Date: Fri, 25 Jun 93 10:31:36 EDT
From: kent AT cme DOT nist DOT gov (Ernie Kent)
To: djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Re: Direct Graphics Memory Writing


>Christopher Matthews Keane, Subject: Direct Graphics Memory Writing
  
> I am using DJGPP 1.09 with GRX in developing some image processing
> programs.  I currently have the screen refresh being done by indexing
> the color to the LUT - and plotting the point via GrPlot(x, y, c);

> Anyone know of a
> routine to move whole lines via direct video writing at a fast
> rate....  I must admit that this garbage is comes up on the screen
> VERY quickly!

I have code for moving blocks (including single lines) of
any x/y extent to the screen by direct write to 0xd0000000
(or the appropriate offset from it if a block goes in the
middle of the screen.) It is written in gas assembler so it
can be assembled and linked by gcc. It works under 1.09
(1.10 not tested.) It is pretty well optimized, and I believe
it is about as fast as it can be done in a reasonably flexible
manner, i.e., no special tricks.  Using a 486DX50 with a
local video bus it is instantaneous perceptually, even for
large images. It moves data in DWORD-sized chunks, so you
either need to move blocks that are multiples of four bytes
in width (anything in height) or pad them with, e.g., zeros.
If this is offensive it is easily changed to write 2-byte
or single-byte increments at some cost in speed.

I wrote it because the library routines were too slow for some
of my applications, and it is indeed significantly faster, at
least on my hardware. It is pretty trivial, but anyone is welcome
to it.

Ernie Kent