From 6b9cc0ed23b15a1b96b322b03feff2153e44a4a9 Mon Sep 17 00:00:00 2001 From: ameerj <52414509+ameerj@users.noreply.github.com> Date: Mon, 6 Mar 2023 21:16:17 -0500 Subject: Refactor AccelerateDMA code --- .../renderer_opengl/gl_texture_cache.cpp | 24 +++++++++++++++------- src/video_core/renderer_opengl/gl_texture_cache.h | 6 ++++++ 2 files changed, 23 insertions(+), 7 deletions(-) (limited to 'src/video_core/renderer_opengl') diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index b047e7b3d..4b13e807d 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -762,14 +762,14 @@ Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBas Image::~Image() = default; -void Image::UploadMemory(const ImageBufferMap& map, +void Image::UploadMemory(GLuint buffer_handle, size_t buffer_offset, std::span copies) { const bool is_rescaled = True(flags & ImageFlagBits::Rescaled); if (is_rescaled) { ScaleDown(true); } - glBindBuffer(GL_PIXEL_UNPACK_BUFFER, map.buffer); - glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, map.offset, unswizzled_size_bytes); + glBindBuffer(GL_PIXEL_UNPACK_BUFFER, buffer_handle); + glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, buffer_offset, unswizzled_size_bytes); glPixelStorei(GL_UNPACK_ALIGNMENT, 1); @@ -788,21 +788,26 @@ void Image::UploadMemory(const ImageBufferMap& map, current_image_height = copy.buffer_image_height; glPixelStorei(GL_UNPACK_IMAGE_HEIGHT, current_image_height); } - CopyBufferToImage(copy, map.offset); + CopyBufferToImage(copy, buffer_offset); } if (is_rescaled) { ScaleUp(); } } -void Image::DownloadMemory(ImageBufferMap& map, +void Image::UploadMemory(const ImageBufferMap& map, + std::span copies) { + UploadMemory(map.buffer, map.offset, copies); +} + +void Image::DownloadMemory(GLuint buffer_handle, size_t buffer_offset, std::span copies) { const bool is_rescaled = True(flags & ImageFlagBits::Rescaled); if (is_rescaled) { ScaleDown(); } glMemoryBarrier(GL_PIXEL_BUFFER_BARRIER_BIT); // TODO: Move this to its own API - glBindBuffer(GL_PIXEL_PACK_BUFFER, map.buffer); + glBindBuffer(GL_PIXEL_PACK_BUFFER, buffer_handle); glPixelStorei(GL_PACK_ALIGNMENT, 1); u32 current_row_length = std::numeric_limits::max(); @@ -820,13 +825,18 @@ void Image::DownloadMemory(ImageBufferMap& map, current_image_height = copy.buffer_image_height; glPixelStorei(GL_PACK_IMAGE_HEIGHT, current_image_height); } - CopyImageToBuffer(copy, map.offset); + CopyImageToBuffer(copy, buffer_offset); } if (is_rescaled) { ScaleUp(true); } } +void Image::DownloadMemory(ImageBufferMap& map, + std::span copies) { + DownloadMemory(map.buffer, map.offset, copies); +} + GLuint Image::StorageHandle() noexcept { switch (info.format) { case PixelFormat::A8B8G8R8_SRGB: diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index e30875496..911e4607a 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -206,9 +206,15 @@ public: Image(Image&&) = default; Image& operator=(Image&&) = default; + void UploadMemory(GLuint buffer_handle, size_t buffer_offset, + std::span copies); + void UploadMemory(const ImageBufferMap& map, std::span copies); + void DownloadMemory(GLuint buffer_handle, size_t buffer_offset, + std::span copies); + void DownloadMemory(ImageBufferMap& map, std::span copies); GLuint StorageHandle() noexcept; -- cgit v1.2.3 From 268942c8fe0c34f193ba3d195a974e91db8af26a Mon Sep 17 00:00:00 2001 From: ameerj <52414509+ameerj@users.noreply.github.com> Date: Mon, 6 Mar 2023 21:53:51 -0500 Subject: gl_rasterizer: Implement AccelerateDMA DmaBufferImageCopy --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 45 ++++++++++++++++++++++-- src/video_core/renderer_opengl/gl_rasterizer.h | 16 +++++---- 2 files changed, 52 insertions(+), 9 deletions(-) (limited to 'src/video_core/renderer_opengl') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 7bced675c..33748762f 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -63,7 +63,7 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra buffer_cache(*this, cpu_memory_, buffer_cache_runtime), shader_cache(*this, emu_window_, device, texture_cache, buffer_cache, program_manager, state_tracker, gpu.ShaderNotify()), - query_cache(*this), accelerate_dma(buffer_cache), + query_cache(*this), accelerate_dma(buffer_cache, texture_cache), fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache), blit_image(program_manager_) {} @@ -1262,7 +1262,8 @@ void RasterizerOpenGL::ReleaseChannel(s32 channel_id) { query_cache.EraseChannel(channel_id); } -AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {} +AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_, TextureCache& texture_cache_) + : buffer_cache{buffer_cache_}, texture_cache{texture_cache_} {} bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) { std::scoped_lock lock{buffer_cache.mutex}; @@ -1274,4 +1275,44 @@ bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) { return buffer_cache.DMAClear(src_address, amount, value); } +template +bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, + const Tegra::DMA::BufferOperand& buffer_operand, + const Tegra::DMA::ImageOperand& image_operand) { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + const auto image_id = texture_cache.DmaImageId(image_operand); + if (image_id == VideoCommon::NULL_IMAGE_ID) { + return false; + } + 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 [buffer, offset] = + buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op); + + const auto [image, copy] = texture_cache.DmaBufferImageCopy( + copy_info, buffer_operand, image_operand, image_id, IS_IMAGE_UPLOAD); + const std::span copy_span{©, 1}; + + if constexpr (IS_IMAGE_UPLOAD) { + image->UploadMemory(buffer->Handle(), offset, copy_span); + } else { + image->DownloadMemory(buffer->Handle(), offset, copy_span); + } + return true; +} + +bool AccelerateDMA::ImageToBuffer(const Tegra::DMA::ImageCopy& copy_info, + const Tegra::DMA::ImageOperand& image_operand, + const Tegra::DMA::BufferOperand& buffer_operand) { + return DmaBufferImageCopy(copy_info, buffer_operand, image_operand); +} + +bool AccelerateDMA::BufferToImage(const Tegra::DMA::ImageCopy& copy_info, + const Tegra::DMA::BufferOperand& buffer_operand, + const Tegra::DMA::ImageOperand& image_operand) { + return DmaBufferImageCopy(copy_info, buffer_operand, image_operand); +} + } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 7e21fc43d..1f6562ef8 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -50,24 +50,26 @@ static_assert(sizeof(BindlessSSBO) * CHAR_BIT == 128); class AccelerateDMA : public Tegra::Engines::AccelerateDMAInterface { public: - explicit AccelerateDMA(BufferCache& buffer_cache); + explicit AccelerateDMA(BufferCache& buffer_cache, TextureCache& texture_cache); bool BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) override; bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override; bool ImageToBuffer(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::ImageOperand& src, - const Tegra::DMA::BufferOperand& dst) override { - return false; - } + const Tegra::DMA::BufferOperand& dst) override; bool BufferToImage(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& src, - const Tegra::DMA::ImageOperand& dst) override { - return false; - } + const Tegra::DMA::ImageOperand& dst) override; private: + template + bool DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, + const Tegra::DMA::BufferOperand& src, + const Tegra::DMA::ImageOperand& dst); + BufferCache& buffer_cache; + TextureCache& texture_cache; }; class RasterizerOpenGL : public VideoCore::RasterizerAccelerated, -- cgit v1.2.3