| // Copyright 2017 The Dawn & Tint Authors | |
| // | |
| // Redistribution and use in source and binary forms, with or without | |
| // modification, are permitted provided that the following conditions are met: | |
| // | |
| // 1. Redistributions of source code must retain the above copyright notice, this | |
| // list of conditions and the following disclaimer. | |
| // | |
| // 2. Redistributions in binary form must reproduce the above copyright notice, | |
| // this list of conditions and the following disclaimer in the documentation | |
| // and/or other materials provided with the distribution. | |
| // | |
| // 3. Neither the name of the copyright holder nor the names of its | |
| // contributors may be used to endorse or promote products derived from | |
| // this software without specific prior written permission. | |
| // | |
| // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" | |
| // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
| // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE | |
| // DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE | |
| // FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | |
| // DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR | |
| // SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER | |
| // CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, | |
| // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | |
| // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
| #include "src/dawn/native/d3d12/TextureCopySplitter.h" | |
| #include "src/dawn/native/Format.h" | |
| #include "src/dawn/native/d3d12/d3d12_platform.h" | |
| #include "src/utils/assert.h" | |
| namespace dawn::native::d3d12 { | |
| namespace { | |
| BlockOrigin3D ComputeBlockOffsets(const TypedTexelBlockInfo& blockInfo, | |
| uint32_t offset, | |
| BlockCount blocksPerRow) { | |
| DAWN_ASSERT(blocksPerRow != BlockCount{0u}); | |
| BlockCount offsetInBlocks = blockInfo.BytesToBlocks(offset); | |
| BlockCount blockOffsetX = offsetInBlocks % blocksPerRow; | |
| BlockCount blockOffsetY = offsetInBlocks / blocksPerRow; | |
| return {blockOffsetX, blockOffsetY, BlockCount{0u}}; | |
| } | |
| uint64_t OffsetToFirstCopiedTexel(const TypedTexelBlockInfo& blockInfo, | |
| BlockCount blocksPerRow, | |
| uint64_t alignedOffset, | |
| BlockOrigin3D bufferOffset) { | |
| DAWN_ASSERT(bufferOffset.z == BlockCount{0u}); | |
| uint64_t offset = | |
| alignedOffset + blockInfo.ToBytes(bufferOffset.x + blocksPerRow * bufferOffset.y); | |
| return offset; | |
| } | |
| uint64_t AlignDownForDataPlacement(uint64_t offset) { | |
| return offset & ~static_cast<uint64_t>(D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT - 1); | |
| } | |
| void Recompute3DTextureCopyRegionWithEmptyFirstRowAndEvenCopyHeight( | |
| BlockOrigin3D origin, | |
| BlockExtent3D copySize, | |
| const TypedTexelBlockInfo& blockInfo, | |
| BlockCount blocksPerRow, | |
| BlockCount rowsPerImage, | |
| TextureCopySubresource& copy, | |
| uint32_t i) { | |
| // Let's assign data and show why copy region generated by ComputeTextureCopySubresource | |
| // is incorrect if there is an empty row at the beginning of the copy block. | |
| // Assuming that bytesPerRow is 256 and we are doing a B2T copy, and copy size is {width: 2, | |
| // height: 4, depthOrArrayLayers: 3}. Then the data layout in buffer is demonstrated | |
| // as below: | |
| // | |
| // |<----- bytes per row ------>| | |
| // | |
| // |----------------------------| | |
| // row (N - 1) | | | |
| // row N | ++~~~~~~~~~| | |
| // row (N + 1) |~~~~~~~~~~~~~~~~~++~~~~~~~~~| | |
| // row (N + 2) |~~~~~~~~~~~~~~~~~++~~~~~~~~~| | |
| // row (N + 3) |~~~~~~~~~~~~~~~~~++~~~~~~~~~| | |
| // row (N + 4) |~~~~~~~~~~~~~~~~~++~~~~~~~~~| | |
| // row (N + 5) |~~~~~~~~~~~~~~~~~++~~~~~~~~~| | |
| // row (N + 6) |~~~~~~~~~~~~~~~~~++~~~~~~~~~| | |
