public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
From: "thomas dot schlichter at web dot de" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug optimization/14121] New: use inline assembler register constraints as home register hints
Date: Wed, 11 Feb 2004 23:56:00 -0000	[thread overview]
Message-ID: <20040211235607.14121.thomas.schlichter@web.de> (raw)

Hi,

gcc 3.4.0 (and 3.3.1) seems not to use the register constraints of the inline
assembler as hints where to place local variables. For example following part of
the file bin2dec.c (will provide it later) :

    int m2 = m;
    unsigned long c = 0;

    do {
      asm("divl\t%2" : "+a"(x[m2]), "+d"(c) : "r"(1000000000));
    } while (--m2 >= 0);

is translated to following x86 assembler instructions (gcc -O2 -S):

        movl    $1000000000, %ebx
        .p2align 4,,15
.L3:
        movl    %edi, -20(%ebp)
        xorl    %ecx, %ecx
        .p2align 4,,15
.L6:
        movl    -20(%ebp), %edx
        movl    (%esi,%edx,4), %eax
        movl    %ecx, %edx
#APP
        divl    %ebx
#NO_APP
        movl    %edx, %ecx
        movl    -20(%ebp), %edx
        movl    %eax, (%esi,%edx,4)
        decl    %edx
        movl    %edx, -20(%ebp)
        jns     .L6

So the variable "c" seems to be placed in the register "ecx" whereas it is
constrained to register "edx". If "c" was placed in "edx" the register "ecx"
would become free and could be used for "m2". So this variable would not have to
be placed on the stack...

gcc 2.95.3 seems to do this placement, but I don't know if it does this
coincidental or intentionally... It generates following code (ebx<>m, ecx<>m2,
edx<>c, edi<>x):

        movl %ebx,%ecx
        xorl %edx,%edx
        .p2align 2,0x90
.L10:
        movl (%edi,%ecx,4),%eax
        movl $1000000000,%esi
#APP
        divl    %esi
#NO_APP
        movl %eax,(%edi,%ecx,4)
        decl %ecx
        jns .L10

So it would be GREAT if 3.4.0 would do anything like that (plus moving the
assignment to esi outside the loop, of course ;-)

Best reagrds
   Thomas Schlichter

-- 
           Summary: use inline assembler register constraints as home
                    register hints
           Product: gcc
           Version: 3.4.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P2
         Component: optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: thomas dot schlichter at web dot de
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14121


             reply	other threads:[~2004-02-11 23:56 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2004-02-11 23:56 thomas dot schlichter at web dot de [this message]
2004-02-11 23:58 ` [Bug optimization/14121] " thomas dot schlichter at web dot de
2004-02-11 23:59 ` thomas dot schlichter at web dot de
2004-02-12  0:00 ` thomas dot schlichter at web dot de
2004-02-12  0:02 ` thomas dot schlichter at web dot de
2004-02-12  2:45 ` pinskia at gcc dot gnu dot org

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=20040211235607.14121.thomas.schlichter@web.de \
    --to=gcc-bugzilla@gcc.gnu.org \
    --cc=gcc-bugs@gcc.gnu.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).