Fix: Swift 6 concurrency - async render + Sendable wrapper
Some checks are pending
Auto Tag on Push / auto-tag (push) Waiting to run

This commit is contained in:
ewen 2026-01-15 23:43:56 +01:00
parent 586f87e222
commit 38868a2aba
2 changed files with 148 additions and 113 deletions

View file

@ -1,4 +1,4 @@
import Metal
@preconcurrency import Metal
import MetalKit
import CoreGraphics
@ -30,7 +30,8 @@ struct RenderParameters {
var randomSeed: UInt32
}
final class MetalImageRenderer: Sendable {
@MainActor
final class MetalImageRenderer {
private let device: MTLDevice
private let commandQueue: MTLCommandQueue
private let pipelineState: MTLComputePipelineState
@ -65,120 +66,147 @@ final class MetalImageRenderer: Sendable {
}
}
func render(input: CGImage, params: RenderParameters) -> CGImage? {
return autoreleasepool {
print("🎨 Metal render started - Image: \(input.width)x\(input.height), Algo: \(params.algorithm)")
func render(input: CGImage, params: RenderParameters) async -> CGImage? {
return await withCheckedContinuation { continuation in
autoreleasepool {
print("🎨 Metal render started - Image: \(input.width)x\(input.height), Algo: \(params.algorithm)")
let textureLoader = MTKTextureLoader(device: device)
let textureLoader = MTKTextureLoader(device: device)
// Load input texture
guard let inputTexture = try? textureLoader.newTexture(cgImage: input, options: [.origin: MTKTextureLoader.Origin.topLeft]) else {
print("❌ Failed to create input texture")
return nil
}
print("✅ Input texture created: \(inputTexture.width)x\(inputTexture.height)")
// Create output texture
let descriptor = MTLTextureDescriptor.texture2DDescriptor(pixelFormat: .rgba8Unorm,
width: inputTexture.width,
height: inputTexture.height,
mipmapped: false)
descriptor.usage = [.shaderWrite, .shaderRead]
guard let outputTexture = device.makeTexture(descriptor: descriptor) else {
print("❌ Failed to create output texture")
return nil
}
// Encode command
guard let commandBuffer = commandQueue.makeCommandBuffer(),
let computeEncoder = commandBuffer.makeComputeCommandEncoder() else {
print("❌ Failed to create command buffer or encoder")
return nil
}
var params = params
if params.algorithm == 7, let pipe1 = pipelineStateFS_Pass1, let pipe2 = pipelineStateFS_Pass2 {
print("🔄 Using Floyd-Steinberg two-pass rendering")
// FLOYD-STEINBERG MULTI-PASS
// Create Error Texture (Float16 or Float32 for precision)
let errorDesc = MTLTextureDescriptor.texture2DDescriptor(pixelFormat: .rgba16Float,
width: inputTexture.width,
height: inputTexture.height,
mipmapped: false)
errorDesc.usage = [.shaderWrite, .shaderRead]
// CRITICAL: Use autoreleasepool check for error texture too
guard let errorTexture = device.makeTexture(descriptor: errorDesc) else {
computeEncoder.endEncoding()
return nil
// Load input texture
guard let inputTexture = try? textureLoader.newTexture(cgImage: input, options: [.origin: MTKTextureLoader.Origin.topLeft]) else {
print("❌ Failed to create input texture")
continuation.resume(returning: nil)
return
}
// PASS 1: Even Rows
computeEncoder.setComputePipelineState(pipe1)
computeEncoder.setTexture(inputTexture, index: 0)
computeEncoder.setTexture(outputTexture, index: 1)
computeEncoder.setTexture(errorTexture, index: 2)
computeEncoder.setBytes(&params, length: MemoryLayout<RenderParameters>.stride, index: 0)
print("✅ Input texture created: \(inputTexture.width)x\(inputTexture.height)")
// Dispatch (1, H/2, 1) -> Each thread handles one full row
let h = (inputTexture.height + 1) / 2
let threadsPerGrid = MTLSizeMake(1, h, 1)
let threadsPerThreadgroup = MTLSizeMake(1, min(h, pipe1.maxTotalThreadsPerThreadgroup), 1)
// Create output texture
let descriptor = MTLTextureDescriptor.texture2DDescriptor(pixelFormat: .rgba8Unorm,
width: inputTexture.width,
height: inputTexture.height,
mipmapped: false)
descriptor.usage = [.shaderWrite, .shaderRead]
computeEncoder.dispatchThreads(threadsPerGrid, threadsPerThreadgroup: threadsPerThreadgroup)
guard let outputTexture = device.makeTexture(descriptor: descriptor) else {
print("❌ Failed to create output texture")
continuation.resume(returning: nil)
return
}
// Memory Barrier (Ensure Pass 1 writes are visible to Pass 2)
computeEncoder.memoryBarrier(scope: .textures)
// Encode command
guard let commandBuffer = commandQueue.makeCommandBuffer(),
let computeEncoder = commandBuffer.makeComputeCommandEncoder() else {
print("❌ Failed to create command buffer or encoder")
continuation.resume(returning: nil)
return
}
// PASS 2: Odd Rows
computeEncoder.setComputePipelineState(pipe2)
computeEncoder.setTexture(inputTexture, index: 0)
computeEncoder.setTexture(outputTexture, index: 1)
computeEncoder.setTexture(errorTexture, index: 2)
computeEncoder.setBytes(&params, length: MemoryLayout<RenderParameters>.stride, index: 0)
var params = params
computeEncoder.dispatchThreads(threadsPerGrid, threadsPerThreadgroup: threadsPerThreadgroup)
if params.algorithm == 7, let pipe1 = pipelineStateFS_Pass1, let pipe2 = pipelineStateFS_Pass2 {
