diff options
author | Fernando Sahmkow <fsahmkow27@gmail.com> | 2022-03-27 05:05:57 +0200 |
---|---|---|
committer | Fernando Sahmkow <fsahmkow27@gmail.com> | 2023-03-05 12:18:00 +0100 |
commit | 8a3411b417f76db786b1d3cfffbd90926abb20ca (patch) | |
tree | 383070bd0d4a33189f38423ceea9a5692d38ba09 /src/video_core/engines | |
parent | Merge pull request #9884 from liamwhite/service-cleanup (diff) | |
download | yuzu-8a3411b417f76db786b1d3cfffbd90926abb20ca.tar yuzu-8a3411b417f76db786b1d3cfffbd90926abb20ca.tar.gz yuzu-8a3411b417f76db786b1d3cfffbd90926abb20ca.tar.bz2 yuzu-8a3411b417f76db786b1d3cfffbd90926abb20ca.tar.lz yuzu-8a3411b417f76db786b1d3cfffbd90926abb20ca.tar.xz yuzu-8a3411b417f76db786b1d3cfffbd90926abb20ca.tar.zst yuzu-8a3411b417f76db786b1d3cfffbd90926abb20ca.zip |
Diffstat (limited to 'src/video_core/engines')
-rw-r--r-- | src/video_core/engines/maxwell_dma.cpp | 107 | ||||
-rw-r--r-- | src/video_core/engines/maxwell_dma.h | 88 |
2 files changed, 117 insertions, 78 deletions
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index 7762c7d96..e68850dc5 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -14,7 +14,13 @@ #include "video_core/textures/decoders.h" MICROPROFILE_DECLARE(GPU_DMAEngine); +MICROPROFILE_DECLARE(GPU_DMAEngineBL); +MICROPROFILE_DECLARE(GPU_DMAEngineLB); +MICROPROFILE_DECLARE(GPU_DMAEngineBB); MICROPROFILE_DEFINE(GPU_DMAEngine, "GPU", "DMA Engine", MP_RGB(224, 224, 128)); +MICROPROFILE_DEFINE(GPU_DMAEngineBL, "GPU", "DMA Engine Block - Linear", MP_RGB(224, 224, 128)); +MICROPROFILE_DEFINE(GPU_DMAEngineLB, "GPU", "DMA Engine Linear - Block", MP_RGB(224, 224, 128)); +MICROPROFILE_DEFINE(GPU_DMAEngineBB, "GPU", "DMA Engine Block - Block", MP_RGB(224, 224, 128)); namespace Tegra::Engines { @@ -72,6 +78,7 @@ void MaxwellDMA::Launch() { memory_manager.FlushCaching(); if (!is_src_pitch && !is_dst_pitch) { // If both the source and the destination are in block layout, assert. + MICROPROFILE_SCOPE(GPU_DMAEngineBB); CopyBlockLinearToBlockLinear(); ReleaseSemaphore(); return; @@ -87,8 +94,10 @@ void MaxwellDMA::Launch() { } } else { if (!is_src_pitch && is_dst_pitch) { + MICROPROFILE_SCOPE(GPU_DMAEngineBL); CopyBlockLinearToPitch(); } else { + MICROPROFILE_SCOPE(GPU_DMAEngineLB); CopyPitchToBlockLinear(); } } @@ -153,21 +162,35 @@ void MaxwellDMA::Launch() { } void MaxwellDMA::CopyBlockLinearToPitch() { - UNIMPLEMENTED_IF(regs.src_params.block_size.width != 0); - UNIMPLEMENTED_IF(regs.src_params.layer != 0); - - const bool is_remapping = regs.launch_dma.remap_enable != 0; - - // Optimized path for micro copies. - const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count; - if (!is_remapping && dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X && - regs.src_params.height > GOB_SIZE_Y) { - FastCopyBlockLinearToPitch(); + UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); + + u32 bytes_per_pixel = 1; + DMA::ImageOperand src_operand; + src_operand.bytes_per_pixel = bytes_per_pixel; + src_operand.params = regs.src_params; + src_operand.address = regs.offset_in; + + DMA::BufferOperand dst_operand; + dst_operand.pitch = regs.pitch_out; + dst_operand.width = regs.line_length_in; + dst_operand.height = regs.line_count; + dst_operand.address = regs.offset_out; + DMA::ImageCopy copy_info{}; + copy_info.length_x = regs.line_length_in; + copy_info.length_y = regs.line_count; + auto& accelerate = rasterizer->AccessAccelerateDMA(); + if (accelerate.ImageToBuffer(copy_info, src_operand, dst_operand)) { return; } + UNIMPLEMENTED_IF(regs.src_params.block_size.width != 0); + UNIMPLEMENTED_IF(regs.src_params.block_size.depth != 0); + UNIMPLEMENTED_IF(regs.src_params.block_size.depth == 0 && regs.src_params.depth != 1); + // Deswizzle the input and copy it