| // row (N + 7) |~~~~~~~~~~~~~~~~~++~~~~~~~~~| | |
| // row (N + 8) |~~~~~~~~~~~~~~~~~++~~~~~~~~~| | |
| // row (N + 9) |~~~~~~~~~~~~~~~~~++~~~~~~~~~| | |
| // row (N + 10) |~~~~~~~~~~~~~~~~~++~~~~~~~~~| | |
| // row (N + 11) |~~~~~~~~~~~~~~~~~++ | | |
| // |----------------------------| | |
| // The copy we mean to do is the following: | |
| // | |
| // - image 0: row N to row (N + 3), | |
| // - image 1: row (N + 4) to row (N + 7), | |
| // - image 2: row (N + 8) to row (N + 11). | |
| // | |
| // Note that alignedOffset is at the beginning of row (N - 1), while buffer offset makes | |
| // the copy start at row N. Row (N - 1) is the empty row between alignedOffset and offset. | |
| // | |
| // The 2D copy region of image 0 we received from Compute2DTextureCopySubresourceAligned() is | |
| // the following: | |
| // | |
| // |-------------------| | |
| // row (N - 1) | | | |
| // row N | ++| | |
| // row (N + 1) |~~~~~~~~~~~~~~~~~++| | |
| // row (N + 2) |~~~~~~~~~~~~~~~~~++| | |
| // row (N + 3) |~~~~~~~~~~~~~~~~~++| | |
| // |-------------------| | |
| // | |
| // However, if we simply expand the copy region of image 0 to all depth ranges of a 3D | |
| // texture, we will copy 5 rows every time, and every first row of each slice will be | |
| // skipped. As a result, the copied data will be: | |
| // | |
| // - image 0: row N to row (N + 3), which is correct. Row (N - 1) is skipped. | |
| // - image 1: row (N + 5) to row (N + 8) because row (N + 4) is skipped. It is incorrect. | |
| // | |
| // Likewise, all other image followed will be incorrect because we wrongly keep skipping | |
| // one row for each depth slice. | |
| // | |
| // Solution: split the copy region to two copies: copy 3 (rowsPerImage - 1) rows and | |
| // expand to all depth slices in the first copy. 3 rows + one skipped rows = 4 rows, which | |
| // equals to rowsPerImage. Then copy the last row in the second copy. However, the copy | |
| // block of the last row of the last image may out-of-bound (see the details below), so | |
| // we need an extra copy for the very last row. | |
| // Copy 0: copy 3 rows, not 4 rows. | |
| // _____________________ | |
| // / /| | |
| // / / | | |
| // |-------------------| | | |
| // row (N - 1) | | | | |
| // row N | ++| | | |
| // row (N + 1) |~~~~~~~~~~~~~~~~~++| / | |
| // row (N + 2) |~~~~~~~~~~~~~~~~~++|/ | |
| // |-------------------| | |
| // Copy 1: move down two rows and copy the last row on image 0, and expand to | |
| // copySize.depthOrArrayLayers - 1 depth slices. Note that if we expand it to all depth | |
| // slices, the last copy block will be row (N + 9) to row (N + 12). Row (N + 11) might | |
| // be the last row of the entire buffer. Then row (N + 12) will be out-of-bound. | |
| // _____________________ | |
| // / /| | |
| // / / | | |
| // |-------------------| | | |
| // row (N + 1) | | | | |
| // row (N + 2) | | | | |
| // row (N + 3) | ++| / | |
| // row (N + 4) |~~~~~~~~~~~~~~~~~~~|/ | |
| // |-------------------| | |
| // | |
| // copy 2: copy the last row of the last image. | |
| // |-------------------| | |
| // row (N + 11)| ++| | |
| // |-------------------| | |
| // Copy 0: copy copySize0.height - 1 rows | |
| TextureCopySubresource::CopyInfo& copy0 = copy.copies[i]; | |
| copy0.copySize.height = copySize.height - BlockCount{1u}; | |
| copy0.bufferSize.height = rowsPerImage; | |
| // Copy 1: move down 2 rows and copy the last row on image 0, and expand to all depth slices | |
| // but the last one. | |
| TextureCopySubresource::CopyInfo* copy1 = copy.AddCopy(); | |
| *copy1 = copy0; | |
| copy1->alignedOffset = copy1->alignedOffset + 2 * blockInfo.ToBytes(blocksPerRow); | |
| copy1->textureOffset.y += copySize.height - BlockCount{1u}; | |
| // Offset two rows from the copy height for bufferOffset1 (See the figure above): | |
| // - one for the row we advanced in the buffer: row (N + 4). | |
