Class MetalTensorOperations

Namespace

DotCompute.Backends.Metal.Execution

Assembly

DotCompute.Backends.Metal.dll

Metal tensor operations using Metal Performance Shaders (MPS) for ML workloads. Provides high-performance matrix operations, convolution, and batch normalization optimized for Apple Silicon and Metal GPUs.

public sealed class MetalTensorOperations : IDisposable

Inheritance

object

MetalTensorOperations

Implements

IDisposable

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

object.ToString()

Extension Methods

BufferExtensions.IsSafeToDispose(IDisposable)

BufferExtensions.SafeDispose(IDisposable)

Constructors

MetalTensorOperations(nint, nint, ILogger<MetalTensorOperations>, MetalPerformanceShadersBackend?)

public MetalTensorOperations(nint device, nint commandQueue, ILogger<MetalTensorOperations> logger, MetalPerformanceShadersBackend? mpsBackend = null)

Parameters

device nint

commandQueue nint

logger ILogger<MetalTensorOperations>

mpsBackend MetalPerformanceShadersBackend

Methods

BatchNormAsync(float[], float[], float[], int, int, int, int, CancellationToken)

Performs batch normalization using MPS.

public Task<float[]> BatchNormAsync(float[] input, float[] scale, float[] bias, int batchSize, int channels, int height, int width, CancellationToken cancellationToken = default)

Parameters

input float[]

scale float[]

bias float[]

batchSize int

channels int

height int

width int

cancellationToken CancellationToken

Returns

Task<float[]>

ConvolutionAsync(float[], float[], ConvolutionConfig, CancellationToken)

Performs 2D convolution using MPS.

public Task<float[]> ConvolutionAsync(float[] input, float[] kernel, ConvolutionConfig config, CancellationToken cancellationToken = default)

Parameters

input float[]

kernel float[]

config ConvolutionConfig

cancellationToken CancellationToken

Returns

Task<float[]>

Dispose()

Performs application-defined tasks associated with freeing, releasing, or resetting unmanaged resources.

public void Dispose()

GetMetrics()

Gets performance metrics for tensor operations.

public TensorPerformanceMetrics GetMetrics()

Returns

TensorPerformanceMetrics

MatrixMultiplyAsync(float[], float[], int, int, int, CancellationToken)

Performs matrix multiplication using MPS: C = A × B.

public Task<float[]> MatrixMultiplyAsync(float[] a, float[] b, int m, int n, int k, CancellationToken cancellationToken = default)

Parameters

a float[]

b float[]

m int

n int

k int

cancellationToken CancellationToken

Returns

Task<float[]>