On Fri, 26 Apr 2019, Martin Liška wrote:
> I'm suggesting to adjust that to:
> - -Os will keep using rep-scasb as -Os means optimize for size
> no matter what speed impact is
I'm not sure it's a good choice, the inline sequence is
xorl %eax, %eax
orq $-1, %rcx
repnz scasb
notq %rcx
decq %rcx
compared to simply
call strlen
it's not even shorter.
> - otherwise use call to strlen
> - when -minline-all-stringops is enabled and -O2+ is used, then
> expand to 4B loop for all possible alignments
But it's not OK to use misaligned loads, because overreading past
end of string may cross a page boundary and cause a segfault. What does
your patch generate for the testcase at -O2 -minline-all-stringops?
Alexander
