合并访存实验

{
int blx=blockIdx.x;
int thx=threadIdx.x*stride;
for (int i=0;i<stride;i++)
{
C[blx*blockDim.x+thx+i]=A[blx*blockDim.x+thx+i]+B[blx*blockDim.x+thx+i];
cudaMemcpy(h_c,d_c,NUM*sizeof(float),cudaMemcpyDeviceToHost);
cudaFree(d_a);
cudaFree(d_b);
cudaFree(d_c);
// for (int i=0;i<NUM;i++)
// {
// printf("%f",h_c[i]);
cudaEventRecord(end,0);
cudaEventSynchronize(end);
cudaEventElapsedTime(&elapsedTime, start, end);
printf("%f ",elor (int i=0;i<NUM;i++)
{
h_a[i]=1;h_b[i]=2;h_c[i]=0;
}
float *d_a,*d_b,*d_c;
cudaMalloc((void **)&d_a,NUM*sizeof(float));
cudaMalloc((void **)&d_b,NUM*sizeof(float));
// matrixmul_gpu.cpp : Defines the entry point for the console application.
//合并访存实验
//
//
#include<stdio.h>
#include"math.h"
#include <stdlib.h>
#include "time.h"
#include <assert.h>
#include "cuda_runtime.h"
#define step 1024
#define stride 2
#define NUM 1024*1024
__global__ void Addwithstride(float *A, float *B, float *C)
cudaMemcpy(d_c,h_c,NUM*sizeof(float),cudaMemcpyHostToDevice);
dim3 Threadperblock_withstride(step/stride);
dim3 Block_withstride(step);
cudaEvent_t start,end;
}
int main( int argc, char** argv)
{
float *h_a,*h_b,*h_c;
h_a=(float *)malloc(NUM*sizeof(float));
h_b=(float *)malloc(NUM*sizeof(float));
h_c=(float *)malloc(NUM*sizeof(float));
}
}
__global__ void Addwithoutstride(float *A, float *B, float *C)
{
int blx=blockIdx.x;
int thx=threadIdx.x;
C[blx*blockDim.x+threadIdx.x]=A[blx*blockDim.x+thx]+B[blx*blockDim.x+thx];
cudaMemcpy(d_b,h_b,NUM*sizeof(float),cudaMemcpyHostToDevice);
cudaMemcpy(d_c,h_c,NUM*sizeof(float),cudaMemcpyHostToDevice);
dim3 Threadperblock_withoutstride(step);
cudaMalloc((void **)&d_c,NUM*sizeof(float));
cudaMemcpy(d_a,h_a,NUM*sizeof(float),cudaMemcpyHostToDevice);
cudaMemcpy(d_b,h_b,NUM*sizeof(float),cudaMemcpyHostToDevice);
cudaEventRecord(end,0);
cudaEventSynchronize(end);
float elapsedTime;
cudaEventElapsedTime(&elapsedTime, start, end);
printf("%f ",elapsedTime);
dim3 Block_withoutstirde(step);
cudaEventCreate(&start);
cudaEventCreate(&end);
cudaEventRecord(start,0);
Addwithoutstride<<<Block_withoutstirde,Threadperblock_withoutstride>>>(d_a,d_b,d_c);
cudaEventCreate(&start);
cudaEventCreate(&end);
cudaEventRecord(start,0);
Addwithstride<<<Block_withstride,Threadperblock_withstride>>>(d_a,d_b,d_c);
// }
cudaMalloc((void **)&d_a,NUM*sizeof(float));
cudaMalloc((void **)&d_b,NUM*sizeof(float));
cudaMalloc((void **)&d_c,NUM*sizeof(float));
cudaMemcpy(d_a,h_a,NUM*sizeof(float),cudaMemcpyHostToDevice);