diff options
10 files changed, 32 insertions, 9 deletions
diff --git a/src/shader_recompiler/frontend/maxwell/translate_program.cpp b/src/shader_recompiler/frontend/maxwell/translate_program.cpp index 928b35561..47183cae1 100644 --- a/src/shader_recompiler/frontend/maxwell/translate_program.cpp +++ b/src/shader_recompiler/frontend/maxwell/translate_program.cpp @@ -298,7 +298,7 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo Optimization::PositionPass(env, program); - Optimization::GlobalMemoryToStorageBufferPass(program); + Optimization::GlobalMemoryToStorageBufferPass(program, host_info); Optimization::TexturePass(env, program, host_info); if (Settings::values.resolution_info.active) { diff --git a/src/shader_recompiler/host_translate_info.h b/src/shader_recompiler/host_translate_info.h index 7d2ded907..1b53404fc 100644 --- a/src/shader_recompiler/host_translate_info.h +++ b/src/shader_recompiler/host_translate_info.h @@ -16,6 +16,7 @@ struct HostTranslateInfo { bool needs_demote_reorder{}; ///< True when the device needs DemoteToHelperInvocation reordered bool support_snorm_render_buffer{}; ///< True when the device supports SNORM render buffers bool support_viewport_index_layer{}; ///< True when the device supports gl_Layer in VS + u32 min_ssbo_alignment{}; ///< Minimum alignment supported by the device for SSBOs bool support_geometry_shader_passthrough{}; ///< True when the device supports geometry ///< passthrough shaders bool support_conditional_barrier{}; ///< True when the device supports barriers in conditional diff --git a/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp b/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp index d1e59f22e..0cea79945 100644 --- a/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp +++ b/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp @@ -11,6 +11,7 @@ #include "shader_recompiler/frontend/ir/breadth_first_search.h" #include "shader_recompiler/frontend/ir/ir_emitter.h" #include "shader_recompiler/frontend/ir/value.h" +#include "shader_recompiler/host_translate_info.h" #include "shader_recompiler/ir_opt/passes.h" namespace Shader::Optimization { @@ -408,7 +409,7 @@ void CollectStorageBuffers(IR::Block& block, IR::Inst& inst, StorageInfo& info) } /// Returns the offset in indices (not bytes) for an equivalent storage instruction -IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer) { +IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer, u32 alignment) { IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; IR::U32 offset; if (const std::optional<LowAddrInfo> low_addr{TrackLowAddress(&inst)}) { @@ -421,7 +422,10 @@ IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer } // Subtract the least significant 32 bits from the guest offset. The result is the storage // buffer offset in bytes. - const IR::U32 low_cbuf{ir.GetCbuf(ir.Imm32(buffer.index), ir.Imm32(buffer.offset))}; + IR::U32 low_cbuf{ir.GetCbuf(ir.Imm32(buffer.index), ir.Imm32(buffer.offset))}; + + // Align the offset base to match the host alignment requirements + low_cbuf = ir.BitwiseAnd(low_cbuf, ir.Imm32(~(alignment - 1U))); return ir.ISub(offset, low_cbuf); } @@ -516,7 +520,7 @@ void Replace(IR::Block& block, IR::Inst& inst, const IR::U32& storage_index, } } // Anonymous namespace -void GlobalMemoryToStorageBufferPass(IR::Program& program) { +void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateInfo& host_info) { StorageInfo info; for (IR::Block* const block : program.post_order_blocks) { for (IR::Inst& inst : block->Instructions()) { @@ -540,7 +544,8 @@ void GlobalMemoryToStorageBufferPass(IR::Program& program) { const IR::U32 index{IR::Value{static_cast<u32>(info.set.index_of(it))}}; IR::Block* const block{storage_inst.block}; IR::Inst* const inst{storage_inst.inst}; - const IR::U32 offset{StorageOffset(*block, *inst, storage_buffer)}; + const IR::U32 offset{ + StorageOffset(*block, *inst, storage_buffer, host_info.min_ssbo_alignment)}; Replace(*block, *inst, index, offset); } } diff --git a/src/shader_recompiler/ir_opt/passes.h b/src/shader_recompiler/ir_opt/passes.h index 629d18fa1..7082fc5f2 100644 --- a/src/shader_recompiler/ir_opt/passes.h +++ b/src/shader_recompiler/ir_opt/passes.h @@ -16,7 +16,7 @@ void CollectShaderInfoPass(Environment& env, IR::Program& program); void ConditionalBarrierPass(IR::Program& program); void ConstantPropagationPass(Environment& env, IR::Program& program); void