Re: [問題] 使用CUDA來擷取矩陣中部分值
→
,
→
,
沒錯
→
,
→
,
→
,
→
,
推
,
→
,
→
,
說實話我完全不知道你的blocks和threads為什麼要取這種數字啊
要是矩陣的大小改了怎麼辦?
→
,
→
,
→
,
從你乘出來的結果看也沒有好像處理完512*512的矩陣啊
下面是我把你之前的範例改成任意大小的矩陣都能處理的type
NNx NNy BLOCK_SIZE 都可以任意設 (只要沒有overflow)
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <cuda_runtime.h>
#include <cutil.h>
#define NNx 16
#define NNy 16
#define BLOCK_SIZE 8
#define ALIGN 3 // 對齊printf用
__global__ void movaIM5(float *input1, float *output1)
{
int col = blockIdx.x*blockDim.x+threadIdx.x;
int row = blockIdx.y*blockDim.y+threadIdx.y;
if(col < NNx && row+1 < NNy)
output1[row*NNx+col] = input1[(row+1)*NNx+col]; // 取下面矩陣
//if(col < NNx && row < (NNy-1))
// output1[row*NNx+col] = input1[row*NNx+col]; // 取上面矩陣
}
__global__ void movaIM6(float *input1, float *output1)
{
int col = blockIdx.x*blockDim.x+threadIdx.x;
int row = blockIdx.y*blockDim.y+threadIdx.y;
if(col+1 < NNx && row < NNy)
output1[row*(NNx-1)+col] = input1[row*NNx+col+1]; // 取右邊矩陣
//if(col < (NNx-1) && row < NNy)
// output1[row*(NNx-1)+col] = input1[row*NNx+col]; // 取左邊矩陣
}
int main(int argc, char* argv[])
{
int i;
float input[NNx*NNy];
float *d_input;
for(i = 0; i < NNx*NNy; i++){
input[i] = i+1;
if(i % NNx == 0) printf("\n");
else printf(" ");
printf("%*.0f", ALIGN, input[i]);
}
printf("\n");
cudaMalloc((void**)&d_input, sizeof(float)*NNx*NNy);
cudaMemcpy( d_input, input, sizeof(float)*NNx*NNy, cudaMemcpyHostToDevice );
float *d_IM5, *d_IM6;
cudaMalloc((void**)&d_IM5, sizeof(float)*NNx*(NNy-1));
cudaMalloc((void**)&d_IM6, sizeof(float)*(NNx-1)*NNy);
int bx = (NNx + BLOCK_SIZE - 1) / BLOCK_SIZE;
int by = (NNy + BLOCK_SIZE - 1) / BLOCK_SIZE;
dim3 blocks(bx,by);
dim3 threads(BLOCK_SIZE,BLOCK_SIZE);
movaIM5<<<blocks, threads>>>( d_input, d_IM5 );
movaIM6<<<blocks, threads>>>( d_input, d_IM6 );
float *IM5, *IM6;
IM5 = (float*) malloc( sizeof(float)*NNx*(NNy-1) );
IM6 = (float*) malloc( sizeof(float)*(NNx-1)*NNy );
cudaMemcpy( IM5, d_IM5, sizeof(float)*NNx*(NNy-1), cudaMemcpyDeviceToHost );
cudaMemcpy( IM6, d_IM6, sizeof(float)*(NNx-1)*NNy, cudaMemcpyDeviceToHost );
for(i = 0; i < NNx*(NNy-1); i++){
if(i % NNx == 0) printf("\n");
else printf(" ");
printf("%*.0f", ALIGN, IM5[i]);
}
printf("\n");
for(i = 0; i < (NNx-1)*NNy; i++){
if(i % (NNx-1) == 0) printf("\n");
else printf(" ");
printf("%*.0f", ALIGN, IM6[i]);
}
printf("\n");
cudaFree(d_input);
cudaFree(d_IM5);
cudaFree(d_IM6);
system("pause");
return 0;
}
--
※ 發信站: 批踢踢實業坊(ptt.cc)
◆ From: 122.120.40.234
→
02/24 02:26, , 1F
02/24 02:26, 1F
→
02/24 02:27, , 2F
02/24 02:27, 2F
→
02/24 07:10, , 3F
02/24 07:10, 3F
→
02/24 07:13, , 4F
02/24 07:13, 4F
→
02/24 07:14, , 5F
02/24 07:14, 5F
※ 編輯: lgen7604 來自: 122.120.40.234 (02/24 07:22)
→
02/24 14:17, , 6F
02/24 14:17, 6F
→
02/24 14:19, , 7F
02/24 14:19, 7F
→
02/25 01:35, , 8F
02/25 01:35, 8F
→
02/25 01:35, , 9F
02/25 01:35, 9F
→
02/25 01:36, , 10F
02/25 01:36, 10F
→
02/25 01:36, , 11F
02/25 01:36, 11F
→
02/25 01:36, , 12F
02/25 01:36, 12F
→
02/25 01:37, , 13F
02/25 01:37, 13F
→
02/25 01:48, , 14F
02/25 01:48, 14F
→
02/25 01:49, , 15F
02/25 01:49, 15F
討論串 (同標題文章)
完整討論串 (本文為第 7 之 10 篇)：
3
10
C_and_CPP 近期熱門文章
PTT數位生活區 即時熱門文章