No error return on invalid cuFuncSetBlockShape

I just inadvertently called cuFuncSetBlockShape with arguments of (35, 360, 1), resulting in a block of size 12600, on a GPU with MAX_BLOCKS_PER_THREAD==512. It came back without an error–why?

(CUDA 2.0 on Ubuntu)

I’d say that there’s nothing specifically illegal about setting bigger blocks than the card can handle. Future cards may run bigger blocks. It’s when you ask your specific card to run these blocks that the you get gunk in the works.