c++怎么在macOS上使用Metal进行GPU编程_C++图形加速与Metal开发实践

#include <metal_stdlib>
using namespace metal;
<p>kernel void vector_add(device const float<em> a,
device const float</em> b,
device float* result,
uint id [[thread_position_in_grid]]) {
result[id] = a[id] + b[id];
}

// MetalBridge.mm
#import <Metal/Metal.h>
<p>void runVectorAddOnGPU(float<em> a, float</em> b, float* result, int count) {
// 获取默认设备
id<MTLDevice> device = MTLCreateSystemDefaultDevice();</p><pre class="brush:php;toolbar:false;"><pre class="brush:php;toolbar:false;">// 创建命令队列
id<MTLCommandQueue> queue = [device newCommandQueue];

// 加载默认库（包含编译后的shader）
id<MTLLibrary> library = [device newDefaultLibrary];
id<MTLFunction> function = [library newFunctionWithName:@"vector_add"];

// 创建管线状态
id<MTLComputePipelineState> pipelineState = 
    [device newComputePipelineStateWithFunction:function error:nil];

// 创建缓冲区
id<MTLBuffer> bufferA = [device newBufferWithBytes:a 
                            length:count * sizeof(float) 
                           options:MTLResourceStorageModeShared];
id<MTLBuffer> bufferB = [device newBufferWithBytes:b 
                            length:count * sizeof(float) 
                           options:MTLResourceStorageModeShared];
id<MTLBuffer> bufferResult = [device newBufferWithLength:count * sizeof(float) 
                                                 options:MTLResourceStorageModeShared];

// 创建命令缓冲区
id<MTLCommandBuffer> commandBuffer = [queue commandBuffer];
id<MTLComputeCommandEncoder> encoder = [commandBuffer computeCommandEncoder];
[encoder setComputePipelineState:pipelineState];
[encoder setBuffer:bufferA offset:0 atIndex:0];
[encoder setBuffer:bufferB offset:0 atIndex:1];
[encoder setBuffer:bufferResult offset:0 atIndex:2];

// 配置线程布局
MTLSize threadsPerGroup = {64, 1, 1};
MTLSize numThreadgroups = {(count + 63) / 64, 1, 1};
[encoder dispatchThreads:numThreadgroups threadsPerThreadgroup:threadsPerGroup];
[encoder endEncoding];

// 提交并等待完成
[commandBuffer commit];
[commandBuffer waitUntilCompleted];

// 拷贝结果回CPU内存
memcpy(result, bufferResult.contents, count * sizeof(float));

// 释放资源（实际中应管理生命周期）

// main.cpp 或 main.mm
extern "C" void runVectorAddOnGPU(float* a, float* b, float* result, int count);
<p>int main() {
const int n = 1024;
float a[n], b[n], result[n];
for (int i = 0; i < n; ++i) {
a[i] = i;
b[i] = i * 2;
}</p><pre class="brush:php;toolbar:false;"><pre class="brush:php;toolbar:false;">runVectorAddOnGPU(a, b, result, n);

// 验证结果
for (int i = 0; i < 10; ++i)
    printf("result[%d] = %f\n", i, result[i]);

return 0;