50 lines
1.6 KiB
Plaintext
50 lines
1.6 KiB
Plaintext
|
#include <cstdio>
|
||
|
|
|
||
|
|
#include "vecadd.h"
|
||
|
|
|
||
|
|
#define CHECK_CUDA(call) \
|
||
|
|
do { \
|
||
|
|
cudaError_t status_ = call; \
|
||
|
|
if (status_ != cudaSuccess) { \
|
||
|
|
fprintf(stderr, "CUDA error (%s:%d): %s:%s\n", __FILE__, __LINE__, \
|
||
|
|
cudaGetErrorName(status_), cudaGetErrorString(status_)); \
|
||
|
|
exit(EXIT_FAILURE); \
|
||
|
|
} \
|
||
|
|
} while (0)
|
||
|
|
|
||
|
|
__global__ void vecadd_kernel(const int N, const float *a, const float *b, float *c) {
|
||
|
|
int tidx = blockIdx.x * blockDim.x + threadIdx.x;
|
||
|
|
c[tidx] = a[tidx] + b[tidx];
|
||
|
|
}
|
||
|
|
|
||
|
|
// Device(GPU) pointers
|
||
|
|
static float *A_gpu, *B_gpu, *C_gpu;
|
||
|
|
|
||
|
|
void vecadd(float *_A, float *_B, float *_C, int N) {
|
||
|
|
// (TODO) Upload A and B vector to GPU
|
||
|
|
|
||
|
|
// Launch kernel on a GPU
|
||
|
|
dim3 gridDim(N / 512);
|
||
|
|
dim3 blockDim(512);
|
||
|
|
vecadd_kernel<<<gridDim, blockDim>>>(N, A_gpu, B_gpu, C_gpu);
|
||
|
|
|
||
|
|
// (TODO) Download C vector from GPU
|
||
|
|
|
||
|
|
// DO NOT REMOVE; NEEDED FOR TIME MEASURE
|
||
|
|
CHECK_CUDA(cudaDeviceSynchronize());
|
||
|
|
}
|
||
|
|
|
||
|
|
void vecadd_init(int N) {
|
||
|
|
// (TODO) Allocate device memory
|
||
|
|
|
||
|
|
// DO NOT REMOVE; NEEDED FOR TIME MEASURE
|
||
|
|
CHECK_CUDA(cudaDeviceSynchronize());
|
||
|
|
}
|
||
|
|
|
||
|
|
void vecadd_cleanup(float *_A, float *_B, float *_C, int N) {
|
||
|
|
// (TODO) Do any post-vecadd cleanup work here.
|
||
|
|
|
||
|
|
// DO NOT REMOVE; NEEDED FOR TIME MEASURE
|
||
|
|
CHECK_CUDA(cudaDeviceSynchronize());
|
||
|
|
}
|