print("🔄 Using Floyd-Steinberg two-pass rendering")
} else {
print("🔄 Using standard dithering algorithm")
// FLOYD-STEINBERG MULTI-PASS
// STANDARD ALGORITHMS
computeEncoder.setComputePipelineState(pipelineState)
computeEncoder.setTexture(inputTexture, index: 0)
computeEncoder.setTexture(outputTexture, index: 1)
computeEncoder.setBytes(&params, length: MemoryLayout<RenderParameters>.stride, index: 0)
// Create Error Texture (Float16 or Float32 for precision)
let errorDesc = MTLTextureDescriptor.texture2DDescriptor(pixelFormat: .rgba16Float,
width: inputTexture.width,
height: inputTexture.height,
mipmapped: false)
errorDesc.usage = [.shaderWrite, .shaderRead]
let w = pipelineState.threadExecutionWidth
let h = pipelineState.maxTotalThreadsPerThreadgroup / w
let threadsPerThreadgroup = MTLSizeMake(w, h, 1)
let threadsPerGrid = MTLSizeMake(inputTexture.width, inputTexture.height, 1)
// CRITICAL: Use autoreleasepool check for error texture too
guard let errorTexture = device.makeTexture(descriptor: errorDesc) else {
computeEncoder.endEncoding()
continuation.resume(returning: nil)
return
}
computeEncoder.dispatchThreads(threadsPerGrid, threadsPerThreadgroup: threadsPerThreadgroup)
// PASS 1: Even Rows
computeEncoder.setComputePipelineState(pipe1)
computeEncoder.setTexture(inputTexture, index: 0)
computeEncoder.setTexture(outputTexture, index: 1)
computeEncoder.setTexture(errorTexture, index: 2)
computeEncoder.setBytes(&params, length: MemoryLayout<RenderParameters>.stride, index: 0)
// Dispatch (1, H/2, 1) -> Each thread handles one full row
let h = (inputTexture.height + 1) / 2
let threadsPerGrid = MTLSizeMake(1, h, 1)
let threadsPerThreadgroup = MTLSizeMake(1, min(h, pipe1.maxTotalThreadsPerThreadgroup), 1)
computeEncoder.dispatchThreads(threadsPerGrid, threadsPerThreadgroup: threadsPerThreadgroup)
// Memory Barrier (Ensure Pass 1 writes are visible to Pass 2)
computeEncoder.memoryBarrier(scope: .textures)
// PASS 2: Odd Rows
computeEncoder.setComputePipelineState(pipe2)
computeEncoder.setTexture(inputTexture, index: 0)
computeEncoder.setTexture(outputTexture, index: 1)
computeEncoder.setTexture(errorTexture, index: 2)
computeEncoder.setBytes(&params, length: MemoryLayout<RenderParameters>.stride, index: 0)
computeEncoder.dispatchThreads(threadsPerGrid, threadsPerThreadgroup: threadsPerThreadgroup)
} else {
print("🔄 Using standard dithering algorithm")
// STANDARD ALGORITHMS
computeEncoder.setComputePipelineState(pipelineState)
computeEncoder.setTexture(inputTexture, index: 0)
computeEncoder.setTexture(outputTexture, index: 1)
computeEncoder.setBytes(&params, length: MemoryLayout<RenderParameters>.stride, index: 0)
let w = pipelineState.threadExecutionWidth
let h = pipelineState.maxTotalThreadsPerThreadgroup / w
let threadsPerThreadgroup = MTLSizeMake(w, h, 1)
let threadsPerGrid = MTLSizeMake(inputTexture.width, inputTexture.height, 1)
computeEncoder.dispatchThreads(threadsPerGrid, threadsPerThreadgroup: threadsPerThreadgroup)
}
computeEncoder.endEncoding()
// Add completion handler properly inside the closure
commandBuffer.addCompletedHandler { buffer in
// We must jump back to MainActor if we want to do UI stuff, but here we just process data.
// However, continuation must be resumed.
// Since the whole function is @MainActor, we should likely resume on main actor?
// Actually, withCheckedContinuation handles the resume context automatically or acts as a bridge.
// But to be safe and strict, let's keep it simple.
if let error = buffer.error {
print("❌ Metal command buffer error: \(error)")
continuation.resume(returning: nil)
} else {
print("✅ Metal render completed successfully")
// Texture -> CGImage conversion is fast enough to do here or dispatch to main
// But since createCGImage creates data copies, it is safe.
// We need the result.
DispatchQueue.main.async {
// We are back on main thread (required for MetalImageRenderer methods if isolated)
// But wait, makeCGImage is private and inside this class.
// If we call self.createCGImage here, we are inside a closure which is NOT isolated to MainActor by default unless specified.
// Let's call a helper or do it carefully.
// BETTER APPROACH:
// Just resume with the texture or nil, and do conversion after await?
// OR: perform conversion here.
// Since `createCGImage` is private and self is MainActor, we must be on MainActor to call it.
let result = self.createCGImage(from: outputTexture)
if result == nil {
print("❌ Failed to create CGImage from output texture")
}
continuation.resume(returning: result)
}
}
}
commandBuffer.commit()
}
computeEncoder.endEncoding()
commandBuffer.commit()
commandBuffer.waitUntilCompleted()
if let error = commandBuffer.error {
print("❌ Metal command buffer error: \(error)")
return nil
}
print("✅ Metal render completed successfully")
let result = createCGImage(from: outputTexture)
if result == nil {
print("❌ Failed to create CGImage from output texture")
}
return result
}
}

