Where I work, we use a number of different machines to encode and decode video streams. The actual code that’s doing the encoding and decoding is a closed-source library we buy from either Intel or AMD, depending upon the codec. The problem is, when I get to rendering the output of the decoders, I get a stutter if I am using two screens on one system. If I use one screen, the stutter doesn’t happen.
I have inserted various timing mechanisms to ascertain what’s going on, using the GL_EXT_timer_query I have found that rendering each screen takes almost 16 ms per screen. The stutter, however, seems to be produced by the renderer showing old frames. I have seen (rarely) frames from at least 40 or 50 frames back reappear all of a sudden, but much more commonly, the video seems to jitter or shake. I have tried various techniques to try to rule out things like context switching, to no avail.
The rendering pipeline I am using is rather simple, I take Y U and V color planes, stick them into individual luminosity textures using the PBO techniques suggested for textures that are often updated, then process those textures in a fragment shader which takes the appropriate texel from each texture into a vector, multiplies that by a YUV to RGB transform, and displays the result.
The file /proc/driver/nvidia/version gives me this output:
NVRM version: NVIDIA UNIX x86 Kernel Module 100.14.09 Sat May 26 00:47:07 PDT 2007
GCC version: gcc version 4.1.2 20070502 (Red Hat 4.1.2-12)
I am running Fedora Core 7 on all machines concerned. I have 3 different machine build-outs that are affected:
Pentium D 3.0 Ghz with GF 6200FX
Core 2 Duo 2.4 Ghz with GF 7600 GS
Core 2 Duo 2.4 Ghz with GF 8600 GT
Does anybody have any idea how I can fix this issue? Might it be a driver issue? I really don’t have “authority” to change out drivers, however, I can push the necessary buttons to try to get that to happen.
Thanks in advance,
Rob Stoddard