https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120360
--- Comment #3 from Linus Torvalds <torva...@linux-foundation.org> ---
(In reply to Linus Torvalds from comment #2)
>
> Stupid test-case:
Actually, that was a bit *too* stupid, since it doesn't show the effect of "we
still need the original value afterwards".
So change it to
void failure(void *ptr)
{
if (MY_IS_ERR_OR_NULL(ptr)) do_something(ptr);
}
to show why the 'add' is bad because it generates extra pointless register
pressure:
movq %rdi, %rax
addq $4095, %rax
jns .L4
(and obviously in this test-case that does nothing else, there is no other
register use, so the "register pressure" is a non-issue and it all just results
in one pointless "mov" instruction - but on real code it tends to result in
more spills and just odd code).
