#include "radv_meta.h"
#include "radv_private.h"
-#include "nir/nir_builder.h"
#include "sid.h"
-/**
- * Vertex attributes used by all pipelines.
- */
-struct vertex_attrs {
- float position[2]; /**< 3DPRIM_RECTLIST */
-};
-
-/* passthrough vertex shader */
-static nir_shader *
-build_nir_vs(void)
-{
- const struct glsl_type *vec4 = glsl_vec4_type();
-
- nir_builder b;
- nir_variable *a_position;
- nir_variable *v_position;
-
- nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_VERTEX, NULL);
- b.shader->info->name = ralloc_strdup(b.shader, "meta_depth_decomp_vs");
-
- a_position = nir_variable_create(b.shader, nir_var_shader_in, vec4,
- "a_position");
- a_position->data.location = VERT_ATTRIB_GENERIC0;
-
- v_position = nir_variable_create(b.shader, nir_var_shader_out, vec4,
- "gl_Position");
- v_position->data.location = VARYING_SLOT_POS;
-
- nir_copy_var(&b, v_position, a_position);
-
- return b.shader;
-}
-
-/* simple passthrough shader */
-static nir_shader *
-build_nir_fs(void)
-{
- nir_builder b;
-
- nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
- b.shader->info->name = ralloc_asprintf(b.shader,
- "meta_depth_decomp_noop_fs");
-
- return b.shader;
-}
static VkResult
-create_pass(struct radv_device *device)
+create_pass(struct radv_device *device,
+ uint32_t samples,
+ VkRenderPass *pass)
{
VkResult result;
VkDevice device_h = radv_device_to_handle(device);
const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
VkAttachmentDescription attachment;
- attachment.format = VK_FORMAT_UNDEFINED;
- attachment.samples = 1;
+ attachment.flags = 0;
+ attachment.format = VK_FORMAT_D32_SFLOAT_S8_UINT;
+ attachment.samples = samples;
attachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
+ attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
+ attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE;
attachment.initialLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
attachment.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
.dependencyCount = 0,
},
alloc,
- &device->meta_state.depth_decomp.pass);
+ pass);
return result;
}
+static VkResult
+create_pipeline_layout(struct radv_device *device, VkPipelineLayout *layout)
+{
+ VkPipelineLayoutCreateInfo pl_create_info = {
+ .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
+ .setLayoutCount = 0,
+ .pSetLayouts = NULL,
+ .pushConstantRangeCount = 0,
+ .pPushConstantRanges = NULL,
+ };
+
+ return radv_CreatePipelineLayout(radv_device_to_handle(device),
+ &pl_create_info,
+ &device->meta_state.alloc,
+ layout);
+}
+
static VkResult
create_pipeline(struct radv_device *device,
- VkShaderModule vs_module_h)
+ VkShaderModule vs_module_h,
+ uint32_t samples,
+ VkRenderPass pass,
+ VkPipelineLayout layout,
+ VkPipeline *decompress_pipeline,
+ VkPipeline *resummarize_pipeline)
{
VkResult result;
VkDevice device_h = radv_device_to_handle(device);
+ struct radv_shader_module vs_module = {0};
+
+ mtx_lock(&device->meta_state.mtx);
+ if (*decompress_pipeline) {
+ mtx_unlock(&device->meta_state.mtx);
+ return VK_SUCCESS;
+ }
+
+ if (!vs_module_h) {
+ vs_module.nir = radv_meta_build_nir_vs_generate_vertices();
+ vs_module_h = radv_shader_module_to_handle(&vs_module);
+ }
struct radv_shader_module fs_module = {
- .nir = build_nir_fs(),
+ .nir = radv_meta_build_nir_fs_noop(),
};
if (!fs_module.nir) {
goto cleanup;
}
+ const VkPipelineSampleLocationsStateCreateInfoEXT sample_locs_create_info = {
+ .sType = VK_STRUCTURE_TYPE_PIPELINE_SAMPLE_LOCATIONS_STATE_CREATE_INFO_EXT,
+ .sampleLocationsEnable = false,
+ };
+
const VkGraphicsPipelineCreateInfo pipeline_create_info = {
.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
.stageCount = 2,
},
.pVertexInputState = &(VkPipelineVertexInputStateCreateInfo) {
.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
- .vertexBindingDescriptionCount = 1,
