[PATCH v2, middle end]: Fix PR 37908, thinko with atomic NAND operation

[Jakub Jelinek]

On Wed, Oct 29, 2008 at 07:37:08PM +0000, Andrew Haley wrote:
> No, because real-world users might depend on the documented behaviour.

Or maybe not.  I've tried codesearch.google.com on both
__sync_fetch_and_nand and __sync_nand_and_fetch and it didn't reveal
anything outside of GCC (and LLVM) sources.  I know that doesn't imply
no such users exist at all, but just that there probably aren't many
at least in the opensource world.

	Jakub