Hi there - Where would a dev go these days to get an initial understanding of SceneKit? The WWDC videos linked in various places seem to be gone?! For example, the SceneKit page at developer.apple.com lists features a session videos link that comes up without any result, https://developer.apple.com/scenekit/ Any advice..? Cheers, Jan

Hello all We would like to use AMD's FidelityFx Downsampler in our custom game engine and we are having difficulties to correctly implement it for Metal due to its use of the globallycoherent keyword. We have done extensive search online but have not succeeded in finding an answer. What we have found is the largely undocumented 'volatile' keyword, so we were hypothesising that marking a texture with 'volatile' (which implies 'device volatile' since it's a texture) could have the same effect but we are far from convinced it would work. Does anyone have insights into this?

I have a fbo generated by last frame, which contain a color texture and a depth texture. Then i want blit that fbo to current MTKView, but failed. Thanks a lot for any suggestions. Here is the code: last frame: id <MTLTexture> src_color; id <MTLTexture> src_depth; drawTo(src_color, src_depth); current frame: first, init a depth texture if null: id<MTLTexture> depth_texture_; if(depth_texture_) { depth_texture_ = [device_ newTextureWithDescriptor:desc]; } second， create desc for encoder: current_desc = metal_view_.currentRenderPassDescriptor; current_desc.depthAttachment.texture = depth_texture_; current_desc.depthAttachment.loadAction = MTLLoadActionClear; current_desc.depthAttachment.clearDepth = 1.0; current_desc.stencilAttachment.texture = depth_texture_; current_desc.stencilAttachment.loadAction = MTLLoadActionClear; current_desc.stencilAttachment.clearStencil = 0; current_desc.colorAttachments[0].loadAction = MTLLoadActionClear; current_desc.colorAttachments[0].clearColor = MTLClearColorMake(1.0, 1, 1, 1.0); third, blit to current MTKView: auto dst_color = current_desc.colorAttachments[0].texture; auto dst_depth = current_desc.depthAttachment.texture; id<MTLCommandBuffer> commandBuffer = [command_queue_ commandBuffer]; commandBuffer.label = @"blit"; id <MTLBlitCommandEncoder> blitEncoder = [commandBuffer blitCommandEncoder]; [blitEncoder copyFromTexture:src_color sourceSlice:0 sourceLevel:0 sourceOrigin:MTLOriginMake(srcRect.x, srcRect.y, 0) sourceSize:MTLSizeMake(srcRect.width, srcRect.height, 1) toTexture:dst_color destinationSlice:0 destinationLevel:0 destinationOrigin:MTLOriginMake(dstRect.x, dstRect.y, 0)]; [blitEncoder copyFromTexture:src_depth sourceSlice:0 sourceLevel:0 sourceOrigin:MTLOriginMake(srcRect.x, srcRect.y, 0) sourceSize:MTLSizeMake(srcRect.width, srcRect.height, 1) toTexture:dst_depth destinationSlice:0 destinationLevel:0 destinationOrigin:MTLOriginMake(dstRect.x, dstRect.y, 0)]; [blitEncoder endEncoding]; [commandBuffer commit]; [commandBuffer waitUntilCompleted];

I'm trying to follow the metal-cpp tutorials I've found at https://developer.apple.com/metal/sample-code/?q=learn The program seems to be launching correctly (I can see the menu bar and interact with it), but nothing is rendered inside the window. I suppose the culprit is somewhere in the following function (I see it binds the device, the view and the window with the object in charge of drawing stuff in the view) void core::Application::applicationDidFinishLaunching(NS::Notification *pNotification) { CGRect frame = (CGRect){{100.0, 100.0}, {512.0, 512.0}}; m_Window->init(frame, NS::WindowStyleMaskClosable | NS::WindowStyleMaskTitled, NS::BackingStoreBuffered, false); m_Device = MTL::CreateSystemDefaultDevice(); m_View = MTK::View::alloc()->init(frame, m_Device); m_View->setColorPixelFormat(MTL::PixelFormat::PixelFormatBGRA8Unorm); m_View->setClearColor(MTL::ClearColor::Make(1.0, 0.0, 0.0, 1.0)); m_ViewDelegate = new graphics::ViewDelegate(m_Device); m_View->setDelegate(m_ViewDelegate); m_Window->setContentView(m_View); m_Window->setTitle(NS::String::string("Template 1", NS::StringEncoding::UTF8StringEncoding)); m_Window->makeKeyAndOrderFront(nullptr); NS::Application* nsApp = reinterpret_cast<NS::Application*>(pNotification->object()); nsApp->activateIgnoringOtherApps(true); } but, as you can infer from the fact that I'm failing at the very first tutorial of the bunch, I'm quite lost. I've tried debugging the app with the Xcode debugger and I saw that it never enters in this function. void ViewDelegate::drawInMTKView(MTK::View *pView) { m_Renderer->Draw(pView); } Can it be a symptom of some call missing from my code? Thank you in advance for your help

