Some exciting news from Matt re: speedups in mjpeg software encoding. -- Professor Lawrence A. Rowe Internet: Rowe@BMRC.Berkeley.EDU Computer Science Division - EECS Phone: 510-642-5117 University of California, Berkeley Fax: 510-642-5615 Berkeley, CA 94720-1776 URL: http://bmrc.berkeley.edu/~larry
-- BEGIN included message
- To: Larry Rowe <larry@cs.berkeley.edu>
- Subject: Late night hacking
- From: Matthew Delco <delco@EECS.Berkeley.EDU>
- Date: Wed, 26 Jul 2000 23:03:46 -0700 (PDT)
After I got vic working again on my local machine I pasted in the MMX code. Without MMX the 422planar operation takes 33ms. With MMX the computation time is reduced to 3.6ms. At 95% quality I can get 15fps (20 with video window closed), and at both 75% and 30% quality I can get 20fps (23 with window closed). I also found that the reported bitrate can exceed the max. bitrate specified by the slider (by a large margin in some cases). In regards to the videoBridge, I found that the sleep code can cause the frame rate to drop from 30fps to 25fps. I'll have to investigate this more. I thought this might be related to the fact that I'm not using the ioctl() calls approach yet. However, the ioctl() calls we discussed appear to be for v4l instead of the jpeg, so I'm not sure how to code things just yet--since we're grabbing jpeg frames instead of raw frames I need to find some code that doesn't use a read() call instead of ioctls. Tomorrow I'll add the summary statics output to the client code of the videoBridge. MD P.S. The videoBridge runs quite nicely at 20Mbps during this time of the day--the occasional frame is lost but other than that the video quality is good.
-- END included message