| // - one for the last row we want to copy: row (N + 3) itself. | |
| copy1->bufferOffset.y = copySize.height - BlockCount{2u}; | |
| copy1->copySize.height = BlockCount{1u}; | |
| copy1->copySize.depthOrArrayLayers--; | |
| copy1->bufferSize.depthOrArrayLayers--; | |
| // Copy 2: copy the last row of the last image. | |
| uint64_t offsetForCopy0 = | |
| OffsetToFirstCopiedTexel(blockInfo, blocksPerRow, copy0.alignedOffset, copy0.bufferOffset); | |
| uint64_t offsetForLastRowOfLastImage = | |
| offsetForCopy0 + | |
| blockInfo.ToBytes(blocksPerRow * | |
| (copy0.copySize.height + | |
| rowsPerImage * (copySize.depthOrArrayLayers - BlockCount{1u}))); | |
| uint64_t alignedOffsetForLastRowOfLastImage = | |
| AlignDownForDataPlacement(offsetForLastRowOfLastImage); | |
| BlockOrigin3D blockOffsetForLastRowOfLastImage = ComputeBlockOffsets( | |
| blockInfo, | |
| static_cast<uint32_t>(offsetForLastRowOfLastImage - alignedOffsetForLastRowOfLastImage), | |
| blocksPerRow); | |
| TextureCopySubresource::CopyInfo* copy2 = copy.AddCopy(); | |
| copy2->alignedOffset = alignedOffsetForLastRowOfLastImage; | |
| copy2->textureOffset = copy1->textureOffset; | |
| copy2->textureOffset.z = origin.z + copySize.depthOrArrayLayers - BlockCount{1u}; | |
| copy2->copySize = copy1->copySize; | |
| copy2->copySize.depthOrArrayLayers = BlockCount{1u}; | |
| copy2->bufferOffset = blockOffsetForLastRowOfLastImage; | |
| copy2->bufferSize.width = copy1->bufferSize.width; | |
| DAWN_ASSERT(copy2->copySize.height == BlockCount{1u}); | |
| copy2->bufferSize.height = copy2->bufferOffset.y + copy2->copySize.height; | |
| copy2->bufferSize.depthOrArrayLayers = BlockCount{1u}; | |
| } | |
| void Recompute3DTextureCopyRegionWithEmptyFirstRowAndOddCopyHeight( | |
| BlockExtent3D copySize, | |
| const TypedTexelBlockInfo& blockInfo, | |
| BlockCount blocksPerRow, | |
| TextureCopySubresource& copy, | |
| uint32_t i) { | |
| // Read the comments of Recompute3DTextureCopyRegionWithEmptyFirstRowAndEvenCopyHeight() for | |
| // the reason why it is incorrect if we simply extend the copy region to all depth slices | |
| // when there is an empty first row at the copy region. | |
| // | |
| // If the copy height is odd, we can use two copies to make it correct: | |
| // - copy 0: only copy the first depth slice. Keep other arguments the same. | |
| // - copy 1: copy all rest depth slices because it will start without an empty row if | |
| // copy height is odd. Odd height + one (empty row) is even. An even row number times | |
| // bytesPerRow (256) will be aligned to D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT (512) | |
| // Copy 0: copy the first depth slice (image 0) | |
| TextureCopySubresource::CopyInfo& copy0 = copy.copies[i]; | |
| copy0.copySize.depthOrArrayLayers = BlockCount{1u}; | |
| const BlockCount kBufferDepth0 = BlockCount{1u}; | |
| copy0.bufferSize.depthOrArrayLayers = kBufferDepth0; | |
| // Copy 1: copy the rest depth slices in one shot | |
| TextureCopySubresource::CopyInfo* copy1 = copy.AddCopy(); | |
| *copy1 = copy0; | |
| DAWN_ASSERT(copySize.height % BlockCount{2u} == BlockCount{1u}); | |
| copy1->alignedOffset += blockInfo.ToBytes((copySize.height + BlockCount{1u}) * blocksPerRow); | |
| DAWN_ASSERT(copy1->alignedOffset % D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT == 0); | |
| // textureOffset1.z should add one because the first slice has already been copied in copy0. | |
| copy1->textureOffset.z++; | |
| // bufferOffset1.y should be 0 because we skipped the first depth slice and there is no empty | |
| // row in this copy region. | |
| copy1->bufferOffset.y = BlockCount{0u}; | |
| copy1->copySize.height = copySize.height; | |
| copy1->copySize.depthOrArrayLayers = copySize.depthOrArrayLayers - BlockCount{1u}; | |
| copy1->bufferSize.height = copySize.height; | |
| copy1->bufferSize.depthOrArrayLayers = copySize.depthOrArrayLayers - BlockCount{1u}; | |
