D3D12: Allow relaxed B2T copy pitch and offset alignment on 2D textures
This patch implements the buffer-texture copy with relaxed copy pitch
and offset alignment (not required to be a multiple of 512 or 256) on
2D textures when `UnrestrictedBufferTextureCopyPitchSupported` is true
on the current D3D12 device so that we no longer need to split one
buffer-texture copy into multiple ones in some cases.
Bug: chromium:381000081
Test: dawn_end2end_tests
Change-Id: Ib4fbc819701ca82aeaaf791fe1b6de42b11b1d18
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/211949
Reviewed-by: Loko Kung <lokokung@google.com>
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
diff --git a/src/dawn/native/Toggles.cpp b/src/dawn/native/Toggles.cpp
index eeef729..d072704 100644
--- a/src/dawn/native/Toggles.cpp
+++ b/src/dawn/native/Toggles.cpp
@@ -589,6 +589,11 @@
"minWaveLaneCount and 8. Some D3D12 drivers is possible to run fragment shader with wave "
"count 8 while reporting minWaveLaneCount 16.",
"https://crbug.com/381969450", ToggleStage::Adapter}},
+ {Toggle::D3D12RelaxBufferTextureCopyPitchAndOffsetAlignment,
+ {"d3d12_relax_buffer_texture_copy_pitch_and_offset_alignment",
+ "Don't require the alignments of D3D12_TEXTURE_DATA_PITCH_ALIGNMENT (256) for row pitch "
+ "and D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT (512) for offset in buffer-texture copies.",
+ "https://crbug.com/381000081", ToggleStage::Device}},
{Toggle::NoWorkaroundSampleMaskBecomesZeroForAllButLastColorTarget,
{"no_workaround_sample_mask_becomes_zero_for_all_but_last_color_target",
"MacOS 12.0+ Intel has a bug where the sample mask is only applied for the last color "
diff --git a/src/dawn/native/Toggles.h b/src/dawn/native/Toggles.h
index 2d327de..f096d35 100644
--- a/src/dawn/native/Toggles.h
+++ b/src/dawn/native/Toggles.h
@@ -140,6 +140,7 @@
VulkanMonolithicPipelineCache,
MetalSerializeTimestampGenerationAndResolution,
D3D12RelaxMinSubgroupSizeTo8,
+ D3D12RelaxBufferTextureCopyPitchAndOffsetAlignment,
// Unresolved issues.
NoWorkaroundSampleMaskBecomesZeroForAllButLastColorTarget,
diff --git a/src/dawn/native/d3d12/D3D12Info.cpp b/src/dawn/native/d3d12/D3D12Info.cpp
index 98910de..e8e7d2b 100644
--- a/src/dawn/native/d3d12/D3D12Info.cpp
+++ b/src/dawn/native/d3d12/D3D12Info.cpp
@@ -113,6 +113,8 @@
D3D12_FEATURE_D3D12_OPTIONS13, &featureOptions13, sizeof(featureOptions13)))) {
info.supportsTextureCopyBetweenDimensions =
featureOptions13.TextureCopyBetweenDimensionsSupported;
+ info.supportsUnrestrictedBufferTextureCopyPitch =
+ featureOptions13.UnrestrictedBufferTextureCopyPitchSupported;
}
info.supportsRootSignatureVersion1_1 = false;
diff --git a/src/dawn/native/d3d12/D3D12Info.h b/src/dawn/native/d3d12/D3D12Info.h
index b9d8e5f..3d6385b 100644
--- a/src/dawn/native/d3d12/D3D12Info.h
+++ b/src/dawn/native/d3d12/D3D12Info.h
@@ -54,6 +54,7 @@
bool use64KBAlignedMSAATexture;
bool supportsHeapFlagCreateNotZeroed;
bool supportsTextureCopyBetweenDimensions;
+ bool supportsUnrestrictedBufferTextureCopyPitch;
// Whether the device support wave intrinsics
bool supportsWaveOps;
uint32_t waveLaneCountMin;
diff --git a/src/dawn/native/d3d12/PhysicalDeviceD3D12.cpp b/src/dawn/native/d3d12/PhysicalDeviceD3D12.cpp
index 45a2679..c46df96 100644
--- a/src/dawn/native/d3d12/PhysicalDeviceD3D12.cpp
+++ b/src/dawn/native/d3d12/PhysicalDeviceD3D12.cpp
@@ -648,6 +648,14 @@
deviceToggles->Default(Toggle::D3D12CreateNotZeroedHeap,
GetDeviceInfo().supportsHeapFlagCreateNotZeroed);
+ // By default allow relaxed row pitch and offset in buffer-texture copies when possible,
+ // otherwise we should never enable this toggle.
