On Wed, Feb 08, 2012 at 08:57:15PM +0100, Thomas Koenig wrote:
> I committed the attached test case after verifying that it passes and
> that only real(4) variables occur (so the requiredment on vect_float
> should be correct.)
This fails e.g. on i686-linux. The problem is that dg-options for */vect/*
tests overrides the standard options, instead of adding to them, so the
-msse2 needed for the test to actually vectorize anything in it is gone.
I think now that we have dg-additional-options, we can use it instead
(at least, the testcase works with that), and perhaps we should eventually
remove the magic testcase names like no-vfa-*, fast-math-*, no-fast-math-*,
wrapv-* etc. and just use dg-additional-options in those vect.exp tests.
Ok for trunk?
2012-02-09 Jakub Jelinek <ja...@redhat.com>
PR fortran/32380
* gfortran.dg/vect/pr32380.f: Use dg-additional-options instead
of dg-options.
--- gcc/testsuite/gfortran.dg/vect/pr32380.f.jj 2012-02-09 12:11:00.000000000
+0100
+++ gcc/testsuite/gfortran.dg/vect/pr32380.f 2012-02-09 12:27:24.860308981
+0100
@@ -1,6 +1,6 @@
! { dg-do compile }
! { dg-require-effective-target vect_float }
-! { dg-options "-O3 -fcray-pointer -ftree-vectorize -fdump-tree-vect-stats" }
+! { dg-additional-options "-O3 -fcray-pointer" }
! PR 32380 - loops were not vectorized due to unaligned store.
subroutine trnfbt(e,f,qs,mte,gm,ihgenf,hgener,lft,llt,sthick,
. fibl,istupd,ies,hoff)
Jakub
