映射内存(Mapped memory)示例

这是CUDA/SDK中的例子

#include <stdio.h>#include <stdlib.h>#include <cuda.h>__global__ void vectorAddGPU(float *a, float *b, float *c, int N){int idx = blockIdx.x*blockDim.x + threadIdx.x;if (idx < N)c[idx] = a[idx] + b[idx];}int main(int argc, char **argv){int n, nelem;float *a, *b, *c; // 页锁定内存的指针float *d_a, *d_b, *d_c; // 映射内存的设备指针 float errorNorm, refNorm, ref, diff;// 把设备标记为可映射内存cudaSetDeviceFlags(cudaDeviceMapHost);//分配映射内存nelem = 1048576;cudaHostAlloc((void **)&a, nelem*sizeof(float), cudaHostAllocMapped);cudaHostAlloc((void **)&b, nelem*sizeof(float), cudaHostAllocMapped);cudaHostAlloc((void **)&c, nelem*sizeof(float), cudaHostAllocMapped);//初始化a,bfor(n = 0; n < nelem; n++){a[n] = rand() / (float)RAND_MAX;b[n] = rand() / (float)RAND_MAX;}//获得映射内存的设备指针cudaHostGetDevicePointer((void **)&d_a, (void *)a, 0);cudaHostGetDevicePointer((void **)&d_b, (void *)b, 0);cudaHostGetDevicePointer((void **)&d_c, (void *)c, 0);printf("使用映射内存的设备指针调用GPU核函数.../n");dim3 block(256);dim3 grid((unsigned int)ceil(nelem/(float)block.x));vectorAddGPU<<<grid, block>>>(d_a, d_b, d_c, nelem); cudaThreadSynchronize();printf("验证结果.../n");errorNorm = 0.f;refNorm = 0.f;for(n = 0; n < nelem; n++){ref = a[n] + b[n];diff = c[n] - ref;errorNorm += diff*diff;refNorm += ref*ref;}errorNorm = (float)sqrt((double)errorNorm);refNorm = (float)sqrt((double)refNorm);printf("释放CPU内存.../n");cudaFreeHost(a);cudaFreeHost(b);cudaFreeHost(c);cudaThreadExit();printf("%s/n", (errorNorm/refNorm < 1.e-6f) ? "PASSED" : "FAILED");}

 

下面是网上的一个例子, 记不清出处啦

#include <stdlib.h>#include <stdio.h>int main() {unsigned int *h_array, *d_array, *h_array_test;int cudaError; int num = 5; int size = num * sizeof(unsigned int); cudaSetDevice(0); cudaSetDeviceFlags( cudaDeviceMapHost ); cudaError=cudaHostAlloc( (void**) &(h_array), size, cudaHostAllocMapped ); cudaError=cudaHostGetDevicePointer( (void**) &(d_array), h_array, 0 ); if (cudaError) printf ("Failed to get device pointer /n"); for (int i = 0; i < num; i ++) h_array[i] = rand(); cudaMemcpy(d_array, h_array, size, cudaMemcpyHostToDevice); h_array_test = (unsigned int*)malloc(size); cudaError=cudaMemcpy( h_array_test, d_array, size, cudaMemcpyDeviceToHost ); printf("在主机上的数组:/n"); for (int i = 0; i < num; i ++) printf("%d: %d/t", i, h_array[i]); printf("/n在设备上的数组:/n"); for (int i = 0; i < num; i ++) printf("%d: %d/t", i, h_array_test[i]);printf("/n");}