From: Brad Spencer <spencer AT ug DOT cs DOT dal DOT ca>
Newsgroups: comp.os.msdos.djgpp
Subject: Optimization flags [was Re: I don't understand djgpp's optimizations]
Date: 6 Apr 1997 02:41:11 GMT
Organization: iSTAR Internet Incorporated
Lines: 37
Message-ID: <5i72g7$7ib@news.istar.ca>
References: <33440E51 DOT 6AE AT imaginet DOT fr> <5i2fm8$b34_001 AT tuwien DOT ac DOT at>
NNTP-Posting-Host: ts20-03.hfx.istar.ca
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp

godzilla wrote:

> i'm using '-O3 -ffast-math -funroll-loops -fomit-frame-pointer' and the
> runtime-difference to watcom-generated code is usually between 10 and 20 %
> (my loops are huge and no so simple)
> nonetheless i prefer djgpp :-)
> greetings

Speaking of optimization flags, I was playing around with my 3D engine
work-in-progress :) and I found the following combination to be 'the
best'.  Mind you, I didn't compare -O2 and -O3 since -O3 and I tend to
get in a lot of arguments.
Anyway, the flags:

The standard stuff

	-O2 -ffast-math -fomit-frame-pointer

plus a little seasoning

	-malign-double -malign-loops=0 -malign-jumps=0

Now, I'm not sure if aligning to 0 is the same as aligning to 1, but it
does what I want and totally throws out any jump alignment.  I'm only
interested in Pentium machines, and AFAIK alignment just slows things
down (NOPs to process), or at least it did for me.

I'm not convinced that -m486 actually does _anything_ (looking briefly
at assembly output, which is what I seem to be spending a lot of time
doing :), so I forget all about it.

Just thought I'd mention this, as I've never seen mention of the jump
alignment switches before...

------------------------------------------------------------------
Brad Spencer - spencer AT ug DOT cs DOT dal DOT ca - in CS @ Dalhousie University
"S.M.A.K.I.B.B.F.B." - A.J. Rimmer | http://ug.cs.dal.ca/~spencer/