diff options
Diffstat (limited to 'src/video_core/renderer_opengl')
4 files changed, 21 insertions, 18 deletions
diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp index f822fa856..44a771d65 100644 --- a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp @@ -220,7 +220,8 @@ GraphicsPipeline::GraphicsPipeline(const Device& device, TextureCache& texture_c ASSERT(num_textures <= MAX_TEXTURES); ASSERT(num_images <= MAX_IMAGES); - const bool assembly_shaders{assembly_programs[0].handle != 0}; + const auto backend = device.GetShaderBackend(); + const bool assembly_shaders{backend == Settings::ShaderBackend::Glasm}; use_storage_buffers = !assembly_shaders || num_storage_buffers <= device.GetMaxGLASMStorageBufferBlocks(); writes_global_memory &= !use_storage_buffers; @@ -230,7 +231,6 @@ GraphicsPipeline::GraphicsPipeline(const Device& device, TextureCache& texture_c GenerateTransformFeedbackState(); } const bool in_parallel = thread_worker != nullptr; - const auto backend = device.GetShaderBackend(); auto func{[this, sources_ = std::move(sources), sources_spirv_ = std::move(sources_spirv), shader_notify, backend, in_parallel, force_context_flush](ShaderContext::Context*) mutable { @@ -559,15 +559,13 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { } void GraphicsPipeline::ConfigureTransformFeedbackImpl() const { - glTransformFeedbackStreamAttribsNV(num_xfb_attribs, xfb_attribs.data(), num_xfb_strides, - xfb_streams.data(), GL_INTERLEAVED_ATTRIBS); + glTransformFeedbackAttribsNV(num_xfb_attribs, xfb_attribs.data(), GL_SEPARATE_ATTRIBS); } void GraphicsPipeline::GenerateTransformFeedbackState() { // TODO(Rodrigo): Inject SKIP_COMPONENTS*_NV when required. An unimplemented message will signal // when this is required. GLint* cursor{xfb_attribs.data()}; - GLint* current_stream{xfb_streams.data()}; for (size_t feedback = 0; feedback < Maxwell::NumTransformFeedbackBuffers; ++feedback) { const auto& layout = key.xfb_state.layouts[feedback]; @@ -575,15 +573,6 @@ void GraphicsPipeline::GenerateTransformFeedbackState() { if (layout.varying_count == 0) { continue; } - *current_stream = static_cast<GLint>(feedback); - if (current_stream != xfb_streams.data()) { - // When stepping one stream, push the expected token - cursor[0] = GL_NEXT_BUFFER_NV; - cursor[1] = 0; - cursor[2] = 0; - cursor += XFB_ENTRY_STRIDE; - } - ++current_stream; const auto& locations = key.xfb_state.varyings[feedback]; std::optional<u32> current_index; @@ -619,7 +608,6 @@ void GraphicsPipeline::GenerateTransformFeedbackState() { } } num_xfb_attribs = static_cast<GLsizei>((cursor - xfb_attribs.data()) / XFB_ENTRY_STRIDE); - num_xfb_strides = static_cast<GLsizei>(current_stream - xfb_streams.data()); } void GraphicsPipeline::WaitForBuild() { diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.h b/src/video_core/renderer_opengl/gl_graphics_pipeline.h index 7b3d7eae8..74fc9cc3d 100644 --- a/src/video_core/renderer_opengl/gl_graphics_pipeline.h +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.h @@ -154,9 +154,7 @@ private: static constexpr std::size_t XFB_ENTRY_STRIDE = 3; GLsizei num_xfb_attribs{}; - GLsizei num_xfb_strides{}; std::array<GLint, 128 * XFB_ENTRY_STRIDE * Maxwell::NumTransformFeedbackBuffers> xfb_attribs{}; - std::array<GLint, Maxwell::NumTransformFeedbackBuffers> xfb_streams{}; std::mutex built_mutex; std::condition_variable built_condvar; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index aadd6967c..dd03efecd 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -380,6 +380,17 @@ void RasterizerOpenGL::DispatchCompute() { pipeline->SetEngine(kepler_compute, gpu_memory); pipeline->Configure(); const auto& qmd{kepler_compute->launch_description}; + auto indirect_address = kepler_compute->GetIndirectComputeAddress(); + if (indirect_address) { + // DispatchIndirect + static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize; + const auto post_op = VideoCommon::ObtainBufferOperation::DiscardWrite; + const auto [buffer, offset] = + buffer_cache.ObtainBuffer(*indirect_address, 12, sync_info, post_op); + glBindBuffer(GL_DISPATCH_INDIRECT_BUFFER, buffer->Handle()); + glDispatchComputeIndirect(static_cast<GLintptr>(offset)); + return; + } glDispatchCompute(qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z); ++num_queued_commands; has_written_global_memory |= pipeline->WritesGlobalMemory(); @@ -1335,7 +1346,8 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, } const u32 buffer_size = static_cast<u32>(buffer_operand.pitch * buffer_operand.height); static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize; - const auto post_op = VideoCommon::ObtainBufferOperation::DoNothing; + 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); @@ -1344,8 +1356,12 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, const std::span copy_span{©, 1}; if constexpr (IS_IMAGE_UPLOAD) { + texture_cache.PrepareImage(image_id, true, false); image->UploadMemory(buffer->Handle(), offset, copy_span); } else { + if (offset % BytesPerBlock(image->info.format)) { + return false; + } texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span, buffer_operand.address, buffer_size); } diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp index 544982d18..c437013e6 100644 --- a/src/video_core/renderer_opengl/util_shaders.cpp +++ b/src/video_core/renderer_opengl/util_shaders.cpp @@ -68,6 +68,7 @@ void UtilShaders::ASTCDecode(Image& image, const StagingBufferMap& map, std::span<const VideoCommon::SwizzleParameters> swizzles) { static constexpr GLuint BINDING_INPUT_BUFFER = 0; static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; + program_manager.LocalMemoryWarmup(); const Extent2D tile_size{ .width = VideoCore::Surface::DefaultBlockWidth(image.info.format), |