http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54402
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-10-29
09:48:59 UTC ---
Haven't gathered var-tracking statistics yet, but from quick glance at
i686 reflect_test..import routine it looks like with LRA more memory accesses
are %ebp based, while with old reload they were %esp based:
- movl %eax, 84(%esp)
+ movl %eax, -7592(%ebp)
etc. As the var-tracking problem seems to be in the memory disambiguation code
(added by Alex for 4.8+ for PR49888) perhaps that could have an effect.
LRA code looks shorter (reflect all_tests.go)).
