From e3a2ca96bd2350471ebb6c2907c67b10254a4f7e Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Fri, 14 Apr 2023 18:07:38 +0200 Subject: Accelerate DMA: Use texture cache async downloads to perform the copies to host. WIP --- src/video_core/renderer_vulkan/vk_rasterizer.cpp | 6 +++--- src/video_core/renderer_vulkan/vk_texture_cache.cpp | 21 +++++++++++++-------- src/video_core/renderer_vulkan/vk_texture_cache.h | 2 +- 3 files changed, 17 insertions(+), 12 deletions(-) (limited to 'src/video_core/renderer_vulkan') diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 8fc783cc0..2559a3aa7 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -781,8 +781,7 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, } const u32 buffer_size = static_cast(buffer_operand.pitch * buffer_operand.height); static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize; - const auto post_op = IS_IMAGE_UPLOAD ? VideoCommon::ObtainBufferOperation::DoNothing - : VideoCommon::ObtainBufferOperation::MarkAsWritten; + const auto post_op = VideoCommon::ObtainBufferOperation::DoNothing; const auto [buffer, offset] = buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op); @@ -793,7 +792,8 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, if constexpr (IS_IMAGE_UPLOAD) { image->UploadMemory(buffer->Handle(), offset, copy_span); } else { - texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span); + texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span, + buffer_operand.address, buffer_size); } return true; } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index e4d077e63..da3841bb3 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -1342,17 +1342,19 @@ void Image::UploadMemory(const StagingBufferRef& map, std::span buffers_span, VkDeviceSize offset, +void Image::DownloadMemory(std::span buffers_span, std::span offsets_span, std::span copies) { const bool is_rescaled = True(flags & ImageFlagBits::Rescaled); if (is_rescaled) { ScaleDown(); } boost::container::small_vector buffers_vector{}; - for (auto& buffer : buffers_span) { - buffers_vector.push_back(buffer); + boost::container::small_vector, 1> vk_copies; + for (size_t index = 0; index < buffers_span.size(); index++) { + buffers_vector.emplace_back(buffers_span[index]); + vk_copies.emplace_back( + TransformBufferImageCopies(copies, offsets_span[index], aspect_mask)); } - std::vector vk_copies = TransformBufferImageCopies(copies, offset, aspect_mask); scheduler->RequestOutsideRenderPassOperationContext(); scheduler->Record([buffers = std::move(buffers_vector), image = *original_image, aspect_mask = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) { @@ -1377,9 +1379,9 @@ void Image::DownloadMemory(std::span buffers_span, VkDeviceSize offset cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, read_barrier); - for (auto buffer : buffers) { - cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffer, - vk_copies); + for (size_t index = 0; index < buffers.size(); index++) { + cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffers[index], + vk_copies[index]); } const VkMemoryBarrier memory_write_barrier{ @@ -1418,7 +1420,10 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span copies); - void DownloadMemory(std::span buffers, VkDeviceSize offset, + void DownloadMemory(std::span buffers, std::span offsets, std::span copies); void DownloadMemory(const StagingBufferRef& map, -- cgit v1.2.3