over. - const Parameters& src_params = regs.src_params; + const DMA::Parameters& src_params = regs.src_params; + + const bool is_remapping = regs.launch_dma.remap_enable != 0; const u32 num_remap_components = regs.remap_const.num_dst_components_minus_one + 1; const u32 remap_components_size = regs.remap_const.component_size_minus_one + 1; @@ -187,7 +210,7 @@ void MaxwellDMA::CopyBlockLinearToPitch() { x_offset >>= bpp_shift; } - const u32 bytes_per_pixel = base_bpp << bpp_shift; + bytes_per_pixel = base_bpp << bpp_shift; const u32 height = src_params.height; const u32 depth = src_params.depth; const u32 block_height = src_params.block_size.height; @@ -195,11 +218,12 @@ void MaxwellDMA::CopyBlockLinearToPitch() { const size_t src_size = CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth); + const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count; read_buffer.resize_destructive(src_size); write_buffer.resize_destructive(dst_size); - memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); - memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size); + memory_manager.ReadBlock(src_operand.address, read_buffer.data(), src_size); + memory_manager.ReadBlockUnsafe(dst_operand.address, write_buffer.data(), dst_size); UnswizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset, src_params.origin.y, x_elements, regs.line_count, block_height, block_depth, @@ -216,6 +240,24 @@ void MaxwellDMA::CopyPitchToBlockLinear() { const u32 num_remap_components = regs.remap_const.num_dst_components_minus_one + 1; const u32 remap_components_size = regs.remap_const.component_size_minus_one + 1; + u32 bytes_per_pixel = 1; + DMA::ImageOperand dst_operand; + dst_operand.bytes_per_pixel = bytes_per_pixel; + dst_operand.params = regs.dst_params; + dst_operand.address = regs.offset_out; + DMA::BufferOperand src_operand; + src_operand.pitch = regs.pitch_in; + src_operand.width = regs.line_length_in; + src_operand.height = regs.line_count; + src_operand.address = regs.offset_in; + DMA::ImageCopy copy_info{}; + copy_info.length_x = regs.line_length_in; + copy_info.length_y = regs.line_count; + auto& accelerate = rasterizer->AccessAccelerateDMA(); + if (accelerate.BufferToImage(copy_info, src_operand, dst_operand)) { + return; + } + const auto& dst_params = regs.dst_params; const u32 base_bpp = !is_remapping ? 1U : num_remap_components * remap_components_size; @@ -233,7 +275,7 @@ void MaxwellDMA::CopyPitchToBlockLinear() { x_offset >>= bpp_shift; } - const u32 bytes_per_pixel = base_bpp << bpp_shift; + bytes_per_pixel = base_bpp << bpp_shift; const u32 height = dst_params.height; const u32 depth = dst_params.depth; const u32 block_height = dst_params.block_size.height; @@ -260,45 +302,14 @@ void MaxwellDMA::CopyPitchToBlockLinear() { memory_manager.WriteBlockCached(regs.offset_out, write_buffer.data(), dst_size); } -void MaxwellDMA::FastCopyBlockLinearToPitch() { - const u32 bytes_per_pixel = 1U; - const size_t src_size = GOB_SIZE; - const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count; - u32 pos_x = regs.src_params.origin.x; - u32 pos_y = regs.src_params.origin.y; - const u64 offset = GetGOBOffset(regs.src_params.width, regs.src_params.height, pos_x, pos_y, - regs.src_params.block_size.height, bytes_per_pixel); - const u32 x_in_gob = 64 / bytes_per_pixel; - pos_x = pos_x % x_in_gob; - pos_y = pos_y % 8; - - read_buffer.resize_destructive(src_size); - write_buffer.resize_destructive(dst_size); - - if (Settings::IsGPULevelExtreme()) { - memory_manager.ReadBlock(regs.offset_in + offset, read_buffer.data(), src_size); - memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size); - } else { - memory_manager.ReadBlockUnsafe(regs.offset_in + offset, read_buffer.data(), src_size); - memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size); - } - - UnswizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, regs.src_params.width, - regs.src_params.height, 1, pos_x, pos_y, regs.line_length_in, regs.line_count, - regs.src_params.block_size.height, regs.src_params.block_size.depth, - regs.pitch_out); - - memory_manager.WriteBlockCached(regs.offset_out, write_buffer.data(), dst_size); -} - void MaxwellDMA::CopyBlockLinearToBlockLinear() { UNIMPLEMENTED_IF(regs.src_params.block_size.width != 0); const bool is_remapping = regs.launch_dma.remap_enable != 0; // Deswizzle the input and copy it over. - const Parameters& src = regs.src_params; - const Parameters& dst = regs.dst_params; + const DMA::Parameters& src = regs.src_params; + const DMA::Parameters& dst = regs.dst_params; const u32 num_remap_components = regs.remap_const.num_dst_components_minus_one + 1; const u32 remap_components_size = regs.remap_const.component_size_minus_one + 1; diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h index 0e594fa74..69e26cb32 100644 --- a/src/video_core/engines/maxwell_dma.h +++ b/src/video_core/engines/maxwell_dma.h @@ -24,6 +24,54 @@ namespace VideoCore { class RasterizerInterface; } +namespace Tegra { +namespace DMA { + +union Origin { + BitField<0, 16, u32> x; + BitField<16, 16, u32> y; +}; +static_assert(sizeof(Origin) == 4); + +struct ImageCopy { + u32 length_x{}; + u32 length_y{}; +}; + +union BlockSize { + BitField<0, 4, u32> width; + BitField<4, 4, u32> height; + BitField<8, 4, u32> depth; + BitField<12, 4, u32> gob_height; +}; +static_assert(sizeof(BlockSize) == 4); + +struct Parameters { + BlockSize block_size; + u32 width; + u32 height; + u32 depth; + u32 layer; + Origin origin; +}; +static_assert(sizeof(Parameters) == 24); + +struct ImageOperand { + u32 bytes_per_pixel; + Parameters params; + GPUVAddr address; +}; + +struct BufferOperand { + u32 pitch; + u32 width; + u32 height; + GPUVAddr address; +}; + +} // namespace DMA +} // namespace Tegra + namespace Tegra::Engines { class AccelerateDMAInterface { @@ -32,6 +80,12 @@ public: virtual bool BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) = 0; virtual bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) = 0; + + virtual bool ImageToBuffer(const DMA::ImageCopy& copy_info, const DMA::ImageOperand& src, + const DMA::BufferOperand& dst) = 0; + + virtual bool BufferToImage(const DMA::ImageCopy& copy_info, const DMA::BufferOperand& src, + const DMA::ImageOperand& dst) = 0; }; /** @@ -51,30 +105,6 @@ public: } }; - union BlockSize { - BitField<0, 4, u32> width; - BitField<4, 4, u32> height; - BitField<8, 4, u32> depth; - BitField<12, 4, u32> gob_height; - }; - static_assert(sizeof(BlockSize) == 4); - - union Origin { - BitField<0, 16, u32> x; - BitField<16, 16, u32> y; - }; - static_assert(sizeof(Origin) == 4); - - struct Parameters { - BlockSize block_size; - u32 width; - u32 height; - u32 depth; - u32 layer; - Origin origin; - }; - static_assert(sizeof(Parameters) == 24); - struct Semaphore { PackedGPUVAddr address; u32 payload; @@ -227,8 +257,6 @@ private: void CopyBlockLinearToBlockLinear(); - void FastCopyBlockLinearToPitch(); - void ReleaseSemaphore(); void ConsumeSinkImpl() override; @@ -261,17 +289,17 @@ private: u32 reserved05[0x3f]; PackedGPUVAddr offset_in; PackedGPUVAddr offset_out; - u32 pitch_in; - u32 pitch_out; + s32 pitch_in; + s32 pitch_out; u32 line_length_in; u32 line_count; u32 reserved06[0xb6]; u32 remap_consta_value; u32 remap_constb_value; RemapConst remap_const; - Parameters dst_params; + DMA::Parameters dst_params; u32 reserved07[0x1]; - Parameters src_params; + DMA::Parameters src_params; u32 reserved08[0x275]; u32 pm_trigger_end; u32 reserved09[0x3ba]; |