cuda-fixnum
cuda-fixnum copied to clipboard
Use warp votes to branch on argument size to select fastest algo
Something like this for example:
if (__all(bits < digit::BITS))
algo_small_params(...);
else
algo_generic(...);