Apple Metal元素矩阵乘法（Hadamard产品）

男儿欲遂平生志，六经勤向窗前读。这篇文章主要讲述Apple Metal元素矩阵乘法（Hadamard产品）相关的知识，希望能为你提供帮助。
是否可以使用Apple的金属性能着色器执行Hadamard产品？我看到可以使用this执行正常的矩阵乘法，但我特意寻找一个逐元素乘法，或者一个巧妙的方法来构造一个。（例如，是否可以将MPSMatrix转换为MPSVector，然后使用向量执行产品？）
更新：我感谢使用着色器的建议！我正在开发一个实现，看起来很有前途！一旦我有工作，我会发布解决方案。
答案好的，根据评论者的建议在这里回答我自己的问题 - 试着写我自己的着色器！
这是着色器代码：

#include < metal_stdlib> using namespace metal; /* hadamardProduct: Perform an element-wise multiplication (hadamard product) of the two input matrices A and B, store the result in C */ kernel void hadamardProductKernel( texture_buffer< float, access::read> A [[texture(0)]], texture_buffer< float, access::read> B [[texture(1)]], texture_buffer< float, access::write> C [[texture(2)]], uint gid [[thread_position_in_grid]]) { // C[i,j] = A[i,j] * B[i,j] C.write(A.read(gid) * B.read(gid), gid); }

在两个4x4矩阵上执行着色器的swift：

import Foundation import Metal import MetalKitguard let gpu = MTLCreateSystemDefaultDevice(), let commandQueue = gpu.makeCommandQueue(), let commandBuffer = commandQueue.makeCommandBuffer(), let defaultLibrary = gpu.makeDefaultLibrary(), let kernelFunction = defaultLibrary.makeFunction(name: "hadamardProductKernel") else {exit(1)}// Create the matrices to multiply (as row-major matrices) var A:[Float] = [2,0,0,0, 0,2,0,0, 0,0,2,0, 0,0,0,2]var B:[Float] = [1,0,0,0, 0,2,0,0, 0,0,3,0, 0,0,0,4]let A_buffer = gpu.makeTexture(descriptor: MTLTextureDescriptor.textureBufferDescriptor(with: .r32Float, width: 16, resourceOptions: .storageModeManaged, usage: .shaderRead)) let B_buffer = gpu.makeTexture(descriptor: MTLTextureDescriptor.textureBufferDescriptor(with: .r32Float, width: 16, resourceOptions: .storageModeManaged, usage: .shaderRead)) let C_buffer = gpu.makeTexture(descriptor: MTLTextureDescriptor.textureBufferDescriptor(with: .r32Float, width: 16, resourceOptions: .storageModeManaged, usage: .shaderWrite)) A_buffer?.replace(region: MTLRegionMake1D(0, 16), mipmapLevel: 0, withBytes: UnsafeRawPointer(A), bytesPerRow: 64) B_buffer?.replace(region: MTLRegionMake1D(0, 16), mipmapLevel: 0, withBytes: UnsafeRawPointer(B), bytesPerRow: 64)let computePipelineState = try gpu.makeComputePipelineState(function: kernelFunction) let computeEncoder = commandBuffer.makeComputeCommandEncoder() computeEncoder?.setComputePipelineState(computePipelineState) computeEncoder?.setTexture(A_buffer, index: 0) computeEncoder?.setTexture(B_buffer, index: 1) computeEncoder?.setTexture(C_buffer, index: 2) let threadGroupSize = MTLSize(width: 16, height: 1, depth: 1) let threadGroupCount = MTLSize(width: 1, height: 1, depth: 1) computeEncoder?.dispatchThreadgroups(threadGroupCount, threadsPerThreadgroup: threadGroupSize) computeEncoder?.endEncoding() commandBuffer.commit() commandBuffer.waitUntilCompleted()print("done")

【Apple Metal元素矩阵乘法（Hadamard产品）】感谢任何链接到资源的评论，以进一步了解这类事情。