To: Olly Betts <olly AT mantis DOT co DOT uk>
Cc: djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Re: Speed tuning programs 
Date: Sun, 14 Aug 94 18:34:25 +0300
From: eliz AT is DOT elta DOT co DOT il

> Compiler           Time (secs)
> 
> BCC 3.1, no opt    5.60
> BCC 3.1, -Os size  5.49
> BCC 3.1, -Ot time  5.44
> DJGPP 1.11m5 -O2   6.87
> DJGPP 1.12 -O2     6.43

	1) Try using -O3, which adds additional optimisations.  GCC
info file has a list of optimization options which you might also
consider (I think, loop-unrolling isn't included even in -O3, so
you might try it, too).
	2) The runnning time is quite short (5-6 seconds), so the
overhead of loading go32 and protected mode set-up before your
program gets control can be significant in relative figures.  Are
these running times characteristic of your typical application?
If so, what difference does 1 second make? If not, try running for
longer time, and see if you get better performance *ratio*.
	3) You didn't mention if you use some software disk cache
like SmartDrive.  Do you?  If you do, you can forget about the claims
that ``DJGPP is a bit weak if you're doing lots of I/O stuff'' (but see
4 below), because the cache all but eliminates I/O performance problems.
	4) On a similar machine, 7 seconds is what takes to DJGPP-compiled
program to read a 4-megabyte file (with SmartDrive installed), so if you
read multi-megabyte files like this, then yes, you might have I/O
bottleneck.  I think, this is somewhat due to the fact that DJGPP
read()'s files in chunks of 4KB, which might be sub-optimal when reading
LARGE files.  Otherwise, your problem is *not* I/O.
	5) If your program was originally born for Borland C, then it
might have some algorithmic problems which are necessary when writing
16-bit programs.  You might consider going through your code and searching
for such things, which may boost performance with GCC.

Hope this helps,
		Eli Zaretskii