On 920365639 seconds since the Beginning of the UNIX epoch
Jonathan Stone wrote:
>
>
>I've been kernel-profiling rcons on a pmax. The machine is a 5000/150
>(50MHz r4000, 12.5MHz 32-bit TC bus) with PMAGB-B sfb" 1280x1024x8-bit board.
>
>The "benchmark' I've been using is
>    head -1000  /usr/share/misc/termcap
>
>to the rcons device; that takes ~7 minutes, catting the whole thing
>takes too long.

This is on the same track, but a bit of an aside.

I rewrote the rcons stuff on the alpha/TGA for my Multia (code
being reviewed by Ross), and got a 16x improvement in speed by
simply replacing copyrow() routine with something that uses the
TGA's blitting.

I've also been thinking about replacing the char painting code with
stipple blitting as that would allow the chars to be drawn a lot
faster, but to actually get much advantage out of that wscons would
have to pass me multiple characters at a time (which I was also
thinking of looking into at some point).

But all in all, I think that the biggest win for all platforms
would be jump scrolling as it reduces the number of times that rows
need to be copied.  (Thus a bit less MI acceleration would be
necessary.)

 == Roland Dowdeswell
 == http://www.imrryr.org/~elric/