[Bug rtl-optimization/91332] ICE: Segfault in gfortran when compiling massive subroutine with -O3

2020-03-01 Thread clogged.drainpipe at gmail dot com 
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91332

--- Comment #3 from Madarpok .  ---
Any movement regarding this?

[Bug rtl-optimization/91332] ICE: Segfault in gfortran when compiling massive subroutine with -O3

2019-08-07 Thread clogged.drainpipe at gmail dot com 
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91332

--- Comment #2 from Madarpok .  ---
I can confirm that raising the stack limit with ulimit -s unlimited does not
prevent this crash.
Something inside gfortran is breaking/overflowing.

[Bug rtl-optimization/91332] ICE: Segfault in gfortran when compiling massive subroutine with -O3

2019-08-05 Thread rguenth at gcc dot gnu.org 
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91332

Richard Biener  changed:

   What|Removed |Added

   Keywords||compile-time-hog
 Status|UNCONFIRMED |NEW
   Last reconfirmed||2019-08-05
  Component|fortran |rtl-optimization
 Ever confirmed|0   |1
  Known to fail||7.4.1, 9.1.1

--- Comment #1 from Richard Biener  ---
Try increasing an eventually present stack limit (ulimit -s unlimited).  OTOH
trying to reprodce I see the compile uses >7GB of memory and the stack is
still <16MB.  Later memory usage climbs beyond my machines 32GB memory and
the process is killed...

The testcase somewhat resembles that of PR46590?  That is, it is a sequence
of many single-nested loops.  Trimming the number down to 1000 shows

 load CSE after reload   :  54.39 (78%) usr   0.00 ( 0%) sys  54.39 (77%) wall 
   274 kB ( 0%) ggc
 TOTAL :  69.30 0.9570.26
330192 kB

for GCC 7 and

 dead store elim1   :   8.89 ( 13%)   0.36 ( 36%)   9.26 ( 14%)
1317578 kB ( 36%)
 dead store elim2   :  16.29 ( 25%)   0.21 ( 21%)  16.51 ( 25%)
2195108 kB ( 60%)
 load CSE after reload  :  23.78 ( 36%)   0.00 (  0%)  23.79 ( 35%)
  0 kB (  0%)
 TOTAL  :  66.06  1.00 67.20   
3672823 kB

103.00user 1.40system 1:44.42elapsed 99%CPU (0avgtext+0avgdata
2465980maxresident)k
0inputs+2208outputs (0major+616413minor)pagefaults 0swaps

for GCC 9.  Upping to 2000 loops makes the above (GCC 9)

 forward prop   :  21.74 ( 12%)   0.01 (  1%)  21.74 ( 11%)
   9208 kB (  0%)
 dead store elim1   :  16.92 (  9%)   0.63 ( 35%)  17.54 (  9%)
2670613 kB ( 37%)
 dead store elim2   :  28.27 ( 15%)   0.36 ( 20%)  28.62 ( 15%)
4321137 kB ( 59%)
 load CSE after reload  :  95.49 ( 51%)   0.00 (  0%)  95.50 ( 50%)
  0 kB (  0%)
 TOTAL  : 188.82  1.78190.61   
7300901 kB

188.82user 1.94system 3:10.77elapsed 99%CPU (0avgtext+0avgdata
4846972maxresident)k
0inputs+4304outputs (0major+1222407minor)pagefaults 0swaps

so it really seems to be all RTL-land.