[Bug target/64753] New: Redundant cmp instruction on x86_64

[Bug target/64753] New: Redundant cmp instruction on x86_64

[rv at rasmusvillemoes dot dk]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64753

            Bug ID: 64753
           Summary: Redundant cmp instruction on x86_64
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rv at rasmusvillemoes dot dk
              Host: x86_64
            Target: x86_64

The linux kernel's library strncmp is this:

int strncmp(const char *cs, const char *ct, size_t count)
{
        unsigned char c1, c2;

        while (count) {
                c1 = *cs++;
                c2 = *ct++;
                if (c1 != c2)
                        return c1 < c2 ? -1 : 1;
                if (!c1)
                        break;
                count--;
        }
        return 0;
}

Compiling with gcc -O2 -S I get this:

strncmp:
.LFB0:
        .cfi_startproc
        testq   %rdx, %rdx
        je      .L10
        movzbl  (%rdi), %ecx
        movzbl  (%rsi), %r8d
#        cmpb    %r8b, %cl
#        jne     .L3
        testb   %cl, %cl
        je      .L10
        subq    $1, %rdx
        xorl    %eax, %eax
        jmp     .L4
        .p2align 4,,10
        .p2align 3
.L6:
        movzbl  1(%rdi,%rax), %ecx
        movzbl  1(%rsi,%rax), %r8d
#        cmpb    %r8b, %cl
#        jne     .L3
        addq    $1, %rax
        testb   %cl, %cl
        je      .L10
.L4:
        cmpq    %rdx, %rax
        jne     .L6
.L10:
        xorl    %eax, %eax
        ret
        .p2align 4,,10
        .p2align 3
.L3:
        cmpb    %r8b, %cl
        sbbl    %eax, %eax
        orl     $1, %eax
        ret
        .cfi_endproc

At the two places marked # we do a cmp and a conditional jump to .L3, where for
good measure the same cmp is done again... there's no other path to .L3, so it
would seem that simply omitting that extra cmp should be ok. 

This is with gcc-5.0 (GCC) 5.0.0 20150112 (experimental), but I see the same
with gcc (Debian 4.7.2-5) 4.7.2.

[Bug target/64753] Redundant cmp instruction on x86_64

[rv at rasmusvillemoes dot dk]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64753

--- Comment #1 from Rasmus Villemoes <rv at rasmusvillemoes dot dk> ---
FWIW, I wrote a stupid script looking for the pattern

  <test/cmp instruction>
  <(conditional) jump to abcd>
...
  abcd: <same test/cmp instruction>

Compiling a defconfig linux kernel with gcc 4.7 gives 201 instances, 5.0 gives
252. The second test/cmp may not be redundant in all cases, but at least the
jump target could be adjusted to the instruction following abcd. I filtered out
cases where the jump itself was a jump target (so the only path to it is
through the preceding test/cmp), though that didn't actually remove anything.

[Bug target/64753] Redundant cmp instruction on x86_64

[ubizjak at gmail dot com]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="UTF-8", Size: 3061 bytes --]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64753

--- Comment #2 from Uroš Bizjak <ubizjak at gmail dot com> ---
(In reply to Rasmus Villemoes from comment #0)

> At the two places marked # we do a cmp and a conditional jump to .L3, where
> for good measure the same cmp is done again... there's no other path to .L3,
> so it would seem that simply omitting that extra cmp should be ok. 

IIRC postreload redundant compare elimination pass removes these redundant
compares. Unfortunately, the pass is not yet enabled for x86 targets.
>From gcc-bugs-return-475242-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org Wed Jan 28 13:31:04 2015
Return-Path: <gcc-bugs-return-475242-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Delivered-To: listarch-gcc-bugs@gcc.gnu.org
Received: (qmail 17995 invoked by alias); 28 Jan 2015 13:31:04 -0000
Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-bugs.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-help@gcc.gnu.org>
Sender: gcc-bugs-owner@gcc.gnu.org
Delivered-To: mailing list gcc-bugs@gcc.gnu.org
Received: (qmail 17468 invoked by uid 48); 28 Jan 2015 13:30:59 -0000
From: "pault at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug fortran/64757] [5 Regression] ICE in fold_convert_loc, at fold-const.c:2353
Date: Wed, 28 Jan 2015 13:31:00 -0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: fortran
X-Bugzilla-Version: 5.0
X-Bugzilla-Keywords: ice-on-valid-code
X-Bugzilla-Severity: normal
X-Bugzilla-Who: pault at gcc dot gnu.org
X-Bugzilla-Status: NEW
X-Bugzilla-Priority: P4
X-Bugzilla-Assigned-To: pault at gcc dot gnu.org
X-Bugzilla-Target-Milestone: 5.0
X-Bugzilla-Flags:
X-Bugzilla-Changed-Fields: cc assigned_to attachments.created
Message-ID: <bug-64757-4-ELTYTvJFpC@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-64757-4@http.gcc.gnu.org/bugzilla/>
References: <bug-64757-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
X-SW-Source: 2015-01/txt/msg03236.txt.bz2
Content-length: 681

https://gcc.gnu.org/bugzilla/show_bug.cgi?idd757

Paul Thomas <pault at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pault at gcc dot gnu.org
           Assignee|unassigned at gcc dot gnu.org      |pault at gcc dot gnu.org

--- Comment #7 from Paul Thomas <pault at gcc dot gnu.org> ---
Created attachment 34607
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id4607&actioníit
Draft patch for the PR

I believe that this regression was caused by the fix for pr60357.

The patch is regtesting right now.

Paul

[Bug target/64753] Redundant cmp instruction on x86_64

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64753

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
   Target Milestone|---                         |9.0
         Resolution|---                         |FIXED

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
r8-647-g4f0473fe89e68b

and some more improvement that was done in GCC 9.