is 4Mbyte the max allowed size for linear int tex texture size

Accelerated Computing CUDA CUDA Programming and Performance
1

problem fixed: i used a too small constant memory.

We should alloc as big as 16k ints (actually 15k is safe) in constMem!

Hi, I use int single-component(red) 1D texture, and when size>1M, computing with it gets wrong number. is 1M int (4Mbyte) the biggest number? thanks.

(my card is g80 GTX)

//main.cu

texture< AttrType, 1, cudaReadModeElementType> tex;

	tex.addressMode[0] = cudaAddressModeClamp;

    tex.filterMode = cudaFilterModePoint;

    tex.normalized = 0;

	cudaChannelFormatDesc desc = cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindUnsigned);

	int *d_tex = 0;

	//asign d_tex..=..

	CUDA_SAFE_CALL(cudaBindTexture("tex", d_tex, &desc, sizeof(int) * N, 0));

//kernel.cu

extern texture< AttrType, 1, cudaReadModeElementType> tex;

..

kernel{

..	result = texfetch(tex[i], offset);

}