[Bug middle-end/57370] New: [4.9 Regression] compile time hog in reassoc

[Joost.VandeVondele at mat dot ethz.ch]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57370

            Bug ID: 57370
           Summary: [4.9 Regression] compile time hog in reassoc
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: Joost.VandeVondele at mat dot ethz.ch

Between rev 199047 and 199075 the attached testcase started using an excessive
amount of time (up from about 4s with 4.7.2). The important option seems to be
-ffast-math (16s without this option).

Reproduce with:

time gfortran -c -g -O2 -ffast-math bug.f90

The stack looks like:

#0  0x0000000000af4b79 in reassociate_bb(basic_block_def*) ()
#1  0x0000000000af4ae8 in reassociate_bb(basic_block_def*) ()
#2  0x0000000000af4ae8 in reassociate_bb(basic_block_def*) ()
#3  0x0000000000af4ae8 in reassociate_bb(basic_block_def*) ()
#4  0x0000000000af4ae8 in reassociate_bb(basic_block_def*) ()
#5  0x0000000000af6701 in execute_reassoc() ()
#6  0x0000000000906550 in execute_one_pass(opt_pass*) ()