Re: c/3913: On Athlon's, gcc 3.0 produces code running 1/2 as fast as gcc 2.95

Re: c/3913: On Athlon's, gcc 3.0 produces code running 1/2 as fast as gcc 2.95

[rth]

Synopsis: On Athlon's, gcc 3.0 produces code running 1/2 as fast as gcc 2.95

State-Changed-From-To: open->closed
State-Changed-By: rth
State-Changed-When: Wed Apr 24 14:41:30 2002
State-Changed-Why:
    From the web page:
    
    NOTE: all of these problems appear to be fixed in the newer 3.1 preleases. I tested gcc-20020225.tar.gz, and found it performed indistinguishably from 2.95.2 for both Athlon and PIII. An ATLAS user has independently verified this observation for PIII/P4.

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=3913

c/3913: On Athlon's, gcc 3.0 produces code running 1/2 as fast as gcc 2.95

[rwhaley]

>Number:         3913
>Category:       c
>Synopsis:       On Athlon's, gcc 3.0 produces code running 1/2 as fast as gcc 2.95
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          pessimizes-code
>Submitter-Id:   net
>Arrival-Date:   Wed Aug 01 17:56:00 PDT 2001
>Closed-Date:
>Last-Modified:
>Originator:     rwhaley@cs.utk.edu
>Release:        gcc version 3.0
>Organization:
>Environment:
RedHat Linux 7.1 on a 1.2Ghz Athlon.  gcc -v gives:
hades. /usr/local/gcc-3.0/bin/gcc -v
Reading specs from /usr/local/gcc-3.0/lib/gcc-lib/i686-pc-linux-gnu/3.0/specs
Configured with: ./configure --prefix=/usr/local/gcc-3.0
Thread model: single
gcc version 3.0
>Description:
The new memory fetch scheduling algorithm reduces Athlon performance by a factor of 2 over the one used by gcc 2.95.x.  It also appears that the x87 register stack utilization algorithm  is worse than 2.95.x as well. 

Problem is too complex to reduce to 1 single preprocessed file.  The full problem is detailed (along with code to reproduce, and assembler outputs) here: 
   http://www.cs.utk.edu/~rwhaley/ATLAS/gcc30.html
    
   
>How-To-Repeat:
 http://www.cs.utk.edu/~rwhaley/ATLAS/gcc30.html#dup
>Fix:
Have option to use 2.95 scheduler for scheduling problem.
Have option to use 2.95 x87 register stack for register stack usage problem.
>Release-Note:
>Audit-Trail:
>Unformatted: