From: Jakub Jelinek <jakub@redhat.com>
To: Richard Biener <rguenther@suse.de>
Cc: gcc-patches@gcc.gnu.org
Subject: [PATCH] match.pd: Fix up gcc.dg/pr54346.c on i686-linux [PR54346]
Date: Thu, 20 Oct 2022 14:48:17 +0200 [thread overview]
Message-ID: <Y1FDkdIxfNGPH7KZ@tucnak> (raw)
Hi!
The pr54346.c testcase FAILs on i686-linux (without -msse*) for multiple
reasons. One is the trivial missing -Wno-psabi which the following patch
adds, but that isn't enough. The thing is that without native vector
support, we have VEC_PERM_EXPRs in the IL and are actually considering
the nested VEC_PERM_EXPRs into one VEC_PERM_EXPR optimization, but punt
because can_vec_perm_const_p (result_mode, op_mode, sel2, false) is false.
Such a test makes sense to prevent "optimizing" two VEC_PERM_EXPRs
that can be handled by the backend natively into one VEC_PERM_EXPR
that can't be handled. But if both of the original VEC_PERM_EXPRs
can't be handled natively either, having just one VEC_PERM_EXPR that will be
lowered by generic vec lowering is IMHO still better than 2.
Or even if we trade just one VEC_PERM_EXPR that can't be handled plus
one that can to one that can't be handled.
Lightly tested so far, ok for trunk if it passes full bootstrap/regtest
on x86_64-linux and i686-linux?
BTW, the testcase also needs to have executable permissions removed...
2022-10-20 <jakub@redhat.com>
PR tree-optimization/54346
* match.pd ((vec_perm (vec_perm@0 @1 @2 VECTOR_CST) @0 VECTOR_CST)):
Optimize nested VEC_PERM_EXPRs even if target can't handle the
new one provided we don't increase number of VEC_PERM_EXPRs the
target can't handle.
* gcc.dg/pr54346.c: Add -Wno-psabi to dg-options.
--- gcc/match.pd.jj 2022-10-19 11:28:35.111654555 +0200
+++ gcc/match.pd 2022-10-20 13:45:57.489512189 +0200
@@ -8118,7 +8118,16 @@ and,
vec_perm_indices sel2 (builder2, 2, nelts);
tree op0 = NULL_TREE;
- if (can_vec_perm_const_p (result_mode, op_mode, sel2, false))
+ /* If the new VEC_PERM_EXPR can't be handled but both
+ original VEC_PERM_EXPRs can, punt.
+ If one or both of the original VEC_PERM_EXPRs can't be
+ handled and the new one can't be either, don't increase
+ number of VEC_PERM_EXPRs that can't be handled. */
+ if (can_vec_perm_const_p (result_mode, op_mode, sel2, false)
+ || (single_use (@0)
+ ? (!can_vec_perm_const_p (result_mode, op_mode, sel0, false)
+ || !can_vec_perm_const_p (result_mode, op_mode, sel1, false))
+ : !can_vec_perm_const_p (result_mode, op_mode, sel1, false)))
op0 = vec_perm_indices_to_tree (TREE_TYPE (@4), sel2);
}
(if (op0)
--- gcc/testsuite/gcc.dg/pr54346.c.jj 2022-10-11 10:00:07.456124822 +0200
+++ gcc/testsuite/gcc.dg/pr54346.c 2022-10-20 13:46:10.933330119 +0200
@@ -1,5 +1,5 @@
/* { dg-do compile } */
-/* { dg-options "-O -fdump-tree-dse1" } */
+/* { dg-options "-O -fdump-tree-dse1 -Wno-psabi" } */
typedef int veci __attribute__ ((vector_size (4 * sizeof (int))));
Jakub
next reply other threads:[~2022-10-20 12:48 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-10-20 12:48 Jakub Jelinek [this message]
2022-10-20 13:05 ` Richard Biener
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=Y1FDkdIxfNGPH7KZ@tucnak \
--to=jakub@redhat.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=rguenther@suse.de \
/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).