Re: Some benchmarking on graphics

----- Original Message -----
From: "Dmitry Karasik" <dmitry_at_karasik.eu.org>
To: <prima_at_prima.eu.org>

 Michael> Using Bitmaps instead of Images gave an breathtaking speed up
 Michael> from 3 to 22 fps (under Windows 2000) Also buffered mode with
 Michael> direct canvas draw (which generally is already = very fast) will
 Michael> again speed up things considerable (in contrast to expectations

D> That's expected. DeviceBitmaps are supposed to correspond to a video
D> device format in best possible way, hence the name. Images may ( and
D> usually do ) undergo conversion, which of course slows things a lot.

Michael> of "buffering")

My surprise was caused by the speed-up on using "buffered mode".
On Windows, the direct outputting of a single graphic operation seems to be more costly
than to pick them up in an intermediate buffer and output it only once. This however only affects
the "direct mode", there is not much change with image copying, as can be ssen in the figures and
*is* expected after some thinking ;-)

The effect ist quite impressive on a windows machione, though - it seem from Dimitry's
figures - disappointig on Unix (see line 2 below).

I add my own figures of a similar hardware (Dell-Laptop 800 MHz), but a different OS.
It seems, on Windows2000 DeviceBitmaps are *very* efficient, bufferes mode is a good
thing even for performance improvement, in contrast to Unix, whereas changing colors is
*much*more expensive than on Unix.

D> Here: FreeBSD v4.3, XFree86 v4.0.3, AMD Athlon 650MHz, 800x800
M> Windows results in brackets()

               Canvas Image DeviceBitmap
0 19(19) 2 (3) 11(23)
B 11(22) 2(3) 9(22)
M 2 1 2
S 21(34) 3(3) 12(45)
BM 2 1 2
BS 13(45) 3 (3) 11(45)
MS 3 2 2
BMS 2 1 2

(all frames per second)

BTW I got a confusing message of your server, indicating the appendix had not been sent.
Interested persons can download the script from
http://www.mpeuser.com/perl/bench_canvas.pl
as well.

Kindly Mike
Received on Sat 22 Feb 2003 - 16:10:51 CET