Andre DRASZIK wrote:
Hi,
On Wed, 2010-05-05 at 12:43 +0200, Matteo Canella wrote:
CPU is near 100% because it's busy to send drawing command to the GPU.
It is usign a FIFO to send those commands and the most of the time
it's waiting for the FIFO to be free. This is not a big deal for the
normal use of the library.
Sure, depending on how much you draw, of course. Just something I
noticed... Being able to batch loads of operations without having to
busy wait and instead do something else while it's drawing is one of our
use cases, so 100% for CPU while drawing would not be acceptable for us.
You can compare with the SH772x results on the web site, quite old news.
Below 10% CPU load in FillRectangle test (100% GPU) should be a good measure,
at least for the default benchmark size of 256x256.
Anyway, you say that the number are not great, but compared to what?
Not knowing where your CPU utilisation came from, it could have been
that CPU use was too high to efficiently drive the GPU (as fast as
possible) ... I.e. the GPU idling waiting for instructions from the CPU
or so, but doesn't seem to be the case.
I would e.g. question why fill spans is so much slower than fill
rectangle.
The CPU bottle neck causes FillSpan to be slow, too much overhead.
Try with df_dok --size 512x512 or higher to determine where is the
break even point, i.e. 99% CPU with 99% GPU load.
--
Best regards,
Denis Oliver Kropp
.------------------------------------------.
| DirectFB - Hardware accelerated graphics |
| http://www.directfb.org/ |
"------------------------------------------"
_______________________________________________
directfb-users mailing list
[email protected]
http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-users
