GTX960 GPU

测试了下性能。计算平方和

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
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
#include <stdio.h>
#define N 1048576
#define  THREAD_PRE_BLOCK 1024
#define BLOCK_PRE_GRID 1



__global__  void squaresSum(int *data, int *sum)
{
	__shared__ int cached[THREAD_PRE_BLOCK];
	int tid = threadIdx.x + blockIdx.x*blockDim.x;
	int cacheIdx = threadIdx.x;

	int tmp = 0;
	while(tid < N)
	{
		tmp +=  data[tid]*data[tid];
		tid += blockDim.x * gridDim.x;
	}
	cached[cacheIdx] = tmp;

	__syncthreads();

	int i = blockDim.x/2;
	while(i!=0)
	{
		if(cacheIdx<i)
		{
			cached[cacheIdx] += cached[cacheIdx+i];
		}
		__syncthreads();
		i  /= 2;
	}
	if(cacheIdx == 0)
	{
		sum[blockIdx.x] = cached[0];
	}
}

void generateData(int * data , int s)
{
	int i;
	for(i=0;i<s;i++)
	{
		data[i] = rand()%5;
	}
}

int cusumCpu(int * data)
{
	int tmp = 0;
	int i;
	for(i=0;i<N;i++)
	{
		tmp += data[i]*data[i];
	}
	return tmp;
}

int main()
{
	int *gpu_data, *gpu_sum;
	int * data,*sum;
	time_t t1,t2,t3,t4;

	data = (int *) malloc(N*sizeof(int));
	cudaMalloc((void**) &gpu_data, sizeof(int) * N);
	sum = (int *) malloc(BLOCK_PRE_GRID*sizeof(int));
	cudaMalloc((void**) &gpu_sum, sizeof(int) * BLOCK_PRE_GRID);

	generateData(data,N);

	cudaMemcpy(gpu_data, data, sizeof(int) * N, cudaMemcpyHostToDevice);
	t1 = clock();
	squaresSum<<<BLOCK_PRE_GRID,THREAD_PRE_BLOCK>>>(gpu_data,gpu_sum);
	t2 = clock();
	cudaMemcpy(sum,gpu_sum,BLOCK_PRE_GRID*sizeof(int),cudaMemcpyDeviceToHost);

	int i,gpu_res = 0,cpu_res;
	for(i=0;i<BLOCK_PRE_GRID;i++)
	{
		gpu_res += sum[i];
	}
	t3 = clock();
	cpu_res = cusumCpu(data);
	t4 = clock();
	printf("gpu res : %d, time : %ld\ncpu res : %d, time : %ld\n",gpu_res,t2-t1,cpu_res,t4-t3);
	cudaFree(gpu_data);
	cudaFree(gpu_sum);
	free(data);
}

如果t2放在内存拷贝后面时间大幅增加：
Imgur
为啥这么费时。。