In the programming guide, I saw “Maximum width for a 1D texture reference bound to linear memory” is 2^27.
But my program,
cudaBindTexture(NULL, texRef, packet.pk.dev, maxDataSize);
ch = tex1Dfetch(texRef, base + i);
I used such code. but if base + i is sufficiently big (not big as maxDataSize ~ 2048 * 4096 = 2^23), it gives me an error.
And if I see the cuda-gdb, it seems like the bug in texture memory.
But packet.pk.dev is global memory. So by the programming guide, it is less then 2^27. I don’t know why it gives me an error?
DO you have any idea?