>Don't bother to try it with Arachne 1,50b2 unless you happen to have a
>DPMI service provider in your path - it won't work.
I can state this directly (after reading through the docs for DJGPP):
Do NOT, I repeat do NOT, use the DPMI server from DR-DOS 7.0x! For best
performence use CWSDPMI.EXE.
>CWSDPMI.exe is included in the 1.6b1 distribution package to fatten it up
>a bit, I guess, because Arachne 1.6b1 seems to work without it.
>Maybe someone can tell me what it is there for.
It's for viewing PNG files.
image/png PNG>BMP|$ecwsdpmi.exe\n$epng2bmp.exe -s -o $2 $1
image/x-png PNG>BMP|$ecwsdpmi.exe\n$epng2bmp.exe -s -o $2 $1
I don't know why Michael loads cwsdpmi.exe before calling png2bmp since it
works fine anyway (all programs that use cwsdpmi.exe will first look in
memmory and then in current dir and then the path AFAIK).
>Which can be seen to be almost exactly twice as fast.
Or more.
>But, I think it would be more useful to reduce that 3.5 second paint job.
>Since the above improvement won't work at all on a '286 - which NEEDS it
>more !
I know people don't want to hear this, but who is trying to use a 286 for
viewing JPGs anyway? I can understand someone who wants to use it for
surfing - but for actually viewing images?
The sollution seems simple to me, this test shows that 32-bit *can* be
faster than 16-bit. So a 32-bit version of core.exe might be able to work
faster (nothing can be guaranteed). And there's a 32-bit WATTCP IIRC. Still
this doesn't increase speed on an XT/AT.
>Thanks, Bernie. Now how do we get enough information about Arachne to be
>able to write some video drivers in ASM ?
Anyone who knows how to write such in ASM can give it a try, the lopif
package should be up now. (Hmm... no it isn't). You'll need to talk to
Michael about that. I could of course upload my older version but I'll have
to assume something's changed since Arachne 1.60b1 is faster in your test.
To get back to djpeg.exe I was wondering if perhaps it's stupid to remove
the diffrent dithering options - perhaps I should atleast test floating
point on some CPU that's very good at it (an AMD Athlon for instance - but
I don't know anyone that has such a beast).
Compiling with 486 code didn't seem to give me more speed (I need to invent
a better way to test here on my end - reducing time from 3 seconds isn't
easy). How do you do Clarence?
Perhaps I should start testing with the 386SX-16 that sits in between my
monitor and keyboard?
//Bernie
http://bernie.arachne.cz/ DOS programs, Star Wars ...
