From: bcrowell AT eos DOT ncsu DOT edu (Bennett Crowell)
Newsgroups: comp.os.msdos.djgpp
Subject: Re: Watcom vs DJGPP
Date: 6 Nov 1997 22:40:46 GMT
Organization: North Carolina State University
Lines: 32
Message-ID: <63th1e$ber$1@uni00nw.unity.ncsu.edu>
References: <1 DOT 5 DOT 4 DOT 32 DOT 19971103185401 DOT 006f5e38 AT dce03 DOT ipt DOT br> <63nhql$ri3 AT news DOT kom DOT tuwien DOT ac DOT at>
NNTP-Posting-Host: c13211-144wl.bae.ncsu.edu
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Precedence: bulk

In article <63nhql$ri3 AT news DOT kom DOT tuwien DOT ac DOT at>,
godzilla <e8725229 AT stud1 DOT tuwien DOT ac DOT at> wrote:
>
>personal experience: watcom-compiled progs (extended dos with dos4/gw 
>extender) are 10 to 15 (20) percent faster then djgpp-compiled ones on a 
>pentium (with all (p5-specific) optimizations turned on), while there was 
>no difference in speed on a 486 (with 486- optimizations) -> the lacking 
>pentium-specific optimization
>might account for the difference.
>
>.. this may have no general significance, since all the programs i wrote at 
>that time where roughly of the same kind (numerical integration of large 
>systems of diff. equations, little I/O-operations)
>
>greetings from vienna
>godzilla

I have had similar experiencs with Watcom 10.6 and djgpp running on
a 486 under DOS and also under Windows 95. The programs involved are
doing operations on matrices of up to 1000 X 2000 four-byte
floating point elements. Profiling indicates that most of the time
is spent in the matrix multiply routine. So far the Watcom product
seems to generate code that is at least equal to and usually a few
percent faster than gcc.

Current system: 100 MHz 486, 16 Mb RAM, Windows 95

-- 


Bennett Crowell
bcrowell AT eos DOT ncsu DOT edu