From: Hongtao Liu <crazylht@gmail.com>
To: Richard Biener <richard.guenther@gmail.com>
Cc: liuhongt <hongtao.liu@intel.com>,
Michael Meissner <meissner@linux.ibm.com>,
Andrew Waterman <andrew@sifive.com>,
Segher Boessenkool <segher@kernel.crashing.org>,
asolokha@gmx.com, Andreas Schwab <schwab@linux-m68k.org>,
GCC Patches <gcc-patches@gcc.gnu.org>
Subject: Re: [PATCH 2/2] validate_subreg before call gen_lowpart to avoid ICE.
Date: Fri, 10 Sep 2021 22:25:45 +0800 [thread overview]
Message-ID: <CAMZc-byeXhdcYV0sbjCtGLY2r26egbOsW8LfZi5y_CkSOgqU1g@mail.gmail.com> (raw)
In-Reply-To: <CAMZc-bxCNLLT9mJ6mYEzJbP4DW58tC1V1-kiOmY8dx-JeRan7g@mail.gmail.com>
[-- Attachment #1: Type: text/plain, Size: 4057 bytes --]
On Fri, Sep 10, 2021 at 9:44 PM Hongtao Liu <crazylht@gmail.com> wrote:
>
> On Fri, Sep 10, 2021 at 9:32 PM Richard Biener
> <richard.guenther@gmail.com> wrote:
> >
> > On September 10, 2021 3:27:09 PM GMT+02:00, Hongtao Liu <crazylht@gmail.com> wrote:
> > >On Fri, Sep 10, 2021 at 9:16 PM Richard Biener via Gcc-patches
> > ><gcc-patches@gcc.gnu.org> wrote:
> > >>
> > >> On Fri, Sep 10, 2021 at 2:58 PM liuhongt <hongtao.liu@intel.com> wrote:
> > >> >
> > >> > gcc/ChangeLog:
> > >> >
> > >> > * expmed.c (extract_bit_field_using_extv): validate_subreg
> > >> > before call gen_lowpart.
> > >> > ---
> > >> > gcc/expmed.c | 6 +++++-
> > >> > 1 file changed, 5 insertions(+), 1 deletion(-)
> > >> >
> > >> > diff --git a/gcc/expmed.c b/gcc/expmed.c
> > >> > index 3143f38e057..10d62d857a8 100644
> > >> > --- a/gcc/expmed.c
> > >> > +++ b/gcc/expmed.c
> > >> > @@ -1571,12 +1571,16 @@ extract_bit_field_using_extv (const extraction_insn *extv, rtx op0,
> > >> >
> > >> > if (GET_MODE (target) != ext_mode)
> > >> > {
> > >> > + machine_mode tmode = GET_MODE (target);
> > >> > /* Don't use LHS paradoxical subreg if explicit truncation is needed
> > >> > between the mode of the extraction (word_mode) and the target
> > >> > mode. Instead, create a temporary and use convert_move to set
> > >> > the target. */
> > >> > if (REG_P (target)
> > >> > - && TRULY_NOOP_TRUNCATION_MODES_P (GET_MODE (target), ext_mode))
> > >>
> > >> ^^^
> > >>
> > >> I wonder if herein lies the problem in that the HFmode "truncation" from SImode
> > >> is considered noop? Note the underlying target hook only looks at the mode
> > >> precision and thus receives 16 and 32, and thus maybe that
> > >> TRULY_NOOP_TRUNCATION_MODES_P query only makes sense for
> > >> integer modes? Though the documentation of the hook only talks about
> > >> "conversion" of "values" ...
> > >>
> > >> So maybe a targetm.modes_tieable_p (GET_MODE (target), extmode) check
> > >> is missing?
> > >
> > >According to document, it should be true for
> > >targetm.modes_tieable_p(HFmode, SImode) since HFmode can be allocated
> > >to gpr.
> > >
> > >----------------
> > >This hook returns true if a value of mode mode1 is accessible in mode
> > >mode2 without
> > >copying
> > >-------------------
> > >
> > >and also here gen_lowpart (SImode, HFmode, target) is called and hit
> > >gcc_assert, not (subreg:HF (reg:SI) 0)
> >
> > I see. Of course that leads to a suggestion to allow the subreg based on modes_tieable_p, but then others will know why that's the wrong thing to do?
> I'm testing this
>
> 1 file changed, 2 insertions(+), 1 deletion(-)
> gcc/expmed.c | 3 ++-
>
> modified gcc/expmed.c
> @@ -1576,7 +1576,8 @@ extract_bit_field_using_extv (const
> extraction_insn *extv, rtx op0,
> mode. Instead, create a temporary and use convert_move to set
> the target. */
> if (REG_P (target)
> - && TRULY_NOOP_TRUNCATION_MODES_P (GET_MODE (target), ext_mode))
> + && TRULY_NOOP_TRUNCATION_MODES_P (GET_MODE (target), ext_mode)
> + && targetm.modes_tieable_p (GET_MODE (target), ext_mode))
> {
> target = gen_lowpart (ext_mode, target);
> if (partial_subreg_p (GET_MODE (spec_target), ext_mode))
>
Updated patch.
