radv: Add on-demand compilation of built-in shaders.

In environments where we cannot cache, e.g. Android (no homedir),
ChromeOS (readonly rootfs) or sandboxes (cannot open cache), the
startup cost of creating a device in radv is rather high, due
to compiling all possible built-in pipelines up front. This meant
depending on the CPU a 1-4 sec cost of creating a Device.

For CTS this cost is unacceptable, and likely for starting random
apps too.

So if there is no cache, with this patch radv will compile shaders
on demand. Once there is a cache from the first run, even if
incomplete, the driver knows that it can likely write the cache
and precompiles everything.

Note that I did not switch the buffer and itob/btoi compute pipelines
to on-demand, since you cannot really do anything in Vulkan without
them and there are only a few.

This reduces the CTS runtime for the no caches scenario on my
threadripper from 32 minutes to 8 minutes.

Reviewed-by: Dave Airlie <airlied@redhat.com>

1 files changed, 24 insertions, 7 deletions

diff --git a/src/amd/vulkan/radv_meta_resolve_fs.c b/src/amd/vulkan/radv_meta_resolve_fs.c
index 6013503b837..21a5922f5df 100644
--- a/src/amd/vulkan/radv_meta_resolve_fs.c
+++ b/src/amd/vulkan/radv_meta_resolve_fs.c
@@ -161,10 +161,18 @@ create_resolve_pipeline(struct radv_device *device,
 			int samples_log2,
 			VkFormat format)
 {
+	mtx_lock(&device->meta_state.mtx);
+
+	unsigned fs_key = radv_format_meta_fs_key(format);
+	VkPipeline *pipeline = &device->meta_state.resolve_fragment.rc[samples_log2].pipeline[fs_key];
+	if (*pipeline) {
+		mtx_unlock(&device->meta_state.mtx);
+		return VK_SUCCESS;
+	}
+
 	VkResult result;
 	bool is_integer = false;
 	uint32_t samples = 1 << samples_log2;
-	unsigned fs_key = radv_format_meta_fs_key(format);
 	const VkPipelineVertexInputStateCreateInfo *vi_create_info;
 	vi_create_info = &normal_vi_create_info;
 	if (vk_format_is_int(format))
@@ -180,9 +188,6 @@ create_resolve_pipeline(struct radv_device *device,
 
 	assert(!*rp);
 
-	VkPipeline *pipeline = &device->meta_state.resolve_fragment.rc[samples_log2].pipeline[fs_key];
-	assert(!*pipeline);
-
 	VkPipelineShaderStageCreateInfo pipeline_shader_stages[] = {
 		{
 			.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
@@ -307,11 +312,12 @@ create_resolve_pipeline(struct radv_device *device,
 	ralloc_free(vs.nir);
 	ralloc_free(fs.nir);
 
+	mtx_unlock(&device->meta_state.mtx);
 	return result;
 }
 
 VkResult
-radv_device_init_meta_resolve_fragment_state(struct radv_device *device)
+radv_device_init_meta_resolve_fragment_state(struct radv_device *device, bool on_demand)
 {
 	VkResult res;
 
@@ -319,6 +325,9 @@ radv_device_init_meta_resolve_fragment_state(struct radv_device *device)
 	if (res != VK_SUCCESS)
 		goto fail;
 
+	if (on_demand)
+		return VK_SUCCESS;
+
 	for (uint32_t i = 0; i < MAX_SAMPLES_LOG2; ++i) {
 		for (unsigned j = 0; j < NUM_META_FS_KEYS; ++j) {
 			res = create_resolve_pipeline(device, i, radv_fs_key_format_exemplars[j]);
@@ -404,10 +413,18 @@ emit_resolve(struct radv_cmd_buffer *cmd_buffer,
 			      push_constants);
 
 	unsigned fs_key = radv_format_meta_fs_key(dest_iview->vk_format);
-	VkPipeline pipeline_h = device->meta_state.resolve_fragment.rc[samples_log2].pipeline[fs_key];
+	VkPipeline* pipeline = &device->meta_state.resolve_fragment.rc[samples_log2].pipeline[fs_key];
+
+	if (*pipeline == VK_NULL_HANDLE) {
+		VkResult ret = create_resolve_pipeline(device, samples_log2, radv_fs_key_format_exemplars[fs_key]);
+		if (ret != VK_SUCCESS) {
+			cmd_buffer->record_result = ret;
+			return;
+		}
+	}
 
 	radv_CmdBindPipeline(cmd_buffer_h, VK_PIPELINE_BIND_POINT_GRAPHICS,
-			     pipeline_h);
+			     *pipeline);
 
 	radv_CmdSetViewport(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkViewport) {
 		.x = dest_offset->x,