This avoids vectorizing and thus chaning loop trip count.
Tested on x86_64, applied.
Richard.
2015-01-16 Richard Biener
PR tree-optimization/61743
* gcc.dg/tree-ssa/pr61743-1.c: Add -fno-tree-vectorize.
* gcc.dg/tree-ssa/pr61743-2.c: Likewise.
Index: gcc/testsuite/g
I am testing the following hack^Wpatch which makes us preserve
range information in some very special case in PRE. This is
important to get number of iteration analysis results in a way
to enable unrolling of loops in some EEMBC testcase (I don't
have access to that benchmark unless it is part of