| } | |
| TextureCopySubresource Compute2DTextureCopySubresourceAligned(BlockOrigin3D origin, | |
| BlockExtent3D copySize, | |
| const TypedTexelBlockInfo& blockInfo, | |
| uint64_t offset, | |
| BlockCount blocksPerRow) { | |
| TextureCopySubresource copy; | |
| // The copies must be 512-aligned. To do this, we calculate the first 512-aligned address | |
| // preceding our data. | |
| uint64_t alignedOffset = AlignDownForDataPlacement(offset); | |
| // If the provided offset to the data was already 512-aligned, we can simply copy the data | |
| // without further translation. | |
| if (offset == alignedOffset) { | |
| TextureCopySubresource::CopyInfo* copyInfo = copy.AddCopy(); | |
| copyInfo->bufferOffset = {}; // 0,0,0 | |
| copyInfo->textureOffset = origin; | |
| copyInfo->copySize = copySize; | |
| copyInfo->alignedOffset = alignedOffset; | |
| copyInfo->bufferSize = copySize; | |
| return copy; | |
| } | |
| DAWN_ASSERT(alignedOffset < offset); | |
| DAWN_ASSERT(offset - alignedOffset < D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT); | |
| // We must reinterpret our aligned offset into X and Y offsets with respect to the row | |
| // pitch. | |
| // | |
| // You can visualize the data in the buffer like this: | |
| // |-----------------------++++++++++++++++++++++++++++++++| | |
| // ^ 512-aligned address ^ Aligned offset ^ End of copy data | |
| // | |
| // Now when you consider the row pitch, you can visualize the data like this: | |
| // |~~~~~~~~~~~~~~~~| | |
| // |~~~~~+++++++++++| | |
| // |++++++++++++++++| | |
| // |+++++~~~~~~~~~~~| | |
| // |<---row pitch-->| | |
| // | |
| // The X and Y offsets calculated in ComputeBlockOffsets can be visualized like this: | |
| // |YYYYYYYYYYYYYYYY| | |
| // |XXXXXX++++++++++| | |
| // |++++++++++++++++| | |
| // |++++++~~~~~~~~~~| | |
| // |<---row pitch-->| | |
| BlockOrigin3D blockOffset = | |
| ComputeBlockOffsets(blockInfo, static_cast<uint32_t>(offset - alignedOffset), blocksPerRow); | |
| DAWN_ASSERT(blockOffset.y <= BlockCount{1u}); | |
| DAWN_ASSERT(blockOffset.z == BlockCount{0u}); | |
| BlockCount copyBlocksPerRowPitch = copySize.width; | |
| BlockCount blockOffsetInRowPitch = blockOffset.x; | |
| if (copyBlocksPerRowPitch + blockOffsetInRowPitch <= blocksPerRow) { | |
| // The region's rows fit inside the bytes per row. In this case, extend the width of the | |
| // PlacedFootprint and copy the buffer with an offset location | |
| // |<------------- bytes per row ------------->| | |
| // | |
| // |-------------------------------------------| | |
| // | | | |
| // | +++++++++++++++++~~~~~~~~~| | |
| // |~~~~~~~~~~~~~~~~~+++++++++++++++++~~~~~~~~~| | |
| // |~~~~~~~~~~~~~~~~~+++++++++++++++++~~~~~~~~~| | |
| // |~~~~~~~~~~~~~~~~~+++++++++++++++++~~~~~~~~~| | |
| // |~~~~~~~~~~~~~~~~~+++++++++++++++++ | | |
| // |-------------------------------------------| | |
| // Copy 0: | |
| // |----------------------------------| | |
| // | | | |
| // | +++++++++++++++++| | |
| // |~~~~~~~~~~~~~~~~~+++++++++++++++++| | |
| // |~~~~~~~~~~~~~~~~~+++++++++++++++++| | |
| // |~~~~~~~~~~~~~~~~~+++++++++++++++++| | |
| // |~~~~~~~~~~~~~~~~~+++++++++++++++++| | |
| // |----------------------------------| | |
| TextureCopySubresource::CopyInfo* copyInfo = copy.AddCopy(); | |
| copyInfo->bufferOffset = blockOffset; | |
| copyInfo->textureOffset = origin; | |
| copyInfo->copySize = copySize; | |
| copyInfo->alignedOffset = alignedOffset; | |
| copyInfo->bufferSize = {copySize.width + blockOffset.x, copySize.height + blockOffset.y, | |
| copySize.depthOrArrayLayers}; | |
| return copy; | |
| } | |
| // The region's rows straddle the bytes per row. Split the copy into two copies | |
| // |<------------- bytes per row ------------->| | |
| // | |
| // |-------------------------------------------| | |
| // | | | |
| // | ++++++++| | |
