A bug when running 3D Benchmark of Quadrant?

62 views
Skip to first unread message

csanta

unread,
May 9, 2011, 5:58:29 PM5/9/11
to Aurora Softworks: Quadrant
Hello,

I'd like to ask whether the Quadrant developers have noticed a bug in
the 3D benchmark, specifically when running the Planets demo (Mars and
a moon). If you look really close you will notice black pixels coming
through the small white planet. This has been reproduced in Droid1,
DroidX and even Samsung Galaxy, all of them having an SGX GPU from Img
Technologies. The real question that I wanted to ask is that when you
look in logcat you will notice that the 3D app has actually selected
an software EGLConfig, and that's why the low FPS in the planet demo.
I know for a fact that the s/w GL renderer is executing as oppose to
the GPU because of the following message:

D/dalvikvm( 2117): GC_EXPLICIT freed 42K, 32% free 7338K/10759K,
paused 1ms+2ms
I/ARMAssembler( 2117): generated
scanline__00000197:03010144_00009501_00000000 [141 ipp] (225 ins) at
[0x598941e8:0x5989456c] in 396729 ns
I/ ( 1820): Allocated 0x599be000 size 294912
I/com.aurorasoftworks.qua( 2117): Video surface created
I/com.aurorasoftworks.qua( 2117): Video surface changed
I/com.aurorasoftworks.qua( 2117): Video surface destroyed

The generated scanline is actually coming from PixelFlinger, which is
Android's s/w GL renderer.
Another test that I did was that if I remove the software GL library
(libGLES_android.so), then both issues dissapear, the planet demo runs
way faster, and the black issues is gone (which imply a bug in
PixelFlinger).

Any comments?

thanks,

CS

Michael Leone

unread,
May 19, 2011, 2:31:09 PM5/19/11
to Aurora Softworks: Quadrant
We see exactly the same issue and have done similar things to prove
it. Since we don't have source and the Quadrant developers won't
respond to the email address on their website, we have to guess but we
believe they are incorrectly trying to use a 16-bit depth and only S/W
GL supports that. Hence they don't use hardware GL and you see the
artifacts. We'd like to learn more about the logic they use in their
EGL config chooser. This should only be a 5 min conversation or email
and the benchmark could be easily fixed.

-Michael
Reply all
Reply to author
Forward
0 new messages