Using Indirect Command Buffers seems to break XCode's ability to debug Metal shaders. Attempting to debug a vertex shader either yields errors like Unable to connect to device (6) or Function argument 'vertexBuffer.0' does not have a valid vertex buffer binding at index '0'. I've tried the XCode versions 13.1 (13A1030d) and 13.2 beta 2 (13C5081f). Is this a known problem/limitation of XCode's Metal Shader debugger? Are there any workarounds for this? Are there better ways to "debug" Metal shaders in general? I thought my code was doing something bad/weird (I'm still new to Metal), but found even the Apple Developer sample code Encoding Indirect Command Buffers on the CPU, XCode cannot debug the shaders. I've reduced a reproduction to a simple, single file Swift MacOS application drawing a triangle (see Github Repo link below). Setting up Render Pipeline: let desc = MTLRenderPipelineDescriptor() desc.label = "RenderPipeline" desc.vertexFunction = vertexFn desc.fragmentFunction = fragFn desc.colorAttachments[0].pixelFormat = COLOR_PIXEL_FORMAT desc.supportIndirectCommandBuffers = true desc.inputPrimitiveTopology = .triangle let vertexDesc = MTLVertexDescriptor() vertexDesc.attributes[0].bufferIndex = 0 vertexDesc.attributes[0].offset = 0 vertexDesc.attributes[0].format = .float4 vertexDesc.layouts[0].stride = MemoryLayout<VertexPosition>.stride vertexDesc.layouts[0].stepRate = 1 vertexDesc.layouts[0].stepFunction = .perVertex desc.vertexDescriptor = vertexDesc Setting up Indirect Command Buffer: let desc = MTLIndirectCommandBufferDescriptor() desc.commandTypes = .draw desc.inheritBuffers = false desc.inheritPipelineState = false desc.maxVertexBufferBindCount = 1 desc.maxFragmentBufferBindCount = 0 let buf = device.makeIndirectCommandBuffer(descriptor: desc, maxCommandCount: 1, options: MTLResourceOptions.storageModeManaged)! buf.label = "IndirectCommandBuffer" let renderEncoder = buf.indirectRenderCommandAt(0) renderEncoder.setRenderPipelineState(pipelineState) renderEncoder.setVertexBuffer(vertexFnArgBuffer, offset: 0, at: 0) renderEncoder.drawPrimitives(.triangle, vertexStart: 0, vertexCount: 3, instanceCount: 1, baseInstance: 0) Issue command: let commandBuffer = commandQueue.makeCommandBuffer()! commandBuffer.label = "@CommandBuffer" let renderEncoder = commandBuffer.makeRenderCommandEncoder(descriptor: view.currentRenderPassDescriptor!)! renderEncoder.label = "@RenderCommandEncoder" renderEncoder.setViewport(drawableViewport) renderEncoder.use(vertexFnArgBuffer, usage: .read, stages: .vertex) renderEncoder.executeCommandsInBuffer(indirectCommandBuffer, range: 0..<1) renderEncoder.endEncoding() commandBuffer.present(view.currentDrawable!) commandBuffer.commit() The XCode project is pushed to this Github Repo peterwmwong/MetalGPUFrameCaptureDebugRepro and includes a GIF and MOV screen capture of the specific errors I encountered attempting to debug the simple shader. Any help would be appreciated! Thanks.

Can't seem to get the Metal HUD to display value range's (pre 26 Tahoe). The documented environment variable MTL_HUD_SHOW_VALUE_RANGE doesn't seem to work. https://developer.apple.com/documentation/xcode/monitoring-your-metal-apps-graphics-performance#Display-the-value-range-of-metrics Anyone having any luck?

