[Bug tree-optimization/46590] long compile time with -O2 and many loops

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46590

--- Comment #42 from Richard Biener <rguenth at gcc dot gnu.org> ---
Another interesting bit - -O0 takes 17s (and also ~1GB memory) but -Og is
not much faster than -O1 (116s).

> ./f951 -quiet -Og t2.f90 -ftime-report 

Execution times (seconds)
 df reaching defs        :  62.51 (54%) usr 104.47 (85%) sys 167.73 (70%) wall 
     0 kB ( 0%) ggc

bah.  I didn't constrain RTL optimizers with -Og (seems to be RTL loop
invariant motion here - after all at -Og no gimple loop opts run and
it _can_ move quite some invariants).

Disabling loop2 for -Og improves this to 50s

Execution times (seconds)
 alias stmt walking      :  10.11 (20%) usr   0.10 (14%) sys  10.29 (20%) wall 
   338 kB ( 0%) ggc
 parser (global)         :   2.34 ( 5%) usr   0.06 ( 8%) sys   2.39 ( 5%) wall 
 70763 kB (11%) ggc
 tree PTA                :   6.12 (12%) usr   0.04 ( 6%) sys   6.16 (12%) wall 
  2735 kB ( 0%) ggc
 loop init               :   5.78 (11%) usr   0.01 ( 1%) sys   5.78 (11%) wall 
 24078 kB ( 4%) ggc
 integrated RA           :   2.68 ( 5%) usr   0.04 ( 6%) sys   2.73 ( 5%) wall 
 64596 kB (10%) ggc
 TOTAL                 :  50.76             0.71            51.71            
646243 kB

I'll push a patch to do that (the above shows PTA which also can be
disabled, and alias stmt walking could be limited some more).