Address code quality concerns for WebGPU after hello-triangle prototype is ready. View in context: https://bugs.webkit.org/attachment.cgi?id=354034&action=review > Source/WebCore/Modules/webgpu/GPUPipelineDescriptorBase.h:37 > + Vector<GPUPipelineStageDescriptor> stages; Are you sure this is the right internal design? We need to differentiate between the vertex shader and the fragment shader when we compile the state. That shouldn't have to iterate through a list of things just to find which item is for the fragment shader. > Source/WebCore/Modules/webgpu/GPUPipelineStageDescriptor.h:37 > + const GPUShaderModule& module; What happens when the module gets garbage collected? Will this be dangling? > Source/WebCore/Modules/webgpu/GPURenderPipelineDescriptor.h:37 > +struct GPURenderPipelineDescriptor : GPUPipelineDescriptorBase { Why struct? > Source/WebCore/Modules/webgpu/GPURenderPipelineDescriptor.h:52 > + int primitiveTopology; This shouldn't be an int. > Source/WebCore/Modules/webgpu/WebGPUPipelineStageDescriptor.h:39 > + const WebGPUShaderModule* module = nullptr; Ditto about the garbage collector > Source/WebCore/Modules/webgpu/WebGPUShaderStage.h:36 > + enum { enum class > Source/WebCore/Modules/webgpu/WebGPUShaderStage.h:39 > + VERTEX = 0, > + FRAGMENT = 1, > + COMPUTE = 2 Can we use different internal names for these? Our style is not to use ALL CAPS. > Source/WebCore/Modules/webgpu/cocoa/GPURenderPipeline.h:50 > +#else > +using PlatformRenderPipeline = void; > +using PlatformRenderPipelineSmartPtr = RefPtr<void>; > +#endif I think it's fine to assume we're not using metal for now. We can break that assumption one day when the linux ports want WebGPU. > Source/WebCore/Modules/webgpu/cocoa/GPURenderPipelineMetal.mm:88 > + BEGIN_BLOCK_OBJC_EXCEPTIONS; We should be deliberate about where we use these. > Source/WebCore/bindings/js/WebCoreBuiltinNames.h:182 > + macro(WebGPURenderPipeline) \ > + macro(WebGPUShaderStage) \ Why aren't the other objects in here? > LayoutTests/webgpu/render-pipelines.html:5 > +<script id="library" type="x-shader/x-metal"> 😐
Note to self: A PipelineDescriptor retains ownership? of the shader modules since the original modules could very well fall out of scope before createRenderPipeline(pipelineDescriptor) is called, so fix the raw refs/ptrs.
Note to past self: The JS ordered dictionaries created in order to represent the PipelineDescriptor will maintain references to any assigned ShaderModules, so the C++ structs do not have to maintain strong references.
Created attachment 354839 [details] Patch
Comment on attachment 354839 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=354839&action=review > Source/WebCore/Modules/webgpu/WebGPUDevice.cpp:70 > + const char* const functionName = "WebGPUDevice::createRenderPipeline()"; > +#if LOG_DISABLED > + UNUSED_PARAM(functionName); > +#endif I think it's ok to duplicate the function name in each LOG statement. > Source/WebCore/Modules/webgpu/WebGPUDevice.cpp:82 > + LOG(WebGPU, "%s: Invalid WebGPUPipelineStageDescriptor!", functionName); Do we need the !!!!s ? :)
I like the !s in my errors messages! Are they bad? Well, pulling out the function name made it a little easier to type at least, especially when I was refactoring stuff to cover error cases.
Comment on attachment 354839 [details] Patch Attachment 354839 [details] did not pass mac-wk2-ews (mac-wk2): Output: https://webkit-queues.webkit.org/results/9995083 New failing tests: http/wpt/beacon/cors/cors-redirect-failure.html
Created attachment 354866 [details] Archive of layout-test-results from ews107 for mac-sierra-wk2 The attached test failures were seen while running run-webkit-tests on the mac-wk2-ews. Bot: ews107 Port: mac-sierra-wk2 Platform: Mac OS X 10.12.6
Comment on attachment 354839 [details] Patch Clearing flags on attachment: 354839 Committed r238208: <https://trac.webkit.org/changeset/238208>
All reviewed patches have been landed. Closing bug.
<rdar://problem/46080587>