How does one check whether XRENDER is accelerated?

[Carsten Haitzler (The Rasterman)]

On Thu, 09 Feb 2006 05:38:02 +0300 Samium Gromoff <_deepfire at mail.ru> babbled:

> > i can't say for others - but i have written 2 tools that basically give u a
> > reasonable idea of overall render accel
> > 
> > http://www.rasterman.com/files/render_bench.tar.gz
> > 
> > is an old benchmarker that pits xrender against software equivalent
> > rendering. it gives one view of gneral render performance as opposed to
> > "diy" in software client-side.
> > 
> > another (more recent and more expansive) is evas - it has full xrender
> > suport as well as software rendering, opengl, directfb, etc. etc.
> > 
> > you basically can just tun the test program (that sits and benchmrks each
> > engine) and see the benchmark values u get at the end.
> > 
> > http://enlightenment.freedesktop.org/fetch.php?file=evas-0.9.9.023.tar.gz
> > 
> > - just bild, install then run the evas_*_test programs and they will spew
> > out a benchmark to stdout. the higher the value - the faster the engine's
> > rendering.
> 
> According to the evas backend tests:
> 
> evas_xrender_x11_test:
> ####################################################
> # Performance Test. Your system scores...
> ####################################################
> # FRAME COUNT: 1057 frames
> # TIME:        20.016 seconds
> # AVERAGE FPS: 52.807 fps
> ####################################################
> # Your system Evas Benchmark:
> #
> # EVAS BENCH: 0.880
> #
> ####################################################
> evas_software_x11_test:
> ####################################################
> # Performance Test. Your system scores...
> ####################################################
> # FRAME COUNT: 6779 frames
> # TIME:        20.004 seconds
> # AVERAGE FPS: 338.885 fps
> ####################################################
> # Your system Evas Benchmark:
> #
> # EVAS BENCH: 5.648
> #
> ####################################################
> evas_buffer_test:
> ####################################################  
> # Performance Test. Your system scores...
> ####################################################
> # FRAME COUNT: 6391 frames
> # TIME:        20.002 seconds
> # AVERAGE FPS: 319.521 fps
> ####################################################
> # Your system Evas Benchmark:
> #
> # EVAS BENCH: 5.325
> #
> ####################################################
> 
> It appears that XRENDER is a dog, for some reason...

it depends on 1. your gfx card chipset, 2. what version of x (x.org) and your
settings, how much video ram etc.

for example - on my 6600gt evas gl gets about 60 or so benchmarks. in
software_x11 on my home box it pushes about 6.0. on my radeon9000 (r250)
laptop, software pushes about 4.x, opengl under the open radeon drivers is
about 9.x and xrender gets about 12.x - but this is a cvs snapshot from a few
months back and it's using exa etc.as accel method. fglrx drivers get opengl
rendering scores of about 20, software_x11 is the same (it's driver independant
really) and xrender gets like 0.2 (it's hopelessly accelerated in fglrx). my
point is - your scores will vary WILDLY based on gfx chipset, cpu, memory bus,
driver, settings and phase of the moon. the software based engines
(software_x11, buffer and fb) will give roughly consistent values on all cpu's
o the same model and clock speed (and same memory clockrate/fsb) as its the
same rendering code everywhwre. when using xrender it may be using software
fallbacks, or partial software fallbacks and partial gfx chip help, or almost
entirely be going via an accelerated pipeline. that test is just an overall
measurment of the routines i tend to need/use a lot from xrender (and it weighs
in heavily needing solid blits and alpha blends WITH filtered and unfilteres
transforms/scaling).

> Also, a side note: the XRENDER test gave a SIGSEGV on the test 36
> called "lines".

yes - i have noticed this. it's ended up always being a segv inside libxrender
inside the trapezoid decomposition routines when passing in a poly outline. i
haven't traced it further than that though but i suspect its a corner case i've
hit inside libxrender.


-- 
------------- Codito, ergo sum - "I code, therefore I am" --------------
The Rasterman (Carsten Haitzler)    raster at rasterman.com
裸好多
Tokyo, Japan (東京 日本)