[RFC] tree combiner

[Andrew Pinski <pinskia at physics dot uc dot edu>]

I am only posting this as a RFC since right now there are a couple 
regressions
due to two latent bugs.  One in fold-const.c which I filed as PR 21850. 
 The
other is that tree-vrp aborts if we don't have either a comparision tree
or SSA_NAME in a COND_EXPR which is not true, we can anything which make
is_gimple_condexpr return true which includes constants.  Now you might 
be
wondering why we still have constants in the COND_EXPR, I don't know.
Cleanup CFG is called but for some reason does not clean it up fully.

Also this does not fully replace forwprop because we don't do a couple
of optimizations in fold and I don't handle also combining references
as of yet and we would run into another latent bug which I wrote up
in <http://gcc.gnu.org/ml/gcc/2005-05/msg01362.html>.

I got tried of looking into the latent bugs so I decided to post the 
patch
as everyone is waiting for it to be posted.

Note I did not update the docs for the passes as of yet.

Thanks,
Andrew Pinski

ChangeLog:

	* Makefile.in (OBJS-common): Add tree-combine.c
	* tree-combine.c: New file.
	* tree-optimize.c (init_tree_optimization_passes): Add 
pass_tree_combine
	after the first and last forwprop.
	* tree-pass.h (pass_tree_combine): Define.

Attachment: combine.diff.txt
Description: Text document