RFA: patch to solve PR38280
Mon Dec 1 16:29:00 GMT 2008
Vladimir Makarov wrote:
> The PR is described on
> In my patch
> I overlooked that a specific order of allocnos in regno_allocno_map
> list can be violated after removing some loops from RA regions. The
> bug occurs in rare cases only when # of loops in function is > 100.
> There are a lot such functions in SPEC2006. So spec2006 is a good
> quality test. It is very hard to make a test for gcc testsuite
> reproducing the bug.
> Is it ok to commit? The patch was successfully bootstrapped on x86,
> x86_64, itanium, and ppc.
> 2008-11-27 Vladimir Makarov <email@example.com>
> PR rtl-optimization/38280
> * ira-build.c (loop_is_inside_p, regno_allocno_order_compare_func,
> ira_rebuild_regno_allocno_list): New functions.
> (regno_allocnos): New static variable.
> (remove_unnecessary_allocnos): Allocate/deallocate regno_allocnos.
> Call ira_rebuild_regno_allocno_list.
More information about the Gcc-patches