Question

我在这里记录了哈里斯记录的代码（并行缩减）。我是cuda编程的新手，我不知道如何为这段代码制作主程序。请帮帮我，谢谢。

这是代码：

template <unsigned int blockSize>
__global__ voidreduce6(int *g_idata, int *g_odata, unsigned int n)
{
extern __shared__ int sdata[];
unsigned int tid = threadIdx.x;
unsigned int i = blockIdx.x*(blockSize*2) + tid;
unsigned int gridSize = blockSize*2*gridDim.x;
sdata[tid] = 0;
do{sdata[tid] += g_idata[i] + g_idata[i+blockSize]; i += gridSize; } while (i < n);
__syncthreads();
if (blockSize >= 512) {if(tid<256) { sdata[tid] += sdata[tid + 256]; } __syncthreads(); }
if (blockSize >= 256) {if(tid<128) { sdata[tid] += sdata[tid + 128]; } __syncthreads(); }
if (blockSize >= 128) {if(tid< 64) { sdata[tid] += sdata[tid + 64]; } __syncthreads(); }
if (tid < 32){
if (blockSize >= 64) sdata[tid] += sdata[tid + 32];
if (blockSize >= 32) sdata[tid] += sdata[tid + 16];
if (blockSize >= 16) sdata[tid] += sdata[tid + 8];
if (blockSize >= 8) sdata[tid] += sdata[tid + 4];
if (blockSize >= 4) sdata[tid] += sdata[tid + 2];
if (blockSize >= 2) sdata[tid] += sdata[tid + 1];
}
if (tid == 0) g_odata[blockIdx.x] = sdata[0];
}

Answer 1

使用此内核的完整示例代码在CUDA sample codes。

中给出

请参阅文件reduction_kernel.cu。内核启动的包装函数将根据特定的简化方法选择一个内核，并且内核也会模拟为threadblock大小：

        case 6:
        default:
            if (isPow2(size))
            {
                switch (threads)
                {
                    case 512:
                        reduce6<T, 512, true><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case 256:
                        reduce6<T, 256, true><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case 128:
                        reduce6<T, 128, true><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case 64:
                        reduce6<T,  64, true><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case 32:
                        reduce6<T,  32, true><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case 16:
                        reduce6<T,  16, true><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case  8:
                        reduce6<T,   8, true><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case  4:
                        reduce6<T,   4, true><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case  2:
                        reduce6<T,   2, true><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case  1:
                        reduce6<T,   1, true><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;
                }
            }
            else
            {
                switch (threads)
                {
                    case 512:
                        reduce6<T, 512, false><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case 256:
                        reduce6<T, 256, false><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case 128:
                        reduce6<T, 128, false><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case 64:
                        reduce6<T,  64, false><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case 32:
                        reduce6<T,  32, false><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case 16:
                        reduce6<T,  16, false><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case  8:
                        reduce6<T,   8, false><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case  4:
                        reduce6<T,   4, false><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case  2:
                        reduce6<T,   2, false><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;

                    case  1:
                        reduce6<T,   1, false><<< dimGrid, dimBlock, smemSize >>>(d_idata, d_odata, size);
                        break;
                }
            }

            break;
    }
}

// Instantiate the reduction function for 3 types
template void
reduce<int>(int size, int threads, int blocks,
            int whichKernel, int *d_idata, int *d_odata);

template void
reduce<float>(int size, int threads, int blocks,
              int whichKernel, float *d_idata, float *d_odata);

template void
reduce<double>(int size, int threads, int blocks,
               int whichKernel, double *d_idata, double *d_odata);

谁可以从mark的文档中为我提供这个代码的主程序？

1 个答案: