Message ID | CAO2gOZWsxUcxy4Azjyhbr9qmv4nUzr5zw7hmKqwWK2_Yz4k6JQ@mail.gmail.com |
---|---|
State | New |
Headers | show |
> This patch makes sure max count is used when merging two basic blocks. > > Bootstrapped and testing on-going. > > OK for trunk if test is ok? > > Thanks, > Dehao > > gcc/ChangeLog: > 2014-05-16 Dehao Chen <dehao@google.com> > > * tree-cfg.c (gimple_merge_blocks): Updates bb count with max count. > > Index: gcc/tree-cfg.c > =================================================================== > --- gcc/tree-cfg.c (revision 210531) > +++ gcc/tree-cfg.c (working copy) > @@ -1877,6 +1877,15 @@ gimple_merge_blocks (basic_block a, basic_block b) > } > } > > + /* When merging two BBs, if their counts are different, the larger count > + is selected as the new bb count. This is to handle inconsistent > + profiles. */ > + if (a->count != b->count) > + { > + a->count = MAX (a->count, b->count); > + a->frequency = MAX (a->frequency, b->frequency); > + } I would do max of frequency indpendently on count. I.e. just remove the conditional around the MAX operation. This seems to make sense even w/o profile feedback. Honza
> > This patch makes sure max count is used when merging two basic blocks. > > > > Bootstrapped and testing on-going. > > > > OK for trunk if test is ok? > > > > Thanks, > > Dehao > > > > gcc/ChangeLog: > > 2014-05-16 Dehao Chen <dehao@google.com> > > > > * tree-cfg.c (gimple_merge_blocks): Updates bb count with max count. > > > > Index: gcc/tree-cfg.c > > =================================================================== > > --- gcc/tree-cfg.c (revision 210531) > > +++ gcc/tree-cfg.c (working copy) > > @@ -1877,6 +1877,15 @@ gimple_merge_blocks (basic_block a, basic_block b) > > } > > } > > > > + /* When merging two BBs, if their counts are different, the larger count > > + is selected as the new bb count. This is to handle inconsistent > > + profiles. */ > > + if (a->count != b->count) > > + { > > + a->count = MAX (a->count, b->count); > > + a->frequency = MAX (a->frequency, b->frequency); > > + } > > I would do max of frequency indpendently on count. I.e. just remove the conditional around > the MAX operation. > This seems to make sense even w/o profile feedback. And patch is OK with that change. Honza > > Honza
Index: gcc/tree-cfg.c =================================================================== --- gcc/tree-cfg.c (revision 210531) +++ gcc/tree-cfg.c (working copy) @@ -1877,6 +1877,15 @@ gimple_merge_blocks (basic_block a, basic_block b) } } + /* When merging two BBs, if their counts are different, the larger count + is selected as the new bb count. This is to handle inconsistent + profiles. */ + if (a->count != b->count) + { + a->count = MAX (a->count, b->count); + a->frequency = MAX (a->frequency, b->frequency); + } + /* Merge the sequences. */ last = gsi_last_bb (a); gsi_insert_seq_after (&last, bb_seq (b), GSI_NEW_STMT);