optimization/7713: Performance regression from gcc 2.95.3

[jlquinn@optonline.net]

>Number:         7713
>Category:       optimization
>Synopsis:       Performance regression from gcc 2.95.3
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          pessimizes-code
>Submitter-Id:   net
>Arrival-Date:   Sat Aug 24 13:56:00 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     jlquinn@optonline.net
>Release:        gcc-3.2
>Organization:
>Environment:
powerpc-unknown-linux-gnu
glibc 2.2.5
binutils 2.12.90.0.9
Debian Woody
240MHz 604e
1MB cache
>Description:
Running C benchmark found here:
http://www.cwi.nl/~tromp/c4/fhour.html

I was looking at runtimes from different compilers, gcc 2.95.4 (.3 plus Debian patches) and released gcc 3.2 included.

If I compile with gcc 2.95 -O2, I get the following performance numbers from the program:

207.2 Kpos/sec, 208.7 Kpos/sec

gcc 2.95 -O2 -mcpu=604e
214.5 Kpos/sec, 215.2 Kpos/sec

gcc 3.2 -O2
197.9 201.2

gcc 3.2 -O2 -mcpu=604e
199.4 201.1 198.7

I've played with some other settings and have been unable to  get 3.2 to match or beat 2.95's performance here.

Note, using the 3.3 mainline of about July 23, I got about equivalent runtime performance on the same benchmark.

>How-To-Repeat:
Build with the same compile-time flags.

Run as:

c4 < input

>Fix:

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