static void lower_vulkan_resource_index(nir_builder *b, nir_intrinsic_instr *instr, struct tu_shader *shader, const struct tu_pipeline_layout *layout) { nir_ssa_def *vulkan_idx = instr->src[0].ssa; unsigned set = nir_intrinsic_desc_set(instr); unsigned binding = nir_intrinsic_binding(instr); struct tu_descriptor_set_layout *set_layout = layout->set[set].layout; struct tu_descriptor_set_binding_layout *binding_layout = &set_layout->binding[binding]; nir_ssa_def *base; if (binding_layout->type == VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK) return; shader->active_desc_sets |= 1u << set; switch (binding_layout->type) { case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC: case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: if (layout->independent_sets) { /* With independent sets, we don't know * layout->set[set].dynamic_offset_start until after link time which * with fast linking means after the shader is compiled. We have to * get it from the const file instead. */ base = nir_imm_int(b, binding_layout->dynamic_offset_offset / (4 * A6XX_TEX_CONST_DWORDS)); nir_ssa_def *dynamic_offset_start = nir_load_uniform(b, 1, 32, nir_imm_int(b, 0), .base = shader->const_state.dynamic_offset_loc + set); base = nir_iadd(b, base, dynamic_offset_start); } else { base = nir_imm_int(b, (layout->set[set].dynamic_offset_start + binding_layout->dynamic_offset_offset) / (4 * A6XX_TEX_CONST_DWORDS)); } set = MAX_SETS; break; default: base = nir_imm_int(b, binding_layout->offset / (4 * A6XX_TEX_CONST_DWORDS)); break; } unsigned stride = binding_layout->size / (4 * A6XX_TEX_CONST_DWORDS); assert(util_is_power_of_two_nonzero(stride)); nir_ssa_def *shift = nir_imm_int(b, util_logbase2(stride)); nir_ssa_def *def = nir_vec3(b, nir_imm_int(b, set), nir_iadd(b, base, nir_ishl(b, vulkan_idx, shift)), shift); nir_ssa_def_rewrite_uses(&instr->dest.ssa, def); nir_instr_remove(&instr->instr); } static void lower_vulkan_resource_reindex(nir_builder *b, nir_intrinsic_instr *instr) { nir_ssa_def *old_index = instr->src[0].ssa; nir_ssa_def *delta = instr->src[1].ssa; nir_ssa_def *shift = nir_channel(b, old_index, 2); nir_ssa_def *new_index = nir_vec3(b, nir_channel(b, old_index, 0), nir_iadd(b, nir_channel(b, old_index, 1), nir_ishl(b, delta, shift)), shift); nir_ssa_def_rewrite_uses(&instr->dest.ssa, new_index); nir_instr_remove(&instr->instr); } static void lower_load_vulkan_descriptor(nir_builder *b, nir_intrinsic_instr *intrin) { nir_ssa_def *old_index = intrin->src[0].ssa; /* Loading the descriptor happens as part of the load/store instruction so * this is a no-op. We just need to turn the shift into an offset of 0. */ nir_ssa_def *new_index = nir_vec3(b, nir_channel(b, old_index, 0), nir_channel(b, old_index, 1), nir_imm_int(b, 0)); nir_ssa_def_rewrite_uses(&intrin->dest.ssa, new_index); nir_instr_remove(&intrin->instr); } static void lower_ssbo_ubo_intrinsic(struct tu_device *dev, nir_builder *b, nir_intrinsic_instr *intrin) { const nir_intrinsic_info *info = &nir_intrinsic_infos[intrin->intrinsic]; /* The bindless base is part of the instruction, which means that part of * the "pointer" has to be constant. We solve this in the same way the blob * does, by generating a bunch of if-statements. In the usual case where * the descriptor set is constant we can skip that, though). */ unsigned buffer_src; if (intrin->intrinsic == nir_intrinsic_store_ssbo) { /* This has the value first */ buffer_src = 1; } else { buffer_src = 0; } nir_ssa_scalar scalar_idx = nir_ssa_scalar_resolved(intrin->src[buffer_src].ssa, 0); nir_ssa_def *descriptor_idx = nir_channel(b, intrin->src[buffer_src].ssa, 1); /* For isam, we need to use the appropriate descriptor if 16-bit storage is * enabled. Descriptor 0 is the 16-bit one, descriptor 1 is the 32-bit one. */ if (dev->physical_device->info->a6xx.storage_16bit && intrin->intrinsic == nir_intrinsic_load_ssbo && (nir_intrinsic_access(intrin) & ACCESS_CAN_REORDER) && intrin->dest.ssa.bit_size > 16) { descriptor_idx = nir_iadd(b, descriptor_idx, nir_imm_int(b, 1)); } nir_ssa_def *results[MAX_SETS + 1] = { NULL }; if (nir_ssa_scalar_is_const(scalar_idx)) { nir_ssa_def *bindless = nir_bindless_resource_ir3(b, 32, descriptor_idx, .desc_set = nir_ssa_scalar_as_uint(scalar_idx)); nir_instr_rewrite_src_ssa(&intrin->instr, &intrin->src[buffer_src], bindless); return; } nir_ssa_def *base_idx = nir_channel(b, scalar_idx.def, scalar_idx.comp); for (unsigned i = 0; i < MAX_SETS + 1; i++) { /* if (base_idx == i) { ... */ nir_if *nif = nir_push_if(b, nir_ieq_imm(b, base_idx, i)); nir_ssa_def *bindless = nir_bindless_resource_ir3(b, 32, descriptor_idx, .desc_set = i); nir_intrinsic_instr *copy = nir_intrinsic_instr_create(b->shader, intrin->intrinsic); copy->num_components = intrin->num_components; for (unsigned src = 0; src < info->num_srcs; src++) { if (src == buffer_src) copy->src[src] = nir_src_for_ssa(bindless); else copy->src[src] = nir_src_for_ssa(intrin->src[src].ssa); } for (unsigned idx = 0; idx < info->num_indices; idx++) { copy->const_index[idx] = intrin->const_index[idx]; } if (info->has_dest) { nir_ssa_dest_init(©->instr, ©->dest, intrin->dest.ssa.num_components, intrin->dest.ssa.bit_size, NULL); results[i] = ©->dest.ssa; } nir_builder_instr_insert(b, ©->instr); /* } else { ... */ nir_push_else(b, nif); } nir_ssa_def *result = nir_ssa_undef(b, intrin->dest.ssa.num_components, intrin->dest.ssa.bit_size); for (int i = MAX_SETS; i >= 0; i--) { nir_pop_if(b, NULL); if (info->has_dest) result = nir_if_phi(b, results[i], result); } if (info->has_dest) nir_ssa_def_rewrite_uses(&intrin->dest.ssa, result); nir_instr_remove(&intrin->instr); } static nir_ssa_def * build_bindless(struct tu_device *dev, nir_builder *b, nir_deref_instr *deref, bool is_sampler, struct tu_shader *shader, const struct tu_pipeline_layout *layout) { nir_variable *var = nir_deref_instr_get_variable(deref); unsigned set = var->data.descriptor_set; unsigned binding = var->data.binding; const struct tu_descriptor_set_binding_layout *bind_layout = &layout->set[set].layout->binding[binding]; /* input attachments use non bindless workaround */ if (bind_layout->type == VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT && !TU_DEBUG(DYNAMIC)) { const struct glsl_type *glsl_type = glsl_without_array(var->type); uint32_t idx = var->data.index * 2; BITSET_SET_RANGE_INSIDE_WORD(b->shader->info.textures_used, idx, (idx + bind_layout->array_size * 2) - 1); /* D24S8 workaround: stencil of D24S8 will be sampled as uint */ if (glsl_get_sampler_result_type(glsl_type) == GLSL_TYPE_UINT) idx += 1; if (deref->deref_type == nir_deref_type_var) return nir_imm_int(b, idx); nir_ssa_def *arr_index = nir_ssa_for_src(b, deref->arr.index, 1); return nir_iadd(b, nir_imm_int(b, idx), nir_imul_imm(b, arr_index, 2)); } shader->active_desc_sets |= 1u << set; nir_ssa_def *desc_offset; unsigned descriptor_stride; unsigned offset = 0; /* Samplers come second in combined image/sampler descriptors, see * write_combined_image_sampler_descriptor(). */ if (is_sampler && bind_layout->type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) { offset = 1; } desc_offset = nir_imm_int(b, (bind_layout->offset / (4 * A6XX_TEX_CONST_DWORDS)) + offset); descriptor_stride = bind_layout->size / (4 * A6XX_TEX_CONST_DWORDS); if (deref->deref_type != nir_deref_type_var) { assert(deref->deref_type == nir_deref_type_array); nir_ssa_def *arr_index = nir_ssa_for_src(b, deref->arr.index, 1); desc_offset = nir_iadd(b, desc_offset, nir_imul_imm(b, arr_index, descriptor_stride)); } return nir_bindless_resource_ir3(b, 32, desc_offset, .desc_set = set); } static void lower_image_deref(struct tu_device *dev, nir_builder *b, nir_intrinsic_instr *instr, struct tu_shader *shader, const struct tu_pipeline_layout *layout) { nir_deref_instr *deref = nir_src_as_deref(instr->src[0]); nir_ssa_def *bindless = build_bindless(dev, b, deref, false, shader, layout); nir_rewrite_image_intrinsic(instr, bindless, true); } static bool lower_intrinsic(nir_builder *b, nir_intrinsic_instr *instr, struct tu_device *dev, struct tu_shader *shader, const struct tu_pipeline_layout *layout) { switch (instr->intrinsic) { case nir_intrinsic_load_push_constant: lower_load_push_constant(dev, b, instr, shader, layout); return true; case nir_intrinsic_load_vulkan_descriptor: lower_load_vulkan_descriptor(b, instr); return true; case nir_intrinsic_vulkan_resource_index: lower_vulkan_resource_index(b, instr, shader, layout); return true; case nir_intrinsic_vulkan_resource_reindex: lower_vulkan_resource_reindex(b, instr); return true; case nir_intrinsic_load_ubo: case nir_intrinsic_load_ssbo: case nir_intrinsic_store_ssbo: case nir_intrinsic_ssbo_atomic: case nir_intrinsic_ssbo_atomic_swap: case nir_intrinsic_get_ssbo_size: lower_ssbo_ubo_intrinsic(dev, b, instr); return true; case nir_intrinsic_image_deref_load: case nir_intrinsic_image_deref_store: case nir_intrinsic_image_deref_atomic: case nir_intrinsic_image_deref_atomic_swap: case nir_intrinsic_image_deref_size: case nir_intrinsic_image_deref_samples: lower_image_deref(dev, b, instr, shader, layout); return true; default: return false; } } static void lower_tex_ycbcr(const struct tu_pipeline_layout *layout, nir_builder *builder, nir_tex_instr *tex) { int deref_src_idx = nir_tex_instr_src_index(tex, nir_tex_src_texture_deref); assert(deref_src_idx >= 0); nir_deref_instr *deref = nir_src_as_deref(tex->src[deref_src_idx].src); nir_variable *var = nir_deref_instr_get_variable(deref); const struct tu_descriptor_set_layout *set_layout = layout->set[var->data.descriptor_set].layout; const struct tu_descriptor_set_binding_layout *binding = &set_layout->binding[var->data.binding]; const struct tu_sampler_ycbcr_conversion *ycbcr_samplers = tu_immutable_ycbcr_samplers(set_layout, binding); if (!ycbcr_samplers) return; /* For the following instructions, we don't apply any change */ if (tex->op == nir_texop_txs || tex->op == nir_texop_query_levels || tex->op == nir_texop_lod) return; assert(tex->texture_index == 0); unsigned array_index = 0; if (deref->deref_type != nir_deref_type_var) { assert(deref->deref_type == nir_deref_type_array); if (!nir_src_is_const(deref->arr.index)) return; array_index = nir_src_as_uint(deref->arr.index); array_index = MIN2(array_index, binding->array_size - 1); } const struct tu_sampler_ycbcr_conversion *ycbcr_sampler = ycbcr_samplers + array_index; if (ycbcr_sampler->ycbcr_model == VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY) return; builder->cursor = nir_after_instr(&tex->instr); uint8_t bits = vk_format_get_component_bits(ycbcr_sampler->format, UTIL_FORMAT_COLORSPACE_RGB, PIPE_SWIZZLE_X); switch (ycbcr_sampler->format) { case VK_FORMAT_G8B8G8R8_422_UNORM: case VK_FORMAT_B8G8R8G8_422_UNORM: case VK_FORMAT_G8_B8R8_2PLANE_420_UNORM: case VK_FORMAT_G8_B8_R8_3PLANE_420_UNORM: /* util_format_get_component_bits doesn't return what we want */ bits = 8; break; default: break; } uint32_t bpcs[3] = {bits, bits, bits}; /* TODO: use right bpc for each channel ? */ nir_ssa_def *result = nir_convert_ycbcr_to_rgb(builder, ycbcr_sampler->ycbcr_model, ycbcr_sampler->ycbcr_range, &tex->dest.ssa, bpcs); nir_ssa_def_rewrite_uses_after(&tex->dest.ssa, result, result->parent_instr); builder->cursor = nir_before_instr(&tex->instr); } static bool lower_tex(nir_builder *b, nir_tex_instr *tex, struct tu_device *dev, struct tu_shader *shader, const struct tu_pipeline_layout *layout) { lower_tex_ycbcr(layout, b, tex); int sampler_src_idx = nir_tex_instr_src_index(tex, nir_tex_src_sampler_deref); if (sampler_src_idx >= 0) { nir_deref_instr *deref = nir_src_as_deref(tex->src[sampler_src_idx].src); nir_ssa_def *bindless = build_bindless(dev, b, deref, true, shader, layout); nir_instr_rewrite_src(&tex->instr, &tex->src[sampler_src_idx].src, nir_src_for_ssa(bindless)); tex->src[sampler_src_idx].src_type = nir_tex_src_sampler_handle; } int tex_src_idx = nir_tex_instr_src_index(tex, nir_tex_src_texture_deref); if (tex_src_idx >= 0) { nir_deref_instr *deref = nir_src_as_deref(tex->src[tex_src_idx].src); nir_ssa_def *bindless = build_bindless(dev, b, deref, false, shader, layout); nir_instr_rewrite_src(&tex->instr, &tex->src[tex_src_idx].src, nir_src_for_ssa(bindless)); tex->src[tex_src_idx].src_type = nir_tex_src_texture_handle; /* for the input attachment case: */ if (bindless->parent_instr->type != nir_instr_type_intrinsic) tex->src[tex_src_idx].src_type = nir_tex_src_texture_offset; } return true; } struct lower_instr_params { struct tu_device *dev; struct tu_shader *shader; const struct tu_pipeline_layout *layout; }; static bool lower_instr(nir_builder *b, nir_instr *instr, void *cb_data) { struct lower_instr_params *params = (struct lower_instr_params *) cb_data; b->cursor = nir_before_instr(instr); switch (instr->type) { case nir_instr_type_tex: return lower_tex(b, nir_instr_as_tex(instr), params->dev, params->shader, params->layout); case nir_instr_type_intrinsic: return lower_intrinsic(b, nir_instr_as_intrinsic(instr), params->dev, params->shader, params->layout); default: return false; } } /* Since we always push inline uniforms into constant memory, lower loads of * them to load_uniform which turns into constant memory loads. */ static bool lower_inline_ubo(nir_builder *b, nir_instr *instr, void *cb_data) { if (instr->type != nir_instr_type_intrinsic) return false; nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr); if (intrin->intrinsic != nir_intrinsic_load_ubo) return false; struct lower_instr_params *params = (struct lower_instr_params *) cb_data; struct tu_shader *shader = params->shader; const struct tu_pipeline_layout *layout = params->layout; nir_binding binding = nir_chase_binding(intrin->src[0]); if (!binding.success) return false; struct tu_descriptor_set_layout *set_layout = layout->set[binding.desc_set].layout; struct tu_descriptor_set_binding_layout *binding_layout = &set_layout->binding[binding.binding]; if (binding_layout->type != VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK) return false; /* lookup the const offset of the inline UBO */ struct tu_const_state *const_state = &shader->const_state; unsigned base = UINT_MAX; bool use_load = false; for (unsigned i = 0; i < const_state->num_inline_ubos; i++) { if (const_state->ubos[i].base == binding.desc_set && const_state->ubos[i].offset == binding_layout->offset) { base = const_state->ubos[i].const_offset_vec4 * 4; use_load = const_state->ubos[i].push_address; break; } } if (base == UINT_MAX) { /* Assume we're loading out-of-bounds from a 0-sized inline uniform * filtered out below. */ nir_ssa_def_rewrite_uses(&intrin->dest.ssa, nir_ssa_undef(b, intrin->num_components, intrin->dest.ssa.bit_size)); return true; } nir_ssa_def *offset = intrin->src[1].ssa; b->cursor = nir_before_instr(instr); nir_ssa_def *val; if (use_load) { nir_ssa_def *base_addr = nir_load_uniform(b, 2, 32, nir_imm_int(b, 0), .base = base); val = nir_load_global_ir3(b, intrin->num_components, intrin->dest.ssa.bit_size, base_addr, nir_ishr_imm(b, offset, 2)); } else { val = nir_load_uniform(b, intrin->num_components, intrin->dest.ssa.bit_size, nir_ishr_imm(b, offset, 2), .base = base); } nir_ssa_def_rewrite_uses(&intrin->dest.ssa, val); nir_instr_remove(instr); return true; } /* Figure out the range of push constants that we're actually going to push to * the shader, and tell the backend to reserve this range when pushing UBO * constants. */ static void gather_push_constants(nir_shader *shader, struct tu_shader *tu_shader) { uint32_t min = UINT32_MAX, max = 0; nir_foreach_function(function, shader) { if (!function->impl) continue; nir_foreach_block(block, function->impl) { nir_foreach_instr_safe(instr, block) { if (instr->type != nir_instr_type_intrinsic) continue; nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr); if (intrin->intrinsic != nir_intrinsic_load_push_constant) continue; uint32_t base = nir_intrinsic_base(intrin); uint32_t range = nir_intrinsic_range(intrin); min = MIN2(min, base); max = MAX2(max, base + range); break; } } } if (min >= max) { tu_shader->const_state.push_consts.lo = 0; tu_shader->const_state.push_consts.dwords = 0; return; } /* CP_LOAD_STATE OFFSET and NUM_UNIT for SHARED_CONSTS are in units of * dwords while loading regular consts is in units of vec4's. * So we unify the unit here as dwords for tu_push_constant_range, then * we should consider correct unit when emitting. * * Note there's an alignment requirement of 16 dwords on OFFSET. Expand * the range and change units accordingly. */ tu_shader->const_state.push_consts.lo = (min / 4) / 4 * 4; tu_shader->const_state.push_consts.dwords = align(max, 16) / 4 - tu_shader->const_state.push_consts.lo; } static bool tu_lower_io(nir_shader *shader, struct tu_device *dev, struct tu_shader *tu_shader, const struct tu_pipeline_layout *layout) { if (!tu6_shared_constants_enable(layout, dev->compiler)) gather_push_constants(shader, tu_shader); struct tu_const_state *const_state = &tu_shader->const_state; unsigned reserved_consts_vec4 = align(DIV_ROUND_UP(const_state->push_consts.dwords, 4), dev->compiler->const_upload_unit); if (layout->independent_sets) { const_state->dynamic_offset_loc = reserved_consts_vec4 * 4; reserved_consts_vec4 += DIV_ROUND_UP(MAX_SETS, 4); } else { const_state->dynamic_offset_loc = UINT32_MAX; } /* Reserve space for inline uniforms, so we can always load them from * constants and not setup a UBO descriptor for them. */ for (unsigned set = 0; set < layout->num_sets; set++) { const struct tu_descriptor_set_layout *desc_layout = layout->set[set].layout; if (!desc_layout || !desc_layout->has_inline_uniforms) continue; for (unsigned b = 0; b < desc_layout->binding_count; b++) { const struct tu_descriptor_set_binding_layout *binding = &desc_layout->binding[b]; if (binding->type != VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK) continue; if (!