| // |+++++++++~~~~~~~~~~~~~~~~~~~~~~~~~~++++++++| | |
| // |+++++++++~~~~~~~~~~~~~~~~~~~~~~~~~~++++++++| | |
| // |+++++++++~~~~~~~~~~~~~~~~~~~~~~~~~~++++++++| | |
| // |+++++++++~~~~~~~~~~~~~~~~~~~~~~~~~~++++++++| | |
| // |+++++++++ | | |
| // |-------------------------------------------| | |
| // Copy 0: | |
| // |-------------------------------------------| | |
| // | | | |
| // | ++++++++| | |
| // |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~++++++++| | |
| // |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~++++++++| | |
| // |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~++++++++| | |
| // |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~++++++++| | |
| // |-------------------------------------------| | |
| // Copy 1: | |
| // |---------| | |
| // | | | |
| // | | | |
| // |+++++++++| | |
| // |+++++++++| | |
| // |+++++++++| | |
| // |+++++++++| | |
| // |+++++++++| | |
| // |---------| | |
| // Copy 0 | |
| DAWN_ASSERT(blocksPerRow > blockOffsetInRowPitch); | |
| const BlockExtent3D copySize0 = {blocksPerRow - blockOffset.x, copySize.height, | |
| copySize.depthOrArrayLayers}; | |
| TextureCopySubresource::CopyInfo* copyInfo0 = copy.AddCopy(); | |
| copyInfo0->bufferOffset = blockOffset; | |
| copyInfo0->textureOffset = origin; | |
| copyInfo0->copySize = copySize0; | |
| copyInfo0->alignedOffset = alignedOffset; | |
| copyInfo0->bufferSize = {blocksPerRow, copySize.height + blockOffset.y, | |
| copySize.depthOrArrayLayers}; | |
| // Copy 1 | |
| const uint64_t offsetForCopy1 = offset + blockInfo.ToBytes(copySize0.width); | |
| const uint64_t alignedOffsetForCopy1 = AlignDownForDataPlacement(offsetForCopy1); | |
| const BlockOrigin3D blockOffsetForCopy1 = ComputeBlockOffsets( | |
| blockInfo, static_cast<uint32_t>(offsetForCopy1 - alignedOffsetForCopy1), blocksPerRow); | |
| DAWN_ASSERT(blockOffsetForCopy1.y <= BlockCount{1u}); | |
| DAWN_ASSERT(blockOffsetForCopy1.z == BlockCount{0u}); | |
| const BlockOrigin3D textureOffset1 = {origin.x + copySize0.width, origin.y, origin.z}; | |
| DAWN_ASSERT(copySize.width > copySize0.width); | |
| const BlockExtent3D copySize1 = {copySize.width - copySize0.width, copySize.height, | |
| copySize.depthOrArrayLayers}; | |
| const BlockOrigin3D bufferOffset1 = blockOffsetForCopy1; | |
| const BlockExtent3D bufferSize1 = {copySize1.width + blockOffsetForCopy1.x, | |
| copySize.height + blockOffsetForCopy1.y, | |
| copySize.depthOrArrayLayers}; | |
| TextureCopySubresource::CopyInfo* copyInfo1 = copy.AddCopy(); | |
| copyInfo1->bufferOffset = bufferOffset1; | |
| copyInfo1->textureOffset = textureOffset1; | |
| copyInfo1->copySize = copySize1; | |
| copyInfo1->alignedOffset = alignedOffsetForCopy1; | |
| copyInfo1->bufferSize = bufferSize1; | |
| return copy; | |
| } | |
| TextureCopySubresource Compute2DTextureCopySubresourceRelaxed(BlockOrigin3D origin, | |
| BlockExtent3D copySize, | |
| const TypedTexelBlockInfo& blockInfo, | |
| uint64_t offset, | |
| BlockCount /*blocksPerRow*/) { | |
| 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------>| |----------| | |
| // | |
| copyInfo->textureOffset = {origin.x, origin.y, BlockCount{0u}}; | |
| copyInfo->bufferOffset = {}; // 0,0,0 | |
| copyInfo->copySize = {copySize.width, copySize.height, BlockCount{1u}}; | |
| copyInfo->alignedOffset = offset; | |
| copyInfo->bufferSize = {copySize.width, copySize.height, BlockCount{1u}}; | |
| return copy; | |
| } | |
| TextureCopySubresource Compute3DTextureCopySubresourceAligned(BlockOrigin3D origin, | |
| BlockExtent3D copySize, | |
| const TypedTexelBlockInfo& blockInfo, | |
| uint64_t offset, | |
| BlockCount blocksPerRow, | |
| BlockCount rowsPerImage) { | |
| // To compute the copy region(s) for 3D textures, we call Compute2DTextureCopySubresourceAligned | |
| // and get copy region(s) for the first slice of the copy, then extend to all depth slices | |
