[Bug tree-optimization/52734] [4.7/4.8 Regression] Incorrect optimization of uClibc sbrk()

[vries at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52734

--- Comment #10 from vries at gcc dot gnu.org 2012-03-27 13:19:12 UTC ---
Created attachment 27014
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27014
Tentative patch

The key element of this patch is:
...
+  if (gvn_used && vuse1 != vuse2)
+    return;
+
...

If gvn was use to prove two bbs equal, and the incoming vuses are not equal,
don't do the substitution. This is a bit too restrictive, I'm pondering about a
more precise fix.