summaryrefslogtreecommitdiffstats
path: root/src/video_core/renderer_vulkan/vk_compute_pass.cpp
diff options
context:
space:
mode:
authorameerj <52414509+ameerj@users.noreply.github.com>2021-02-13 22:49:24 +0100
committerameerj <52414509+ameerj@users.noreply.github.com>2021-03-13 18:16:03 +0100
commit20eb368e147e1c27f05d6923c51596f8dfe24e89 (patch)
treea8b1c8eb79eb55e189a10dfd43b8b6bb1449220f /src/video_core/renderer_vulkan/vk_compute_pass.cpp
parenthost_shaders: Modify shader cmake integration to allow for larger shaders (diff)
downloadyuzu-20eb368e147e1c27f05d6923c51596f8dfe24e89.tar
yuzu-20eb368e147e1c27f05d6923c51596f8dfe24e89.tar.gz
yuzu-20eb368e147e1c27f05d6923c51596f8dfe24e89.tar.bz2
yuzu-20eb368e147e1c27f05d6923c51596f8dfe24e89.tar.lz
yuzu-20eb368e147e1c27f05d6923c51596f8dfe24e89.tar.xz
yuzu-20eb368e147e1c27f05d6923c51596f8dfe24e89.tar.zst
yuzu-20eb368e147e1c27f05d6923c51596f8dfe24e89.zip
Diffstat (limited to 'src/video_core/renderer_vulkan/vk_compute_pass.cpp')
-rw-r--r--src/video_core/renderer_vulkan/vk_compute_pass.cpp298
1 files changed, 298 insertions, 0 deletions
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
index 2f9a7b028..7587ab1e0 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
@@ -11,18 +11,38 @@
#include "common/assert.h"
#include "common/common_types.h"
#include "common/div_ceil.h"
+#include "video_core/host_shaders/astc_decoder_comp_spv.h"
#include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
#include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
+#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
+#include "video_core/texture_cache/accelerated_swizzle.h"
+#include "video_core/texture_cache/types.h"
+#include "video_core/textures/astc.h"
+#include "video_core/textures/decoders.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
+
+using Tegra::Texture::SWIZZLE_TABLE;
+using Tegra::Texture::ASTC::EncodingsValues;
+
namespace {
+
+constexpr u32 ASTC_BINDING_SWIZZLE_BUFFER = 0;
+constexpr u32 ASTC_BINDING_INPUT_BUFFER = 1;
+constexpr u32 ASTC_BINDING_ENC_BUFFER = 2;
+constexpr u32 ASTC_BINDING_6_TO_8_BUFFER = 3;
+constexpr u32 ASTC_BINDING_7_TO_8_BUFFER = 4;
+constexpr u32 ASTC_BINDING_8_TO_8_BUFFER = 5;
+constexpr u32 ASTC_BINDING_BYTE_TO_16_BUFFER = 6;
+constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 7;
+
VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
return {
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
@@ -50,6 +70,67 @@ std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBinding
}};
}
+std::array<VkDescriptorSetLayoutBinding, 8> BuildASTCDescriptorSetBindings() {
+ return {{
+ {
+ .binding = ASTC_BINDING_SWIZZLE_BUFFER, // Swizzle buffer
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ {
+ .binding = ASTC_BINDING_INPUT_BUFFER, // ASTC Img data buffer
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ {
+ .binding = ASTC_BINDING_ENC_BUFFER, // Encodings buffer
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ {
+ .binding = ASTC_BINDING_6_TO_8_BUFFER, // BINDING_6_TO_8_BUFFER
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ {
+ .binding = ASTC_BINDING_7_TO_8_BUFFER, // BINDING_7_TO_8_BUFFER
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ {
+ .binding = ASTC_BINDING_8_TO_8_BUFFER, // BINDING_8_TO_8_BUFFER
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ {
+ .binding = ASTC_BINDING_BYTE_TO_16_BUFFER, // BINDING_BYTE_TO_16_BUFFER
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ {
+ .binding = ASTC_BINDING_OUTPUT_IMAGE, // Output image
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ }};
+}
+
VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
return {
.dstBinding = 0,
@@ -61,6 +142,90 @@ VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
};
}
+std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateTemplateEntry() {
+ return {{
+ {
+ .dstBinding = ASTC_BINDING_SWIZZLE_BUFFER,
+ .dstArrayElement = 0,
+ .descriptorCount = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .offset = 0 * sizeof(DescriptorUpdateEntry),