| // and become a 3D copy. However, this doesn't work as easily as that due to some corner | |
| // cases. | |
| // | |
| // For example, if bufferHeight is greater than rowsPerImage in each generated copy | |
| // region and we simply extend the 2D copy region to all copied depth slices, copied data | |
| // will be incorrectly offset for each depth slice except the first one. | |
| // | |
| // For these special cases, we need to recompute the copy regions for 3D textures by | |
| // splitting the incorrect copy region to a couple more copy regions. | |
| // Call Compute2DTextureCopySubresourceAligned and get copy regions. This function has already | |
| // forwarded "copySize.depthOrArrayLayers" to all depth slices. | |
| TextureCopySubresource copySubresource = | |
| Compute2DTextureCopySubresourceAligned(origin, copySize, blockInfo, offset, blocksPerRow); | |
| DAWN_ASSERT(copySubresource.count <= 2); | |
| // If copySize.depthOrArrayLayers is 1, we can return copySubresource. Because we don't need to | |
| // extend the copy region(s) to other depth slice(s). | |
| if (copySize.depthOrArrayLayers == BlockCount{1u}) { | |
| return copySubresource; | |
| } | |
| // The copy region(s) generated by Compute2DTextureCopySubresourceAligned might be incorrect. | |
| // However, we may append a couple more copy regions in the for loop below. We don't need | |
| // to revise these new added copy regions. | |
| uint32_t originalCopyCount = copySubresource.count; | |
| for (uint32_t i = 0; i < originalCopyCount; ++i) { | |
| // There can be one empty row at most in a copy region. | |
| BlockCount bufferHeight = copySubresource.copies[i].bufferSize.height; | |
| DAWN_ASSERT(bufferHeight <= rowsPerImage + BlockCount{1u}); | |
| if (bufferHeight == rowsPerImage) { | |
| // If the copy region's bufferHeight equals to rowsPerImage, we can use this | |
| // copy region without any modification. | |
| continue; | |
| } | |
| if (bufferHeight < rowsPerImage) { | |
| // If we are copying multiple depth slices, we should skip rowsPerImage rows for | |
| // each slice even though we only copy partial rows in each slice sometimes. | |
| copySubresource.copies[i].bufferSize.height = rowsPerImage; | |
| } else { | |
| // bufferHeight > rowsPerImage. There is an empty row in this copy region due to | |
| // alignment adjustment. | |
| // bytesPerRow is definitely 256, and it is definitely a full copy on height. | |
| // Otherwise, bufferHeight won't be greater than rowsPerImage and there won't be | |
| // an empty row at the beginning of this copy region. | |
| uint64_t bytesPerRow = blockInfo.ToBytes(blocksPerRow); | |
| DAWN_ASSERT(bytesPerRow == D3D12_TEXTURE_DATA_PITCH_ALIGNMENT); | |
| DAWN_ASSERT(copySize.height == rowsPerImage); | |
| const BlockCount copyHeight = copySize.height; | |
| if (dchecked_cast<uint32_t>(copyHeight) % 2 == 0) { | |
| // If copyHeight is even and there is an empty row at the beginning of the | |
| // first slice of the copy region, the offset of all depth slices will never be | |
| // aligned to D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT (512) and there is always | |
| // an empty row at each depth slice. We need a totally different approach to | |
| // split the copy region. | |
| Recompute3DTextureCopyRegionWithEmptyFirstRowAndEvenCopyHeight( | |
| origin, copySize, blockInfo, blocksPerRow, rowsPerImage, copySubresource, i); | |
| } else { | |
| // If copyHeight is odd and there is an empty row at the beginning of the | |
| // first slice of the copy region, we can split the copy region into two copies: | |
| // copy0 to copy the first slice, copy1 to copy the rest slices because the | |
| // offset of slice 1 is aligned to D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT (512) | |
| // without an empty row. This is an easier case relative to cases with even copy | |
