I'm trying to build an MDLMesh then add normals let mdlMesh = MDLMesh.newBox(withDimensions: SIMD3<Float>(1, 1, 1), segments: SIMD3<UInt32>(2, 2, 2), geometryType: MDLGeometryType.triangles, inwardNormals:false, allocator: allocator) mdlMesh.addNormals(withAttributeNamed: MDLVertexAttributeNormal, creaseThreshold: 0) When I render the mesh, some normals are (0,0,0). I don't know if the problem is in the mesh, or in the conversion to MTKMesh. Is there a way to examine an MDLMesh with the geometry viewer? When I look at the variable values for my mdlMesh I get this: Not too useful. I don't know how to track down the normals. What's the best way to find out where the normals getting broken?

I am building a MacOS desktop app (https://anukari.com) that is using Metal compute to do real-time audio/DSP processing, as I have a problem that is highly parallelizable and too computationally expensive for the CPU. However it seems that the way in which I am using the GPU, even when my app is fully compute-limited, the OS never increases the power/performance state. Because this is a real-time audio synthesis application, it's a huge problem to not be able to take advantage of the full clock speeds that the GPU is capable of, because the app can't keep up with real-time. I discovered this issue while profiling the app using Instrument's Metal tracing (and Game tracing) modes. In the profiling configuration under "Metal Application" there is a drop-down to select the "Performance State." If I run the application under Instruments with Performance State set to Maximum, it runs amazingly well, and all my problems go away. For comparison, when I run the app on its own, outside of Instruments, the expensive GPU computation it's doing takes around 2x as long to complete, meaning that the app performs half as well. I've done a ton of work to micro-optimize my Metal compute code, based on every scrap of information from the WWDC videos, etc. A problem I'm running into is that I think that the more efficient I make my code, the less it signals to the OS that I want high GPU clock speeds! I think part of why the OS is confused is that in most use cases, my computation can be done using only a small number of Metal threadgroups. I'm guessing that the OS heuristics see that only a small fraction of the GPU is saturated and fail to scale up the power/clock state. I'm not sure what to do here; I'm in a bit of a bind. One possibility is that I intentionally schedule busy work -- spin threadgroups just to waste energy and signal to the OS that I need higher clock speeds. This is obviously a really bad idea, but it might work. Is there any other (better) way for my app to signal to the OS that it is doing real-time latency-sensitive computation on the GPU and needs the clock speeds to be scaled up? Note that game mode is not really an option, as my app also runs as an AU plugin inside hosts like Garageband, so it can't be made fullscreen, etc.

I made a box with MDLMesh.newBox(). I added normals. let mdlMesh = MDLMesh.newBox(withDimensions: SIMD3<Float>(1, 1, 1), segments: SIMD3<UInt32>(2, 2, 2), geometryType: MDLGeometryType.triangles, inwardNormals:false, allocator: allocator) mdlMesh.addNormals(withAttributeNamed: MDLVertexAttributeNormal, creaseThreshold: 0.25) After I convert to MTKMesh the normals are (0,0,0) for a group of vertices. I can only inspect the geometry after I convert to MTKMesh. Is there a way you can use Geometry Viewer on a MDLMesh?

hi When analyzing our game using Instruments, I've always been confused about the two items "Drawable Present" and "Drawable Presented" in the GPU column. The timing of Drawable Present seems to be when the CPU layer calls commandbuffer:present, rather than when the actual encoding is completed on the GPU. Also, what does drawable presented specifically mean? In our case, when a CPU stall occurs, it appears that the vsync interval changes in the next frame, and a surface that has already been calculated is not displayed. Why is this happening?

I am new to Xcode and trying to learn how to use Metal for my internship. I am trying to link the binaries of Foundation.framework, Metal.framework, and Quartcore.framework. But whenever I try to build it always fails to find any of them. I have my Header Search Path as $(PROJECT_DIR)/metal-cpp, I tried adding some for the Frameworks but that did not work either. I do have the binaries linked in the Build Phases, so I don't know what else I could be missing.

In the Creating A 3D Application With Hydra Rendering tutorial on the Apple Developer website, on the last step where I execute this command: cmake -S ~/Users/macuser/CreatingA3DApplicationWithHydraRendering/ -B ~/Users/macuser/CreatingA3DApplicationWithHydraRendering/ I keep getting an error: CMake Error at CMakeLists.txt:5 (include): include could not find requested file: /Users/macuser/USDInstall/bin/pxrConfig.cmake I've tried to follow the instructions as mentioned in the README.md file included in the project files at least 5 times as well as moving the pxrConfig.cmake file around and copying it in different folders, then executed the command and was still unsuccessful into generating the proper file expected to compile and render the HydraPlayer renderer. How do I get cmake to generate the Xcode file to create the HydraPlayer renderer?