- .pVertexBindingDescriptions = (VkVertexInputBindingDescription[]) {
- {
- .binding = 0,
- .stride = sizeof(struct vertex_attrs),
- .inputRate = VK_VERTEX_INPUT_RATE_VERTEX
- },
- },
- .vertexAttributeDescriptionCount = 1,
- .pVertexAttributeDescriptions = (VkVertexInputAttributeDescription[]) {
- {
- /* Position */
- .location = 0,
- .binding = 0,
- .format = VK_FORMAT_R32G32_SFLOAT,
- .offset = offsetof(struct vertex_attrs, position),
- },
- },
+ .vertexBindingDescriptionCount = 0,
+ .vertexAttributeDescriptionCount = 0,
},
.pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
},
.pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
- .rasterizationSamples = 1,
+ .pNext = &sample_locs_create_info,
+ .rasterizationSamples = samples,
.sampleShadingEnable = false,
.pSampleMask = NULL,
.alphaToCoverageEnable = false,
},
.pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
- .dynamicStateCount = 2,
+ .dynamicStateCount = 3,
.pDynamicStates = (VkDynamicState[]) {
VK_DYNAMIC_STATE_VIEWPORT,
VK_DYNAMIC_STATE_SCISSOR,
+ VK_DYNAMIC_STATE_SAMPLE_LOCATIONS_EXT,
},
},
- .renderPass = device->meta_state.depth_decomp.pass,
+ .layout = layout,
+ .renderPass = pass,
.subpass = 0,
};
.db_flush_stencil_inplace = true,
},
&device->meta_state.alloc,
- &device->meta_state.depth_decomp.decompress_pipeline);
+ decompress_pipeline);
if (result != VK_SUCCESS)
goto cleanup;
.db_resummarize = true,
},
&device->meta_state.alloc,
- &device->meta_state.depth_decomp.resummarize_pipeline);
+ resummarize_pipeline);
if (result != VK_SUCCESS)
goto cleanup;
cleanup:
ralloc_free(fs_module.nir);
+ if (vs_module.nir)
+ ralloc_free(vs_module.nir);
+ mtx_unlock(&device->meta_state.mtx);
return result;
}
radv_device_finish_meta_depth_decomp_state(struct radv_device *device)
{
struct radv_meta_state *state = &device->meta_state;
- VkDevice device_h = radv_device_to_handle(device);
- VkRenderPass pass_h = device->meta_state.depth_decomp.pass;
- const VkAllocationCallbacks *alloc = &device->meta_state.alloc;
-
- if (pass_h)
- radv_DestroyRenderPass(device_h, pass_h,
- &device->meta_state.alloc);
- VkPipeline pipeline_h = state->depth_decomp.decompress_pipeline;
- if (pipeline_h) {
- radv_DestroyPipeline(device_h, pipeline_h, alloc);
- }
- pipeline_h = state->depth_decomp.resummarize_pipeline;
- if (pipeline_h) {
- radv_DestroyPipeline(device_h, pipeline_h, alloc);
+ for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
+ radv_DestroyRenderPass(radv_device_to_handle(device),
+ state->depth_decomp[i].pass,
+ &state->alloc);
+ radv_DestroyPipelineLayout(radv_device_to_handle(device),
+ state->depth_decomp[i].p_layout,
+ &state->alloc);
+ radv_DestroyPipeline(radv_device_to_handle(device),
+ state->depth_decomp[i].decompress_pipeline,
+ &state->alloc);
+ radv_DestroyPipeline(radv_device_to_handle(device),
+ state->depth_decomp[i].resummarize_pipeline,
+ &state->alloc);
}
}
VkResult
-radv_device_init_meta_depth_decomp_state(struct radv_device *device)
+radv_device_init_meta_depth_decomp_state(struct radv_device *device, bool on_demand)
{
+ struct radv_meta_state *state = &device->meta_state;
VkResult res = VK_SUCCESS;
- zero(device->meta_state.depth_decomp);
-
- struct radv_shader_module vs_module = { .nir = build_nir_vs() };
+ struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
if (!vs_module.nir) {
/* XXX: Need more accurate error */
res = VK_ERROR_OUT_OF_HOST_MEMORY;
goto fail;
}
- res = create_pass(device);
- if (res != VK_SUCCESS)
- goto fail;
-
VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
- res = create_pipeline(device, vs_module_h);
- if (res != VK_SUCCESS)
- goto fail;
+
+ for (uint32_t i = 0; i < ARRAY_SIZE(state->depth_decomp); ++i) {
+ uint32_t samples = 1 << i;
+
+ res = create_pass(device, samples, &state->depth_decomp[i].pass);
