OpenACC 简单的直方图

 1 #include <stdio.h>
 2 #include <stdlib.h>
 3 #include <openacc.h>
 4 
 5 int main()
 6 {
 7     const int length = 1024, basket = 10;
 8     int data[length], histgram[basket];
 9 
10     srand(97);
11     for (int i = 0; i < basket; histgram[i++] = 0);
12     for (int i = 0; i < length; data[i++] = rand() % 10);
13 
14 #pragma acc parallel loop
15     for (int i = 0; i < length; i++)
16     {
17 #pragma acc atomic update
18         histgram[data[i]]+=1;
19     }
20 
21     for (int i = 0; i < basket; i++)
22         printf("histgram[%d] = %d\n", i, histgram[i]);
23 
24     getchar();
25     return 0;
26 }

D:\Code\OpenACC\OpenACCProject\OpenACCProject>pgcc -acc -Minfo main.c -o main_acc.exe
main:
     14, Accelerator kernel generated
         Generating Tesla code
         15, #pragma acc loop gang, vector(128) /* blockIdx.x threadIdx.x */
     14, Generating implicit copyout(histgram[data])
         Generating implicit copyin(data[:1024])

D:\Code\OpenACC\OpenACCProject\OpenACCProject>main_acc.exe
launch CUDA kernel  file=D:\Code\OpenACC\OpenACCProject\OpenACCProject\main.c function=main line=14 device=0 threadid=1 num_gangs=8 num_workers=1 vector_length=128 grid=8 block=128
histgram[0] = 0
histgram[1] = 0
histgram[2] = 0
histgram[3] = 0
histgram[4] = 0
histgram[5] = 100
histgram[6] = 0
histgram[7] = 0
histgram[8] = 0
histgram[9] = 0

cuan@CUAN:/mnt/d/Code/OpenACC/OpenACCProject/OpenACCProject$ pgcc -acc -Minfo main.c -o main_acc_ubuntu.exe
main:
     14, Accelerator kernel generated
         Generating Tesla code
         15, #pragma acc loop gang, vector(128) /* blockIdx.x threadIdx.x */
     14, Generating implicit copyout(histgram[data])
         Generating implicit copyin(data[:1024])
cuan@CUAN:/mnt/d/Code/OpenACC/OpenACCProject/OpenACCProject$ ./main_acc_ubuntu.exe
histgram[0] = 123
histgram[1] = 79
histgram[2] = 90
histgram[3] = 109
histgram[4] = 102
histgram[5] = 96
histgram[6] = 117
histgram[7] = 104
histgram[8] = 110
histgram[9] = 94

 1 #include <stdio.h>
 2 #include <stdlib.h>
 3 #include <openacc.h>
 4 
 5 int main()
 6 {
 7     const int length = 1024, basket = 10;
 8     int data[length], histgram[basket];
 9 
10     srand(97);
11     for (int i = 0; i < basket; histgram[i++] = 0);
12     for (int i = 0; i < length; data[i++] = rand() % 10);
13 
14     int temp;               // 新定义一个变量
15 #pragma acc parallel loop
16     for (int i = 0; i < length; i++)
17     {
18         temp = data[i];     // 原子操作之前先取地址
19 #pragma acc atomic update
20         histgram[temp] += 1;// 使用 temp 作为下标
21     }
22 
23     for (int i = 0; i < basket; i++)
24         printf("histgram[%d] = %d\n", i, histgram[i]);
25 
26     getchar();
27     return 0;
28 }