From:	"Andrew Crabtree" <andrewc AT rosemail DOT rose DOT hp DOT com>
Newsgroups:	comp.os.msdos.djgpp
Subject:	Re: I Thankyor
Date:	Wed, 7 Jan 1998 10:04:16 -0800
Organization:	Hewlett Packard
Lines:	22
Message-ID:	<690g2l$di9$1@rosenews.rose.hp.com>
References:	<01bd1b87$ddabca40$093163c3 AT default>
NNTP-Posting-Host:	ros51675cra.rose.hp.com
Mime-Version:	1.0
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp

>Mindyou by what you have said i don't think its worth it, i've used -03
>optimization, which does increase performance a great deal, however, i'm a
>speed freak so i want more :)
Well, you can try pgcc (www.goof.com) on -O6, it optimizes much better than
gcc for x86 targets.

> which seems like the compilier
>optimizations.Thats handy because that's generally a lot easier to do and
>understand  (and doesn't include pointers!).
I wouldn't discount the use of the register keyword as useless as some
others suggested.  Really, c optimizers can be very stupid.  Unless you
profile your data and keep track of branch flow, and then feed the profile
data back into the compiler to optimize reg use based on it (which I doubt
you are doing), then
using the register keyword should help.  It also has the benefit of
documenting your code.  You are effectively telling the compiler which
branches you think will be hit most often.

Andy

- Raw text -