1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
|
#include <cstdio>
__global__ void add(int N, int* a, int* b, int* out) {
const int id = blockIdx.x;
out[id] = a[id] + b[id];
}
int main() {
constexpr int N = 100;
bool success = false;
int host_array[N] = {0};
int* dev_arrays[3] = {nullptr};
// Allocate device arrays.
for (int i = 0; i < 3; ++i) {
if (cudaMalloc(&dev_arrays[i], N * sizeof(int)) != cudaSuccess) {
goto cleanup;
}
}
// Fill the host array with values 0..N-1.
for (int i = 0; i < N; ++i) {
host_array[i] = i;
}
// Copy the host array to each of the first two device arrays.
for (int i = 0; i < 2; ++i) {
if (cudaMemcpy(
dev_arrays[i], host_array, N * sizeof(int),
cudaMemcpyHostToDevice) != cudaSuccess) {
goto cleanup;
}
}
// Add the first two arrays.
// N blocks, 1 thread per block.
add<<<N, 1>>>(N, dev_arrays[0], dev_arrays[1], dev_arrays[2]);
// Copy the result from the third array to the host.
if (cudaMemcpy(
host_array, dev_arrays[2], N * sizeof(int), cudaMemcpyDeviceToHost) !=
cudaSuccess) {
goto cleanup;
}
// Print the result.
for (int i = 0; i < N; ++i) {
printf("%d ", host_array[i]);
}
printf("\n");
success = true;
cleanup:
for (int i = 0; i < 3; ++i) {
if (dev_arrays[i] != nullptr) {
cudaFree(dev_arrays[i]);
}
}
return success ? 0 : 1;
}
|