+ if (res != VK_SUCCESS)
+ goto fail;
+
+ res = create_pipeline_layout(device,
+ &state->depth_decomp[i].p_layout);
+ if (res != VK_SUCCESS)
+ goto fail;
+
+ if (on_demand)
+ continue;
+
+ res = create_pipeline(device, vs_module_h, samples,
+ state->depth_decomp[i].pass,
+ state->depth_decomp[i].p_layout,
+ &state->depth_decomp[i].decompress_pipeline,
+ &state->depth_decomp[i].resummarize_pipeline);
+ if (res != VK_SUCCESS)
+ goto fail;
+ }
goto cleanup;
return res;
}
-static void
-emit_depth_decomp(struct radv_cmd_buffer *cmd_buffer,
- const VkOffset2D *dest_offset,
- const VkExtent2D *depth_decomp_extent,
- VkPipeline pipeline_h)
-{
- struct radv_device *device = cmd_buffer->device;
- VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
- uint32_t offset;
- const struct vertex_attrs vertex_data[3] = {
- {
- .position = {
- -1.0,
- -1.0,
- },
- },
- {
- .position = {
- -1.0,
- 1.0,
- },
- },
- {
- .position = {
- 1.0,
- -1.0,
- },
- },
- };
-
- radv_cmd_buffer_upload_data(cmd_buffer, sizeof(vertex_data), 16, vertex_data, &offset);
- struct radv_buffer vertex_buffer = {
- .device = device,
- .size = sizeof(vertex_data),
- .bo = cmd_buffer->upload.upload_bo,
- .offset = offset,
- };
-
- VkBuffer vertex_buffer_h = radv_buffer_to_handle(&vertex_buffer);
-
- radv_CmdBindVertexBuffers(cmd_buffer_h,
- /*firstBinding*/ 0,
- /*bindingCount*/ 1,
- (VkBuffer[]) { vertex_buffer_h },
- (VkDeviceSize[]) { 0 });
-
- RADV_FROM_HANDLE(radv_pipeline, pipeline, pipeline_h);
+enum radv_depth_op {
+ DEPTH_DECOMPRESS,
+ DEPTH_RESUMMARIZE,
+};
- if (cmd_buffer->state.pipeline != pipeline) {
- radv_CmdBindPipeline(cmd_buffer_h, VK_PIPELINE_BIND_POINT_GRAPHICS,
- pipeline_h);
+static VkPipeline *
+radv_get_depth_pipeline(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_image *image, enum radv_depth_op op)
+{
+ struct radv_meta_state *state = &cmd_buffer->device->meta_state;
+ uint32_t samples = image->info.samples;
+ uint32_t samples_log2 = ffs(samples) - 1;
+ VkPipeline *pipeline;
+
+ if (!state->depth_decomp[samples_log2].decompress_pipeline) {
+ VkResult ret;
+
+ ret = create_pipeline(cmd_buffer->device, VK_NULL_HANDLE, samples,
+ state->depth_decomp[samples_log2].pass,
+ state->depth_decomp[samples_log2].p_layout,
+ &state->depth_decomp[samples_log2].decompress_pipeline,
+ &state->depth_decomp[samples_log2].resummarize_pipeline);
+ if (ret != VK_SUCCESS) {
+ cmd_buffer->record_result = ret;
+ return NULL;
+ }
+ }
+
+ switch (op) {
+ case DEPTH_DECOMPRESS:
+ pipeline = &state->depth_decomp[samples_log2].decompress_pipeline;
+ break;
+ case DEPTH_RESUMMARIZE:
+ pipeline = &state->depth_decomp[samples_log2].resummarize_pipeline;
+ break;
+ default:
+ unreachable("unknown operation");
}
- radv_CmdSetViewport(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkViewport) {
- .x = dest_offset->x,
- .y = dest_offset->y,
- .width = depth_decomp_extent->width,
- .height = depth_decomp_extent->height,
- .minDepth = 0.0f,
- .maxDepth = 1.0f
- });
-
- radv_CmdSetScissor(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkRect2D) {
- .offset = *dest_offset,
- .extent = *depth_decomp_extent,
- });
-
- radv_CmdDraw(cmd_buffer_h, 3, 1, 0, 0);
+ return pipeline;
}
-
static void radv_process_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
struct radv_image *image,
VkImageSubresourceRange *subresourceRange,
- VkPipeline pipeline_h)
+ struct radv_sample_locations_state *sample_locs,
+ enum radv_depth_op op)
{
struct radv_meta_saved_state saved_state;
- struct radv_meta_saved_pass_state saved_pass_state;
VkDevice device_h = radv_device_to_handle(cmd_buffer->device);
VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
- uint32_t width = radv_minify(image->extent.width,
+ uint32_t width = radv_minify(image->info.width,
subresourceRange->baseMipLevel);
