Modified: trunk/Source/WebGPU/WebGPU/Queue.mm (292200 => 292201)
--- trunk/Source/WebGPU/WebGPU/Queue.mm 2022-04-01 04:50:58 UTC (rev 292200)
+++ trunk/Source/WebGPU/WebGPU/Queue.mm 2022-04-01 05:00:52 UTC (rev 292201)
@@ -30,6 +30,7 @@
#import "Buffer.h"
#import "CommandBuffer.h"
#import "Device.h"
+#import "Texture.h"
namespace WebGPU {
@@ -236,21 +237,257 @@
ensureBlitCommandEncoder();
// FIXME(PERFORMANCE): Suballocate, so the common case doesn't need to hit the kernel.
+ // FIXME(PERFORMANCE): Should this temporary buffer really be shared?
id<MTLBuffer> temporaryBuffer = [m_device.device() newBufferWithBytes:data length:static_cast<NSUInteger>(size) options:MTLResourceStorageModeShared];
if (!temporaryBuffer)
return;
- [m_blitCommandEncoder copyFromBuffer:temporaryBuffer sourceOffset:0 toBuffer:buffer.buffer() destinationOffset:static_cast<NSUInteger>(bufferOffset) size:static_cast<NSUInteger>(size)];
+ [m_blitCommandEncoder
+ copyFromBuffer:temporaryBuffer
+ sourceOffset:0
+ toBuffer:buffer.buffer()
+ destinationOffset:static_cast<NSUInteger>(bufferOffset)
+ size:static_cast<NSUInteger>(size)];
}
-void Queue::writeTexture(const WGPUImageCopyTexture& destination, const void* data, size_t dataSize, const WGPUTextureDataLayout& dataLayout, const WGPUExtent3D& writeSize)
+static bool validateWriteTexture(const WGPUImageCopyTexture& destination, const WGPUTextureDataLayout& dataLayout, const WGPUExtent3D& size, size_t dataByteSize, const WGPUTextureDescriptor& textureDesc)
{
- UNUSED_PARAM(destination);
- UNUSED_PARAM(data);
- UNUSED_PARAM(dataSize);
- UNUSED_PARAM(dataLayout);
- UNUSED_PARAM(writeSize);
+ // "validating GPUImageCopyTexture(destination, size) returns true."
+ if (!Texture::validateImageCopyTexture(destination, size))
+ return false;
+
+ // "textureDesc.usage includes COPY_DST."
+ if (!(textureDesc.usage & WGPUTextureUsage_CopyDst))
+ return false;
+
+ // "textureDesc.sampleCount is 1."
+ if (textureDesc.sampleCount != 1)
+ return false;
+
+ // "validating texture copy range(destination, size) return true."
+ if (!Texture::validateTextureCopyRange(destination, size))
+ return false;
+
+ // "destination.aspect must refer to a single aspect of textureDesc.format."
+ if (!Texture::refersToSingleAspect(textureDesc.format, destination.aspect))
+ return false;
+
+ // "That aspect must be a valid image copy destination according to § 25.1.2 Depth-stencil formats."
+ if (!Texture::isValidImageCopyDestination(textureDesc.format, destination.aspect))
+ return false;
+
+ // "Let aspectSpecificFormat = textureDesc.format."
+ auto aspectSpecificFormat = textureDesc.format;
+
+ // "If textureDesc.format is a depth-or-stencil format:"
+ if (Texture::isDepthOrStencilFormat(textureDesc.format)) {
+ // "Set aspectSpecificFormat to the aspect-specific format of textureDesc.format according to § 25.1.2 Depth-stencil formats."
+ aspectSpecificFormat = Texture::aspectSpecificFormat(textureDesc.format, destination.aspect);
+ }
+
+ // "validating linear texture data(dataLayout, dataByteSize, aspectSpecificFormat, size) succeeds."
+ if (!Texture::validateLinearTextureData(dataLayout, dataByteSize, aspectSpecificFormat, size))
+ return false;
+
+ return true;
}
+void Queue::writeTexture(const WGPUImageCopyTexture& destination, const void* data, size_t dataSize, const WGPUTextureDataLayout& dataLayout, const WGPUExtent3D& size)
+{
+ if (destination.nextInChain || dataLayout.nextInChain)
+ return;
+
+ // https://gpuweb.github.io/gpuweb/#dom-gpuqueue-writetexture
+
+ // "Let dataBytes be a copy of the bytes held by the buffer source data."