Hi, I have a CUDA program that I want to convert to Metal Compute so that we can support Apple hardware. When I wrote the CUDA version, I was able to write efficient code because I learned first about the Cuda-core architecture. The way the cores can access memory for instance is very important information so that I could write code that efficiently access the memory. Now I want to do the same for the Metal Compute software. But I can not find any information about the low level architecture and especially the things you should know to be able to write efficient code. Do I miss something? Is there some guide giving hints for the most efficient way to access memory for instance?

I am working on an application where we are planning to use Metal for directly rendering custom content. When user looks at something on the rendered image, I want to get the position or ray of cursor (the point where the user is currently looking at) to render something else like a crosshair. Is it possible to get the cursor position information on VisionOS to accomplish this? How can I know if something is being hovered on by the eyes?

Platform 'METAL' is experimental and not all JAX functionality may be correctly supported! 2024-03-23 22:04:38.947506: W pjrt_plugin/src/mps_client.cc:563] WARNING: JAX Apple GPU support is experimental and not all JAX functionality is correctly supported! Metal device set to: Apple M1 Pro systemMemory: 16.00 GB maxCacheSize: 5.33 GB loc("-":0:0): error: current mps dialect version is 1.0.0, can't parse version 1.1.0 /AppleInternal/Library/BuildRoots/495c257e-668e-11ee-93ce-926038f30c31/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShadersGraph/mpsgraph/MetalPerformanceShadersGraph/Core/Files/MPSGraphExecutable.mm:1097: failed assertion `Error importing MLIR bytecode. ' zsh: abort python -c 'import jax; print(jax.numpy.arange(10))'

in Swift languange, CVMetalTextureCacheCreateTextureFromImage return CVMetalTexture, and CVMetalTexture is Swift class, so. it doesn't need to call CVBufferRelease manually. My question is : should I use a variable to keep strong reference until GPU finished (until addCompleteHandler callback ) ？ cvmetaltexturecachecreatetexture