I’m trying to follow Apple’s “WWDC24: Bring your machine learning and AI models to Apple Silicon” session to convert the Mistral-7B-Instruct-v0.2 model into a Core ML package, but I’ve run into a roadblock that I can’t seem to overcome. I’ve uploaded my full conversion script here for reference: https://pastebin.com/T7Zchzfc When I run the script, it progresses through tracing and MIL conversion but then fails at the backend_mlprogram stage with this error: https://pastebin.com/fUdEzzKM The core of the error is: ValueError: Op "keyCache_tmp" (op_type: identity) Input x="keyCache" expects list, tensor, or scalar but got state[tensor[1,32,8,2048,128,fp16]] I’ve registered my KV-cache buffers in a StatefulMistralWrapper subclass of nn.Module, matching the keyCache and valueCache state names in my ct.StateType definitions, but Core ML’s backend pass reports the state tensor as an invalid input. I’m using Core ML Tools 8.3.0 on Python 3.9.6, targeting iOS18, and forcing CPU conversion (MPS wasn’t available). Any pointers on how to satisfy the handle_unused_inputs pass or properly declare/cache state for GQA models in Core ML would be greatly appreciated! Thanks in advance for your help, Usman Khan

Hi, seems MSL is missing support for a clock() shader instruction available in other graphics APIs like Vulkan or OpenGL for example.. useful for counting cost in number of clock cycles of some code insider shader with much finer granularity than launching a micro kernel with same instructions and measuring cycles cost from CPU.. also useful for MoltenVK to support that extensions.. thanks..

Hi, Introducing Swift Concurrency to my Metal app has been a bit challenging as Swift Concurrency is limited by the cooperative thread pool. GPU work is obviously not CPU bound and can block forward moving progress, especially when using waitUntilCompleted on the command buffer. For concurrent render work this has the potential of under utilizing the CPU and even creating dead locks. My question is, what is the Metal's teams general recommendation when it comes to concurrency? It seems to me that Dispatch or OperationQueues are still the preferred way for Metal bound tasks in order to gain maximum performance? To integrate with Swift Concurrency my idea is to use continuations that kick off render jobs via Dispatch or Queues? Would this be the best solution to bridge async tasks with Metal work? Thanks!

Starting with iOS 18.0 beta 1, I've noticed that RealityKit frequently crashes in the simulator when an app launches and presents an ARView. I was able to create a small sample app with repro steps that demonstrates the issue, and I've submitted feedback: FB16144085 I've included a crash log with the feedback. If possible, I'd appreciate it if an Apple engineer could investigate and suggest a workaround. It's awkward to be restricted to the iOS 17 simulator, which does not exhibit this behavior. Please let me know if there's anything I can do to help. Thank you.

