From: Segher Boessenkool <segher@kernel.crashing.org>
To: Terry Guo <flameroc@gmail.com>
Cc: richard.sandiford@arm.com, rdsandiford@googlemail.com,
GCC Patches <gcc-patches@gcc.gnu.org>,
Hale Wang <hale.wang@arm.com>
Subject: Re: Ping : [PATCH] [gcc, combine] PR46164: Don't combine the insns if a volatile register is contained.
Date: Thu, 26 Feb 2015 06:23:00 -0000 [thread overview]
Message-ID: <20150226055537.GA15492@gate.crashing.org> (raw)
In-Reply-To: <CAGbRaL4HOT9jB0wJ+eS+BoAVmuULz-rv0tRVJKsZFLYESUsm+w@mail.gmail.com>
On Tue, Feb 17, 2015 at 11:39:34AM +0800, Terry Guo wrote:
> On Sun, Feb 15, 2015 at 7:35 PM, Segher Boessenkool
> <segher@kernel.crashing.org> wrote:
> > Hi Terry,
> >
> > I still think this is stage1 material.
> >
> >> + /* Don't combine if dest contains a user specified register and i3 contains
> >> + ASM_OPERANDS, because the user specified register (same with dest) in i3
> >> + would be replaced by the src of insn which might be different with
> >> + the user's expectation. */
> >
> > "Do not eliminate a register asm in an asm input" or similar? Text
> > explaining why REG_USERVAR_P && HARD_REGISTER_P works here would be
> > good to have, too.
> diff --git a/gcc/combine.c b/gcc/combine.c
> index f779117..aeb2854 100644
> --- a/gcc/combine.c
> +++ b/gcc/combine.c
> @@ -1779,7 +1779,7 @@ can_combine_p (rtx_insn *insn, rtx_insn *i3, rtx_insn *pred ATTRIBUTE_UNUSED,
> {
> int i;
> const_rtx set = 0;
> - rtx src, dest;
> + rtx src, dest, asm_op;
> rtx_insn *p;
> #ifdef AUTO_INC_DEC
> rtx link;
> @@ -1914,6 +1914,14 @@ can_combine_p (rtx_insn *insn, rtx_insn *i3, rtx_insn *pred ATTRIBUTE_UNUSED,
> set = expand_field_assignment (set);
> src = SET_SRC (set), dest = SET_DEST (set);
>
> + /* Use REG_USERVAR_P and HARD_REGISTER_P to check whether DEST is a user
> + specified register, and do not eliminate such register if it is in an
> + asm input because we may end up with something different with user's
> + expectation. */
That doesn't explain why this will hit (almost) only on register asms.
The user's expectation doesn't matter that much either: GCC would violate
its own documentation / promises, that matters more ;-)
> + if (REG_P (dest) && REG_USERVAR_P (dest) && HARD_REGISTER_P (dest)
> + && ((asm_op = extract_asm_operands (PATTERN (i3))) != NULL))
You do not need the temporary variable, nor the != 0 or the extra parens;
just write
&& extract_asm_operands (PATTERN (i3))
Cheers,
Segher
next prev parent reply other threads:[~2015-02-26 5:55 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-02-09 1:42 Hale Wang
2015-02-12 15:54 ` Richard Sandiford
2015-02-12 22:16 ` Segher Boessenkool
2015-02-12 22:46 ` Jeff Law
2015-02-13 3:18 ` Hale Wang
2015-02-13 9:07 ` Richard Sandiford
2015-02-15 10:26 ` Terry Guo
2015-02-15 11:35 ` Segher Boessenkool
2015-02-17 3:39 ` Terry Guo
2015-02-26 6:23 ` Segher Boessenkool [this message]
2015-02-26 12:12 ` Terry Guo
2015-02-26 20:26 ` 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=20150226055537.GA15492@gate.crashing.org \
--to=segher@kernel.crashing.org \
--cc=flameroc@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=hale.wang@arm.com \
--cc=rdsandiford@googlemail.com \
--cc=richard.sandiford@arm.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).