[pretty-ipa, PATCH 1/3] Make new intra-SRA and IPA-SRA work after a-i merge

Martin Jambor mjambor@suse.cz
Tue Apr 7 15:49:00 GMT 2009


Hi,

On Tue, Apr 07, 2009 at 10:07:06AM +0200, Richard Guenther wrote:
> On Mon, Apr 6, 2009 at 8:32 PM, Martin Jambor <mjambor@suse.cz> wrote:
> > Note that IPA-SRA is still disabled for benchmarking purposes.  But
> > this patch makes it bootstrap and test ok if uncommented.
> 
> Looks good.  Btw,
> 
> > 2009-04-06  Martin Jambor  <mjambor@suse.cz>
> >
> >        * tree-ssa-alias.c (refs_may_alias_p_1): Check for
> >        is_gimple_min_invariant rather than CONSTANT_CLASS_P so that invariant
> >        ADDR_EXPRS are include too.
> 
> this change is also ok for trunk.

OK, so I have bootstrapped and tested the following on trunk on
x86_64-linux and committed it as revision 145685.

No testcases, but again my SRA will not even bootstrap without this.

Thanks,

Martin



2009-04-07  Martin Jambor  <mjambor@suse.cz>

	* tree-ssa-alias.c (refs_may_alias_p_1): Check for
	is_gimple_min_invariant rather than CONSTANT_CLASS_P so that invariant
	ADDR_EXPRS are include too.


Index: mine/gcc/tree-ssa-alias.c
===================================================================
--- mine.orig/gcc/tree-ssa-alias.c
+++ mine/gcc/tree-ssa-alias.c
@@ -688,9 +688,9 @@ refs_may_alias_p_1 (tree ref1, tree ref2
      *D.1663_44 = VIEW_CONVERT_EXPR<struct DB_LSN>(__tmp$B0F64_59);
      which is seen as a struct copy.  */
   if (TREE_CODE (base1) == SSA_NAME
-      || CONSTANT_CLASS_P (base1)
       || TREE_CODE (base2) == SSA_NAME
-      || CONSTANT_CLASS_P (base2))
+      || is_gimple_min_invariant (base1)
+      || is_gimple_min_invariant (base2))
     return false;
 
   var1_p = SSA_VAR_P (base1);



More information about the Gcc-patches mailing list