I am trying to achieve an animated gradient effect that changes values over time based on the current seconds. I am also using AVPlayer and AVMutableVideoComposition along with custom instruction and class to generate the effect. I didn't want to load any video file, but rather generate a custom video with my own set of instructions. I used Metal Compute shaders to generate the effects and make the video to be 20 seconds. However, when I run the code, I get a frozen player with the gradient applied, but when I try to play the video, I get this warning in the console :- Visual isTranslatable: NO; reason: observation failure: noObservations Here is the screenshot :- My entire code :- import AVFoundation import Metal class GradientVideoCompositorTest: NSObject, AVVideoCompositing { var sourcePixelBufferAttributes: [String: Any]? = [ kCVPixelBufferPixelFormatTypeKey as String: kCVPixelFormatType_32BGRA ] var requiredPixelBufferAttributesForRenderContext: [String: Any] = [ kCVPixelBufferPixelFormatTypeKey as String: kCVPixelFormatType_32BGRA ] private var renderContext: AVVideoCompositionRenderContext? private var metalDevice: MTLDevice! private var metalCommandQueue: MTLCommandQueue! private var metalLibrary: MTLLibrary! private var metalPipeline: MTLComputePipelineState! override init() { super.init() setupMetal() } func setupMetal() { guard let device = MTLCreateSystemDefaultDevice(), let queue = device.makeCommandQueue(), let library = try? device.makeDefaultLibrary(), let function = library.makeFunction(name: "gradientShader") else { fatalError("Metal setup failed") } self.metalDevice = device self.metalCommandQueue = queue self.metalLibrary = library self.metalPipeline = try? device.makeComputePipelineState(function: function) } func renderContextChanged(_ newRenderContext: AVVideoCompositionRenderContext) { renderContext = newRenderContext } func startRequest(_ request: AVAsynchronousVideoCompositionRequest) { guard let outputPixelBuffer = renderContext?.newPixelBuffer(), let metalTexture = createMetalTexture(from: outputPixelBuffer) else { request.finish(with: NSError(domain: "com.example.gradient", code: -1, userInfo: nil)) return } var time = Float(request.compositionTime.seconds) renderGradient(to: metalTexture, time: time) request.finish(withComposedVideoFrame: outputPixelBuffer) } private func createMetalTexture(from pixelBuffer: CVPixelBuffer) -> MTLTexture? { var texture: MTLTexture? let width = CVPixelBufferGetWidth(pixelBuffer) let height = CVPixelBufferGetHeight(pixelBuffer) let textureDescriptor = MTLTextureDescriptor.texture2DDescriptor( pixelFormat: .bgra8Unorm, width: width, height: height, mipmapped: false ) textureDescriptor.usage = [.shaderWrite, .shaderRead] CVPixelBufferLockBaseAddress(pixelBuffer, .readOnly) if let textureCache = createTextureCache(), let cvTexture = createCVMetalTexture(from: pixelBuffer, cache: textureCache) { texture = CVMetalTextureGetTexture(cvTexture) } CVPixelBufferUnlockBaseAddress(pixelBuffer, .readOnly) return texture } private func renderGradient(to texture: MTLTexture, time: Float) { guard let commandBuffer = metalCommandQueue.makeCommandBuffer(), let commandEncoder = commandBuffer.makeComputeCommandEncoder() else { return } commandEncoder.setComputePipelineState(metalPipeline) commandEncoder.setTexture(texture, index: 0) var mutableTime = time commandEncoder.setBytes(&mutableTime, length: MemoryLayout<Float>.size, index: 0) let threadsPerGroup = MTLSize(width: 16, height: 16, depth: 1) let threadGroups = MTLSize( width: (texture.width + 15) / 16, height: (texture.height + 15) / 16, depth: 1 ) commandEncoder.dispatchThreadgroups(threadGroups, threadsPerThreadgroup: threadsPerGroup) commandEncoder.endEncoding() commandBuffer.commit() } private func createTextureCache() -> CVMetalTextureCache? { var cache: CVMetalTextureCache? CVMetalTextureCacheCreate(kCFAllocatorDefault, nil, metalDevice, nil, &cache) return cache } private func createCVMetalTexture(from pixelBuffer: CVPixelBuffer, cache: CVMetalTextureCache) -> CVMetalTexture? { var cvTexture: CVMetalTexture? let width = CVPixelBufferGetWidth(pixelBuffer) let height = CVPixelBufferGetHeight(pixelBuffer) CVMetalTextureCacheCreateTextureFromImage( kCFAllocatorDefault, cache, pixelBuffer, nil, .bgra8Unorm, width, height, 0, &cvTexture ) return cvTexture } } class GradientCompositionInstructionTest: NSObject, AVVideoCompositionInstructionProtocol { var timeRange: CMTimeRange var enablePostProcessing: Bool = true var containsTweening: Bool = true var requiredSourceTrackIDs: [NSValue]? = nil var passthroughTrackID: CMPersistentTrackID = kCMPersistentTrackID_Invalid init(timeRange: CMTimeRange) { self.timeRange = timeRange } } func createGradientVideoComposition(duration: CMTime, size: CGSize) -> AVMutableVideoComposition { let composition = AVMutableComposition() let instruction = GradientCompositionInstructionTest(timeRange: CMTimeRange(start: .zero, duration: duration)) let videoComposition = AVMutableVideoComposition() videoComposition.customVideoCompositorClass = GradientVideoCompositorTest.self videoComposition.renderSize = size videoComposition.frameDuration = CMTime(value: 1, timescale: 30) // 30 FPS videoComposition.instructions = [instruction] return videoComposition } #include <metal_stdlib> using namespace metal; kernel void gradientShader(texture2d<float, access::write> output [[texture(0)]], constant float &time [[buffer(0)]], uint2 id [[thread_position_in_grid]]) { float2 uv = float2(id) / float2(output.get_width(), output.get_height()); // Animated colors based on time float3 color1 = float3(sin(time) * 0.8 + 0.1, 0.6, 1.0); float3 color2 = float3(0.12, 0.99, cos(time) * 0.9 + 0.3); // Linear interpolation for gradient float3 gradientColor = mix(color1, color2, uv.y); output.write(float4(gradientColor, 1.0), id); }