I'm trying to use an Argument Buffer with a Buffer pointer and noticed crashes XCode 13.2 (19585) while debugging (Frame Capture) on a MBP M1 Max (macOS 12.2.1): Crashed Thread: 24 Dispatch queue: gputools.GPUMTLVariablesViewContentProvider.0x2d0638a40 Exception Type: EXC_BAD_ACCESS (SIGSEGV) Exception Codes: KERN_INVALID_ADDRESS at 0x0000000000000a10 Exception Codes: 0x0000000000000001, 0x0000000000000a10 Exception Note: EXC_CORPSE_NOTIFY I've created a small reproduction project (with screen recordings): https://github.com/peterwmwong/ArgumentBufferXCodeCrashRepro It's essentially a stripped down version of Apple Developer sample code: https://developer.apple.com/documentation/metal/buffers/managing_groups_of_resources_with_argument_buffers?language=objc Basically, this is the setup of the Argument Buffer (SceneArgumentBuffer) and Buffer pointer (Rectangle *) it references: typedef struct Rectangle { vector_float4 color; vector_float2 size; } Rectangle; struct SceneArgumentBuffer { device Rectangle *rects [[ id(SceneArgumentBufferIDRectangles) ]]; }; vertex VertexOut vertexShader( uint instanceID [[ instance_id ]], uint vertexID [[ vertex_id ]], constant SceneArgumentBuffer &args [[ buffer(VertexBufferIndexArgumentBuffer) ]]) { const Rectangle rect = args.rects[instanceID]; // ... } uint16_t numRects = 2; _rectanglesBuffer = [_device newBufferWithLength:sizeof(Rectangle) * numRects options:MTLResourceStorageModeShared]; _rectanglesBuffer.label = @"Rects Buffer"; Rectangle * const rects = _rectanglesBuffer.contents; rects[0].color = simd_make_float4(0.0, 1.0, 0.0, 1.0); rects[0].size = simd_make_float2(0.75, 0.75); rects[1].color = simd_make_float4(0.0, 0.0, 1.0, 1.0); rects[1].size = simd_make_float2(0.5, 0.25); // Argument buffer creation and encoding id<MTLArgumentEncoder> argumentEncoder = [vertexFunction newArgumentEncoderWithBufferIndex:VertexBufferIndexArgumentBuffer]; _argumentBuffer = [_device newBufferWithLength:argumentEncoder.encodedLength options:0]; _argumentBuffer.label = @"Argument Buffer"; [argumentEncoder setArgumentBuffer:_argumentBuffer offset:0]; [argumentEncoder setBuffer:_rectanglesBuffer offset:0 atIndex:SceneArgumentBufferIDRectangles]; // Encoding into a Render Command id<MTLRenderCommandEncoder> renderEncoder = [commandBuffer renderCommandEncoderWithDescriptor:renderPassDescriptor]; [renderEncoder useResource:_rectanglesBuffer usage:MTLResourceUsageRead stages:MTLRenderStageVertex]; [renderEncoder setVertexBuffer:_argumentBuffer offset:0 atIndex:VertexBufferIndexArgumentBuffer]; My application "works". Runs as expected, access to argument buffer and referenced buffer yields correct information to draw the screen correctly... it's just stinks (maintenance/development nightmare) that debugging is hosed. I can't debug shaders or look at pipeline statistics. Is this suppose to work? Are there any weird XCode/Metal environment variables or settings that I should be using to get this to work? Any help/suggestions would be appreciated, thanks!

Is there a limit to the number MTLIO load texture commands that can safely be executed in a single command buffer? When I tried loading each face (6) of a cube texture using a single command buffer, some of the faces will have bad data. Basically... let cmdBuf = ioCommandQueue.makeCommandBuffer() for slice in 0...5 { cmdBuf.load(cubeTexture, slice: slice, level: 0, size: size, sourceBytesPerRow:bytesPerRow, sourceBytesPerImage: bytesPerImage, destinationOrigin: destOrigin, sourceHandle: fileHandles[slice], sourceHandleOffset: 0) } cmdBuf.commit() cmdBuf.waitUntilCompleted() /* This passes, even though my texture is messed up...? */ assert(cmdBuf.status == .complete) // Also, no errors or warnings when XCode Diagnostics / Metal Validation is turned on. If I use multiple command buffers (one load texture command each), everything is fine... for slice in 0...5 { // IMPORTANT: Moved inside the LOOP let cmdBuf = ioCommandQueue.makeCommandBuffer() cmdBuf.load(cubeTexture, slice: slice, level: 0, size: size, sourceBytesPerRow:bytesPerRow, sourceBytesPerImage: bytesPerImage, destinationOrigin: destOrigin, sourceHandle: fileHandles[slice], sourceHandleOffset: 0) // IMPORTANT: Moved inside the LOOP cmdBuf.commit() cmdBuf.waitUntilCompleted() assert(cmdBuf.status == .complete) } I'm assuming this a bug (opened FB10582329)... Or at the very least, some documentation or Metal Validation log would be helpful to understanding what's going wrong/workaround. I've uploaded reduced reproduction (Single Swift file, command line application, XCode project) to this GitHub Repository: https://github.com/peterwmwong/x-mtlio-load-multiple-textures Thanks.

I tried with the latest Xcode 14 RC, but still am unable to extract a pipeline script from a harvested binary archive, as per WWDC 2022 session video. Here's what I've tried (Runnable XCode project on Github): // Create a simple render pipeline let device = MTLCreateSystemDefaultDevice()! let lib = device.makeDefaultLibrary()! let pipelineDesc = MTLRenderPipelineDescriptor() pipelineDesc.vertexFunction = lib.makeFunction(name: "main_vertex") pipelineDesc.fragmentFunction = lib.makeFunction(name: "main_fragment") pipelineDesc.colorAttachments[0]?.pixelFormat = .bgra8Unorm // Harvest GPU Archive let archivePath = NSTemporaryDirectory().appending("harvested-archive.metallib") let archiveDesc = MTLBinaryArchiveDescriptor() let archive = try device.makeBinaryArchive(descriptor: archiveDesc) try archive.addRenderPipelineFunctions(descriptor: pipelineDesc) try archive.serialize(to: NSURL.fileURL(withPath: archivePath)) // In a terminal... > xcrun metal-source -flatbuffers=json harvested-archive.metallib -o descriptors.mtlp-json Unfortunately, metal-source produces a directory of metallibs and no pipeline JSON script in sight. Did I miss something? Has the process changed since the WWDC 2022? Thanks.

Anyone else unable to download the "Rendering a Scene with Deferred Lighting in C++" (https://developer.apple.com/documentation/metal/rendering-a-scene-with-deferred-lighting-in-c++?language=objc)? I just an error page: Is there another place to download this sample?

With the Xcode 15 RC, the documentation for Metal Pipelines Script (man metal-pipelines-script) doesn't mention anything about defining a Mesh Render Pipeline (MTLMeshRenderPipelineDescriptor). Is there way to do offline compilation OR harvest binary archives for mesh shaders?