-
Notifications
You must be signed in to change notification settings - Fork 5
/
reduce-cuda-large.cu
73 lines (52 loc) · 1.67 KB
/
reduce-cuda-large.cu
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
63
64
65
66
67
68
69
70
71
72
/*
* A very simple cuda implementation of reduce. Uses an array of 1024x1024
* items which are summed into a 1024 array and then summed into a value.
*/
#include <stdio.h>
#include <stdlib.h>
/*
* this kernel will sum all of the data from in into out - at
* least as far as the block will carry you
*/
__global__ void reduce(float* out, float* in, int size);
void startClock(char*);
void stopClock(char*);
void printClock(char*);
int main(int argc, char** argv) {
int size = 1024*1024;
printf("size = %d\n",size);
void *d_in; // device data
void *d_mid; // device data - middle results
void *d_out; // device data - the answer
float *h_in; // host data
float h_out;
int numBlocks = 1024;
cudaMalloc(&d_in,size*sizeof(float));
cudaMalloc(&d_mid,numBlocks*sizeof(float));
cudaMalloc(&d_out,sizeof(float));
h_in = (float*) malloc(size*sizeof(float));
for (int i = 0; i < size; i++) {
h_in[i] = 1;
}
startClock("copy data to device");
cudaMemcpy(d_in,h_in,size*sizeof(float),cudaMemcpyHostToDevice);
stopClock("copy data to device");
startClock("compute");
// use max threads/block and the required # of blocks AND
// ask for some shared memory
reduce<<<1024,1024,1024>>>((float*) d_mid,(float*) d_in,size);
reduce<<<1,1024,1024>>>((float*)d_out,(float*)d_mid,1024);
cudaThreadSynchronize();
stopClock("compute");
startClock("copy data to host");
h_out = -17;
cudaMemcpy(&h_out,d_out,sizeof(float),cudaMemcpyDeviceToHost);
stopClock("copy data to host");
printf("The total is %f\n",h_out);
free(h_in);
cudaFree(d_in);
cudaFree(d_out);
printClock("copy data to device");
printClock("compute");
printClock("copy data to host");
}