Hi Apple, In VisionOS, for real-time streaming of large 3D scenes, I plan to create Metal buffers and textures in multiple threads and then use a compute shader on the main thread to copy the Metal resources into RealityKit, minimizing main thread usage. Given that most of RealityKit's default APIs require execution on the main actor (main thread), it is not ideal for streaming data. Is this approach the best way to handle streaming data and real-time rendering? Thank you very much.

Hello I am trying to get thread group memory access in fragment shader. In essence, I would like to have all the fragments in a tile to bitwiseOR some value. My idea was to use simd_or across the SIMD group, then make each SIMD group thread 0 to atomic or the value into thread group memory. Finally very first thread of the tile would be tasked with writing the value down to texture with write access. Now, I can allocate the thread group memory argument to the fragment function all right. MTLRenderEncoder has setThreadgroupMemoryLength call, which I am using the following way [renderEncoder setThreagroupMemoryLength: 16 offset: 0 atIndex:0] Unfortunately, all I am getting is the following error (runtime assertion) -[MTLDebugRenderCommandEncoder setThreadgroupMemoryLength:offset:atIndex:]:3487: failed assertion Set Threadgroup Memory Length Validation offset + length(16) must be <= threadgroupMemoryLength(0).` What I am doing wrong? How I can get thread group memory in the fragment shader? I know I could use tile shading and compute function but the problem is that here I really like to use fragment stuff. Will be grateful for help.

Hi, there's this point at which a beginner needs to beg for help. Unable to open mach-O at path: /Library/Caches/com.apple.xbs/Binaries/RenderBox/install/Root/System/Library/PrivateFrameworks/RenderBox.framework/default.metallib Error:2 I get this everytime I select a month and year on a custom date picker, I believe because I try to force the ".generateChartData()" for the chart to update. I guess the problem might be that the ".onAppear" and ".onChange" are conflicting with each other? } .onChange(of: showDatePicker) { viewModel.startDate = selectedDate viewModel.generateChartData() } } .onAppear { viewModel.generateChartData() }

I am working on a project for macOS where I am taking an AVCaptureSession's CVPixelBuffer and I need to convert it into a MTLTexture for rendering. On macOS the pixel format is 2vuy, there does not seem to be a clear format conversion while converting to a metal texture. I have been able to convert it to a texture but the color space seems to be off as it is rendering distorted colors with a double image. I believe 2vuy is a single pane color space and I have tried to account for that, but I am unaware of what is off. I have attached The CVPixelBuffer and The distorted MTLTexture along with a laundry list of errors. On iOS my conversions are fine, it is only the macOS 2vuy pixel format that seems to have issues. My code for the conversion is also attached. If there are any suggestions or guidance on how to properly convert a 2vuy CVPixelBuffer to a MTLTexture I would greatly appreciate it. Many Thanks Conversion_Logs.txt ConversionCode.swift

I’m building a professional camera app where users can customize the video recording format and color grading. In the func captureOutput(_ output: AVCaptureOutput, didOutput sampleBuffer: CMSampleBuffer, from connection: AVCaptureConnection) method, I handle video frames and use Metal for real-time color grading. This works well when device.activeColorSpace is sRGB or P3, and the results are great. However, when the color space is HLG_BT2020 or appleLog, the MTKTextureLoader.newTexture(cgImage: cgImage, options: options) method throws an error. After researching, I found that the video frame in these color spaces has a bit-per-channel (bpc) greater than 8 after being converted to CGImage, causing the texture creation to fail. I tried converting the CGImage to a lower bpc to successfully create the texture, but the final output image is garbled and not as expected. Is there a solution to this issue?

Currently I am using mixed style immersive view to place both my WindowView(plain style) and ImmersiveView content together. The issue is that the rendering depth testing may always let the virtual content block my normal WindowView. Is it possible to manually set windowedVIew always displays in the front of my virtual view in mixed style immersion? (I know modelSortGroup but it doesn't quite fits here) Or if I can dynamically change the .progressive value when the immersive space is open (set the value to zero means .mixed itself right?)