+ .stride = sizeof(DescriptorUpdateEntry),
+ },
+ {
+ .dstBinding = ASTC_BINDING_INPUT_BUFFER,
+ .dstArrayElement = 0,
+ .descriptorCount = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .offset = 1 * sizeof(DescriptorUpdateEntry),
+ .stride = sizeof(DescriptorUpdateEntry),
+ },
+ {
+ .dstBinding = ASTC_BINDING_ENC_BUFFER,
+ .dstArrayElement = 0,
+ .descriptorCount = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .offset = 2 * sizeof(DescriptorUpdateEntry),
+ .stride = sizeof(DescriptorUpdateEntry),
+ },
+ {
+ .dstBinding = ASTC_BINDING_6_TO_8_BUFFER,
+ .dstArrayElement = 0,
+ .descriptorCount = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .offset = 3 * sizeof(DescriptorUpdateEntry),
+ .stride = sizeof(DescriptorUpdateEntry),
+ },
+ {
+ .dstBinding = ASTC_BINDING_7_TO_8_BUFFER,
+ .dstArrayElement = 0,
+ .descriptorCount = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .offset = 4 * sizeof(DescriptorUpdateEntry),
+ .stride = sizeof(DescriptorUpdateEntry),
+ },
+ {
+ .dstBinding = ASTC_BINDING_8_TO_8_BUFFER,
+ .dstArrayElement = 0,
+ .descriptorCount = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .offset = 5 * sizeof(DescriptorUpdateEntry),
+ .stride = sizeof(DescriptorUpdateEntry),
+ },
+ {
+ .dstBinding = ASTC_BINDING_BYTE_TO_16_BUFFER,
+ .dstArrayElement = 0,
+ .descriptorCount = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .offset = 6 * sizeof(DescriptorUpdateEntry),
+ .stride = sizeof(DescriptorUpdateEntry),
+ },
+ {
+ .dstBinding = ASTC_BINDING_OUTPUT_IMAGE,
+ .dstArrayElement = 0,
+ .descriptorCount = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+ .offset = 7 * sizeof(DescriptorUpdateEntry),
+ .stride = sizeof(DescriptorUpdateEntry),
+ },
+ }};
+}
+
+struct AstcPushConstants {
+ std::array<u32, 2> num_image_blocks;
+ std::array<u32, 2> blocks_dims;
+ u32 layer;
+ VideoCommon::Accelerated::BlockLinearSwizzle2DParams params;
+};
+
+struct AstcBufferData {
+ decltype(SWIZZLE_TABLE) swizzle_table_buffer = SWIZZLE_TABLE;
+ decltype(EncodingsValues) encoding_values = EncodingsValues;
+ decltype(REPLICATE_6_BIT_TO_8_TABLE) replicate_6_to_8 = REPLICATE_6_BIT_TO_8_TABLE;
+ decltype(REPLICATE_7_BIT_TO_8_TABLE) replicate_7_to_8 = REPLICATE_7_BIT_TO_8_TABLE;
+ decltype(REPLICATE_8_BIT_TO_8_TABLE) replicate_8_to_8 = REPLICATE_8_BIT_TO_8_TABLE;
+ decltype(REPLICATE_BYTE_TO_16_TABLE) replicate_byte_to_16 = REPLICATE_BYTE_TO_16_TABLE;
+} constexpr ASTC_BUFFER_DATA;
} // Anonymous namespace
VKComputePass::VKComputePass(const Device& device, VKDescriptorPool& descriptor_pool,
@@ -238,4 +403,137 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
return {staging.buffer, staging.offset};
}
+using namespace Tegra::Texture::ASTC;
+ASTCDecoderPass::ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_,
+ VKDescriptorPool& descriptor_pool_,
+ StagingBufferPool& staging_buffer_pool_,
+ VKUpdateDescriptorQueue& update_descriptor_queue_,
+ MemoryAllocator& memory_allocator_)
+ : VKComputePass(device_, descriptor_pool_, BuildASTCDescriptorSetBindings(),
+ BuildASTCPassDescriptorUpdateTemplateEntry(),
+ BuildComputePushConstantRange(sizeof(AstcPushConstants)),
+ ASTC_DECODER_COMP_SPV),
+ device{device_}, scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
+ update_descriptor_queue{update_descriptor_queue_}, memory_allocator{memory_allocator_} {}
+
+ASTCDecoderPass::~ASTCDecoderPass() = default;
+
+void ASTCDecoderPass::MakeDataBuffer() {
+ data_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
+ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .size = sizeof(ASTC_BUFFER_DATA),
+ .usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+ .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+ .queueFamilyIndexCount = 0,
+ .pQueueFamilyIndices = nullptr,
+ });
+ data_buffer_commit = memory_allocator.Commit(data_buffer, MemoryUsage::Upload);
+
+ const auto staging_ref =
+ staging_buffer_pool.Request(sizeof(ASTC_BUFFER_DATA), MemoryUsage::Upload);
+ std::memcpy(staging_ref.mapped_span.data(), &ASTC_BUFFER_DATA, sizeof(ASTC_BUFFER_DATA));