| // height. | |
| Recompute3DTextureCopyRegionWithEmptyFirstRowAndOddCopyHeight( | |
| copySize, blockInfo, blocksPerRow, copySubresource, i); | |
| } | |
| } | |
| } | |
| return copySubresource; | |
| } | |
| TextureCopySubresource Compute3DTextureCopySubresourceRelaxed(BlockOrigin3D origin, | |
| BlockExtent3D copySize, | |
| const TypedTexelBlockInfo& blockInfo, | |
| uint64_t offset, | |
| BlockCount blocksPerRow, | |
| BlockCount rowsPerImage) { | |
| TextureCopySubresource copy; | |
| BlockOrigin3D bufferOffset{BlockCount{0u}, BlockCount{0u}, BlockCount{0u}}; | |
| // You can visualize the data in the buffer (bufferLocation) like the inline comments. | |
| // * copy data is visualized as '+'. | |
| const BlockCount depthInCopy1 = copySize.depthOrArrayLayers - BlockCount{1u}; | |
| if (depthInCopy1 > BlockCount{0u}) { | |
| // `bufferLocation` in the 1st copy (first `depthInCopy1` images, optional): | |
| // | |
| // bufferOffset(0, 0, 0) | |
| // ^ | |
| // | | |
| // |<-------Offset1------>|<-----------RowPitch----------->|----------|------------| | |
| // |----------------------|++++++++++++++++++++++~~~~~~~~~~| | | | | | |
| // |++++++++++++++++++++++~~~~~~~~~~|CopyHeight| | | | |
| // |++++++++++++++++++++++~~~~~~~~~~| | |RowsPerImage| | |
| // |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~|----------| | | | |
| // |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~| | | | | |
| // |---End of 1st image-->|--------------------------------|----------|------------| | |
| // |++++++++++++++++++++++~~~~~~~~~~| | | | | |
| // |++++++++++++++++++++++~~~~~~~~~~| | | | | |
| // |++++++++++++++++++++++~~~~~~~~~~| |RowsPerImage| | |
| // |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~| | | | | |
| // |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~| | | | | |
| // |---End of 2nd image-->|--------------------------------|----------|------------| | |
| // |<-----CopyWidth------>| | |
| // | |
| auto* copyInfo1 = copy.AddCopy(); | |
| copyInfo1->bufferOffset = bufferOffset; | |
| copyInfo1->textureOffset = origin; | |
| copyInfo1->copySize = {copySize.width, copySize.height, depthInCopy1}; | |
| copyInfo1->alignedOffset = offset; | |
| copyInfo1->bufferSize = {copySize.width, rowsPerImage, depthInCopy1}; | |
| } | |
| { | |
| // We have to use the 2nd copy because there may not be enough memory to hold | |
| // (RowPitch * RowsPerImage) data for the last image in the buffer. | |
| // | |
| // `bufferLocation` in the 2nd copy (the last image): | |
| // | |
| // bufferOffset (0, 0, 0) | |
| // Begin of the last image | |
| // ^ | |
| // | | |
| // |<-------Offset2------>|<-----------RowPitch----------->|----------| | |
| // |----------------------|++++++++++++++++++++++~~~~~~~~~~| | | | |
| // |++++++++++++++++++++++~~~~~~~~~~|CopyHeight| | |
| // |++++++++++++++++++++++| | | | | |
| // |----------------------|---------|----------| | |
| // |<-----CopyWidth------>| | |
| // ^ | |
| // End of all buffer data | |
| // | |
| DAWN_ASSERT(copySize.depthOrArrayLayers >= BlockCount{1u}); | |
| constexpr BlockCount depthInCopy2{1u}; | |
| const BlockCount rowsPerImageInTexels2 = copySize.height; | |
| auto* copyInfo2 = copy.AddCopy(); | |
| copyInfo2->bufferOffset = bufferOffset; | |
| copyInfo2->textureOffset = {origin.x, origin.y, origin.z + depthInCopy1}; | |
| copyInfo2->copySize = {copySize.width, copySize.height, depthInCopy2}; | |
| copyInfo2->alignedOffset = | |
| offset + blockInfo.ToBytes(blocksPerRow * rowsPerImage * depthInCopy1); | |
| copyInfo2->bufferSize = {copySize.width, rowsPerImageInTexels2, depthInCopy2}; | |
| } | |
| return copy; | |
| } | |
| } // namespace | |
| TextureCopySubresource::CopyInfo* TextureCopySubresource::AddCopy() { | |
| DAWN_ASSERT(this->count < kMaxTextureCopyRegions); | |
| return &this->copies[this->count++]; | |
| } | |
