for information : what is the size of the pixel buffer?

thx Cyrille

Le 06/06/2010 16:49, Claude Heiland-Allen a écrit :

Hi list,

Is there any way in current Gem to use asynchronous readback of pixel buffer objects to speed up recording (or other GPU<->CPU transfer)?

In one of my projects in plain C/OpenGL I used this technique successfully:

1. naive (blocking) glReadPixels direct to CPU ram and piped via

stdout to ppmtoy4m|y4mscaler|mpeg2enc -> about 5fps

2. async glReadPixels to pixel buffer object mapped into the CPU

with one frame latency and piped via stdout to same encoder -> about 20fps

3. async glReadPixels to PBO to CPU piped via stdout to /dev/null

-> about 25fps

(laptop with 4GB ram, intel core2 dual core cpu, nvidia g105m gpu, tested with 32bit gnu/linux/ubuntu/karmic generic-pae kernel)

Thanks for any tips,

Claude

On 2010-06-06 16:49, Claude Heiland-Allen wrote:

Hi list,

Is there any way in current Gem to use asynchronous readback of pixel buffer objects to speed up recording (or other GPU<->CPU transfer)?

[pix_texture] can use asynchronous CPU->GPU transfers. they can be set via the "pbo" message (e.g. [pbo 2( will create 2 pbos for buffering)

there is no such thing for [pix_snap] yet.

fgmasdr IOhannes