+ scheduler.Record([src = staging_ref.buffer, dst = *data_buffer](vk::CommandBuffer cmdbuf) {
+ cmdbuf.CopyBuffer(src, dst,
+ VkBufferCopy{
+ .srcOffset = 0,
+ .dstOffset = 0,
+ .size = sizeof(ASTC_BUFFER_DATA),
+ });
+ cmdbuf.PipelineBarrier(
+ VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0,
+ VkMemoryBarrier{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = 0,
+ .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
+ },
+ {}, {});
+ });
+}
+
+void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
+ std::span<const VideoCommon::SwizzleParameters> swizzles) {
+ using namespace VideoCommon::Accelerated;
+ const VideoCommon::Extent2D tile_size{
+ .width = VideoCore::Surface::DefaultBlockWidth(image.info.format),
+ .height = VideoCore::Surface::DefaultBlockHeight(image.info.format),
+ };
+ scheduler.RequestOutsideRenderPassOperationContext();
+ if (!data_buffer) {
+ MakeDataBuffer();
+ }
+ const std::array<u32, 2> block_dims{tile_size.width, tile_size.height};
+ for (s32 layer = 0; layer < image.info.resources.layers; layer++) {
+ for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
+ const size_t input_offset = swizzle.buffer_offset + map.offset;
+ const auto num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
+ const auto num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
+ const std::array num_image_blocks{swizzle.num_tiles.width, swizzle.num_tiles.height};
+ const u32 layer_image_size =
+ image.guest_size_bytes - static_cast<u32>(swizzle.buffer_offset);
+
+ update_descriptor_queue.Acquire();
+ update_descriptor_queue.AddBuffer(*data_buffer,
+ offsetof(AstcBufferData, swizzle_table_buffer),
+ sizeof(AstcBufferData::swizzle_table_buffer));
+ update_descriptor_queue.AddBuffer(map.buffer, input_offset, image.guest_size_bytes);
+ update_descriptor_queue.AddBuffer(*data_buffer,
+ offsetof(AstcBufferData, encoding_values),
+ sizeof(AstcBufferData::encoding_values));
+ update_descriptor_queue.AddBuffer(*data_buffer,
+ offsetof(AstcBufferData, replicate_6_to_8),
+ sizeof(AstcBufferData::replicate_6_to_8));
+ update_descriptor_queue.AddBuffer(*data_buffer,
+ offsetof(AstcBufferData, replicate_7_to_8),
+ sizeof(AstcBufferData::replicate_7_to_8));
+ update_descriptor_queue.AddBuffer(*data_buffer,
+ offsetof(AstcBufferData, replicate_8_to_8),
+ sizeof(AstcBufferData::replicate_8_to_8));
+ update_descriptor_queue.AddBuffer(*data_buffer,
+ offsetof(AstcBufferData, replicate_byte_to_16),
+ sizeof(AstcBufferData::replicate_byte_to_16));
+ update_descriptor_queue.AddImage(image.StorageImageView());
+
+ const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
+ // To unswizzle the ASTC data
+ const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
+ scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = map.buffer,
+ num_dispatches_x, num_dispatches_y, layer_image_size,
+ num_image_blocks, block_dims, layer, params, set,
+ image = image.Handle(), input_offset,
+ aspect_mask = image.AspectMask()](vk::CommandBuffer cmdbuf) {
+ const AstcPushConstants uniforms{num_image_blocks, block_dims, layer, params};
+
+ cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
+ cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
+ cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms);
+ cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, 1);
+
+ const VkImageMemoryBarrier image_barrier{
+ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
+ .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+ .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+ .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+ .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+ .image = image,
+ .subresourceRange{
+ .aspectMask = aspect_mask,
+ .baseMipLevel = 0,
+ .levelCount = VK_REMAINING_MIP_LEVELS,
+ .baseArrayLayer = 0,
+ .layerCount = VK_REMAINING_ARRAY_LAYERS,
+ },
+ };
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+ VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, image_barrier);
+ });
+ }
+ }
+}
+
} // namespace Vulkan