+ if (!GetDeviceInfo().supportsUnrestrictedBufferTextureCopyPitch) {
+ deviceToggles->ForceSet(Toggle::D3D12RelaxBufferTextureCopyPitchAndOffsetAlignment, false);
+ }
+ deviceToggles->Default(Toggle::D3D12RelaxBufferTextureCopyPitchAndOffsetAlignment,
+ GetDeviceInfo().supportsUnrestrictedBufferTextureCopyPitch);
+
// Native support of packed 4x8 integer dot product required shader model 6.4 or higher, and
// DXC 1.4 or higher.
if (!(GetAppliedShaderModelUnderToggles(*deviceToggles) >= 64) ||
diff --git a/src/dawn/native/d3d12/TextureCopySplitter.cpp b/src/dawn/native/d3d12/TextureCopySplitter.cpp
index 67501cd..9c80fc4 100644
--- a/src/dawn/native/d3d12/TextureCopySplitter.cpp
+++ b/src/dawn/native/d3d12/TextureCopySplitter.cpp
@@ -682,4 +682,40 @@
return copySubresource;
}
+
+TextureCopySubresource Compute2DTextureCopySubresourceWithRelaxedRowPitchAndOffset(
+ BufferTextureCopyDirection direction,
+ Origin3D origin,
+ Extent3D copySize,
+ const TexelBlockInfo& blockInfo,
+ uint64_t offset,
+ uint32_t bytesPerRow) {
+ TextureCopySubresource copy;
+ auto* copyInfo = copy.AddCopy();
+
+ // You can visualize the data in the buffer (bufferLocation) like this:
+ // * copy data is visualized as '+'.
+ //
+ // bufferOffset(0, 0, 0)
+ // ^
+ // |
+ // |<-------Offset------->|<-----------RowPitch----------->|----------|
+ // |----------------------|++++++++++++++++++++++~~~~~~~~~~| | |
+ // |++++++++++++++++++++++~~~~~~~~~~|CopyHeight|
+ // |++++++++++++++++++++++| | | |
+ // |<-----CopyWidth------>| |----------|
+ //
+ Origin3D textureOffset = {origin.x, origin.y, 0};
+ Origin3D bufferOffset = {0, 0, 0};
+ Extent3D copySizeOneLayer = {copySize.width, copySize.height, 1};
+ ComputeSourceRegionForCopyInfo(copyInfo, direction, bufferOffset, textureOffset,
+ copySizeOneLayer);
+
+ Extent3D bufferSize = {copySize.width, copySize.height, 1};
+
+ FillFootprintAndOffsetOfBufferLocation(©Info->bufferLocation, offset, bufferSize,
+ bytesPerRow);
+
+ return copy;
+}
} // namespace dawn::native::d3d12
diff --git a/src/dawn/native/d3d12/TextureCopySplitter.h b/src/dawn/native/d3d12/TextureCopySplitter.h
index 4d0f5b8..5c6939d 100644
--- a/src/dawn/native/d3d12/TextureCopySplitter.h
+++ b/src/dawn/native/d3d12/TextureCopySplitter.h
@@ -117,6 +117,17 @@
uint64_t offset,
uint32_t bytesPerRow,
uint32_t rowsPerImage);
+
+// Compute the `TextureCopySubresource` for one subresource of a 2D texture with relaxed row pitch
+// and offset.
