From:	"David Wilson" <s335194 AT student DOT uq DOT edu DOT au>
Newsgroups:	comp.os.msdos.djgpp
Subject:	Pentium-optimizing gcc: speed test results
Date:	10 May 1997 13:49:59 GMT
Organization:	University of Queensland
Lines:	34
Message-ID:	<01bc5d49$0641c2a0$71db6682@s335194.student.uq.edu.au>
NNTP-Posting-Host:	ccjanew.slip.cc.uq.edu.au
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp

I downloaded an updated cc1.exe from ftp://ftp.goof.com/pub/pcg/cc1.exe.gz
the other day (2.7.2p snapshot 970426). I wanted to see if it produced
better-performing code with Pentium optimization enabled.

I compiled and ran a small program I got from Dr. Dobb's Journal a few
months back that generates a tilable texture image. It's very floating
point intensive and usually runs in about 2000 iterations - hence a perfect
candidate for Pentium optimizations.

The system configuration was:

* Pentium 100
* 16MB of RAM
* Running under Windows 95

Here are the results, compared to stock-standard DJGPP 2.01 with gcc
2.7.2.1:

Relative performance comparisons:

(1.00 == standard 2.7.2.1 no optimization)
				  +---------------------+---------+----------+
				  |   No Optimization   |   -O2   |   -O6    |
+-----------------+---------------------+---------+----------|
| gcc 2.7.2.1     |       1.00          |  2.80   |   2.82   |
+-----------------+---------------------+---------+----------|
| gcc 2.7.2p	  |		 1.0005         |  2.81   |   2.81   |
| snapshot 970426 |                     |		  |			 |
+-----------------+---------------------+---------+----------+


--David

- Raw text -