Currently (jax-metal 0.0.5) doesn't support matrix decompositions like Cholesky, LU, or Eigen, on Metal. I have a 64GB M1 Max on Sonoma 14.2.1 (23C71). Eigen raises a NotImplemented error, but Cholesky and jax.linalg.inv error out. When using the CPU by setting export JAX_PLATFORM_NAME=CPU then the decompositions work. The error I am getting: XlaRuntimeError Traceback (most recent call last) Cell In[33], line 1 ----> 1 jsl.cho_factor(Sigma_0_inv) File ~/miniconda3/envs/jaxmetal/lib/python3.10/site-packages/jax/_src/scipy/linalg.py:61, in cho_factor(failed resolving arguments) 56 @_wraps(scipy.linalg.cho_factor, 57 lax_description=_no_overwrite_and_chkfinite_doc, skip_params=('overwrite_a', 'check_finite')) 58 def cho_factor(a: ArrayLike, lower: bool = False, overwrite_a: bool = False, 59 check_finite: bool = True) -> tuple[Array, bool]: 60 del overwrite_a, check_finite # Unused ---> 61 return (cholesky(a, lower=lower), lower) File ~/miniconda3/envs/jaxmetal/lib/python3.10/site-packages/jax/_src/scipy/linalg.py:54, in cholesky(failed resolving arguments) 49 @_wraps(scipy.linalg.cholesky, 50 lax_description=_no_overwrite_and_chkfinite_doc, skip_params=('overwrite_a', 'check_finite')) 51 def cholesky(a: ArrayLike, lower: bool = False, overwrite_a: bool = False, 52 check_finite: bool = True) -> Array: 53 del overwrite_a, check_finite # Unused ---> 54 return _cholesky(a, lower) File ~/miniconda3/envs/jaxmetal/lib/python3.10/site-packages/jax/_src/compiler.py:255, in backend_compile(backend, module, options, host_callbacks) 250 return backend.compile(built_c, compile_options=options, 251 host_callbacks=host_callbacks) 252 # Some backends don't have host_callbacks option yet 253 # TODO(sharadmv): remove this fallback when all backends allow compile 254 # to take in host_callbacks --> 255 return backend.compile(built_c, compile_options=options) XlaRuntimeError: UNKNOWN: /var/folders/9d/0035yr7j3bx84h3ghpp_86pc0000gn/T/ipykernel_40684/3046650730.py:1:0: error: failed to legalize operation 'mhlo.cholesky' /var/folders/9d/0035yr7j3bx84h3ghpp_86pc0000gn/T/ipykernel_40684/3046650730.py:1:0: note: see current operation: %5 = "mhlo.cholesky"(%4) {lower = true} : (tensor<200x200xf32>) -> tensor<200x200xf32>

I wanna draw a pixel buffer directly on the screen with the Metal API. in OpenGL I can use glDrawPixels how to do it in Metal?

I'm using DrawableQueue to create textures that I apply to my ShaderGraphMaterial texture. My metal render is using a range of alpha values as a test. My objects displayed with the DrawableQueue texture are working as expected, but the alpha component is not working. Is this an issue with my DrawableQueue descriptor? My ShaderGraphMaterial? A missing setting on my scene objects? or some limitation in visionOS? DrawableQueue descriptor let descriptor = await TextureResource.DrawableQueue.Descriptor( pixelFormat: .rgba8Unorm, width: textureResource!.width, height: textureResource!.height, usage: [.renderTarget, .shaderRead, .shaderWrite], // Usage should match the requirements for how the texture will be used //usage: [.renderTarget], // Usage should match the requirements for how the texture will be used mipmapsMode: .none // Assuming no mipmaps are needed for the text texture ) let queue = try await TextureResource.DrawableQueue(descriptor) queue.allowsNextDrawableTimeout = true await textureResource!.replace(withDrawables: queue) Draw frame: guard let drawable = try? drawableQueue!.nextDrawable(), let commandBuffer = commandQueue?.makeCommandBuffer()//, //let renderPipelineState = renderPipelineState else { return } let renderPassDescriptor = MTLRenderPassDescriptor() renderPassDescriptor.colorAttachments[0].texture = drawable.texture renderPassDescriptor.colorAttachments[0].loadAction = .clear renderPassDescriptor.colorAttachments[0].storeAction = .store renderPassDescriptor.colorAttachments[0].clearColor = clearColor /*renderPassDescriptor.colorAttachments[0].clearColor = MTLClearColor( red: clearColor.red, green: clearColor.green, blue: clearColor.blue, alpha: 0.5 )*/ renderPassDescriptor.renderTargetHeight = drawable.texture.height renderPassDescriptor.renderTargetWidth = drawable.texture.width guard let renderEncoder = commandBuffer.makeRenderCommandEncoder(descriptor: renderPassDescriptor) else { return } renderEncoder.pushDebugGroup("DrawNextFrameWithColor") //renderEncoder.setRenderPipelineState(renderPipelineState) // No need to create a render command encoder with shaders, as we are only clearing the drawable. // Since we are just clearing the drawable to a solid color, no need to draw primitives renderEncoder.endEncoding() commandBuffer.commit() commandBuffer.waitUntilCompleted() drawable.present() }

On macOS 14.2.1 Sonoma, when our app is displayed in full screen, there is a problem where the operations at hand are not reflected at all on the connection destination. This issue does not occur on macOS 13 Ventura. If you stop displaying full screen and reduce the window size, you will be able to operate it to some extent. Were there any specifications or technical changes around the screen between Ventura and Sonoma? Mac at hand* - Relay server - Windows to connect to *This does not occur on windows, android, or ios.

