From: "Roger Sayle" <roger@nextmovesoftware.com>
To: "'Uros Bizjak'" <ubizjak@gmail.com>
Cc: <gcc-patches@gcc.gnu.org>
Subject: RE: [x86 PATCH] PR tree-optimization/98335: New peephole2 xorl; movb -> movzbl
Date: Mon, 7 Mar 2022 11:51:50 -0000 [thread overview]
Message-ID: <001401d83219$bc707f10$35517d30$@nextmovesoftware.com> (raw)
In-Reply-To: <CAFULd4Zjc8p3Rb5fbwMEu3iOA9_o3Q5swj4tEFzsOxFv3uHZPA@mail.gmail.com>
Hi Uros,
> Is there a reason that only inserts to DImode registers are implemented?
> IMO, these peepholes should also handle inserts to SImode.
I wasn't able to construct a test case that produced a byte or word insert
into an SImode register. The front-ends and middle-end end up producing
different code sequences, and -m32 changes the ABI so that small structs
get passed in memory rather than in registers.
Here's the expanded testcase that I investigated:
struct DataCL { char a; int b; };
struct DataWL { short a; int b; };
struct DataIL { int a; int b; };
struct DataCI { char a; short b; };
struct DataWI { short a; short b; };
char c;
short w;
int i;
DataCL foo_cl(int idx) { return { c }; }
DataCL bar_cl(int idx) { return { c, 0 }; }
DataWL foo_wl(int idx) { return { w }; }
DataWL bar_wl(int idx) { return { w, 0 }; }
DataIL foo_il(int idx) { return { i }; }
DataIL bar_il(int idx) { return { i, 0 }; }
DataCI foo_ci(int idx) { return { c }; }
DataCI bar_ci(int idx) { return { c, 0 }; }
DataWI foo_wi(int idx) { return { w }; }
DataWI bar_wi(int idx) { return { w, 0 }; }
I agree that for completeness similar peepholes handling inserts into
SImode would be a good thing, but these wouldn't be restricted by
TARGET_64BIT and would therefore require additional -m32 testing.
The DImode peepholes I can justify for stage4 as PR tree-opt/98335
is a regression, SImode peepholes would be more of a "leap of faith".
If you’d be willing to accept a patch without a testcase, let me know.
It's also a pity that subreg handling in combine doesn't allow merging
these inserts into zero registers to be combined to zero_extends in a
machine independent way. My recent patch for PR 95126 (awaiting
review) should also allow front-ends and middle-end passes more
flexibility in optimizing small struct constructors.
Thanks (as always) for reviewing patches so quickly.
Roger
--
next prev parent reply other threads:[~2022-03-07 11:51 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-03-07 10:55 Roger Sayle
2022-03-07 11:20 ` [x86 PATCH] PR tree-optimization/98335: New peephole2 xorl;movb " Uros Bizjak
2022-03-07 11:51 ` Roger Sayle [this message]
2022-03-07 12:02 ` [x86 PATCH] PR tree-optimization/98335: New peephole2 xorl; movb " Jakub Jelinek
2022-03-09 7:36 ` [x86 PATCH] PR tree-optimization/98335: New peephole2 xorl;movb " Uros Bizjak
2022-03-09 18:10 ` [x86 PATCH] PR tree-optimization/98335: New peephole2 xorl; movb " Roger Sayle
2022-03-10 7:17 ` [x86 PATCH] PR tree-optimization/98335: New peephole2 xorl;movb " Uros Bizjak
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='001401d83219$bc707f10$35517d30$@nextmovesoftware.com' \
--to=roger@nextmovesoftware.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=ubizjak@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).