https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101720
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Component|rtl-optimization |middle-end
Last reconfirmed| |2021-08-02
Ever confirmed|0 |1
--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
For the C++ testcase I see (-O2 -g):
callgraph ipa passes : 32.87 ( 18%) 4.33 ( 30%) 37.20 ( 19%)
1103M ( 17%)
template instantiation : 14.95 ( 8%) 2.60 ( 18%) 17.37 ( 9%)
1110M ( 17%)
integration : 8.35 ( 5%) 1.47 ( 10%) 10.26 ( 5%)
854M ( 13%)
variable tracking : 2.35 ( 1%) 0.02 ( 0%) 2.44 ( 1%)
98M ( 2%)
var-tracking dataflow : 2.67 ( 1%) 0.02 ( 0%) 2.91 ( 1%)
1150k ( 0%)
var-tracking emit : 3.24 ( 2%) 0.03 ( 0%) 3.03 ( 2%)
122M ( 2%)
TOTAL : 184.28 14.25 198.55
6365M
so it's a quite flat profile and dropping -g from -O2 -g changes it like:
callgraph ipa passes : 27.96 ( 20%) 3.51 ( 31%) 31.48 ( 21%)
888M ( 21%)
template instantiation : 14.53 ( 10%) 2.51 ( 22%) 17.01 ( 11%)
1105M ( 26%)
integration : 4.91 ( 4%) 0.62 ( 6%) 5.55 ( 4%)
596M ( 14%)
TOTAL : 140.12 11.26 151.39
4224M
so there is probably some extra non-trivial overhead to debug stmts that
would need to be profiled in more detail. ('integration' is inline
instantiation) Adding -fno-var-tracking-assignments to -O2 -g reduced the
compile-time to 155s.
