passing mutable MTLBuffer between two Metal compute encoders swift - metal

Question

In some objective c classes I have a buffer that is used by two sequential metal pipelines. One kernel processes values based on the rgb luminance of the picture and writes to the mutable buffer and the next draws out a waveform graph based on the data in the first. It's based on the osx shader described here USING METAL PERFORMANCE SHADERS WITH CORE IMAGE.
I managed to convert this from osx to IOS which is working great but I am now trying to convert it to swift to make it more flexible for me in the future. I have other filters I'd like to write and the translating process is also teaching me a lot about how metal pipelines work.
This is where it's breaking down. The graph it draws has one horizontal coloured line fluttering up and down. Close but no cigar. I wondered if it had something to do with how the buffers were created in swift vs objective c. Here's an example of the original:
size_t columnBufSize = sizeof(UInt64)*inputTexture.width*inputTexture.height;
id<MTLBuffer> columnDataRed = [kDevice newBufferWithLength:columnBufSize options:0];
but when I translate this in swift It's not behaving as expected.
let columnBufSize: size_t = MemoryLayout<UInt64>.size * inputTexture!.width * inputTexture!.height
let columnDataRed: MTLBuffer = kDevice.makeBuffer( length: columnBufSize, options: .storageModeShared)!
It looks like only the last value is passed into the second computeEncoder.
What am I missing?
for completion, here is the full code in objective c
+ (BOOL)processWithInputs:(NSArray<id<CIImageProcessorInput>> *)inputs arguments:(NSDictionary<NSString *,id> *)arguments output:(id<CIImageProcessorOutput>)output error:(NSError * _Nullable *)error
{
id<MTLComputePipelineState> renderComputerState = kParadeComputePipelineState;
id<CIImageProcessorInput> input = inputs.firstObject;
id<MTLCommandBuffer> commandBuffer = output.metalCommandBuffer;
commandBuffer.label = #"com.martinhering.WaveformKernel";
id<MTLTexture> inputTexture = input.metalTexture;
id<MTLTexture> outputTexture = output.metalTexture;
MTLSize threadgroupCount = MTLSizeMake(inputTexture.width, inputTexture.height, 1);
MTLSize _threadgroupSize = MTLSizeMake(16, 16, 1);
threadgroupCount.width = (inputTexture.width + _threadgroupSize.width - 1) / _threadgroupSize.width;
threadgroupCount.height = (inputTexture.height + _threadgroupSize.height - 1) / _threadgroupSize.height;
size_t columnBufSize = sizeof(UInt64)*inputTexture.width*inputTexture.height;
id<MTLBuffer> columnDataRed = [kDevice newBufferWithLength:columnBufSize options:0];
id<MTLBuffer> columnDataGreen = [kDevice newBufferWithLength:columnBufSize options:0];
id<MTLBuffer> columnDataBlue = [kDevice newBufferWithLength:columnBufSize options:0];
id<MTLComputeCommandEncoder> computeEncoder;
computeEncoder = [commandBuffer computeCommandEncoder];
[computeEncoder setComputePipelineState:kWaveformComputePipelineState];
[computeEncoder setTexture:inputTexture atIndex:0];
[computeEncoder setBuffer:columnDataRed offset:0 atIndex:0];
[computeEncoder setBuffer:columnDataGreen offset:0 atIndex:1];
[computeEncoder setBuffer:columnDataBlue offset:0 atIndex:2];
[computeEncoder setSamplerState:kSamplerState atIndex:0];
[computeEncoder dispatchThreadgroups:threadgroupCount
threadsPerThreadgroup:_threadgroupSize];
[computeEncoder endEncoding];
computeEncoder = [commandBuffer computeCommandEncoder];
[computeEncoder setComputePipelineState:renderComputerState];
[computeEncoder setTexture:inputTexture atIndex:0];
[computeEncoder setTexture:outputTexture atIndex:1];
[computeEncoder setBuffer:columnDataRed offset:0 atIndex:0];
[computeEncoder setBuffer:columnDataGreen offset:0 atIndex:1];
[computeEncoder setBuffer:columnDataBlue offset:0 atIndex:2];
