[Bug ipa/103227] 58% exchange2 regression with -Ofast -march=native on zen3 between g:1ae8edf5f73ca5c3 and g:2af63f0f53a12a72

2021-11-13 Thread hubicka at gcc dot gnu.org via Gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103227 Jan Hubicka changed: What|Removed |Added CC||hubicka at gcc dot gnu.org,

[Bug ipa/103227] 58% exchange2 regression with -Ofast -march=native on zen3 between g:1ae8edf5f73ca5c3 and g:2af63f0f53a12a72

2021-11-13 Thread hubicka at gcc dot gnu.org via Gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103227 --- Comment #2 from Jan Hubicka --- There is difference in inlier decision. Since all clones are of same size it depends on the order inliner picks them and combines together before hitting large-function-growth. It seems that with isra orderi

[Bug ipa/103227] 58% exchange2 regression with -Ofast -march=native on zen3 between g:1ae8edf5f73ca5c3 and g:2af63f0f53a12a72

2021-11-13 Thread hubicka at gcc dot gnu.org via Gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103227 --- Comment #3 from Jan Hubicka --- The heuristics here is quite simplistic - it all happens after small function inlining while trying to inline function called once. Each of clones is called once and it really depends on the order we walk the