- uint32_t height = radv_minify(image->extent.height,
+ uint32_t height = radv_minify(image->info.height,
subresourceRange->baseMipLevel);
+ uint32_t samples = image->info.samples;
+ uint32_t samples_log2 = ffs(samples) - 1;
+ struct radv_meta_state *meta_state = &cmd_buffer->device->meta_state;
+ VkPipeline *pipeline;
- if (!image->surface.htile_size)
+ if (!radv_image_has_htile(image))
return;
- radv_meta_save_pass(&saved_pass_state, cmd_buffer);
- radv_meta_save_graphics_reset_vport_scissor(&saved_state, cmd_buffer);
+ radv_meta_save(&saved_state, cmd_buffer,
+ RADV_META_SAVE_GRAPHICS_PIPELINE |
+ RADV_META_SAVE_SAMPLE_LOCATIONS |
+ RADV_META_SAVE_PASS);
+
+ pipeline = radv_get_depth_pipeline(cmd_buffer, image, op);
+
+ radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
+ VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
+
+ radv_CmdSetViewport(cmd_buffer_h, 0, 1, &(VkViewport) {
+ .x = 0,
+ .y = 0,
+ .width = width,
+ .height = height,
+ .minDepth = 0.0f,
+ .maxDepth = 1.0f
+ });
+
+ radv_CmdSetScissor(cmd_buffer_h, 0, 1, &(VkRect2D) {
+ .offset = { 0, 0 },
+ .extent = { width, height },
+ });
+
+ if (sample_locs) {
+ assert(image->flags & VK_IMAGE_CREATE_SAMPLE_LOCATIONS_COMPATIBLE_DEPTH_BIT_EXT);
+
+ /* Set the sample locations specified during explicit or
+ * automatic layout transitions, otherwise the depth decompress
+ * pass uses the default HW locations.
+ */
+ radv_CmdSetSampleLocationsEXT(cmd_buffer_h, &(VkSampleLocationsInfoEXT) {
+ .sampleLocationsPerPixel = sample_locs->per_pixel,
+ .sampleLocationGridSize = sample_locs->grid_size,
+ .sampleLocationsCount = sample_locs->count,
+ .pSampleLocations = sample_locs->locations,
+ });
+ }
for (uint32_t layer = 0; layer < radv_get_layerCount(image, subresourceRange); layer++) {
struct radv_image_view iview;
&(VkImageViewCreateInfo) {
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.image = radv_image_to_handle(image),
+ .viewType = radv_meta_get_view_type(image),
.format = image->vk_format,
.subresourceRange = {
.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
.baseArrayLayer = subresourceRange->baseArrayLayer + layer,
.layerCount = 1,
},
- },
- cmd_buffer, VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT);
+ });
VkFramebuffer fb_h;
radv_CmdBeginRenderPass(cmd_buffer_h,
&(VkRenderPassBeginInfo) {
.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
- .renderPass = cmd_buffer->device->meta_state.depth_decomp.pass,
+ .renderPass = meta_state->depth_decomp[samples_log2].pass,
.framebuffer = fb_h,
.renderArea = {
.offset = {
},
VK_SUBPASS_CONTENTS_INLINE);
- emit_depth_decomp(cmd_buffer, &(VkOffset2D){0, 0 }, &(VkExtent2D){width, height}, pipeline_h);
+ radv_CmdDraw(cmd_buffer_h, 3, 1, 0, 0);
radv_CmdEndRenderPass(cmd_buffer_h);
radv_DestroyFramebuffer(device_h, fb_h,
&cmd_buffer->pool->alloc);
}
radv_meta_restore(&saved_state, cmd_buffer);
- radv_meta_restore_pass(&saved_pass_state, cmd_buffer);
}
void radv_decompress_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
struct radv_image *image,
- VkImageSubresourceRange *subresourceRange)
+ VkImageSubresourceRange *subresourceRange,
+ struct radv_sample_locations_state *sample_locs)
{
assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange,
- cmd_buffer->device->meta_state.depth_decomp.decompress_pipeline);
+ sample_locs, DEPTH_DECOMPRESS);
}
void radv_resummarize_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
struct radv_image *image,
- VkImageSubresourceRange *subresourceRange)
+ VkImageSubresourceRange *subresourceRange,
+ struct radv_sample_locations_state *sample_locs)
{
assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
radv_process_depth_image_inplace(cmd_buffer, image, subresourceRange,
- cmd_buffer->device->meta_state.depth_decomp.resummarize_pipeline);
+ sample_locs, DEPTH_RESUMMARIZE);
}