diff options
| -rw-r--r-- | src/shader_recompiler/ir_opt/texture_pass.cpp | 34 | ||||
| -rw-r--r-- | src/shader_recompiler/shader_info.h | 9 | ||||
| -rw-r--r-- | src/video_core/renderer_opengl/gl_compute_pipeline.cpp | 25 | ||||
| -rw-r--r-- | src/video_core/renderer_opengl/gl_graphics_pipeline.cpp | 22 | ||||
| -rw-r--r-- | src/video_core/renderer_vulkan/vk_compute_pipeline.cpp | 25 | ||||
| -rw-r--r-- | src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp | 22 |
6 files changed, 82 insertions, 55 deletions
diff --git a/src/shader_recompiler/ir_opt/texture_pass.cpp b/src/shader_recompiler/ir_opt/texture_pass.cpp index 514cd9b2b6..db8f1ac716 100644 --- a/src/shader_recompiler/ir_opt/texture_pass.cpp +++ b/src/shader_recompiler/ir_opt/texture_pass.cpp @@ -345,7 +345,8 @@ static inline u32 GetTextureHandleCached(Environment& env, const ConstBufferAddr // Cached variants of existing helpers // A handle computed per draw has no descriptor to read at translation time: assume an ordinary -// floating-point 2D texture, the instruction's own type deciding the rest. +// floating-point 2D texture, the instruction's own type deciding the rest (see the TexturePass +// ImageQueryDimensions case, which has none). static inline TextureType ReadTextureTypeCached(Environment& env, const ConstBufferAddr& cbuf) { if (cbuf.handle_program != ~0U) return TextureType::Color2D; return env.ReadTextureType(GetTextureHandleCached(env, cbuf)); @@ -1010,7 +1011,8 @@ public: desc.secondary_cbuf_offset == existing.secondary_cbuf_offset && desc.secondary_shift_left == existing.secondary_shift_left && desc.count == existing.count && desc.size_shift == existing.size_shift && - desc.has_secondary == existing.has_secondary; + desc.has_secondary == existing.has_secondary && + desc.handle_program == existing.handle_program; }); } @@ -1018,7 +1020,8 @@ public: const u32 index{Add(image_buffer_descriptors, desc, [&desc](const auto& existing) { return desc.format == existing.format && desc.cbuf_index == existing.cbuf_index && desc.cbuf_offset == existing.cbuf_offset && desc.count == existing.count && - desc.size_shift == existing.size_shift; + desc.size_shift == existing.size_shift && + desc.handle_program == existing.handle_program; })}; image_buffer_descriptors[index].is_written |= desc.is_written; image_buffer_descriptors[index].is_read |= desc.is_read; @@ -1050,7 +1053,8 @@ public: return desc.type == existing.type && desc.format == existing.format && desc.cbuf_index == existing.cbuf_index && desc.cbuf_offset == existing.cbuf_offset && desc.count == existing.count && - desc.size_shift == existing.size_shift; + desc.size_shift == existing.size_shift && + desc.handle_program == existing.handle_program; })}; image_descriptors[index].is_written |= desc.is_written; image_descriptors[index].is_read |= desc.is_read; @@ -1186,6 +1190,18 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo has_last_is_integer = true; return last_is_integer; }}; + // A handle program has no descriptor to read a type from, and ImageQueryDimensions carries + // none: it takes the type of another instruction using the same handle. + std::unordered_map<u32, TextureType> handle_program_types; + for (const TextureInst& texture_inst : to_replace) { + const IR::Opcode opcode{IndexedInstruction(*texture_inst.inst)}; + if (texture_inst.cbuf.handle_program != ~0U && opcode != IR::Opcode::ImageQueryDimensions && + !IsStorageImageOpcode(opcode)) { + handle_program_types.try_emplace( + texture_inst.cbuf.handle_program, + texture_inst.inst->Flags<IR::TextureInstInfo>().type.Value()); + } + } for (TextureInst& texture_inst : to_replace) { // TODO: Handle arrays IR::Inst* const inst{texture_inst.inst}; @@ -1196,7 +1212,12 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo bool is_multisample{false}; switch (inst->GetOpcode()) { case IR::Opcode::ImageQueryDimensions: - flags.type.Assign(ReadTextureTypeCached(env, cbuf)); + if (const auto it{handle_program_types.find(cbuf.handle_program)}; + it != handle_program_types.end()) { + flags.type.Assign(it->second); + } else { + flags.type.Assign(ReadTextureTypeCached(env, cbuf)); + } inst->SetFlags(flags); break; case IR::Opcode::ImageSampleImplicitLod: @@ -1261,6 +1282,7 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo .cbuf_offset = cbuf.offset, .count = count, .size_shift = size_shift, + .handle_program = cbuf.handle_program, }); } else { index = descriptors.Add(ImageDescriptor{ @@ -1273,6 +1295,7 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo .cbuf_offset = cbuf.offset, .count = count, .size_shift = size_shift, + .handle_program = cbuf.handle_program, }); } break; @@ -1289,6 +1312,7 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo .secondary_shift_left = cbuf.secondary_shift_left, .count = count, .size_shift = size_shift, + .handle_program = cbuf.handle_program, }); } else { count = std::min(count, sampled_dynamic_cap); diff --git a/src/shader_recompiler/shader_info.h b/src/shader_recompiler/shader_info.h index eeb8d768d7..e2b341b0c3 100644 --- a/src/shader_recompiler/shader_info.h +++ b/src/shader_recompiler/shader_info.h @@ -189,6 +189,9 @@ struct TextureBufferDescriptor { u32 secondary_shift_left; u32 count; u32 size_shift; + /// Index in Info::handle_programs when the handle is computed at draw time (read from a + /// storage buffer), ~0 when it is the constant buffer word. + u32 handle_program{~0U}; auto operator<=>(const TextureBufferDescriptor&) const = default; }; @@ -203,6 +206,9 @@ struct ImageBufferDescriptor { u32 cbuf_offset; u32 count; u32 size_shift; + /// Index in Info::handle_programs when the handle is computed at draw time (read from a + /// storage buffer), ~0 when it is the constant buffer word. + u32 handle_program{~0U}; auto operator<=>(const ImageBufferDescriptor&) const = default; }; @@ -240,6 +246,9 @@ struct ImageDescriptor { u32 cbuf_offset; u32 count; u32 size_shift; + /// Index in Info::handle_programs when the handle is computed at draw time (read from a + /// storage buffer), ~0 when it is the constant buffer word. + u32 handle_program{~0U}; auto operator<=>(const ImageDescriptor&) const = default; }; diff --git a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp index eb134e86b5..5fc10cfce2 100644 --- a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp +++ b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp @@ -106,19 +106,18 @@ void ComputePipeline::Configure() { const auto& cbufs{qmd.const_buffer_config}; const bool via_header_index{qmd.linked_tsc != 0}; const auto read_handle{[&](const auto& desc, u32 index) { - if constexpr (std::is_same_v<decltype(desc), const Shader::TextureDescriptor&>) { - if (desc.handle_program != ~0U) { - // Read by the shader from a storage buffer: computed for this dispatch. - return TexturePair(VideoCommon::EvaluateTextureHandle( - info, desc.handle_program, *gpu_memory, - [&](u32 cbuf) -> GPUVAddr { - return cbuf < cbufs.size() && - ((qmd.const_buffer_enable_mask >> cbuf) & 1) != 0 - ? cbufs[cbuf].Address() - : GPUVAddr{}; - }), - via_header_index); - } + if (desc.handle_program != ~0U) { + // Read by the shader from a storage buffer: computed for this dispatch. + return TexturePair(VideoCommon::EvaluateTextureHandle( + info, desc.handle_program, *gpu_memory, + [&](u32 cbuf) -> GPUVAddr { + return cbuf < cbufs.size() && + ((qmd.const_buffer_enable_mask >> cbuf) & + 1) != 0 + ? cbufs[cbuf].Address() + : GPUVAddr{}; + }), + via_header_index); } ASSERT(((qmd.const_buffer_enable_mask >> desc.cbuf_index) & 1) != 0); const u32 index_offset{index << desc.size_shift}; diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp index 81950279d5..4e02f0569b 100644 --- a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp @@ -309,18 +309,16 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) { } const auto& cbufs{maxwell3d->state.shader_stages[stage].const_buffers}; const auto read_handle{[&](const auto& desc, u32 index) { - if constexpr (std::is_same_v<decltype(desc), const Shader::TextureDescriptor&>) { - if (desc.handle_program != ~0U) { - // Read by the shader from a storage buffer: computed for this draw. - return TexturePair(VideoCommon::EvaluateTextureHandle( - info, desc.handle_program, *gpu_memory, - [&](u32 cbuf) -> GPUVAddr { - return cbuf < cbufs.size() && cbufs[cbuf].enabled - ? cbufs[cbuf].address - : GPUVAddr{}; - }), - via_header_index); - } + if (desc.handle_program != ~0U) { + // Read by the shader from a storage buffer: computed for this draw. + return TexturePair(VideoCommon::EvaluateTextureHandle( + info, desc.handle_program, *gpu_memory, + [&](u32 cbuf) -> GPUVAddr { + return cbuf < cbufs.size() && cbufs[cbuf].enabled + ? cbufs[cbuf].address + : GPUVAddr{}; + }), + via_header_index); } ASSERT(cbufs[desc.cbuf_index].enabled); const u32 index_offset{index << desc.size_shift}; diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp index 1cfe391ba6..3a3861930c 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp @@ -176,19 +176,18 @@ bool ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, const auto& cbufs{qmd.const_buffer_config}; const bool via_header_index{qmd.linked_tsc != 0}; const auto read_handle{[&](const auto& desc, u32 index) { - if constexpr (std::is_same_v<decltype(desc), const Shader::TextureDescriptor&>) { - if (desc.handle_program != ~0U) { - // Read by the shader from a storage buffer: computed for this dispatch. - return TexturePair(VideoCommon::EvaluateTextureHandle( - info, desc.handle_program, gpu_memory, - [&](u32 cbuf) -> GPUVAddr { - return cbuf < cbufs.size() && - ((qmd.const_buffer_enable_mask >> cbuf) & 1) != 0 - ? cbufs[cbuf].Address() - : GPUVAddr{}; - }), - via_header_index); - } + if (desc.handle_program != ~0U) { + // Read by the shader from a storage buffer: computed for this dispatch. + return TexturePair(VideoCommon::EvaluateTextureHandle( + info, desc.handle_program, gpu_memory, + [&](u32 cbuf) -> GPUVAddr { + return cbuf < cbufs.size() && + ((qmd.const_buffer_enable_mask >> cbuf) & + 1) != 0 + ? cbufs[cbuf].Address() + : GPUVAddr{}; + }), + via_header_index); } ASSERT(((qmd.const_buffer_enable_mask >> desc.cbuf_index) & 1) != 0); const u32 index_offset{index << desc.size_shift}; diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index fa273f6ef9..4bd2e2735e 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -400,18 +400,16 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) { } const auto& cbufs{maxwell3d->state.shader_stages[stage].const_buffers}; const auto read_handle{[&](const auto& desc, u32 index) { - if constexpr (std::is_same_v<decltype(desc), const Shader::TextureDescriptor&>) { - if (desc.handle_program != ~0U) { - // Read by the shader from a storage buffer: computed for this draw. - return TexturePair(VideoCommon::EvaluateTextureHandle( - info, desc.handle_program, *gpu_memory, - [&](u32 cbuf) -> GPUVAddr { - return cbuf < cbufs.size() && cbufs[cbuf].enabled - ? cbufs[cbuf].address - : GPUVAddr{}; - }), - via_header_index); - } + if (desc.handle_program != ~0U) { + // Read by the shader from a storage buffer: computed for this draw. + return TexturePair(VideoCommon::EvaluateTextureHandle( + info, desc.handle_program, *gpu_memory, + [&](u32 cbuf) -> GPUVAddr { + return cbuf < cbufs.size() && cbufs[cbuf].enabled + ? cbufs[cbuf].address + : GPUVAddr{}; + }), + via_header_index); } ASSERT(cbufs[desc.cbuf_index].enabled); const u32 index_offset{index << desc.size_shift}; |