Hi, I am working with a large project. We are compiling each material to its own .metallib. They all include many common files full of inline functions. Finally we link it all together at the end with a single big pathtrace kernel. Everything works as expected, however the compile times have gotten completely out of hand and it takes multiple minutes to compile at runtime (to native code). I have gathered that I can do this offline by using metal-tt however if I am wondering if there is a way to reduce the compile times in such a scenario, and how to investigate what the root cause of the problem is. I suspect it could have to do with the fact that every materials metallib contains duplications of all the inline functions. Any ideas on how to profile and debug this? Thanks, Rasmus

Hello. In the iOS app i'm working on we are very tight on memory budget and I was looking at ways to reduce our texture memory usage. However I noticed that comparing ASTC8x8 to ASTC12x12, there is no actual difference in allocated memory for most of our textures despite ASTC12x12 having less than half the bpp of 8x8. The difference between the two only becomes apparent for textures 1024x1024 and larger, and even in that case the actual texture data is sometimes only 60% of the allocation size. I understand there must be some alignment and padding going on, but this seems extreme. For an example scene in my app with astc12x12 for most textures there is over a 100mb difference in astc size on disk versus when loaded, so I would love to be able to recover even a portion of that memory. Here is some test code with some measurements i've taken using an iphone 11: for(int i = 0; i &lt; 11; i++) { MTLTextureDescriptor *texDesc = [[MTLTextureDescriptor alloc] init]; texDesc.pixelFormat = MTLPixelFormatASTC_12x12_LDR; int dim = 12; int n = 2 &lt;&lt; i; int mips = i+1; texDesc.width = n; texDesc.height = n; texDesc.mipmapLevelCount = mips; texDesc.resourceOptions = MTLResourceStorageModeShared; texDesc.usage = MTLTextureUsageShaderRead; // Calculate the equivalent astc texture size int blocks = 0; if(mips == 1) { blocks = n/dim + (n%dim&gt;0? 1 : 0); blocks *= blocks; } else { for(int j = 0; j &lt; mips; j++) { int a = 2 &lt;&lt; j; int cur = a/dim + (a%dim&gt;0? 1 : 0); blocks += cur*cur; } } auto tex = [objCObj newTextureWithDescriptor:texDesc]; printf("%dx%d, mips %d, Astc: %d, Metal: %d\n", n, n, mips, blocks*16, (int)tex.allocatedSize); } MTLPixelFormatASTC_12x12_LDR 128x128, mips 7, Astc: 2768, Metal: 6016 256x256, mips 8, Astc: 10512, Metal: 32768 512x512, mips 9, Astc: 40096, Metal: 98304 1024x1024, mips 10, Astc: 158432, Metal: 262144 128x128, mips 1, Astc: 1936, Metal: 4096 256x256, mips 1, Astc: 7744, Metal: 16384 512x512, mips 1, Astc: 29584, Metal: 65536 1024x1024, mips 1, Astc: 118336, Metal: 147456 MTLPixelFormatASTC_8x8_LDR 128x128, mips 7, Astc: 5488, Metal: 6016 256x256, mips 8, Astc: 21872, Metal: 32768 512x512, mips 9, Astc: 87408, Metal: 98304 1024x1024, mips 10, Astc: 349552, Metal: 360448 128x128, mips 1, Astc: 4096, Metal: 4096 256x256, mips 1, Astc: 16384, Metal: 16384 512x512, mips 1, Astc: 65536, Metal: 65536 1024x1024, mips 1, Astc: 262144, Metal: 262144 I also tried using MTLHeaps (placement and automatic) hoping they might be better, but saw nearly the same numbers. Is there any way to have metal allocate these textures in a more compact way to save on memory?

I am working on a project for macOS where I am taking an AVCaptureSession's CVPixelBuffer and I need to convert it into a MTLTexture for rendering. On macOS the pixel format is 2vuy, there does not seem to be a clear format conversion while converting to a metal texture. I have been able to convert it to a texture but the color space seems to be off as it is rendering distorted colors with a double image. I believe 2vuy is a single pane color space and I have tried to account for that, but I am unaware of what is off. I have attached The CVPixelBuffer and The distorted MTLTexture along with a laundry list of errors. On iOS my conversions are fine, it is only the macOS 2vuy pixel format that seems to have issues. My code for the conversion is also attached. If there are any suggestions or guidance on how to properly convert a 2vuy CVPixelBuffer to a MTLTexture I would greatly appreciate it. Many Thanks Conversion_Logs.txt ConversionCode.swift