Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- #include <iostream>
- #include <cassert>
- #include <cuda.h>
- #define HANDLE_ERROR(x) assert(x == cudaSuccess);
- #define THREADS 1
- #define BLK 1000
- #define N BLK * THREADS
- #define MEMCPY 0
- using namespace std;
- __global__ void add(int *da, int *db, int *dc)
- {
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i < N)
- dc[i] = da[i] + db[i];
- }
- int main(void)
- {
- // ensure device supports mapped pinned memory & set appropriate flag
- cudaDeviceProp deviceProp;
- cudaGetDeviceProperties(&deviceProp, 0);
- if (!deviceProp.canMapHostMemory)
- exit(1);
- cudaSetDeviceFlags(cudaDeviceMapHost);
- // valloc for page-aligned memory
- int *a, *b, *c, *da, *db, *dc;
- a = (int*)valloc(N * sizeof(int));
- b = (int*)valloc(N * sizeof(int));
- c = (int*)valloc(N * sizeof(int));
- // register host memory & get device pointers to it
- int bytes = (N * sizeof(int) + 4095)/4096;
- HANDLE_ERROR(cudaHostRegister((void*)a, bytes, cudaHostRegisterMapped));
- HANDLE_ERROR(cudaHostRegister((void*)b, bytes, cudaHostRegisterMapped));
- HANDLE_ERROR(cudaHostRegister((void*)c, bytes, cudaHostRegisterMapped));
- HANDLE_ERROR(cudaHostGetDevicePointer((void**)&da, a, 0));
- HANDLE_ERROR(cudaHostGetDevicePointer((void**)&db, b, 0));
- HANDLE_ERROR(cudaHostGetDevicePointer((void**)&dc, c, 0));
- // initialise vectors on host end
- for (int i = 0; i < N; i++)
- {
- a[i] = 2*i;
- b[i] = -i;
- }
- add<<<BLK,THREADS>>>(da, db, dc);
- // ensure vector addition happened properly
- for (int i = 0; i < N; i++)
- {
- if (c[i] != i)
- {
- cout << "Failed @ idx " << i << ", c[i] = " << c[i] << endl;
- assert(c[i] == i);
- }
- }
- // clean up resources
- HANDLE_ERROR(cudaHostUnregister(a));
- HANDLE_ERROR(cudaHostUnregister(b));
- HANDLE_ERROR(cudaHostUnregister(c));
- free(a);
- free(b);
- free(c);
- cout << "Success." << endl;
- return 0;
- }
Advertisement
Add Comment
Please, Sign In to add comment