https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125294
Bug ID: 125294
Summary: -fanalyzer increases compile time by several orders of
magnitude
Product: gcc
Version: 16.1.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: analyzer
Assignee: dmalcolm at gcc dot gnu.org
Reporter: andres at anarazel dot de
Target Milestone: ---
Created attachment 64449
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=64449&action=edit
Reproducer
Hi,
While setting up -fanalyzer to find bugs in postgres, I noticed that the
overall build time with -fanalyzer was dominated by a handful of files.
Some of them are huge complicated files, but others aren't that big. E.g. for
the attached stem_UTF_8_serbian.i I see compile times < 0.01s without
-fanalyzer, and ~120s with -fanalyzer.
andres@alap6:/srv/dev/build/m$ gcc-16 -ftime-report -ftime-report-details -c -o
/dev/null stem_UTF_8_serbian.i
Time variable wall GGC
phase setup : 0.00 ( 3%) 1976k ( 13%)
phase parsing : 0.03 ( 49%) 7452k ( 50%)
phase opt and generate : 0.03 ( 48%) 5392k ( 36%)
preprocessing : 0.01 ( 13%) 1115k ( 8%)
lexical analysis : 0.01 ( 10%) 0 ( 0%)
`- preprocessing : 0.01 ( 13%) 1115k ( 8%)
parser (global) : 0.01 ( 22%) 5187k ( 35%)
expand : 0.00 ( 3%) 1342k ( 9%)
integrated RA : 0.01 ( 9%) 576k ( 4%)
TOTAL : 0.06 14M
andres@alap6:/srv/dev/build/m$ gcc-16 -ftime-report -ftime-report-details
-fanalyzer -c -o /dev/null stem_UTF_8_serbian.i
Time variable wall GGC
phase setup : 0.00 ( 0%) 1976k ( 13%)
phase parsing : 0.03 ( 0%) 7452k ( 50%)
phase opt and generate : 119.24 (100%) 5408k ( 36%)
callgraph optimization : 0.00 ( 0%) 0 ( 0%)
`- analyzer : 0.77 ( 1%) 0 ( 0%)
callgraph ipa passes : 119.21 (100%) 435k ( 3%)
preprocessing : 0.01 ( 0%) 1115k ( 8%)
lexical analysis : 0.01 ( 0%) 0 ( 0%)
`- preprocessing : 0.01 ( 0%) 1115k ( 8%)
parser (global) : 0.01 ( 0%) 5188k ( 35%)
expand : 0.00 ( 0%) 1342k ( 9%)
analyzer : 0.77 ( 1%) 3352 ( 0%)
`- analyzer: processing worklist : 118.44 ( 99%) 13k ( 0%)
analyzer: processing worklist : 118.44 ( 99%) 13k ( 0%)
TOTAL : 119.27 14M
Of course that's with -O0, where non-analyzer gcc doesn't need to do much, but
it's not much slower with -O3 at 0.155s...
I suspect there's some unfortunate quadratic growth going on due to the number
of variables and case statements in that file.
I manually reduced the attached reproducer to remove a bunch of include related
cruft that seems unrelated to the matter, otherwise the .i file was on the too
large side.
Greetings,
Andres