View file

@ -4,6 +4,11 @@ import ImageIO
import AppKit
import UniformTypeIdentifiers
// Helper for Swift 6 Concurrency
struct SendableCGImage: @unchecked Sendable {
let image: CGImage
}
enum DitherAlgorithm: Int, CaseIterable, Identifiable {
case noDither = 0
case bayer2x2 = 1
@ -175,24 +180,26 @@ class DitherViewModel {
print("🔄 Processing image with algorithm: \(self.selectedAlgorithm.name)")
self.renderTask = Task.detached(priority: .userInitiated) { [input, renderer, params] in
// Wrap CGImage in a Sendable wrapper to satisfy strict concurrency
let inputWrapper = SendableCGImage(image: input)
self.renderTask = Task { @MainActor [renderer, params, inputWrapper] in
if Task.isCancelled {
print("⚠️ Render task cancelled before starting")
return
}
let result = renderer.render(input: input, params: params)
// Call async render method
let result = await renderer.render(input: inputWrapper.image, params: params)
if Task.isCancelled {
print("⚠️ Render task cancelled after render")
return
}
await MainActor.run {
if Task.isCancelled { return }
print("✅ Render complete, updating UI")
self.processedImage = result
}
if Task.isCancelled { return }
print("✅ Render complete, updating UI")
self.processedImage = result
}
}
}