From: Ira Rosen <IRAR@il.ibm.com>
To: Richard Guenther <richard.guenther@gmail.com>
Cc: gcc-patches@gcc.gnu.org
Subject: Re: [patch] Fix PR tree-optimization/45902
Date: Mon, 11 Oct 2010 13:05:00 -0000 [thread overview]
Message-ID: <OFA2C5EF1B.57DA87D7-ONC22577B9.00450843-C22577B9.0047BE43@il.ibm.com> (raw)
In-Reply-To: <AANLkTikeCWFKmyPo6U-aj9yksurWQB71gAN6mzYiC23x@mail.gmail.com>
Richard Guenther <richard.guenther@gmail.com> wrote on 11/10/2010 02:32:08
PM:
>
> On Mon, Oct 11, 2010 at 2:19 PM, Ira Rosen <IRAR@il.ibm.com> wrote:
> >
> > Hi,
> >
> > This patch fixes a bug in creation of a vector of constants in SLP. The
> > problem is in the type of created vector. It should be set to the
vector
> > type of the statement unless it's a pointer.
> >
> > Bootstrapped and tested on x86_64-suse-linux and checked that the
failure
> > is fixed on powerpc64-suse-linux.
> >
> > Committed to mainline, ok for 4.5?
>
> I don't think this makes sense. If at all the selection of which type
> to use should be based on the operation code and the operand
> position, but not on CONSTANT_CLASS_P or pointer-type-ness.
>
> So - what operation code / operand position is currently mishandled?
This function creates vectors of constants or loop invariants. If it's a
variable (loop invariant), we know exactly the type, hence the check for
CONSTANT_CLASS_P.
I guess instead of checking the type we can check if it's POINTER_PLUS_EXPR
to catch the cases where we need the operand's type instead of the stmt's
vectype.
Does this make sense?
Index: tree-vect-slp.c
===================================================================
--- tree-vect-slp.c (revision 165302)
+++ tree-vect-slp.c (working copy)
@@ -1836,10 +1836,10 @@ vect_get_constant_vectors (slp_tree slp_
VEC (tree, heap) *voprnds = VEC_alloc (tree, heap, number_of_vectors);
bool constant_p, is_store;
tree neutral_op = NULL;
+ enum tree_code code = gimple_assign_rhs_code (stmt);
if (STMT_VINFO_DEF_TYPE (stmt_vinfo) == vect_reduction_def)
{
- enum tree_code code = gimple_assign_rhs_code (stmt);
if (reduc_index == -1)
{
VEC_free (tree, heap, *vec_oprnds);
@@ -1895,18 +1895,14 @@ vect_get_constant_vectors (slp_tree slp_
}
if (CONSTANT_CLASS_P (op))
- {
- constant_p = true;
- if (POINTER_TYPE_P (TREE_TYPE (gimple_assign_lhs (stmt))))
- vector_type = get_vectype_for_scalar_type (TREE_TYPE (op));
- else
- vector_type = STMT_VINFO_VECTYPE (stmt_vinfo);
- }
+ constant_p = true;
else
- {
- constant_p = false;
- vector_type = get_vectype_for_scalar_type (TREE_TYPE (op));
- }
+ constant_p = false;
+
+ if (code == POINTER_PLUS_EXPR)
+ vector_type = get_vectype_for_scalar_type (TREE_TYPE (op));
+ else
+ vector_type = STMT_VINFO_VECTYPE (stmt_vinfo);
gcc_assert (vector_type);
nunits = TYPE_VECTOR_SUBPARTS (vector_type);
Thanks,
Ira
next prev parent reply other threads:[~2010-10-11 13:03 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-10-11 12:32 Ira Rosen
2010-10-11 12:56 ` Richard Guenther
2010-10-11 13:05 ` Ira Rosen [this message]
2010-10-11 14:09 ` Richard Guenther
2010-10-11 20:55 ` Ira Rosen
2010-10-12 10:13 ` Richard Guenther
2010-10-13 7:52 ` Ira Rosen
2010-10-16 18:48 ` H.J. Lu
2010-10-17 0:30 ` H.J. Lu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=OFA2C5EF1B.57DA87D7-ONC22577B9.00450843-C22577B9.0047BE43@il.ibm.com \
--to=irar@il.ibm.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=richard.guenther@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).