From:	Geza Herman <geza AT inf DOT bme DOT hu>
Newsgroups:	comp.os.msdos.djgpp
Subject:	Re: Getting the fastest running
Date:	Tue, 6 Jan 1998 14:58:49 +0100
Organization:	Technical University of Budapest
Lines:	32
Message-ID:	<Pine.GSO.3.95.980106144703.24767A-100000@kempelen.inf.bme.hu>
References:	<1B174837C8E AT fs2 DOT mt DOT umist DOT ac DOT uk>
NNTP-Posting-Host:	kempelen.inf.bme.hu
Mime-Version:	1.0
In-Reply-To:	<1B174837C8E@fs2.mt.umist.ac.uk>
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp

>   Geza Herman <geza AT inf DOT bme DOT hu> replied:-
> > Why must you copy? You can write directly to vid.mem. ...
> 
> I must copy because setting up the screen display involves writing and then
> overwriting (1) parts of the background which are hidden by objects, (2) parts
> of objects which are hidden by other objects. Compare the old and troublesome
> `hidden line' problem. If I wrote directly to screen, this hidden matter would
> keep flashing in and out of view. How can I tell the PC's graphics card not to
> redisplay the screen while I am writing a display to it?
> 
Ah. If you use VESA modes, you are able to write to a backscreen. Use the
SetDisplayStart function. Of course, this works if you have enough video
memory( for example in 640x480x256c you have to have 1Mb ). But of course,
buffering sometimes makes better results( for example, if you read this
buffer as well ). 

I disagree with Eli, because
the video memory is very fast in today machines( not 10 times slower ).
I experienced the following results( 486Dx2-66, Cirrus 5428 vid.card ):
	system mem->system mem: 15Mb/sec
	system mem filling: 55Mb/sec
	system mem->video mem: 6.5Mb/sec
	video mem filling: 10.5Mb/sec
( and Cirrus is an average card. S3 is much better )

Of course, with no cacheing. And the dosmemput/get functions is very
optimized. But the fastest is memcpy( because memcpy is inlined ) if you
copy little buffers.( if you copy large buffers, the speed is the same )

Regards,
Geza

- Raw text -