------- Comment #18 from sebpop at gmail dot com 2007-06-26 23:54 -------
Subject: Re: [4.3 Regression] Complete program optimized away (i686,
-ftree-vectorize)
Hi,
> I'm bootstrapping this patch on both amd64-linux and i686-linux.
The patch passed bootstrap and test on both machines with no new
regressions. I had to modify the checking line for the testcase, as I
used -fdump-vect-all-all to prepare the testcase, and when I checked
the vectorizer dump of the checker, it was only two lines long:
/home/pop/gcc/trunk/gcc/testsuite/gfortran.dg/vect/pr32457.f90:6:
note: not vectorized: can't determine dependence between
(*rbound_9(D))[D.137
7_8] and (*rbound_9(D))[D.1375_5]
/home/pop/gcc/trunk/gcc/testsuite/gfortran.dg/vect/pr32457.f90:4:
note: vectorized 0 loops in function.
So I just removed the check for the second pattern:
! { dg-final { scan-tree-dump-times "possible dependence between
data-refs" 1 "vect" } }
and the new version of the testcase is like this:
! { dg-do compile }
! { dg-require-effective-target vect_float }
SUBROUTINE KEEL(RBOUND)
REAL, DIMENSION(0:100) :: RBOUND
DO N = 1, NP1
RBOUND(N) = RBOUND(N-1) + 1
END DO
DO N = 1, NS
WRITE (16,'(I5)') SRAD(N)
END DO
END SUBROUTINE KEEL
! { dg-final { scan-tree-dump-times "vectorized 0 loops" 1 "vect" } }
! { dg-final { cleanup-tree-dump "vect" } }
Sebastian
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32457
