https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119010
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|ASSIGNED |NEW
Priority|P1 |P2
Summary|[15 Regression] 444.namd |[12/13/14/15 Regression]
|shows a huge compile-time |444.namd shows a huge
|regression with |compile-time regression
|-mtune=znver5 |with -mtune=znver5
Target Milestone|15.0 |12.5
Assignee|rguenth at gcc dot gnu.org |unassigned at gcc dot
gnu.org
--- Comment #14 from Richard Biener <rguenth at gcc dot gnu.org> ---
We lack any reservations for "sse" (sqrtpd) and "sseishift" (avx2_ashlvv8si).
We also still have DF moves w/o reservation, not fixed after my last avalanche
of patches :/
I'm unassigning myself now, the P1 compile-time-hog is fixed (which needed the
issue_rate bump which IIRC isn't on branches?), the bad state of the automaton
is now only a regression with -march=native but on all active branches we
backported it to. I'm not sure to what extend we want to fix things there, I'd
suggest to SPEC-test a set of backports to GCC 14 to see if random insn changes
pop up anything.
