Bernardo Innocenti wrote:
> This patch is likely to make a big difference in the EXA codepath
> I benchmarked, which is heavily used by the OLPC GTK theme.
>
> The mini-benchmark attached to #1837 mimics it by drawing rounded
> edges shapes with Cairo.
> [...]
> -------- Original Message --------
> Subject: [PATCH] Avoid an unwanted pixmap migration in EXA when
> compositing trapezoids
> Date: Wed, 15 Aug 2007 19:58:56 +0200
> From: Fredrik Höglund <[EMAIL PROTECTED]>
I made quite some testing with the attached benchmark: I tried with a few
different configurations (24 and 16 bit display, different snapshots of
xserver 1.4): speed increase is consistently at 30%.
Definitely not bad, and the patch is already committed in the server-1.4-branch.
Looking at oprofile output is evident that the main optimization is avoiding
memcpy'ing the pixmaps.
Benchmark result without the patch (16 bpp):
op: rect
user: 2.600
sys: 0.050
real: 7.930
op/s: 126.103
Oprofile report:
samples| %|
------------------
534 56.8085 Xorg
TIMER:0|
samples| %|
------------------
243 45.5056 libc-2.6.so
211 39.5131 libpixman-1.so.0.9.4
38 7.1161 amd_drv.so
17 3.1835 libexa.so
16 2.9963 Xorg
8 1.4981 libfb.so
1 0.1873 libextmod.so
[...]
samples % linenr info image name app name
symbol name
225 23.9362 (no location information) libc-2.6.so Xorg
memcpy
213 22.6596 (no location information) libcairo.so.2.11.5 python
(no symbols)
168 17.8723 pixman-edge.c:322 libpixman-1.so.0.9.4 Xorg
pixman_rasterize_edges
63 6.7021 (no location information) no-vmlinux
no-vmlinux (no symbols)
56 5.9574 (no location information) libpython2.5.so.1.0 python
(no symbols)
19 2.0213 cim_gp.c:3305 amd_drv.so Xorg
gp_wait_until_idle
16 1.7021 (no location information) libc-2.6.so python
msort_with_tmp
12 1.2766 pixman-mmx.c:0 libpixman-1.so.0.9.4 Xorg
__divdi3
11 1.1702 cim_gp.c:1228 amd_drv.so Xorg
gp_color_bitmap_to_screen_blt
** With the patch applied:
op: rect
user: 2.550
sys: 0.020
real: 6.070
op/s: 164.745
samples| %|
------------------
709 47.1096 Xorg
TIMER:0|
samples| %|
------------------
511 72.0733 libpixman-1.so.0.9.4
71 10.0141 amd_drv.so
43 6.0649 Xorg
40 5.6417 libexa.so
36 5.0776 libc-2.6.so
7 0.9873 libfb.so
1 0.1410 librt-2.6.so
[...]
samples % linenr info image name app name
symbol name
407 27.0432 (no location information) libcairo.so.2.11.5 python
(no symbols)
322 21.3953 pixman-edge.c:322 libpixman-1.so.0.9.4 Xorg
pixman_rasterize_edges
130 8.6379 (no location information) no-vmlinux no-vmlinux
(no symbols)
103 6.8439 (no location information) libpython2.5.so.1.0 python
(no symbols)
90 5.9801 pixman-utils.c:165 libpixman-1.so.0.9.4 Xorg
pixman_fill
40 2.6578 (no location information) libc-2.6.so python
msort_with_tmp
32 2.1262 cim_gp.c:1228 amd_drv.so Xorg
gp_color_bitmap_to_screen_blt
#!/usr/bin/python
from __future__ import division
import __main__
import cairo
import gtk
import math
import os
NFRAMES = 1000
WIDTH = 150
HEIGHT = 50
def result(name, start, end):
elapsed_user = end[0] - start[0]
elapsed_sys = end[1] - start[1]
elapsed_real = end[4] - start[4]
print " op: %s" % name
print "user: %.3f" % elapsed_user
print " sys: %.3f" % elapsed_sys
print "real: %.3f" % elapsed_real
print "op/s: %.3f" % (NFRAMES / elapsed_real)
def bench(op, drawarea):
cr = drawingarea.window.cairo_create()
start = os.times()
for i in range(0, NFRAMES):
op(cr, i)
end = os.times()
result(op.__name__, start, end)
def arc(cr, frame):
cr.set_source_rgb(frame % 2, (frame // 2) % 2, (frame // 3) % 2)
cr.arc(150 + frame / 500, 150, 50 + (frame / 10), 0, 2 * math.pi)
cr.fill()
def rect(cr, frame):
cr.set_source_rgb(frame % 2, (frame // 2) % 2, (frame // 3) % 2)
cr.arc(150 + frame / 500, 150, 50 + (frame / 10), 0, 2 * math.pi)
cr.fill()
def rounded_rect(cr, frame):
cr.set_source_rgb(frame % 2, (frame // 2) % 2, (frame // 3) % 2)
cr.arc(WIDTH - 20, 20, 20, - math.pi / 2, 0);
cr.arc(WIDTH - 20, HEIGHT - 20, 20, 0, math.pi / 2);
cr.arc(20, HEIGHT - 20, 20, math.pi / 2, math.pi);
cr.arc(20, 20, 20, - math.pi, - math.pi / 2);
cr.close_path()
cr.fill()
win = gtk.Window()
drawingarea = gtk.DrawingArea()
win.add(drawingarea)
drawingarea.set_size_request(1200,900)
win.show_all()
bench(rect, drawingarea)
del win
_______________________________________________
Devel mailing list
[email protected]
http://lists.laptop.org/listinfo/devel
