diff options
Diffstat (limited to 'src/video_core/renderer_opengl/gl_rasterizer.cpp')
-rw-r--r-- | src/video_core/renderer_opengl/gl_rasterizer.cpp | 125 |
1 files changed, 86 insertions, 39 deletions
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index e960a0ef1..03e82c599 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -139,6 +139,18 @@ void oglEnable(GLenum cap, bool state) { (state ? glEnable : glDisable)(cap); } +void UpdateBindlessPointers(GLenum target, GLuint64EXT* pointers, std::size_t num_entries) { + if (num_entries == 0) { + return; + } + if (num_entries % 2 == 1) { + pointers[num_entries] = 0; + } + const GLsizei num_vectors = static_cast<GLsizei>((num_entries + 1) / 2); + glProgramLocalParametersI4uivNV(target, 0, num_vectors, + reinterpret_cast<const GLuint*>(pointers)); +} + } // Anonymous namespace RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWindow& emu_window, @@ -149,7 +161,8 @@ RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWind shader_cache{*this, system, emu_window, device}, query_cache{system, *this}, buffer_cache{*this, system, device, STREAM_BUFFER_SIZE}, fence_manager{system, *this, texture_cache, buffer_cache, query_cache}, system{system}, - screen_info{info}, program_manager{program_manager}, state_tracker{state_tracker} { + screen_info{info}, program_manager{program_manager}, state_tracker{state_tracker}, + async_shaders{emu_window} { CheckExtensions(); unified_uniform_buffer.Create(); @@ -162,6 +175,23 @@ RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWind nullptr, 0); } } + + if (device.UseAsynchronousShaders()) { + // Max worker threads we should allow + constexpr auto MAX_THREADS = 2u; + // Amount of threads we should reserve for other parts of yuzu + constexpr auto RESERVED_THREADS = 6u; + // Get the amount of threads we can use(this can return zero) + const auto cpu_thread_count = + std::max(RESERVED_THREADS, std::thread::hardware_concurrency()); + // Deduce how many "extra" threads we have to use. + const auto max_threads_unused = cpu_thread_count - RESERVED_THREADS; + // Always allow at least 1 thread regardless of our settings + const auto max_worker_count = std::max(1u, max_threads_unused); + // Don't use more than MAX_THREADS + const auto worker_count = std::min(max_worker_count, MAX_THREADS); + async_shaders.AllocateWorkers(worker_count); + } } RasterizerOpenGL::~RasterizerOpenGL() { @@ -306,7 +336,6 @@ GLintptr RasterizerOpenGL::SetupIndexBuffer() { void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { MICROPROFILE_SCOPE(OpenGL_Shader); auto& gpu = system.GPU().Maxwell3D(); - std::size_t num_ssbos = 0; u32 clip_distances = 0; for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { @@ -329,31 +358,15 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { } // Currently this stages are not supported in the OpenGL backend. - // Todo(Blinkhawk): Port tesselation shaders from Vulkan to OpenGL - if (program == Maxwell::ShaderProgram::TesselationControl) { - continue; - } else if (program == Maxwell::ShaderProgram::TesselationEval) { + // TODO(Blinkhawk): Port tesselation shaders from Vulkan to OpenGL + if (program == Maxwell::ShaderProgram::TesselationControl || + program == Maxwell::ShaderProgram::TesselationEval) { continue; } - Shader* const shader = shader_cache.GetStageProgram(program); - - if (device.UseAssemblyShaders()) { - // Check for ARB limitation. We only have 16 SSBOs per context state. To workaround this - // all stages share the same bindings. - const std::size_t num_stage_ssbos = shader->GetEntries().global_memory_entries.size(); - ASSERT_MSG(num_stage_ssbos == 0 || num_ssbos == 0, "SSBOs on more than one stage"); - num_ssbos += num_stage_ssbos; - } + Shader* const shader = shader_cache.GetStageProgram(program, async_shaders); - // Stage indices are 0 - 5 - const std::size_t stage = index == 0 ? 0 : index - 1; - SetupDrawConstBuffers(stage, shader); - SetupDrawGlobalMemory(stage, shader); - SetupDrawTextures(stage, shader); - SetupDrawImages(stage, shader); - - const GLuint program_handle = shader->GetHandle(); + const GLuint program_handle = shader->IsBuilt() ? shader->GetHandle() : 0; switch (program) { case Maxwell::ShaderProgram::VertexA: case Maxwell::ShaderProgram::VertexB: @@ -370,6 +383,13 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { shader_config.enable.Value(), shader_config.offset); } + // Stage indices are 0 - 5 + const std::size_t stage = index == 0 ? 