Currently I’m implementing CUFFT in a big software package. The software package came with a test program for FFT. The CUFFT failed as the test program was passing an input array of size 1 to be calculated by CUFFT. Yea I know that it doesn’t really make sense to calculate FFT of array with size 1, but I still kinda expect it to give the correct answer (even if it is trivial) instead of giving me the CUFFT_INVALID_PLAN error.
So is this considered a bug? From what I can see, this can be addressed by a simple IF case and doesn’t need to go through the kernel code when the input size is 1. I guessed programmer shouldn’t be expected to be aware of such limitation of CUFFT.