optimization/4919: gcse inhibits loop optimization

optimization/4919: gcse inhibits loop optimization

[ottoni]

>Number:         4919
>Category:       optimization
>Synopsis:       gcse inhibits loop optimization
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          pessimizes-code
>Submitter-Id:   net
>Arrival-Date:   Wed Nov 21 06:26:00 PST 2001
>Closed-Date:
>Last-Modified:
>Originator:     Guilherme Ottoni
>Release:        gcc version 3.1 20011029 (experimental)
>Organization:
>Environment:
RedHat Linux 6.2 running on a PC (K6-II)
>Description:
Compiling with gcc -O2 the source attached, the loop optimizer fails to detect 
BIVs, and so misses the opportunity to apply all the loop 
optimizations. If you add the -no-gcse flag to the command line,
the loop optimizer works properly.
To check what the loop optimizer is doing, I'm looking at 
the debug file generated by -dL.
>How-To-Repeat:
Simply compile the source attached, with -O2, and with and 
without the -fno-gcse. Then look at the .loop debug files to
check the problem.
I'm using a dsp16xx target, but I think this problem will occur 
with any target, as it is before the backend.
>Fix:
Maybe enhancing the BIV detection method to get the BIVs in 
these cases. Another (dirty) solution would be to inhibit the 
GCSE to optimize the code in the LOOP_CONT.
>Release-Note:
>Audit-Trail:
>Unformatted:
----gnatsweb-attachment----
Content-Type: application/octet-stream; name="cc2001.c"
Content-Transfer-Encoding: base64
Content-Disposition: attachment; filename="cc2001.c"

bWFpbigpCnsKICBpbnQgIGksIGF2ZywgYVsxMDBdLCBlcnJvcjsKICBmb3IgKGk9MDsgaSA8IDkw
OyBpKyspCiAgewogICAgaWYgKGklMikKICAgIHsgCiAgICAgIGF2ZyArPSBhW2krMV0gPDwgMjsK
ICAgICAgYVtpKzJdID0gYXZnICogMzsKICAgIH0KICAgIGlmIChhdmcgPCBlcnJvcikKICAgICAg
YXZnIC09IGFbaSsxXSAtIGVycm9yLzI7CiAgICBlbHNlCiAgICAgIGF2ZyAtPSBhW2krMl0gLSBl
cnJvcjsKICB9Cn0KCg==

Re: optimization/4919: gcse inhibits loop optimization

[rth]

The following reply was made to PR optimization/4919; it has been noted by GNATS.

From: rth@gcc.gnu.org
To: gcc-bugs@gcc.gnu.org, gcc-gnats@gcc.gnu.org, gcc-prs@gcc.gnu.org,
  nobody@gcc.gnu.org, ottoni@ic.unicamp.br
Cc:  
Subject: Re: optimization/4919: gcse inhibits loop optimization
Date: 27 Nov 2001 07:03:49 -0000

 Synopsis: gcse inhibits loop optimization
 
 State-Changed-From-To: open->suspended
 State-Changed-By: rth
 State-Changed-When: Mon Nov 26 23:03:48 2001
 State-Changed-Why:
     GCSE did exactly what it was supposed to do -- the expression
     I+1, which appears twice in the body of the loop, is partially
     redundant with the increment of I at the end of the loop.
     
     This isn't really a problem with GCSE, but with the extreme lameness
     of our loop optimizer.  Hopefully we'll have time to rewrite it from
     scratch for gcc 3.2; Jan Hubicka has already volunteered to try.
     
     
 
 http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&pr=4919&database=gcc

Re: optimization/4919: gcse inhibits loop optimization

[rth]

Synopsis: gcse inhibits loop optimization

State-Changed-From-To: open->suspended
State-Changed-By: rth
State-Changed-When: Mon Nov 26 23:03:48 2001
State-Changed-Why:
    GCSE did exactly what it was supposed to do -- the expression
    I+1, which appears twice in the body of the loop, is partially
    redundant with the increment of I at the end of the loop.
    
    This isn't really a problem with GCSE, but with the extreme lameness
    of our loop optimizer.  Hopefully we'll have time to rewrite it from
    scratch for gcc 3.2; Jan Hubicka has already volunteered to try.
    
    

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&pr=4919&database=gcc