[Bug c/50693] New: Slightly different loop body leads to 5.5x slower performance

["alex.gaynor at gmail dot com" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50693

             Bug #: 50693
           Summary: Slightly different loop body leads to 5.5x slower
                    performance
    Classification: Unclassified
           Product: gcc
           Version: 4.6.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: alex.gaynor@gmail.com


Created attachment 25460
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25460
Script which reproduces the speed difference.

Given the two loops in the attached file, the first generates excellent code,
which performs the same as a memset, however the second results in very poor
code which is 5.5x slower than the first. If anyone is curious why we have such
strange looking code: a compiler which targets C.

This was tested by compiling at -O3 with -std=gnu99 on my machine:


alex@alex-gaynor-laptop:/tmp$ gcc -O3 test.c -std=gnu99
alex@alex-gaynor-laptop:/tmp$ time ./a.out > /dev/null 

real    0m0.427s
user    0m0.412s
sys    0m0.016s
alex@alex-gaynor-laptop:/tmp$ time ./a.out > /dev/null 

real    0m0.428s
user    0m0.416s
sys    0m0.008s
alex@alex-gaynor-laptop:/tmp$ time ./a.out > /dev/null 

real    0m0.432s
user    0m0.404s
sys    0m0.024s
alex@alex-gaynor-laptop:/tmp$ 
alex@alex-gaynor-laptop:/tmp$ time ./a.out 0 > /dev/null 

real    0m2.225s
user    0m2.200s
sys    0m0.020s
alex@alex-gaynor-laptop:/tmp$ time ./a.out 0 > /dev/null 

real    0m2.217s
user    0m2.196s
sys    0m0.016s
alex@alex-gaynor-laptop:/tmp$ time ./a.out 0 > /dev/null 

real    0m2.268s
user    0m2.252s
sys    0m0.012s