+TextureCopySubresource Compute2DTextureCopySubresourceWithRelaxedRowPitchAndOffset(
+ BufferTextureCopyDirection direction,
+ Origin3D origin,
+ Extent3D copySize,
+ const TexelBlockInfo& blockInfo,
+ uint64_t offset,
+ uint32_t bytesPerRow);
+
} // namespace dawn::native::d3d12
#endif // SRC_DAWN_NATIVE_D3D12_TEXTURECOPYSPLITTER_H_
diff --git a/src/dawn/native/d3d12/UtilsD3D12.cpp b/src/dawn/native/d3d12/UtilsD3D12.cpp
index f2b55c5..50602da 100644
--- a/src/dawn/native/d3d12/UtilsD3D12.cpp
+++ b/src/dawn/native/d3d12/UtilsD3D12.cpp
@@ -234,6 +234,31 @@
}
}
+void Record2DBufferTextureCopyWithRelaxedOffsetAndPitch(BufferTextureCopyDirection direction,
+ ID3D12GraphicsCommandList* commandList,
+ ID3D12Resource* bufferResource,
+ const uint64_t offset,
+ const uint32_t bytesPerRow,
+ const uint32_t rowsPerImage,
+ const TextureCopy& textureCopy,
+ const TexelBlockInfo& blockInfo,
+ const Extent3D& copySize) {
+ TextureCopySubresource copySubresource =
+ Compute2DTextureCopySubresourceWithRelaxedRowPitchAndOffset(
+ direction, textureCopy.origin, copySize, blockInfo, offset, bytesPerRow);
+
+ uint64_t bytesPerLayer = bytesPerRow * rowsPerImage;
+ uint64_t bufferOffsetForNextLayer = 0;
+ for (uint32_t copyLayer = 0; copyLayer < copySize.depthOrArrayLayers; ++copyLayer) {
+ uint32_t copyTextureLayer = copyLayer + textureCopy.origin.z;
+ RecordBufferTextureCopyFromSplits(direction, commandList, copySubresource, bufferResource,
+ bufferOffsetForNextLayer, bytesPerRow,
+ textureCopy.texture.Get(), textureCopy.mipLevel,
+ copyTextureLayer, textureCopy.aspect);
+ bufferOffsetForNextLayer += bytesPerLayer;
+ }
+}
+
void RecordBufferTextureCopyWithBufferHandle(BufferTextureCopyDirection direction,
ID3D12GraphicsCommandList* commandList,
ID3D12Resource* bufferResource,
@@ -247,6 +272,9 @@
TextureBase* texture = textureCopy.texture.Get();
const TexelBlockInfo& blockInfo = texture->GetFormat().GetAspectInfo(textureCopy.aspect).block;
+ bool useRelaxedRowPitchAndOffset = texture->GetDevice()->IsToggleEnabled(
+ Toggle::D3D12RelaxBufferTextureCopyPitchAndOffsetAlignment);
+
switch (texture->GetDimension()) {
case wgpu::TextureDimension::Undefined:
DAWN_UNREACHABLE();
@@ -255,9 +283,14 @@
// 1D textures copy splits are a subset of the single-layer 2D texture copy splits,
// at least while 1D textures can only have a single array layer.
DAWN_ASSERT(texture->GetArrayLayers() == 1);
-
- TextureCopySubresource copyRegions = Compute2DTextureCopySubresource(
- direction, textureCopy.origin, copySize, blockInfo, offset, bytesPerRow);
+ TextureCopySubresource copyRegions;
+ if (useRelaxedRowPitchAndOffset) {
+ copyRegions = Compute2DTextureCopySubresourceWithRelaxedRowPitchAndOffset(
+ direction, textureCopy.origin, copySize, blockInfo, offset, bytesPerRow);
+ } else {
+ copyRegions = Compute2DTextureCopySubresource(
+ direction, textureCopy.origin, copySize, blockInfo, offset, bytesPerRow);
+ }
RecordBufferTextureCopyFromSplits(direction, commandList, copyRegions, bufferResource,
0, bytesPerRow, texture, textureCopy.mipLevel, 0,
textureCopy.aspect);
@@ -267,9 +300,15 @@
// Record the CopyTextureRegion commands for 2D textures, with special handling of array
// layers since each require their own set of copies.
case wgpu::TextureDimension::e2D:
- Record2DBufferTextureCopyWithSplit(direction, commandList, bufferResource, offset,
- bytesPerRow, rowsPerImage, textureCopy, blockInfo,
- copySize);
+ if (useRelaxedRowPitchAndOffset) {
+ Record2DBufferTextureCopyWithRelaxedOffsetAndPitch(
+ direction, commandList, bufferResource, offset, bytesPerRow, rowsPerImage,
+ textureCopy, blockInfo, copySize);
+ } else {
+ Record2DBufferTextureCopyWithSplit(direction, commandList, bufferResource, offset,
+ bytesPerRow, rowsPerImage, textureCopy,
+ blockInfo, copySize);
+ }
break;
case wgpu::TextureDimension::e3D: {
