Hi all,

I have a question on how to use texture memory in the context I will describe below:

I have a structure:

typedef struct {

    double x;

    double y;

    double z;

} VecR;

I want to use the texture memory for a vector of elements VecR.

The size of the vector of VecR elements is N.

Is it correct to use the texture like this:

int widthTex = ceil(sqrt(N));

texture<double, 2, cudaReadModeElementType> texCoord;

texCoord.addressMode[0] = cudaAddressModeClamp;

texCoord.addressMode[1] = cudaAddressModeClamp;

texCoord.filterMode = cudaFilterModePoint;

texCoord.normalized = false;

cudaChannelFormatDesc channelDesc_V;

channelDesc_V = cudaCreateChannelDesc<double>();

// coordDev is a vector of N elements of type VecR which resides on the device memory

cudaArray *coordCuArray;

cudaMallocArray (&coordCuArray, &channelDesc_V, 3*widthTex, widthTex);

cudaMemcpyToArray(coordCuArray, 0, 0, coordDev, sizeof(VecR)*(N), cudaMemcpyDeviceToDevice);

cudaBindTextureToArray (texCoord, coordCuArray, channelDesc_V );

and then for retrieving the elements of VecR of the element index of the vector (coordDev[index]):

VecR temp;

temp.x = tex2D(texCoord, (index)%(3*widthTex), index/widthTex);

temp.y = tex2D(texCoord, (index+1)%(3*widthTex), index/widthTex);

temp.z = tex2D(texCoord, (index+2)%(3*widthTex), index/widthTex);

I cannot even compile using texture memory like this. How can I do to resolve my problem?

Does anyone have any idea for that?

Thank you in advance,

Ardita