DeadCodeEliminationPass(IR::Program& program); -void GlobalMemoryToStorageBufferPass(IR::Program& program); +void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateInfo& host_info); void IdentityRemovalPass(IR::Program& program); void LowerFp64ToFp32(IR::Program& program); void LowerFp16ToFp32(IR::Program& program); diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 081a574e8..9202e53c7 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -1770,6 +1770,7 @@ template <class P> Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index, bool is_written) const { const GPUVAddr gpu_addr = gpu_memory->Read<u64>(ssbo_addr); + const u32 alignment = runtime.GetStorageBufferAlignment(); const auto size = [&]() { const bool is_nvn_cbuf = cbuf_index == 0; // The NVN driver buffer (index 0) is known to pack the SSBO address followed by its size. @@ -1785,15 +1786,19 @@ Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index, const u32 memory_layout_size = static_cast<u32>(gpu_memory->GetMemoryLayoutSize(gpu_addr)); return std::min(memory_layout_size, static_cast<u32>(8_MiB)); }(); - const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); + const GPUVAddr aligned_gpu_addr = Common::AlignDown(gpu_addr, alignment); + const u32 aligned_size = + Common::AlignUp(static_cast<u32>(gpu_addr - aligned_gpu_addr) + size, alignment); + + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(aligned_gpu_addr); if (!cpu_addr || size == 0) { LOG_WARNING(HW_GPU, "Failed to find storage buffer for cbuf index {}", cbuf_index); return NULL_BINDING; } - const VAddr cpu_end = Common::AlignUp(*cpu_addr + size, YUZU_PAGESIZE); + const VAddr cpu_end = Common::AlignUp(*cpu_addr + aligned_size, Core::Memory::YUZU_PAGESIZE); const Binding binding{ .cpu_addr = *cpu_addr, - .size = is_written ? size : static_cast<u32>(cpu_end - *cpu_addr), + .size = is_written ? aligned_size : static_cast<u32>(cpu_end - *cpu_addr), .buffer_id = BufferId{}, }; return binding; diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h index 41b746f3b..e8dbbd3a2 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.h +++ b/src/video_core/renderer_opengl/gl_buffer_cache.h @@ -182,6 +182,10 @@ public: return device.CanReportMemoryUsage(); } + u32 GetStorageBufferAlignment() const { + return static_cast<u32>(device.GetShaderStorageBufferAlignment()); + } + private: static constexpr std::array PABO_LUT{ GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV, diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 2888e0238..69f6759e6 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -240,6 +240,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo .needs_demote_reorder = device.IsAmd(), .support_snorm_render_buffer = false, .support_viewport_index_layer = device.HasVertexViewportLayer(), + .min_ssbo_alignment = static_cast<u32>(device.GetShaderStorageBufferAlignment()), .support_geometry_shader_passthrough = device.HasGeometryShaderPassthrough(), .support_conditional_barrier = device.SupportsConditionalBarriers(), } { diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index d8148e89a..976c3f6a6 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -355,6 +355,10 @@ bool BufferCacheRuntime::CanReportMemoryUsage() const { return device.CanReportMemoryUsage(); } +u32 BufferCacheRuntime::GetStorageBufferAlignment() const { + return static_cast<u32>(device.GetStorageBufferAlignment()); +} + void BufferCacheRuntime::Finish() { scheduler.Finish(); } diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index 95446c732..833dfac45 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -75,6 +75,8 @@ public: bool CanReportMemoryUsage() const; + u32 GetStorageBufferAlignment() const; + [[nodiscard]] StagingBufferRef UploadStagingBuffer(size_t size); [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size, bool deferred = false); diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 22bf8cc77..5e4380175 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -369,6 +369,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE, .support_snorm_render_buffer = true, .support_viewport_index_layer = device.IsExtShaderViewportIndexLayerSupported(), + .min_ssbo_alignment = static_cast<u32>(device.GetStorageBufferAlignment()), .support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(), .support_conditional_barrier = device.SupportsConditionalBarriers(), }; |