0 : index - 1; + SetupDrawConstBuffers(stage, shader); + SetupDrawGlobalMemory(stage, shader); + SetupDrawTextures(stage, shader); + SetupDrawImages(stage, shader); + // Workaround for Intel drivers. // When a clip distance is enabled but not set in the shader it crops parts of the screen // (sometimes it's half the screen, sometimes three quarters). To avoid this, enable the @@ -731,6 +751,8 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { current_cbuf = 0; auto kernel = shader_cache.GetComputeKernel(code_addr); + program_manager.BindCompute(kernel->GetHandle()); + SetupComputeTextures(kernel); SetupComputeImages(kernel); @@ -745,7 +767,6 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { buffer_cache.Unmap(); const auto& launch_desc = system.GPU().KeplerCompute().launch_description; - program_manager.BindCompute(kernel->GetHandle()); glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z); ++num_queued_commands; } @@ -1005,40 +1026,66 @@ void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding, } void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader) { + static constexpr std::array TARGET_LUT = { + GL_VERTEX_PROGRAM_NV, GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV, + GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV, + }; + auto& gpu{system.GPU()}; auto& memory_manager{gpu.MemoryManager()}; - const auto cbufs{gpu.Maxwell3D().state.shader_stages[stage_index]}; + const auto& cbufs{gpu.Maxwell3D().state.shader_stages[stage_index]}; + const auto& entries{shader->GetEntries().global_memory_entries}; + + std::array<GLuint64EXT, 32> pointers; + ASSERT(entries.size() < pointers.size()); - u32 binding = - device.UseAssemblyShaders() ? 0 : device.GetBaseBindings(stage_index).shader_storage_buffer; - for (const auto& entry : shader->GetEntries().global_memory_entries) { + const bool assembly_shaders = device.UseAssemblyShaders(); + u32 binding = assembly_shaders ? 0 : device.GetBaseBindings(stage_index).shader_storage_buffer; + for (const auto& entry : entries) { const GPUVAddr addr{cbufs.const_buffers[entry.cbuf_index].address + entry.cbuf_offset}; const GPUVAddr gpu_addr{memory_manager.Read<u64>(addr)}; const u32 size{memory_manager.Read<u32>(addr + 8)}; - SetupGlobalMemory(binding++, entry, gpu_addr, size); + SetupGlobalMemory(binding, entry, gpu_addr, size, &pointers[binding]); + ++binding; + } + if (assembly_shaders) { + UpdateBindlessPointers(TARGET_LUT[stage_index], pointers.data(), entries.size()); } } void RasterizerOpenGL::SetupComputeGlobalMemory(Shader* kernel) { auto& gpu{system.GPU()}; auto& memory_manager{gpu.MemoryManager()}; - const auto cbufs{gpu.KeplerCompute().launch_description.const_buffer_config}; + const auto& cbufs{gpu.KeplerCompute().launch_description.const_buffer_config}; + const auto& entries{kernel->GetEntries().global_memory_entries}; + + std::array<GLuint64EXT, 32> pointers; + ASSERT(entries.size() < pointers.size()); u32 binding = 0; - for (const auto& entry : kernel->GetEntries().global_memory_entries) { - const auto addr{cbufs[entry.cbuf_index].Address() + entry.cbuf_offset}; - const auto gpu_addr{memory_manager.Read<u64>(addr)}; - const auto size{memory_manager.Read<u32>(addr + 8)}; - SetupGlobalMemory(binding++, entry, gpu_addr, size); + for (const auto& entry : entries) { + const GPUVAddr addr{cbufs[entry.cbuf_index].Address() + entry.cbuf_offset}; + const GPUVAddr gpu_addr{memory_manager.Read<u64>(addr)}; + const u32 size{memory_manager.Read<u32>(addr + 8)}; + SetupGlobalMemory(binding, entry, gpu_addr, size, &pointers[binding]); + ++binding; + } + if (device.UseAssemblyShaders()) { + UpdateBindlessPointers(GL_COMPUTE_PROGRAM_NV, pointers.data(), entries.size()); } } void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, - GPUVAddr gpu_addr, std::size_t size) { - const auto alignment{device.GetShaderStorageBufferAlignment()}; + GPUVAddr gpu_addr, std::size_t size, + GLuint64EXT* pointer) { + const std::size_t alignment{device.GetShaderStorageBufferAlignment()}; const auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.is_written); - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, info.handle, info.offset, - static_cast<GLsizeiptr>(size)); + if (device.UseAssemblyShaders()) { + *pointer = info.address + info.offset; + } else { + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, info.handle, info.offset, + static_cast<GLsizeiptr>(size)); + } } void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, Shader* shader) { |