Hi!
I have an improvement that handles tree-ssa/pr81588.c optimization
even on branch cost 1, but as it is effectively an extension to the
range var_bound optimization, I'm not sure it should be backported.
Therefore, this patch treats this testcase similarly to other
testcases that rely on branch cost of 2 (so that there is BIT_IOR_EXPR
and not 2 separate jumps).
Tested on x86_64-linux, ok for 7.2?
2017-08-02 Jakub Jelinek <ja...@redhat.com>
PR tree-optimization/81655
PR tree-optimization/81588
* gcc.dg/tree-ssa/pr81588.c: Use -mbranch-cost=2 where possible,
don't run the test on branch-cost=1 targets.
--- gcc/testsuite/gcc.dg/tree-ssa/pr81588.c.jj 2017-08-01 10:28:50.000000000
+0200
+++ gcc/testsuite/gcc.dg/tree-ssa/pr81588.c 2017-08-02 11:02:30.966184545
+0200
@@ -1,7 +1,8 @@
/* PR tree-optimization/81588 */
-/* { dg-do compile } */
+/* { dg-do compile { target { ! "m68k*-*-* mmix*-*-* bfin*-*-* v850*-*-*
moxie*-*-* cris*-*-* m32c*-*-* fr30*-*-* mcore*-*-* powerpc*-*-* xtensa*-*-*
hppa*-*-* nios2*-*-*" } } } */
/* { dg-options "-O2 -fdump-tree-reassoc1-details" } */
-
+/* { dg-additional-options "-mbranch-cost=2" { target mips*-*-* avr-*-*
s390*-*-* i?86-*-* x86_64-*-* } } */
+
extern long long int a, c;
extern unsigned short b;
Jakub
