At 08:04 �� 13/2/2002 +0100, you wrote:

>Marcel Kilgus makes some magical things to make me read
>}
>} What? That 32bit per pixel is far too memory consuming for a windowing
>} system that does the saving of backgrounds itself? Nothing proprietary
>} in this knowledge ;-)
>}
>} The PE was designed for 2 bit per pixel with a small resolution and
>} the concept was good for that purpose. It's however a very bad design
>} if you have more bits per pixel (one single(!) 800x600 window almost
>} needs 2MB in 32 bit mode!).
>
>It would be better if PE was able to compress the saved area on the fly,
>or even better, perform no saving and require a (re-)drawing routine
>from each application. But then this would be trading memory for CPU
>and more problems for the applications.

Well I don't think memory is of any concern for people with hi-colour mode 
capabilities...
I don't think there are users of QPC2 and Qx0  without at least 16 Megs of 
RAM...

As for compression of SCR, I am currently working on a modified variable 
RLE model (3 bytes or 4 bytes long) with the following specs:
1 bit (RLE code / 0 for <128 repetitions and 1 for >128 repetitions.)
7 bits (repetitions)  (if RLE=0) or
15 bits (repetions) (if RLE>0)
16 bits (pixel data)

However it's insanely slow with Basic (Peculiar thing here... VB 6.0 
compiled running Atop a 1GHz Athlon is slower compression-wise than a plain 
QL working with Peeks and Pokes in Memory compiled with Turbo - The QL 
version is at least twice as fast!!! Yep the QL always had kick-a## number 
crunching capabilities). Decompression is extremely fast both places which 
is really good. Especially if the file is loaded using FS.LOAD it's 
practically non-noticeable...


Phoebus

Reply via email to