[Bug rtl-optimization/86685] New: [8/9 Regression] 436.cactusADM regression on aarch64

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86685

            Bug ID: 86685
           Summary: [8/9 Regression] 436.cactusADM regression on aarch64
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: rtl-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ktkachov at gcc dot gnu.org
                CC: rsandifo at gcc dot gnu.org
  Target Milestone: ---

For GCC 8 since r254378 we use SCHED_PRESSURE_MODEL in aarch64 during
scheduling. This gives an improvement in SPEC2017 but we have noticed it
regresses 436.cactusADM from SPEC2006 significantly. On a Cortex-A72 we see a
10% regression from that change.

The vector loop in the hot function bench_staggeredleapfrog2_ contains
significantly more stack spills. I appreciate the goal of that scheduling model
isn't necessarily to minimise stack spills, but it should do a better job at
keeping them under reasonable control

Given that SCHED_PRESSURE_MODEL works quite well for us in other cases it would
be good to fix it going haywire here to get the performance back up.