With the advent of the third dimension, I wanted to know wether if it's currently possible to display the flat swiftUI Views with some thickness in xrOS? While the .frame(depth: CGFloat?) does the job for Views in general, I am eager for a more granular level of control at the pixel-specific level. I was hoping that there are lower level APIs to achieve this & I've looked into the fairly new layerEffect shader API, yet it seems it's incapable of setting the depths of pixels...

I'm trying to debug my metal shaders in Xcode 14.2. However clicking "Capture metal GPU" while debugging recently started showing the following error: Capturing MTLPipelineLibrary is not supported. Unsupported method: -[MTLDevice newPipelineLibraryWithFilePath:error:] To enable capturing, disable calls to unsupported APIs and relaunch your application. I can't find any info about MTLPipelineLibrary or how to disable it. I've also confirmed that Metal GPU Frame Capture is enabled in my build What's causing this issue and how can I work around it so I can debug my shaders again?

Hi, I'm displaying linear gray by CAMetalLayer with the shader below. fragment float4 fragmentShader(VertexOut in [[stage_in]], texture2d<float, access::sample> BGRATexture [[ texture(0) ]]) { float color = in.texCoordinates.x; return float4(float3(color), 1.0); } And my CAMetalLayer has been set to linearSRGB. metalLayer.colorspace = CGColorSpace(name: CGColorSpace.linearSRGB) metalLayer.pixelFormat = .bgra8Unorm Why the display seem add gamma? Apparently the middle gray is 187 but not 128.

Hello, I have been following the excellent/informative "Metal for Machine Learning" from WWDC19 to learn how to do on device training (I have a specific use case for this) and it is all working really well using the MPSNNGraph. However, I would like to call my own metal compute/render function/pipeline to transform the inference result before calculating the loss, does anyone know if this possible and what would this look like in code? Please see my current code below, at the comment I need to call an intermediate compute/render function to transform the inference result image before passing to the MPSNNForwardLossNode. let rgbImageNode = MPSNNImageNode(handle: nil) let inferGraph = makeInferenceGraph() let reshape = MPSNNReshapeNode(source: inferGraph.resultImage, resultWidth: 64, resultHeight: 64, resultFeatureChannels: 4) //Need to call render or compute pipeline to post process in the inference result image let rgbLoss = MPSNNForwardLossNode(source:reshape.resultImage, labels:rgbImageNode, lossDescriptor:lossDescriptor) let initGrad = MPSNNInitialGradientNode(source:rgbLoss.resultImage) let gradNodes = initGrad.trainingGraph(withSourceGradient:nil, nodeHandler:nil) guard let trainGraph = MPSNNGraph(device: device, resultImage: gradNodes![0].resultImage, resultImageIsNeeded: true) else{ fatalError("Unable to get training graph.") } Thanks

Hi, I am looking for a routine to perform complex-valued linear algebra on the GPU in python for scientific programming, in particular quantum physics simulations. At the moment I am looking for a routine for complex-valued matrix multiplication. I found MLX has a routine for float matrix multiplication, but it does not directly work for complex-valued matrices. I figured a work-around by splitting the complex valued matrix into real and imaginary part and working with the pair, but it makes it cumbersome to integrate with the remainder of the code. I was hoping for a library-based implementation similar to cupy. I also tried out using the tensorflow linear algebra routines, but I couldn't get them to run on the GPU by now. Specifically, a testfile with a tensorflow.keras.applications.ResNet50 routine runs on the GPU, but the routines from tensorflow.linalg and tensorflow.math that I tested (matmul, expm, eigh) were not running on the GPU. Any advice on how to make linear algebra calculations on mac GPUs work is highly appreciated! For my application the unified memory might be especially beneficial. Thank you!