Support texture readback through copyTextureRegion

[IsaacMarovitz]

Explicitly adds support for copyTextureRegion calls in the other direction.

[squidbus]

I can't vouch for the D3D12 but the others look okay to me.

[dcvz]

One thing I noticed is that it ignores dstX/Y/Z and sizes the copy from the footprint instead of srcBox which is not how D3D12 does it.

[dcvz]

This takes the copy size from the footprint, D3D12 uses srcBox or the whole subresource when there's no box:

Suggested change

	const MTL::Size size = { dstLocation.placedFootprint.width, dstLocation.placedFootprint.height, dstLocation.placedFootprint.depth };
	const uint32_t mipLevel = srcLocation.subresource.mipLevel;
	const MTL::Size size = (srcBox != nullptr)
	? MTL::Size { NS::UInteger(srcBox->right - srcBox->left), NS::UInteger(srcBox->bottom - srcBox->top), NS::UInteger(srcBox->back - srcBox->front) }
	: MTL::Size { std::max(srcTexture->desc.width >> mipLevel, 1U), std::max(srcTexture->desc.height >> mipLevel, 1U), std::max(srcTexture->desc.depth >> mipLevel, 1U) };

[dcvz]

dstX/Y/Z place the region inside the footprint, so they need to go into the buffer offset:

Suggested change

	const uint32_t bytesPerImage = bytesPerRow * verticalBlocks;
	const uint32_t bytesPerImage = bytesPerRow * verticalBlocks;
	const uint64_t dstOffset = dstLocation.placedFootprint.offset + dstZ * bytesPerImage + (dstY / blockWidth) * bytesPerRow + (dstX / blockWidth) * RenderFormatSize(srcTexture->desc.format);

[dcvz]

Suggested change

	dstLocation.placedFootprint.offset,
	dstOffset,

[dcvz]

Same two things on Vulkan, the offset:

Suggested change

	VkBufferImageCopy imageCopy = {};
	imageCopy.bufferOffset = dstLocation.placedFootprint.offset;
	const uint32_t blockSize = RenderFormatSize(srcTexture->desc.format);
	const uint32_t bytesPerRow = ((dstLocation.placedFootprint.rowWidth + blockWidth - 1) / blockWidth) * blockSize;
	const uint32_t bytesPerImage = bytesPerRow * ((dstLocation.placedFootprint.height + blockWidth - 1) / blockWidth);
	const uint32_t mipLevel = srcLocation.subresource.mipLevel;
	VkBufferImageCopy imageCopy = {};
	imageCopy.bufferOffset = dstLocation.placedFootprint.offset + dstZ * bytesPerImage + (dstY / blockWidth) * bytesPerRow + (dstX / blockWidth) * blockSize;

[dcvz]

and the size:

Suggested change

	imageCopy.imageExtent.width = dstLocation.placedFootprint.width;
	imageCopy.imageExtent.height = dstLocation.placedFootprint.height;
	imageCopy.imageExtent.depth = dstLocation.placedFootprint.depth;
	imageCopy.imageExtent.width = srcBox != nullptr ? uint32_t(srcBox->right - srcBox->left) : std::max(srcTexture->desc.width >> mipLevel, 1U);
	imageCopy.imageExtent.height = srcBox != nullptr ? uint32_t(srcBox->bottom - srcBox->top) : std::max(srcTexture->desc.height >> mipLevel, 1U);
	imageCopy.imageExtent.depth = srcBox != nullptr ? uint32_t(srcBox->back - srcBox->front) : std::max(srcTexture->desc.depth >> mipLevel, 1U);