[Bug tree-optimization/126545] [16/17 Regression] Wrong code with ranger and FP comparisons and NaN handling since r16-1191
"jakub at gcc dot gnu.org via Gcc-bugs" <[email protected]>
| Newsgroups | gmane.comp.gcc.bugs |
|---|---|
| Message-ID | <[email protected]/bugzilla/> |
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126545
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |aldyh at gcc dot gnu.org,
| |amacleod at redhat dot com
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Guess this is because we haven't added VREL_LTGT, VREL_{,UN}ORDERED,
VREL_UN{LT,LE,GT,GE,EQ} which were discussed back in 2023, the
https://gcc.gnu.org/pipermail/gcc-patches/2023-August/thread.html#628262
thread. The relation is VREL_LT y, x on one edge, but not VREL_GE on the
other, but VREL_UNGE (unless the ranges don't allow NaNs for either operands).
y can be NaN, so from VREL_UNGE y, x and VREL_UNGE z, y we can't imply VREL_GE
z, x.