Date:	Wed, 30 Dec 1998 16:40:28 +0200 (IST)
From:	Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
X-Sender:	eliz AT is
To:	Christian Hofrichter <ChristianHofrichter AT gmx DOT de>
cc:	djgpp AT delorie DOT com
Subject:	Re: pairable instructions much faster than the string operations on a Pentium and above ?!
In-Reply-To:	<368A195D.F315167E@gmx.de>
Message-ID:	<Pine.SUN.3.91.981230163234.15292A-100000@is>
MIME-Version:	1.0
Reply-To:	djgpp AT delorie DOT com

On Wed, 30 Dec 1998, Christian Hofrichter wrote:

> So how about a new optimation-switch in djgpp, called pairable
> instructions ?

It is not clear what exactly would you want that switch to do, and in 
which circumstances.

If you want the compiler to emit Pentium-optimized code when it sees e.g. 
a call to memcpy or memset, then this suggestion should go to the 
maintainers of GCC, since the DJGPP port doesn't change the way GCC 
generates code.  There is already -mpentium and -march= switches which 
should trigger such code generation.

If you want the DJGPP library functions to use the faster method, then a
GCC switch cannot be used for that (well, not easily, anyway).  Perhaps a
specialized library of Pentium-optimized functions is required (provided
there are several other cases that can be optimized in this way). 

> After all  it can often double the speed of the program.

The ration 3:2 will not double the speed, it will only give you a 50% 
speed boost.

> I can also be used to improve graphic-performence, can't it ?

I'm not sure, since the bus speed and constraints are an important factor 
when you move data to/from the video RAM.

- Raw text -