+ // "Let dataByteSize be the number of bytes in dataBytes."
+ auto dataByteSize = dataSize;
+
+ // "Let textureDesc be destination.texture.[[descriptor]]."
+ const auto& textureDesc = fromAPI(destination.texture).descriptor();
+
+ // "Let contents be the contents of the images seen by viewing dataBytes with dataLayout and size."
+
+ // "If any of the following conditions are unsatisfied"
+ if (!validateWriteTexture(destination, dataLayout, size, dataByteSize, textureDesc)) {
+ // "generate a validation error and stop."
+ m_device.generateAValidationError("Validation failure."_s);
+ return;
+ }
+
+ // "Write contents into destination."
+ if (!dataSize)
+ return;
+
+ NSUInteger bytesPerRow = dataLayout.bytesPerRow;
+
+ // FIXME: Use checked arithmetic.
+ NSUInteger bytesPerImage = dataLayout.bytesPerRow * dataLayout.rowsPerImage;
+
+ MTLBlitOption options = MTLBlitOptionNone;
+ switch (destination.aspect) {
+ case WGPUTextureAspect_All:
+ options = MTLBlitOptionNone;
+ break;
+ case WGPUTextureAspect_StencilOnly:
+ options = MTLBlitOptionStencilFromDepthStencil;
+ break;
+ case WGPUTextureAspect_DepthOnly:
+ options = MTLBlitOptionDepthFromDepthStencil;
+ break;
+ case WGPUTextureAspect_Force32:
+ return;
+ }
+
+ // FIXME(PERFORMANCE): Instead of checking whether or not the whole queue is idle,
+ // we could detect whether this specific resource is idle, if we tracked every resource.
+ if (isIdle() && options == MTLBlitOptionNone) {
+ switch (fromAPI(destination.texture).texture().storageMode) {
+ case MTLStorageModeShared:
+#if PLATFORM(MAC) || PLATFORM(MACCATALYST)
+ case MTLStorageModeManaged:
+#endif
+ {
+ switch (textureDesc.dimension) {
+ case WGPUTextureDimension_1D: {
+ auto region = MTLRegionMake1D(destination.origin.x, size.width);
+ for (uint32_t layer = 0; layer < size.depthOrArrayLayers; ++layer) {
+ // FIXME: Use checked arithmetic.
+ auto sourceOffset = static_cast<NSUInteger>(dataLayout.offset + layer * bytesPerImage);
+ NSUInteger destinationSlice = destination.origin.z + layer;
+ [fromAPI(destination.texture).texture()
+ replaceRegion:region
+ mipmapLevel:destination.mipLevel
+ slice:destinationSlice
+ withBytes:static_cast<const char*>(data) + sourceOffset
+ bytesPerRow:bytesPerRow
+ bytesPerImage:bytesPerImage];
+ }
+ break;
+ }
+ case WGPUTextureDimension_2D: {
+ auto region = MTLRegionMake2D(destination.origin.x, destination.origin.y, size.width, size.height);
+ for (uint32_t layer = 0; layer < size.depthOrArrayLayers; ++layer) {
+ // FIXME: Use checked arithmetic.
+ auto sourceOffset = static_cast<NSUInteger>(dataLayout.offset + layer * bytesPerImage);
+ NSUInteger destinationSlice = destination.origin.z + layer;
+ [fromAPI(destination.texture).texture()
+ replaceRegion:region
+ mipmapLevel:destination.mipLevel
+ slice:destinationSlice
+ withBytes:static_cast<const char*>(data) + sourceOffset
+ bytesPerRow:bytesPerRow
+ bytesPerImage:bytesPerImage];
+ }
+ break;
+ }
+ case WGPUTextureDimension_3D: {
+ auto region = MTLRegionMake3D(destination.origin.x, destination.origin.y, destination.origin.z, size.width, size.height, size.depthOrArrayLayers);
+ auto sourceOffset = static_cast<NSUInteger>(dataLayout.offset);
+ [fromAPI(destination.texture).texture()
+ replaceRegion:region
+ mipmapLevel:destination.mipLevel
+ slice:0
+ withBytes:static_cast<const char*>(data) + sourceOffset
+ bytesPerRow:bytesPerRow
+ bytesPerImage:bytesPerImage];
+ break;
+ }
+ case WGPUTextureDimension_Force32:
+ ASSERT_NOT_REACHED();
+ return;
+ }
+ return;
+ }
+ case MTLStorageModePrivate:
+ // The only way to get data into a private resource is to tell the GPU to copy it in.
