[PATCH] Fix up vector CONSTRUCTOR verification ICE (PR tree-optimization/54889)
Jakub Jelinek
jakub@redhat.com
Tue Oct 16 15:23:00 GMT 2012
On Mon, Oct 15, 2012 at 10:38:02AM +0200, Richard Biener wrote:
> On Fri, Oct 12, 2012 at 8:16 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> > Apparently vectorizable_load is another spot that could create vector
> > CONSTRUCTORs that wouldn't pass the new CONSTRUCTOR verification.
> >
> > Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux,
> > ok for trunk?
>
> You should only need this on the ARRAY_REF path (I wonder what are
> the types that have a mismatch?), for MEM_REF simply use
> build2 (MEM_REF, TREE_TYPE (vectype), ...).
>
> Ok with that change.
Thanks, here is what I'll commit later today, bootstrapped/regtested on
x86_64-linux and i686-linux:
2012-10-15 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/54889
* tree-vect-stmts.c (vectorizable_load): Add VIEW_CONVERT_EXPR if
ARRAY_REF newref doesn't have compatible type with vectype element
type, use vectype element type for MEM_REF.
* gfortran.dg/pr54889.f90: New test.
--- gcc/tree-vect-stmts.c.jj 2012-10-12 20:18:47.568041703 +0200
+++ gcc/tree-vect-stmts.c 2012-10-15 11:47:57.874019752 +0200
@@ -4743,12 +4743,18 @@ vectorizable_load (gimple stmt, gimple_s
tree newref, newoff;
gimple incr;
if (TREE_CODE (ref) == ARRAY_REF)
- newref = build4 (ARRAY_REF, TREE_TYPE (ref),
- unshare_expr (TREE_OPERAND (ref, 0)),
- running_off,
- NULL_TREE, NULL_TREE);
+ {
+ newref = build4 (ARRAY_REF, TREE_TYPE (ref),
+ unshare_expr (TREE_OPERAND (ref, 0)),
+ running_off,
+ NULL_TREE, NULL_TREE);
+ if (!useless_type_conversion_p (TREE_TYPE (vectype),
+ TREE_TYPE (newref)))
+ newref = build1 (VIEW_CONVERT_EXPR, TREE_TYPE (vectype),
+ newref);
+ }
else
- newref = build2 (MEM_REF, TREE_TYPE (ref),
+ newref = build2 (MEM_REF, TREE_TYPE (vectype),
running_off,
TREE_OPERAND (ref, 1));
--- gcc/testsuite/gfortran.dg/pr54889.f90.jj 2012-10-15 11:46:13.252593971 +0200
+++ gcc/testsuite/gfortran.dg/pr54889.f90 2012-10-15 11:46:13.252593971 +0200
@@ -0,0 +1,10 @@
+! PR tree-optimization/54889
+! { dg-do compile }
+! { dg-options "-O3" }
+! { dg-additional-options "-mavx" { target { i?86-*-* x86_64-*-* } } }
+
+subroutine foo(x,y,z)
+ logical, pointer :: x(:,:)
+ integer :: y, z
+ x=x(1:y,1:z)
+end subroutine
Jakub
More information about the Gcc-patches
mailing list