Bootstrapped and regtested on x86_64-linux-gnu{-m32,}, do I need to
run this patch on other targets machine, or the patch is supposed to
have minimal impact on other targets?
Then, ok for trunk?
> >
> > Richard.
> >
> > >>
> > >> > + && TRULY_NOOP_TRUNCATION_MODES_P (tmode, ext_mode)
> > >> > + && validate_subreg (ext_mode, tmode,
> > >> > + target,
> > >> > + subreg_lowpart_offset (ext_mode, tmode)))
> > >> > {
> > >> > target = gen_lowpart (ext_mode, target);
> > >> > if (partial_subreg_p (GET_MODE (spec_target), ext_mode))
> > >> > --
> > >> > 2.27.0
> > >> >
> > >
> > >
> > >
> >
>
>
> --
> BR,
> Hongtao
--
BR,
Hongtao
[-- Attachment #2: v2-0001-Check-modes_tieable_p-before-call-gen_lowpart-to-.patch --]
[-- Type: application/octet-stream, Size: 1069 bytes --]
From e1a63d41ad733e46f28cff540cfacc4742488782 Mon Sep 17 00:00:00 2001
From: liuhongt <hongtao.liu@intel.com>
Date: Fri, 10 Sep 2021 22:15:36 +0800
Subject: [PATCH v2] Check modes_tieable_p before call gen_lowpart to avoid ICE
in validate_subreg.
gcc/ChangeLog:
* expmed.c (extract_bit_field_using_extv): Check
modes_tieable_p before call gen_lowpart.
---
gcc/expmed.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/gcc/expmed.c b/gcc/expmed.c
index 3143f38e057..fc1c973033c 100644
--- a/gcc/expmed.c
+++ b/gcc/expmed.c
@@ -1576,7 +1576,8 @@ extract_bit_field_using_extv (const extraction_insn *extv, rtx op0,
mode. Instead, create a temporary and use convert_move to set
the target. */
if (REG_P (target)
- && TRULY_NOOP_TRUNCATION_MODES_P (GET_MODE (target), ext_mode))
+ && TRULY_NOOP_TRUNCATION_MODES_P (GET_MODE (target), ext_mode)
+ && targetm.modes_tieable_p (GET_MODE (target), ext_mode))
{
target = gen_lowpart (ext_mode, target);
if (partial_subreg_p (GET_MODE (spec_target), ext_mode))
--
2.27.0
next prev parent reply other threads:[~2021-09-10 14:25 UTC|newest]
Thread overview: 28+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-09-10 12:58 [PATCH 0/2] Revert r12-3277 since it caused regressions on many other targets liuhongt
2021-09-10 12:58 ` [PATCH 1/2] Revert "Get rid of all float-int special cases in validate_subreg." liuhongt
2021-09-10 13:05 ` Richard Biener
2021-09-10 12:58 ` [PATCH 2/2] validate_subreg before call gen_lowpart to avoid ICE liuhongt
2021-09-10 13:15 ` Richard Biener
2021-09-10 13:27 ` Hongtao Liu
2021-09-10 13:32 ` Richard Biener
2021-09-10 13:44 ` Hongtao Liu
2021-09-10 14:25 ` Hongtao Liu [this message]
2021-09-10 21:19 ` Segher Boessenkool
2021-09-11 0:29 ` Hongtao Liu
2021-09-11 1:04 ` Hongtao Liu
2021-09-10 13:52 ` Hongtao Liu
2021-09-10 13:39 ` Hongtao Liu
2021-09-10 13:30 ` Segher Boessenkool
2021-09-10 13:58 ` Richard Biener
2021-09-10 16:24 ` Segher Boessenkool
2021-09-10 18:36 ` Richard Biener
2021-09-10 21:27 ` Segher Boessenkool
2021-09-11 8:25 ` Richard Biener
2021-09-11 9:51 ` Hongtao Liu
2021-09-11 11:09 ` Hongtao Liu
2021-09-13 6:10 ` Richard Biener
2021-09-13 6:32 ` Hongtao Liu
2021-09-13 9:15 ` Richard Biener
2021-09-13 11:14 ` Hongtao Liu
2021-09-13 11:44 ` Tobias Burnus
2021-09-13 11:45 ` 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=CAMZc-byeXhdcYV0sbjCtGLY2r26egbOsW8LfZi5y_CkSOgqU1g@mail.gmail.com \
--to=crazylht@gmail.com \
--cc=andrew@sifive.com \
--cc=asolokha@gmx.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=hongtao.liu@intel.com \
--cc=meissner@linux.ibm.com \
--cc=richard.guenther@gmail.com \
--cc=schwab@linux-m68k.org \
--cc=segher@kernel.crashing.org \
/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).