+ break;
+ default:
+ ASSERT_NOT_REACHED();
+ return;
+ }
+ }
+
+ ensureBlitCommandEncoder();
+ // FIXME(PERFORMANCE): Suballocate, so the common case doesn't need to hit the kernel.
+ // FIXME(PERFORMANCE): Should this temporary buffer really be shared?
+ id<MTLBuffer> temporaryBuffer = [m_device.device() newBufferWithBytes:static_cast<const char*>(data) + dataLayout.offset length:static_cast<NSUInteger>(dataByteSize) options:MTLResourceStorageModeShared];
+ if (!temporaryBuffer)
+ return;
+
+ auto logicalSize = fromAPI(destination.texture).logicalMiplevelSpecificTextureExtent(destination.mipLevel);
+ auto widthForMetal = std::min(size.width, logicalSize.width);
+ auto heightForMetal = std::min(size.height, logicalSize.height);
+ auto depthForMetal = std::min(size.depthOrArrayLayers, logicalSize.depthOrArrayLayers);
+
+ switch (textureDesc.dimension) {
+ case WGPUTextureDimension_1D: {
+ // https://developer.apple.com/documentation/metal/mtlblitcommandencoder/1400771-copyfrombuffer?language=objc
+ // "When you copy to a 1D texture, height and depth must be 1."
+ auto sourceSize = MTLSizeMake(widthForMetal, 1, 1);
+ auto destinationOrigin = MTLOriginMake(destination.origin.x, 1, 1);
+ for (uint32_t layer = 0; layer < size.depthOrArrayLayers; ++layer) {
+ // FIXME: Use checked arithmetic.
+ NSUInteger sourceOffset = layer * bytesPerImage;
+ NSUInteger destinationSlice = destination.origin.z + layer;
+ [m_blitCommandEncoder
+ copyFromBuffer:temporaryBuffer
+ sourceOffset:sourceOffset
+ sourceBytesPerRow:bytesPerRow
+ sourceBytesPerImage:bytesPerImage
+ sourceSize:sourceSize
+ toTexture:fromAPI(destination.texture).texture()
+ destinationSlice:destinationSlice
+ destinationLevel:destination.mipLevel
+ destinationOrigin:destinationOrigin
+ options:options];
+ }
+ break;
+ }
+ case WGPUTextureDimension_2D: {
+ // https://developer.apple.com/documentation/metal/mtlblitcommandencoder/1400771-copyfrombuffer?language=objc
+ // "When you copy to a 2D texture, depth must be 1."
+ auto sourceSize = MTLSizeMake(widthForMetal, heightForMetal, 1);
+ auto destinationOrigin = MTLOriginMake(destination.origin.x, destination.origin.y, 1);
+ for (uint32_t layer = 0; layer < size.depthOrArrayLayers; ++layer) {
+ // FIXME: Use checked arithmetic.
+ NSUInteger sourceOffset = layer * bytesPerImage;
+ NSUInteger destinationSlice = destination.origin.z + layer;
+ [m_blitCommandEncoder
+ copyFromBuffer:temporaryBuffer
+ sourceOffset:sourceOffset
+ sourceBytesPerRow:bytesPerRow
+ sourceBytesPerImage:bytesPerImage
+ sourceSize:sourceSize
+ toTexture:fromAPI(destination.texture).texture()
+ destinationSlice:destinationSlice
+ destinationLevel:destination.mipLevel
+ destinationOrigin:destinationOrigin
+ options:options];
+ }
+ break;
+ }
+ case WGPUTextureDimension_3D: {
+ auto sourceSize = MTLSizeMake(widthForMetal, heightForMetal, depthForMetal);
+ auto destinationOrigin = MTLOriginMake(destination.origin.x, destination.origin.y, destination.origin.z);
+ NSUInteger sourceOffset = 0;
+ [m_blitCommandEncoder
+ copyFromBuffer:temporaryBuffer
+ sourceOffset:sourceOffset
+ sourceBytesPerRow:bytesPerRow
+ sourceBytesPerImage:bytesPerImage
+ sourceSize:sourceSize
+ toTexture:fromAPI(destination.texture).texture()
+ destinationSlice:0
+ destinationLevel:destination.mipLevel
+ destinationOrigin:destinationOrigin
+ options:options];
+ break;
+ }
+ case WGPUTextureDimension_Force32:
+ return;
+ }
+}
+
void Queue::setLabel(String&& label)
{
m_commandQueue.label = label;
