Re: assembly language and unrolling loops
Thomas Davis wrote:
> Keep both, one for non-intel stuff, one for everything else.
> I think the code for the bi-directional is efficient enough. In testing
> my new (and unreleased) kernel driver, I can get 24fps out of the bloody
> thing (80x60, 6bpp) 17fps at 240x120, 6bpp, and 4fps at 320x240, 6bpp!
> These are the exact numbers I'm seeing on my P133.
I get these numbers inside a horribly mutilated version of xfqcam that
is totally coded ineffeciently, outputs its image to an xwindow, and
yields time to XForms to process a control panel. So if I can get this
performance under these circumstances, I suspect that unrolling loops
and writing assembly language stuff isn't going to buy us anything.
Before premature omtimizations, there are other problems that need to be
solved: most importantly in my book is that, except in 6bpp
bidirectional mode, we still don't know what sequence is needed to get
the camera to output a frame and then be in a state ready to output
another frame with only and xfer image command.
I'd like to know these so I can release xfqcam '95 deluxe (tm) ;^)