| TextureCopySubresource Compute2DTextureCopySubresource(BlockOrigin3D origin, | |
| BlockExtent3D copySize, | |
| const TypedTexelBlockInfo& blockInfo, | |
| uint64_t offset, | |
| BlockCount blocksPerRow, | |
| bool relaxedRowAndPitchOffset) { | |
| if (relaxedRowAndPitchOffset) { | |
| return Compute2DTextureCopySubresourceRelaxed(origin, copySize, blockInfo, offset, | |
| blocksPerRow); | |
| } | |
| return Compute2DTextureCopySubresourceAligned(origin, copySize, blockInfo, offset, | |
| blocksPerRow); | |
| } | |
| TextureCopySubresource Compute3DTextureCopySubresource(BlockOrigin3D origin, | |
| BlockExtent3D copySize, | |
| const TypedTexelBlockInfo& blockInfo, | |
| uint64_t offset, | |
| BlockCount blocksPerRow, | |
| BlockCount rowsPerImage, | |
| bool relaxedRowAndPitchOffset) { | |
| if (relaxedRowAndPitchOffset) { | |
| return Compute3DTextureCopySubresourceRelaxed(origin, copySize, blockInfo, offset, | |
| blocksPerRow, rowsPerImage); | |
| } | |
| return Compute3DTextureCopySubresourceAligned(origin, copySize, blockInfo, offset, blocksPerRow, | |
| rowsPerImage); | |
| } | |
| TextureCopySplits Compute2DTextureCopySplits(BlockOrigin3D origin, | |
| BlockExtent3D copySize, | |
| const TypedTexelBlockInfo& blockInfo, | |
| uint64_t offset, | |
| BlockCount blocksPerRow, | |
| BlockCount rowsPerImage) { | |
| TextureCopySplits copies; | |
| // The function Compute2DTextureCopySubresourceAligned() decides how to split the copy based on: | |
| // - the alignment of the buffer offset with D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT (512) | |
| // - the alignment of the buffer offset with D3D12_TEXTURE_DATA_PITCH_ALIGNMENT (256) | |
| // Each layer of a 2D array might need to be split, but because of the WebGPU | |
| // constraint that "bytesPerRow" must be a multiple of 256, all odd (resp. all even) layers | |
| // will be at an offset multiple of 512 of each other, which means they will all result in | |
| // the same 2D split. Thus we can just compute the copy splits for the first and second | |
| // layers, and reuse them for the remaining layers by adding the related offset of each | |
| // layer. Moreover, if "rowsPerImage" is even, both the first and second copy layers can | |
| // share the same copy split, so in this situation we just need to compute copy split once | |
| // and reuse it for all the layers. | |
| BlockExtent3D copyOneLayerSize = copySize; | |
| BlockOrigin3D copyFirstLayerOrigin = origin; | |
| copyOneLayerSize.depthOrArrayLayers = BlockCount{1u}; | |
| copyFirstLayerOrigin.z = BlockCount{0u}; | |
| copies.copySubresources[0] = Compute2DTextureCopySubresourceAligned( | |
| copyFirstLayerOrigin, copyOneLayerSize, blockInfo, offset, blocksPerRow); | |
| // When the copy only refers one texture 2D array layer, | |
| // copies.copySubresources[1] will never be used so we can safely early return here. | |
| if (copySize.depthOrArrayLayers == BlockCount{1u}) { | |
| return copies; | |
| } | |
| const uint64_t bytesPerLayer = blockInfo.ToBytes(blocksPerRow * rowsPerImage); | |
| if (bytesPerLayer % D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT == 0) { | |
| copies.copySubresources[1] = copies.copySubresources[0]; | |
| uint64_t alignedOffset0 = | |
| copies.copySubresources[1].copies[0].alignedOffset + bytesPerLayer; | |
| uint64_t alignedOffset1 = | |
| copies.copySubresources[1].copies[1].alignedOffset + bytesPerLayer; | |
| copies.copySubresources[1].copies[0].alignedOffset = alignedOffset0; | |
| copies.copySubresources[1].copies[1].alignedOffset = alignedOffset1; | |
| } else { | |
| const uint64_t bufferOffsetNextLayer = offset + bytesPerLayer; | |
| copies.copySubresources[1] = Compute2DTextureCopySubresourceAligned( | |
| copyFirstLayerOrigin, copyOneLayerSize, blockInfo, bufferOffsetNextLayer, blocksPerRow); | |
| } | |
| return copies; | |
| } | |
| } // namespace dawn::native::d3d12 |