diff options
author | Fernando Sahmkow <fsahmkow27@gmail.com> | 2019-09-15 17:48:54 +0200 |
---|---|---|
committer | FernandoS27 <fsahmkow27@gmail.com> | 2019-09-19 17:41:27 +0200 |
commit | ba02d564f8a0b0167b96f247b6ad9d2bde05b6c8 (patch) | |
tree | 015edafb7deacabf4fea3c547534bb00ed60781f /src/video_core/renderer_opengl | |
parent | Merge pull request #2784 from ReinUsesLisp/smem (diff) | |
download | yuzu-ba02d564f8a0b0167b96f247b6ad9d2bde05b6c8.tar yuzu-ba02d564f8a0b0167b96f247b6ad9d2bde05b6c8.tar.gz yuzu-ba02d564f8a0b0167b96f247b6ad9d2bde05b6c8.tar.bz2 yuzu-ba02d564f8a0b0167b96f247b6ad9d2bde05b6c8.tar.lz yuzu-ba02d564f8a0b0167b96f247b6ad9d2bde05b6c8.tar.xz yuzu-ba02d564f8a0b0167b96f247b6ad9d2bde05b6c8.tar.zst yuzu-ba02d564f8a0b0167b96f247b6ad9d2bde05b6c8.zip |
Diffstat (limited to 'src/video_core/renderer_opengl')
-rw-r--r-- | src/video_core/renderer_opengl/gl_rasterizer.cpp | 89 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_rasterizer.h | 6 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_shader_decompiler.cpp | 2 |
3 files changed, 88 insertions, 9 deletions
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 4dd08bccb..5df7f3f56 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -405,6 +405,12 @@ bool RasterizerOpenGL::AccelerateDrawBatch(bool is_indexed) { return true; } +bool RasterizerOpenGL::AccelerateDrawMultiBatch(bool is_indexed) { + accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays; + DrawMultiArrays(); + return true; +} + template <typename Map, typename Interval> static constexpr auto RangeFromInterval(Map& map, const Interval& interval) { return boost::make_iterator_range(map.equal_range(interval)); @@ -688,7 +694,7 @@ void RasterizerOpenGL::Clear() { } } -void RasterizerOpenGL::DrawArrays() { +void RasterizerOpenGL::DrawPrelude() { if (accelerate_draw == AccelDraw::Disabled) return; @@ -743,10 +749,7 @@ void RasterizerOpenGL::DrawArrays() { // Upload vertex and index data. SetupVertexBuffer(vao); SetupVertexInstances(vao); - const GLintptr index_buffer_offset = SetupIndexBuffer(); - - // Setup draw parameters. It will automatically choose what glDraw* method to use. - const DrawParameters params = SetupDraw(index_buffer_offset); + index_buffer_offset = SetupIndexBuffer(); // Prepare packed bindings. bind_ubo_pushbuffer.Setup(0); @@ -754,7 +757,8 @@ void RasterizerOpenGL::DrawArrays() { // Setup shaders and their used resources. texture_cache.GuardSamplers(true); - SetupShaders(params.primitive_mode); + const auto primitive_mode = MaxwellToGL::PrimitiveTopology(gpu.regs.draw.topology); + SetupShaders(primitive_mode); texture_cache.GuardSamplers(false); ConfigureFramebuffers(state); @@ -778,11 +782,80 @@ void RasterizerOpenGL::DrawArrays() { if (texture_cache.TextureBarrier()) { glTextureBarrier(); } +} - params.DispatchDraw(); +void RasterizerOpenGL::DrawArrays() { + DrawPrelude(); + + auto& maxwell3d = system.GPU().Maxwell3D(); + auto& regs = maxwell3d.regs; + auto current_instance = maxwell3d.state.current_instance; + auto primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology); + if (accelerate_draw == AccelDraw::Indexed) { + auto index_format = MaxwellToGL::IndexFormat(regs.index_array.format); + auto count = regs.index_array.count; + auto base_vertex = static_cast<GLint>(regs.vb_element_base); + const auto index_buffer_ptr = reinterpret_cast<const void*>(index_buffer_offset); + if (current_instance > 0) { + glDrawElementsInstancedBaseVertexBaseInstance(primitive_mode, count, index_format, + index_buffer_ptr, 1, base_vertex, + current_instance); + } else { + glDrawElementsBaseVertex(primitive_mode, count, index_format, index_buffer_ptr, + base_vertex); + } + } else { + auto count = regs.vertex_buffer.count; + auto vertex_first = regs.vertex_buffer.first; + if (current_instance > 0) { + glDrawArraysInstancedBaseInstance(primitive_mode, vertex_first, count, 1, + current_instance); + } else { + glDrawArrays(primitive_mode, vertex_first, count); + } + } + + accelerate_draw = AccelDraw::Disabled; + maxwell3d.dirty.memory_general = false; +} + +#pragma optimize("", off) + +void RasterizerOpenGL::DrawMultiArrays() { + DrawPrelude(); + + auto& maxwell3d = system.GPU().Maxwell3D(); + auto& regs = maxwell3d.regs; + auto& draw_setup = maxwell3d.mme_draw; + auto num_instances = draw_setup.instance_count; + auto base_instance = static_cast<GLint>(regs.vb_base_instance); + auto primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology); + if (draw_setup.current_mode == Tegra::Engines::Maxwell3D::MMMEDrawMode::Indexed) { + auto index_format = MaxwellToGL::IndexFormat(regs.index_array.format); + auto count = regs.index_array.count; + auto base_vertex = static_cast<GLint>(regs.vb_element_base); + const auto index_buffer_ptr = reinterpret_cast<const void*>(index_buffer_offset); + if (num_instances > 1) { + glDrawElementsInstancedBaseVertexBaseInstance(primitive_mode, count, index_format, + index_buffer_ptr, num_instances, + base_vertex, base_instance); + } else { + glDrawElementsBaseVertex(primitive_mode, count, index_format, index_buffer_ptr, + base_vertex); + } + } else { + auto count = regs.vertex_buffer.count; + auto vertex_first = regs.vertex_buffer.first; + if (num_instances > 1) { + glDrawArraysInstancedBaseInstance(primitive_mode, vertex_first, count, num_instances, + base_instance); + } else { + glDrawArrays(primitive_mode, vertex_first, count); + } + } accelerate_draw = AccelDraw::Disabled; - gpu.dirty.memory_general = false; + maxwell3d.dirty.memory_general = false; } void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index eada752e0..63a914ff9 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -58,6 +58,7 @@ public: ~RasterizerOpenGL() override; void DrawArrays() override; + void DrawMultiArrays() override; void Clear() override; void DispatchCompute(GPUVAddr code_addr) override; void FlushAll() override; @@ -72,6 +73,7 @@ public: bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; bool AccelerateDrawBatch(bool is_indexed) override; + bool AccelerateDrawMultiBatch(bool is_indexed) override; void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) override; void LoadDiskResources(const std::atomic_bool& stop_loading, const VideoCore::DiskResourceLoadCallback& callback) override; @@ -136,6 +138,8 @@ private: void SetupGlobalMemory(const GLShader::GlobalMemoryEntry& entry, GPUVAddr gpu_addr, std::size_t size); + void DrawPrelude(); + /// Configures the current textures to use for the draw command. Returns shaders texture buffer /// usage. TextureBufferUsage SetupDrawTextures(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage, @@ -252,6 +256,8 @@ private: DrawParameters SetupDraw(GLintptr index_buffer_offset); + GLintptr index_buffer_offset; + void SetupShaders(GLenum primitive_mode); enum class AccelDraw { Disabled, Arrays, Indexed }; diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 14834d86a..62e32697e 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -964,7 +964,7 @@ private: switch (element) { case 2: // Config pack's first value is instance_id. - return {"config_pack[0]", Type::Uint}; + return {"gl_InstanceID", Type::Uint}; case 3: return {"gl_VertexID", Type::Int}; } |