CUDA 1D texture fetch always return 0

问题

I am trying to test the CUDA 1D texture with a piece of simple code. It is quite straightforward: first allocates a global memory, then bind it to a texture reference; access the texture from within a kernel by tex1D(); print out the value returned by the texture fetch.

The code is as follows:

#include "cuda.h"
#include "cuda_runtime.h"
#include <iostream>
#include <vector>
#include <cstdio>

using namespace std;

texture<float, cudaTextureType1D, cudaReadModeElementType> texX;

__global__ void kernel(float *X)
{
    int i = threadIdx.x ;
    if ( i >= 128 ) return;
    printf("%.3f\t%.3f\n", tex1D( texX, i*1.0 ), X[i] );
}

int main()
{
    float *devX;
    vector<float> X(128, 3.1415926 );
    cudaMalloc( &devX, 128 * sizeof(float) );
    cudaMemcpy( devX, &X[0], 128*sizeof(float), cudaMemcpyDefault );
    cudaDeviceSynchronize();

    cudaBindTexture( (size_t)0, texX, devX, 128 * sizeof(float) );
    cudaDeviceSynchronize();

    kernel<<<1,128>>>( devX );
    cudaDeviceSynchronize();
    cout<<endl;
    cout<< cudaGetErrorString( cudaGetLastError() ) <<endl;
}

But all I got was like this:

0.000    3.142
0.000    3.142
...
0.000    3.142

no error

Could anyone explain why?

回答1:

You need to use tex1Dfetch() since your texture is bound to linear memory:

printf("%.3f\t%.3f\n", tex1Dfetch( texX, i ), X[i] );

回答2:

From the " CUDA C Programming Guide"

tex1D is used to fetch from the CUDA array specified by the one-dimensional texture object ...

tex1Dfetch is used to fetch from the region of linear memory specified by the one-dimension texture ...

So, if you use cudaMalloc to malloc linear memory but not cuda array, u should select tex1Dfetch

来源：https://stackoverflow.com/questions/12409581/cuda-1d-texture-fetch-always-return-0

标签

cuda

textures