https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119465
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 60877
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60877&action=edit
gcc15-pr119465.patch
That change actually looked reasonable, having long/unsigned long type used and
the shifts going on with 64 - n was definitely weird.
And the testcase actually succeeds e.g. with -m32 -march=x86-64, in that case
some of those are done using SSE2 instructions and not scanned for.
I think we should just disable SSE/MMX and then the instruction count is the
same on all arches.
