Array size restriction in cudaMallocArray?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello All,

I need to allocate a large one-dimensional array, say, of million numbers to further use it as a look up table with linear interpolation. but the largest array I could allocate was one of 65536 numbers, and to do that I had to allocate actually a two-dimensional array of 65536x2 size. Otherwise, cudaMallocArray() returns “invalid argument”… So are there any restriction on the array width/height parameters?

2

There are limitations on the size of texture memory ( it was discussed in the forum few days ago).

If you want to allocate a big 1D array, use regular device memory and cudaMalloc.

Massimiliano

3

Got it, thanks. However as I understand, textures bound to the linear device memory do not support linear filtering, which would be very useful for interpolation…