optimization/6585: useless memory store instructions on x86

optimization/6585: useless memory store instructions on x86

[Bruno Haible]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 3135 bytes --]


>Number:         6585
>Category:       optimization
>Synopsis:       useless memory store instructions on x86
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    unassigned
>State:          open
>Class:          pessimizes-code
>Submitter-Id:   net
>Arrival-Date:   Mon May 06 13:06:00 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     Bruno Haible
>Release:        3.1 20020423 (prerelease)
>Organization:
GNU hackers
>Environment:
System: Linux linuix 2.4.18-4GB #1 Wed Mar 27 13:57:05 UTC 2002 i686 unknown
Architecture: i686

	
host: i686-pc-linux-gnu
build: i686-pc-linux-gnu
target: i686-pc-linux-gnu
configured with: ../configure --prefix=/packages/gnu-snapshot --enable-shared --enable-version-specific-runtime-libs --enable-nls
>Description:

A case of a useless memory store and of a useless register spill.

=========================== mul.c ===============================
long long mul (long long a, long long b) {
  return a * b;
}
=================================================================

$ gcc -v
Lecture des spécification à partir de /packages/gnu-snapshot/lib/gcc-lib/i686-pc-linux-gnu/3.1/specs
Configuré avec: ../configure --prefix=/packages/gnu-snapshot --enable-shared --enable-version-specific-runtime-libs --enable-nls
Modèle de thread: single
version gcc 3.1 20020423 (prerelease)

$ gcc -O6 -fomit-frame-pointer -S mul.c
$ cat mul.s
        .file   "mul.c"
        .text
        .align 2
        .p2align 4,,15
.globl mul
        .type   mul,@function
mul:
        subl    $28, %esp
        movl    40(%esp), %ecx      b0
        movl    %ebx, 12(%esp)                        save %ebx
        movl    32(%esp), %ebx      a0
        movl    %ebp, 24(%esp)                        save %ebp
        movl    44(%esp), %ebp      b1
        movl    %ebx, %eax          a0
        mull    %ecx                %edx:%eax := a0*b0
        movl    %edi, 20(%esp)                        save %edi
        movl    36(%esp), %edi      a1
        movl    %esi, 16(%esp)                        save %esi USELESS!
        movl    16(%esp), %esi                        restore %esi USELESS!
        imull   %edi, %ecx          a1*b0
        movl    %eax, (%esp)
        movl    20(%esp), %edi                        restore %edi
        movl    %ebx, %eax          a0
        movl    %edx, 4(%esp)                         USELESS!
        imull   %ebp, %eax          a0*b1
        movl    12(%esp), %ebx                        restore %ebx
        movl    24(%esp), %ebp                        restore %ebp
        addl    %eax, %edx          hi+a0*b1
        movl    (%esp), %eax        lo
        movl    %edx, 4(%esp)       hi+a0*b1          ! Could be simplified
        addl    %ecx, 4(%esp)       hi+a0*b1+a1*b0    ! to a single insn:
        movl    4(%esp), %edx                         ! addl %ecx, $edx
        addl    $28, %esp
        ret
.Lfe1:
        .size   mul,.Lfe1-mul
        .ident  "GCC: (GNU) 3.1 20020423 (prerelease)"

>How-To-Repeat:

gcc -O6 -fomit-frame-pointer -S mul.c

>Fix:

more optimizations
>Release-Note:
>Audit-Trail:
>Unformatted:

Re: optimization/6585: useless memory store instructions on x86

[bangerth]

Synopsis: useless memory store instructions on x86

State-Changed-From-To: feedback->analyzed
State-Changed-By: bangerth
State-Changed-When: Wed Apr 23 14:05:36 2003
State-Changed-Why:
    Code is better but not good

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=6585

Re: optimization/6585: useless memory store instructions on x86

[Bruno Haible]

The following reply was made to PR optimization/6585; it has been noted by GNATS.

From: Bruno Haible <bruno@clisp.org>
To: bangerth@dealii.org, gcc-bugs@gcc.gnu.org, gcc-prs@gcc.gnu.org,
   nobody@gcc.gnu.org, gcc-gnats@gcc.gnu.org
Cc:  
Subject: Re: optimization/6585: useless memory store instructions on x86
Date: Wed, 23 Apr 2003 14:53:49 +0200 (CEST)

 >     Bruno, I can verify your problem with 3.2, but I get
 >     significantly better code with 3.2.2. Can you say a word
 >     whether the new code satisfies you?
 
 No, the new code is somewhat better, but still exhibits three problems:
 - The dead store is still there.
 - The prologue still saves and restores an unused register.
 - The final reshuffling of registers before return is suboptimal.
 
 =========================== mul.c ===============================
 long long mul (long long a, long long b) {
   return a * b;
 }
 =================================================================
 
 $ gcc -v
 Reading specs from /packages/gnu-inst-gcc/3.2.2/lib/gcc-lib/i686-pc-linux-gnu/3.2.2/specs
 Configured with: ../configure --prefix=/packages/gnu-inst-gcc/3.2.2 --enable-shared --enable-version-specific-runtime-libs --enable-nls --enable-threads=posix --enable-__cxa_atexit
 Thread model: posix
 gcc version 3.2.2
 
 $ gcc -O6 -fomit-frame-pointer -S mul.c
 $ cat mul.s
         .file   "mul.c"
         .text
         .p2align 4,,15
 .globl mul
         .type   mul,@function
 mul:
         subl    $20, %esp
         movl    24(%esp), %eax		a0
         movl    %ebx, 8(%esp)				save %ebx
         movl    32(%esp), %ebx		b0
         movl    %esi, 12(%esp)				save %esi
         movl    36(%esp), %ecx		b1
         movl    %edi, 16(%esp)				save %edi USELESS!
         movl    16(%esp), %edi				restore %edi USELESS!
         mull    %ebx			%edx:%eax := a0*b0
         movl    %edx, %esi
         movl    %eax, (%esp)
         movl    24(%esp), %eax		a0
         movl    %edx, 4(%esp)				USELESS!
         imull   %ecx, %eax		a0*b1
         addl    %eax, %esi
         movl    28(%esp), %eax		a1
         imull   %eax, %ebx		a1*b0
         leal    (%ebx,%esi), %eax
         movl    12(%esp), %esi
         movl    8(%esp), %ebx
         movl    %eax, 4(%esp)		! Could be simplified to 2 instructions:
         movl    (%esp), %eax		! movl    %eax, %edx
         movl    4(%esp), %edx		! movl    (%esp), %eax
         addl    $20, %esp
         ret
 .Lfe1:
         .size   mul,.Lfe1-mul
         .ident  "GCC: (GNU) 3.2.2"

Re: optimization/6585: useless memory store instructions on x86

[bangerth]

Synopsis: useless memory store instructions on x86

State-Changed-From-To: open->feedback
State-Changed-By: bangerth
State-Changed-When: Sat Mar 15 04:58:27 2003
State-Changed-Why:
    Bruno, I can verify your problem with 3.2, but I get
    significantly better code with 3.2.2. Can you say a word
    whether the new code satisfies you?
    
    Thanks
      Wolfgang

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=6585