LNT shows what looks like a clear recent 6.5% regression on the 464.h264ref benchmark from SPEC 2006 CPU suite when compiled with PGO, LTO, -Ofast and -march=native on a Neoverse-N1 CPU: https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=586.220.0 In the same time period, I cannot (immediately) see any similar regressions on any other CPU (all others that we monitor are x86_64) or option combination. Unfortunately I don't have the time and resources to further investigate, but the commit range bad177e848787258..2fc55f51f9953b45 includes 8e26ac4749c (AArch64: Fix codegen regressions around tbz), among other things.
I can't reproduce that. on a Neoverse-N1 I see between those two commits: ./bench-compare.sh 2fc55f51f99 bad177e8487 A 1457 files D 0 files M 0 files Extracted 'loose/2fc55f51f99:2fc55f51f99' A 1457 files D 0 files M 0 files Extracted 'loose/bad177e8487:bad177e8487' difference: 0.083458946886789971215830257097% Compiled with -Ofast -march=native -g -flto=auto and PGO enabled. as extracted from https://lnt.opensuse.org/db_default/v4/SPEC/34263?show_stddev=yes&show_all=yes&show_all_samples=yes&num_comparison_runs=0&test_filter=&test_min_value_filter=&aggregation_fn=min&MW_confidence_lv=0.05&compare_to=34183&baseline=23835&submit=Update our non PGO runs also didn't see a regression.
GCC 13.1 is being released, retargeting bugs to GCC 13.2.
GCC 13.2 is being released, retargeting bugs to GCC 13.3.
GCC 13.3 is being released, retargeting bugs to GCC 13.4.