[computeEncoder setSamplerState:kSamplerState atIndex:0];
[computeEncoder dispatchThreadgroups:threadgroupCount
threadsPerThreadgroup:_threadgroupSize];
[computeEncoder endEncoding];
return YES;
}
Here is my translation
override class func process(with inputs: [CIImageProcessorInput]?, arguments: [String : Any]?, output: CIImageProcessorOutput) throws {
guard
let kDevice = device,
let commandBuffer = output.metalCommandBuffer,
let input = inputs?.first,
let defaultLibrary: MTLLibrary = kDevice.makeDefaultLibrary()
else {
return
}
let samplerDescriptor = MTLSamplerDescriptor()
let kSamplerState = kDevice.makeSamplerState(descriptor: samplerDescriptor)
samplerDescriptor.sAddressMode = .clampToEdge
samplerDescriptor.tAddressMode = .clampToEdge
samplerDescriptor.minFilter = .nearest
samplerDescriptor.magFilter = .nearest
samplerDescriptor.normalizedCoordinates = false
var kWaveformComputePipelineState: MTLComputePipelineState?
var kParadeComputePipelineState: MTLComputePipelineState?
if let aFunction = defaultLibrary.makeFunction(name: "scope_waveform_compute") {
kWaveformComputePipelineState = try? kDevice.makeComputePipelineState(function: aFunction)
}
if let aFunction = defaultLibrary.makeFunction(name: "scope_waveform_parade") {
kParadeComputePipelineState = try? kDevice.makeComputePipelineState(function: aFunction)
}
commandBuffer.label = "com.martinhering.WaveformKernel"
weak var inputTexture: MTLTexture? = input.metalTexture
weak var outputTexture: MTLTexture? = output.metalTexture
var threadgroupCount: MTLSize = MTLSizeMake(inputTexture!.width, inputTexture!.height, 1)
let threadgroupSize: MTLSize = MTLSizeMake(16, 16, 1)
threadgroupCount.width = (inputTexture!.width + threadgroupSize.width - 1) / threadgroupSize.width
threadgroupCount.height = (inputTexture!.height + threadgroupSize.height - 1) / threadgroupSize.height
let columnBufSize: size_t = MemoryLayout<UInt64>.size * inputTexture!.width * inputTexture!.height
let columnDataRed: MTLBuffer = kDevice.makeBuffer( length: columnBufSize, options: .storageModeShared)!
let columnDataGreen: MTLBuffer = kDevice.makeBuffer( length: columnBufSize, options: .storageModeShared)!
let columnDataBlue: MTLBuffer = kDevice.makeBuffer( length: columnBufSize, options: .storageModeShared)!
weak var computeEncoder: MTLComputeCommandEncoder?
computeEncoder = commandBuffer.makeComputeCommandEncoder()
computeEncoder?.setComputePipelineState(kWaveformComputePipelineState!)
computeEncoder?.setTexture(inputTexture, index: 0)
computeEncoder?.setBuffer(columnDataRed, offset: 0, index: 0)
computeEncoder?.setBuffer(columnDataGreen, offset: 0, index: 1)
computeEncoder?.setBuffer(columnDataBlue, offset: 0, index: 2)
computeEncoder?.setSamplerState(kSamplerState, index: 0)
computeEncoder?.dispatchThreadgroups(threadgroupCount, threadsPerThreadgroup: threadgroupSize)
computeEncoder?.endEncoding()
computeEncoder = commandBuffer.makeComputeCommandEncoder()
computeEncoder?.setComputePipelineState(kParadeComputePipelineState!)
computeEncoder?.setTexture(inputTexture, index: 0)
computeEncoder?.setTexture(outputTexture, index: 1)
computeEncoder?.setBuffer(columnDataRed, offset: 0, index: 0)
computeEncoder?.setBuffer(columnDataGreen, offset: 0, index: 1)
computeEncoder?.setBuffer(columnDataBlue, offset: 0, index: 2)
computeEncoder?.setSamplerState(kSamplerState, index: 0)
computeEncoder?.dispatchThreadgroups(threadgroupCount, threadsPerThreadgroup: threadgroupSize)
computeEncoder?.endEncoding()
// return true
}

Answer 1

So thanks to Ken Thomases for this; it was simple in the end.
You have to set the properties of the sampler Descriptor before you create the samplerState.
let samplerDescriptor = MTLSamplerDescriptor()
samplerDescriptor.normalizedCoordinates = false
let kSamplerState = kDevice.makeSamplerState(descriptor: samplerDescriptor)