diff options
| author | Jakub Jelinek <jakub@redhat.com> | 2021-05-18 10:08:51 +0200 |
|---|---|---|
| committer | Jakub Jelinek <jakub@redhat.com> | 2021-05-18 10:08:51 +0200 |
| commit | 65061ea287a80cfb214e402cfd2373a14bfec95a (patch) | |
| tree | 1f99a7d4ffd69642018df6da8b634301e0fce7ef /gcc/function.c | |
| parent | b2a09773155892e426b90ce8367e2ed6996b44ef (diff) | |
| download | gcc-65061ea287a80cfb214e402cfd2373a14bfec95a.zip gcc-65061ea287a80cfb214e402cfd2373a14bfec95a.tar.gz gcc-65061ea287a80cfb214e402cfd2373a14bfec95a.tar.bz2 | |
phiopt: Optimize partial_ordering spaceship >= 0 -ffinite-math-only [PR94589]
As mentioned earlier, spaceship_replacement didn't optimize partial_ordering
>= 0 comparisons, because the possible values are -1, 0, 1, 2 and the
>= comparison is implemented as (res & 1) == res to choose the 0 and 1
cases from that. As we optimize that only with -ffinite-math-only, the
2 case is assumed not to happen and my earlier match.pd change optimizes
(res & 1) == res into (res & ~1) == 0, so this patch pattern matches
that case and handles it like res >= 0.
2021-05-18 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/94589
* tree-ssa-phiopt.c (spaceship_replacement): Pattern match
phi result used in (res & ~1) == 0 comparison as res >= 0 as
res == 2 would be UB with -ffinite-math-only.
* g++.dg/opt/pr94589-2.C: Adjust scan-tree-dump count from 14 to 12.
Diffstat (limited to 'gcc/function.c')
0 files changed, 0 insertions, 0 deletions