(binding->shader_stages & mesa_to_vk_shader_stage(shader->info.stage))) continue; /* Workaround a CTS bug by ignoring zero-sized inline uniform * blocks that aren't being properly filtered out when creating the * descriptor set layout, see * https://gitlab.khronos.org/Tracker/vk-gl-cts/-/issues/4115 */ if (binding->size == 0) continue; /* If we don't know the size at compile time due to a variable * descriptor count, then with descriptor buffers we cannot know * how much space the real inline uniform has. In this case we fall * back to pushing the address and using ldg, which is slower than * setting up a descriptor but setting up our own descriptor with * descriptor_buffer is also painful and has to be done on the GPU * and doesn't avoid the UBO getting pushed anyway and faulting if a * out-of-bounds access is hidden behind an if and not dynamically * executed. Given the small max size, there shouldn't be much reason * to use variable size anyway. */ bool push_address = desc_layout->has_variable_descriptors && b == desc_layout->binding_count - 1; if (push_address) { perf_debug(dev, "falling back to ldg for variable-sized inline " "uniform block"); } assert(const_state->num_inline_ubos < ARRAY_SIZE(const_state->ubos)); unsigned size_vec4 = push_address ? 1 : DIV_ROUND_UP(binding->size, 16); const_state->ubos[const_state->num_inline_ubos++] = (struct tu_inline_ubo) { .base = set, .offset = binding->offset, .push_address = push_address, .const_offset_vec4 = reserved_consts_vec4, .size_vec4 = size_vec4, }; reserved_consts_vec4 += align(size_vec4, dev->compiler->const_upload_unit); } } tu_shader->reserved_user_consts_vec4 = reserved_consts_vec4; struct lower_instr_params params = { .dev = dev, .shader = tu_shader, .layout = layout, }; bool progress = false; if (const_state->num_inline_ubos) { progress |= nir_shader_instructions_pass(shader, lower_inline_ubo, nir_metadata_none, ¶ms); } progress |= nir_shader_instructions_pass(shader, lower_instr, nir_metadata_none, ¶ms); /* Remove now-unused variables so that when we gather the shader info later * they won't be counted. */ if (progress) nir_opt_dce(shader); progress |= nir_remove_dead_variables(shader, nir_var_uniform | nir_var_mem_ubo | nir_var_mem_ssbo, NULL); return progress; } struct lower_fdm_options { unsigned num_views; bool adjust_fragcoord; bool multiview; }; static bool lower_fdm_filter(const nir_instr *instr, const void *data) { const struct lower_fdm_options *options = (const struct lower_fdm_options *)data; if (instr->type != nir_instr_type_intrinsic) return false; nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr); return intrin->intrinsic == nir_intrinsic_load_frag_size || (intrin->intrinsic == nir_intrinsic_load_frag_coord && options->adjust_fragcoord); } static nir_ssa_def * lower_fdm_instr(struct nir_builder *b, nir_instr *instr, void *data) { const struct lower_fdm_options *options = (const struct lower_fdm_options *)data; nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr); nir_ssa_def *view; if (options->multiview) { nir_variable *view_var = nir_find_variable_with_location(b->shader, nir_var_shader_in, VARYING_SLOT_VIEW_INDEX); if (view_var == NULL) { view_var = nir_variable_create(b->shader, nir_var_shader_in, glsl_int_type(), NULL); view_var->data.location = VARYING_SLOT_VIEW_INDEX; view_var->data.interpolation = INTERP_MODE_FLAT; view_var->data.driver_location = b->shader->num_inputs++; } view = nir_load_var(b, view_var); } else { view = nir_imm_int(b, 0); } nir_ssa_def *frag_size = nir_load_frag_size_ir3(b, view, .range = options->num_views); if (intrin->intrinsic == nir_intrinsic_load_frag_coord) { nir_ssa_def *frag_offset = nir_load_frag_offset_ir3(b, view, .range = options->num_views); nir_ssa_def *unscaled_coord = nir_load_frag_coord_unscaled_ir3(b); nir_ssa_def *xy = nir_channels(b, unscaled_coord, 0x3); xy = nir_fmul(b, nir_fsub(b, xy, frag_offset), nir_i2f32(b, frag_size)); return nir_vec4(b, nir_channel(b, xy, 0), nir_channel(b, xy, 1), nir_channel(b, unscaled_coord, 2), nir_channel(b, unscaled_coord, 3)); } assert(intrin->intrinsic == nir_intrinsic_load_frag_size); return frag_size; } static bool tu_nir_lower_fdm(nir_shader *shader, const struct lower_fdm_options *options) { return nir_shader_lower_instructions(shader, lower_fdm_filter, lower_fdm_instr, (void *)options); } static void shared_type_info(const struct glsl_type *type, unsigned *size, unsigned *align) { assert(glsl_type_is_vector_or_scalar(type)); unsigned comp_size = glsl_type_is_boolean(type) ? 4 : glsl_get_bit_size(type) / 8; unsigned length = glsl_get_vector_elements(type); *size = comp_size * length; *align = comp_size; } static void tu_gather_xfb_info(nir_shader *nir, struct ir3_stream_output_info *info) { nir_shader_gather_xfb_info(nir); if (!nir->xfb_info) return; nir_xfb_info *xfb = nir->xfb_info; uint8_t output_map[VARYING_SLOT_TESS_MAX]; memset(output_map, 0, sizeof(output_map)); nir_foreach_shader_out_variable(var, nir) { unsigned slots = var->data.compact ? DIV_ROUND_UP(glsl_get_length(var->type), 4) : glsl_count_attribute_slots(var->type, false); for (unsigned i = 0; i < slots; i++) output_map[var->data.location + i] = var->data.driver_location + i; } assert(xfb->output_count <= IR3_MAX_SO_OUTPUTS); info->num_outputs = xfb->output_count; for (int i = 0; i < IR3_MAX_SO_BUFFERS; i++) { info->stride[i] = xfb->buffers[i].stride / 4; info->buffer_to_stream[i] = xfb->buffer_to_stream[i]; } info->streams_written = xfb->streams_written; for (int i = 0; i < xfb->output_count; i++) { info->output[i].register_index = output_map[xfb->outputs[i].location]; info->output[i].start_component = xfb->outputs[i].component_offset; info->output[i].num_components = util_bitcount(xfb->outputs[i].component_mask); info->output[i].output_buffer = xfb->outputs[i].buffer; info->output[i].dst_offset = xfb->outputs[i].offset / 4; info->output[i].stream = xfb->buffer_to_stream[xfb->outputs[i].buffer]; } } struct tu_shader * tu_shader_create(struct tu_device *dev, nir_shader *nir, const struct tu_shader_key *key, struct tu_pipeline_layout *layout, const VkAllocationCallbacks *alloc) { struct tu_shader *shader; shader = (struct tu_shader *) vk_zalloc2( &dev->vk.alloc, alloc, sizeof(*shader), 8, VK_SYSTEM_ALLOCATION_SCOPE_COMMAND); if (!shader) return NULL; const nir_opt_access_options access_options = { .is_vulkan = true, }; NIR_PASS_V(nir, nir_opt_access, &access_options); if (nir->info.stage == MESA_SHADER_FRAGMENT) { const nir_input_attachment_options att_options = { .use_fragcoord_sysval = true, .use_layer_id_sysval = false, /* When using multiview rendering, we must use * gl_ViewIndex as the layer id to pass to the texture * sampling function. gl_Layer doesn't work when * multiview is enabled. */ .use_view_id_for_layer = key->multiview_mask != 0, .unscaled_input_attachment_ir3 = key->unscaled_input_fragcoord, }; NIR_PASS_V(nir, nir_lower_input_attachments, &att_options); } /* This has to happen before lower_input_attachments, because we have to * lower input attachment coordinates except if unscaled. */ const struct lower_fdm_options fdm_options = { .num_views = MAX2(util_last_bit(key->multiview_mask), 1), .adjust_fragcoord = key->fragment_density_map, }; NIR_PASS_V(nir, tu_nir_lower_fdm, &fdm_options); /* This needs to happen before multiview lowering which rewrites store * instructions of the position variable, so that we can just rewrite one * store at the end instead of having to rewrite every store specified by * the user. */ ir3_nir_lower_io_to_temporaries(nir); if (nir->info.stage == MESA_SHADER_VERTEX && key->multiview_mask) { tu_nir_lower_multiview(nir, key->multiview_mask, dev); } if (nir->info.stage == MESA_SHADER_FRAGMENT && key->force_sample_interp) { nir_foreach_shader_in_variable(var, nir) { if (!var->data.centroid) var->data.sample = true; } } NIR_PASS_V(nir, nir_lower_explicit_io, nir_var_mem_push_const, nir_address_format_32bit_offset); NIR_PASS_V(nir, nir_lower_explicit_io, nir_var_mem_ubo | nir_var_mem_ssbo, nir_address_format_vec2_index_32bit_offset); NIR_PASS_V(nir, nir_lower_explicit_io, nir_var_mem_global, nir_address_format_64bit_global); if (nir->info.stage == MESA_SHADER_COMPUTE) { NIR_PASS_V(nir, nir_lower_vars_to_explicit_types, nir_var_mem_shared, shared_type_info); NIR_PASS_V(nir, nir_lower_explicit_io, nir_var_mem_shared, nir_address_format_32bit_offset); if (nir->info.zero_initialize_shared_memory && nir->info.shared_size > 0) { const unsigned chunk_size = 16; /* max single store size */ /* Shared memory is allocated in 1024b chunks in HW, but the zero-init * extension only requires us to initialize the memory that the shader * is allocated at the API level, and it's up to the user to ensure * that accesses are limited to those bounds. */ const unsigned shared_size = ALIGN(nir->info.shared_size, chunk_size); NIR_PASS_V(nir, nir_zero_initialize_shared_memory, shared_size, chunk_size); } const struct nir_lower_compute_system_values_options compute_sysval_options = { .has_base_workgroup_id = true, }; NIR_PASS_V(nir, nir_lower_compute_system_values, &compute_sysval_options); } nir_assign_io_var_locations(nir, nir_var_shader_in, &nir->num_inputs, nir->info.stage); nir_assign_io_var_locations(nir, nir_var_shader_out, &nir->num_outputs, nir->info.stage); /* Gather information for transform feedback. This should be called after: * - nir_split_per_member_structs. * - nir_remove_dead_variables with varyings, so that we could align * stream outputs correctly. * - nir_assign_io_var_locations - to have valid driver_location */ struct ir3_stream_output_info so_info = {}; if (nir->info.stage == MESA_SHADER_VERTEX || nir->info.stage == MESA_SHADER_TESS_EVAL || nir->info.stage == MESA_SHADER_GEOMETRY) tu_gather_xfb_info(nir, &so_info); /* Temporary stopgap until legacy atomics are removed */ NIR_PASS_V(nir, nir_lower_legacy_atomics); NIR_PASS_V(nir, tu_lower_io, dev, shader, layout); nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir)); ir3_finalize_nir(dev->compiler, nir); bool shared_consts_enable = tu6_shared_constants_enable(layout, dev->compiler); if (shared_consts_enable) assert(!shader->const_state.push_consts.dwords); const struct ir3_shader_options options = { .reserved_user_consts = shader->reserved_user_consts_vec4, .api_wavesize = key->api_wavesize, .real_wavesize = key->real_wavesize, .shared_consts_enable = shared_consts_enable, }; shader->ir3_shader = ir3_shader_from_nir(dev->compiler, nir, &options, &so_info); return shader; } void tu_shader_destroy(struct tu_device *dev, struct tu_shader *shader, const VkAllocationCallbacks *alloc) { ir3_shader_destroy(shader->ir3_shader); vk_free2(&dev->vk.alloc, alloc, shader); }