On 940800267 seconds since the Beginning of the UNIX epoch
Andrew Gillham wrote:
>
>[New Tests]
>NetBSD 1.4.1 (UVM patch) + stock XalphaNetBSD
>   120 reps @  57.4746 msec (    17.4/sec): Copy 500x500 from window to window
>  2000 reps @   4.6990 msec (   213.0/sec): 500x500 rectangle
>    70 reps @  81.6900 msec (    12.2/sec): (xor) 500x500 rectangle
>  5000 reps @   1.1911 msec (   840.0/sec): Fill 100x100 equivalent cmplx poly
>160000 reps @   0.0394 msec ( 25400.0/sec): Char in 80-char image line (6x13)
>300000 reps @   0.0217 msec ( 46100.0/sec): 10-pixel line segment
>
>NetBSD 1.4.1 (UVM patch) + R. C. Dowdeswell XalphaNetBSD
>   800 reps @   8.8409 msec (   113.0/sec): Copy 500x500 from window to window
>  3000 reps @   2.4917 msec (   401.0/sec): 500x500 rectangle
>  1000 reps @   5.2983 msec (   189.0/sec): (xor) 500x500 rectangle
>  8000 reps @   0.6773 msec (  1480.0/sec): Fill 100x100 equivalent cmplx poly
> 56000 reps @   0.0992 msec ( 10100.0/sec): Char in 80-char image line (6x13)
>100000 reps @   0.0509 msec ( 19600.0/sec): 10-pixel line segment

So, you'll notice that my server is a slower on the last two.
Well, now that read-mod-write is less expensive, I need to check
each of the accelerations to see if it still makes sense (in the
case of the last couple I lost out...)

So, I'll put off committing my changes until I get time to disable
some of the stuff.  And, I'll probably run a couple of experiments
to see if I can reclaim some higher performance by using another
tweak or two.

(For the interested, it turns out that using miFoo routines that
called pGC->ops->FillSpans were quicker than the cfb code before,
but now all that has changed so I need to go back to a few cfbFoo's
here and there.)

I hope to do that sometime in the coming week.

 == Roland Dowdeswell                      http://www.Imrryr.ORG/~elric/  ==
 == The Unofficial NetBSD Web Pages        http://www.Imrryr.ORG/NetBSD/  ==
 == The NetBSD Project                            http://www.NetBSD.ORG/  ==