From: Richard Sandiford <richard.sandiford@arm.com>
To: Richard Biener via Gcc-patches <gcc-patches@gcc.gnu.org>
Cc: Richard Biener <rguenther@suse.de>
Subject: Re: [PATCH] tree-optimization/105250 - adjust fold_convertible_p PR105140 fix
Date: Wed, 13 Apr 2022 09:04:02 +0100 [thread overview]
Message-ID: <mptzgkpo1e5.fsf@arm.com> (raw)
In-Reply-To: <20220413074636.9269E13A91@imap2.suse-dmz.suse.de> (Richard Biener via Gcc-patches's message of "Wed, 13 Apr 2022 09:46:36 +0200 (CEST)")
Richard Biener via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> The following reverts the original PR105140 fix and goes for instead
> applying the additional fold_convert constraint for VECTOR_TYPE
> conversions also to fold_convertible_p. I did not try sanitizing
> all of this at this point.
>
> Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
>
> 2022-04-13 Richard Biener <rguenther@suse.de>
>
> PR tree-optimization/105250
> * fold-const.cc (fold_convertible_p): Revert
> r12-7979-geaaf77dd85c333, instead check for size equality
> of the vector types involved.
This doesn't look right, and I think it'll break SVE. For one
thing, the tree_int_cst_equal check is bound to fail for
variable-length vectors.
But also, the idea was to allow element-wise conversions between
different vector sizes. For example, you can do a nop/convert
from V4SI to V4DI, which converts 4 SIs to 4 DIs. This is used
a lot for conversions to and from “partial” SVE vectors, where smaller
elements are stored in wider containers.
Thanks,
Richard
>
> * gcc.dg/pr105250.c: New testcase.
> ---
> gcc/fold-const.cc | 7 +++----
> gcc/testsuite/gcc.dg/pr105250.c | 29 +++++++++++++++++++++++++++++
> 2 files changed, 32 insertions(+), 4 deletions(-)
> create mode 100644 gcc/testsuite/gcc.dg/pr105250.c
>
> diff --git a/gcc/fold-const.cc b/gcc/fold-const.cc
> index 7226bc5af01..a57ad0739fb 100644
> --- a/gcc/fold-const.cc
> +++ b/gcc/fold-const.cc
> @@ -2379,13 +2379,12 @@ build_zero_vector (tree type)
> return build_vector_from_val (type, t);
> }
>
> -/* Returns true, if ARG, an operand or a type, is convertible to TYPE
> - using a NOP_EXPR. */
> +/* Returns true, if ARG is convertible to TYPE using a NOP_EXPR. */
>
> bool
> fold_convertible_p (const_tree type, const_tree arg)
> {
> - const_tree orig = TYPE_P (arg) ? arg : TREE_TYPE (arg);
> + const_tree orig = TREE_TYPE (arg);
>
> if (type == orig)
> return true;
> @@ -2417,7 +2416,7 @@ fold_convertible_p (const_tree type, const_tree arg)
> return (VECTOR_TYPE_P (orig)
> && known_eq (TYPE_VECTOR_SUBPARTS (type),
> TYPE_VECTOR_SUBPARTS (orig))
> - && fold_convertible_p (TREE_TYPE (type), TREE_TYPE (orig)));
> + && tree_int_cst_equal (TYPE_SIZE (type), TYPE_SIZE (orig)));
>
> default:
> return false;
> diff --git a/gcc/testsuite/gcc.dg/pr105250.c b/gcc/testsuite/gcc.dg/pr105250.c
> new file mode 100644
> index 00000000000..665dd95d8cb
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/pr105250.c
> @@ -0,0 +1,29 @@
> +/* { dg-do compile } */
> +/* { dg-options "-w -Wno-psabi -O2" } */
> +
> +typedef int __attribute__((__vector_size__(4))) T;
> +typedef int __attribute__((__vector_size__(8))) U;
> +typedef int __attribute__((__vector_size__(16))) V;
> +typedef int __attribute__((__vector_size__(32))) W;
> +typedef _Float32 __attribute__((__vector_size__(16))) F;
> +typedef _Float64 __attribute__((__vector_size__(32))) G;
> +void foo();
> +
> +foo(int, int, int, int, U, U, V, V, W, W, int,
> + T, int, U, U, V, V, W, W, T,
> + T, int, U, U, V, V, W, W, T,
> + T, int, W, W, T, T, int, int, int,
> + int, int, int, W, int, int, int, int, int, int,
> + V, W, T, int, int, U, F, int, int, int,
> + int, int, int, G)
> +{
> + foo(0, 0, 0, 0, (U){}, (U){}, (V){}, (V){}, (W){},
> + (W){}, 2, (T){}, 0, 0, 0, 0, (U){}, (U){},
> + (V){}, (V){}, (W){}, (W){}, (T){},
> + (T){}, 0, 0, 0, 0, (U){}, (U){}, (V){},
> + (V){}, (W){}, (W){}, (T){}, (T){}, 0, 0, 0,
> + 0, 0, 0, (T){},
> + (T){}, (W){},
> + (W){}, (T){}, (T){}, 0, 0, 0, 0, 0, 0, (W){},
> + (V){}, (W){}, (T){}, 0, 0, (U){}, (F){});
> +}
next prev parent reply other threads:[~2022-04-13 8:04 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-04-13 7:46 Richard Biener
2022-04-13 8:04 ` Richard Sandiford [this message]
2022-04-13 8:48 ` Richard Biener
2022-04-13 8:55 ` Richard Biener
2022-04-13 9:06 ` Richard Biener
2022-04-13 10:59 ` Richard Sandiford
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=mptzgkpo1e5.fsf@arm.com \
--to=richard.sandiford@arm.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).