Help needed to reduce unusually high CPU utilization in v4l2src, caused by memory copy.
Locked
 
|
Hi,
I am optimizing a V4L2 driver for video capture device. In the v4l2 driver, the DMA is performed based on frames using contingent memory buffers. After a frame is captured, the frame data is copied to a application buffer in v4l2src. gst_v4l2src_grab_frame() -> *buf = gst_buffer_copy (pool_buffer); However, the time spent on copying the buffer is very costly, about 0.01 seconds, which is very large considering frame time is only 0.033 seconds. I do not know whether this is the right forum. Any suggestions will be appreciated. Thanks, Yiliang |
Re: [gst-devel] Help needed to reduce unusually high CPU utilization in v4l2src, caused by memory copy.
|
|
Hi,
would it ber possible to mmap the memory are from the driver so that v4l2src passes that very buffer to the remaining of the pipeline? The only problem would be that, in case of a circular buffer in the driver, each memory area will remain locked until the pipeline has ended processing it. On Thu, Jan 21, 2010 at 12:57 AM, yiliang <[hidden email]> wrote: > > Hi, > > I am optimizing a V4L2 driver for video capture device. In the v4l2 driver, > the DMA is performed based on frames using contingent memory buffers. After > a frame is captured, the frame data is copied to a application buffer in > v4l2src. > > gst_v4l2src_grab_frame() > -> *buf = gst_buffer_copy (pool_buffer); > > However, the time spent on copying the buffer is very costly, about 0.01 > seconds, which is very large considering frame time is only 0.033 seconds. > > I do not know whether this is the right forum. Any suggestions will be > appreciated. > > Thanks, > Yiliang > > -- > View this message in context: http://n4.nabble.com/Help-needed-to-reduce-unusually-high-CPU-utilization-in-v4l2src-caused-by-memory-copy-tp1019508p1019508.html > Sent from the GStreamer-devel mailing list archive at Nabble.com. > > ------------------------------------------------------------------------------ > Throughout its 18-year history, RSA Conference consistently attracts the > world's best and brightest in the field, creating opportunities for Conference > attendees to learn about information security's most important issues through > interactions with peers, luminaries and emerging and established companies. > http://p.sf.net/sfu/rsaconf-dev2dev > _______________________________________________ > gstreamer-devel mailing list > [hidden email] > https://lists.sourceforge.net/lists/listinfo/gstreamer-devel > ------------------------------------------------------------------------------ Throughout its 18-year history, RSA Conference consistently attracts the world's best and brightest in the field, creating opportunities for Conference attendees to learn about information security's most important issues through interactions with peers, luminaries and emerging and established companies. http://p.sf.net/sfu/rsaconf-dev2dev _______________________________________________ gstreamer-devel mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/gstreamer-devel |
|
|
Hi Gibro,
Thanks for the kind reply. Even I remove the memcpy in v4l2src, the overhead is still there. I guess the overhead is just shifted to another stage that consumes the data. I found the problem was with the coherent buffer used in the driver, since the v4l2 driver I am investigating uses the buffer queue defined in videobuf-dma-contig.c. After I replace the coherent buffers with streaming buffers, the problem goes away. One correction to my previous email. The memory should "contiguous" instead of "contingent" Best regards, Yiliang On Fri, Jan 22, 2010 at 11:48 PM, Gibro Vacco <[hidden email]> wrote:
------------------------------------------------------------------------------ Throughout its 18-year history, RSA Conference consistently attracts the world's best and brightest in the field, creating opportunities for Conference attendees to learn about information security's most important issues through interactions with peers, luminaries and emerging and established companies. http://p.sf.net/sfu/rsaconf-dev2dev _______________________________________________ gstreamer-devel mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/gstreamer-devel |
Re: [gst-devel] Help needed to reduce unusually high CPU utilization in v4l2src, caused by memory copy.
|
|
What device is this coming from? Do you experience this for all v4l2
video capture sources? I think thisq is probably typical of a lot of mediocre "web cams." On Sat, Jan 23, 2010 at 8:24 PM, Yiliang Bao <[hidden email]> wrote: > Hi Gibro, > > Thanks for the kind reply. Even I remove the memcpy in v4l2src, the overhead > is still there. I guess the overhead is just shifted to another stage that > consumes the data. I found the problem was with the coherent buffer used in > the driver, since the v4l2 driver I am investigating uses the buffer queue > defined in videobuf-dma-contig.c. After I replace the coherent buffers with > streaming buffers, the problem goes away. > > One correction to my previous email. The memory should "contiguous" instead > of "contingent" > > Best regards, > Yiliang > > On Fri, Jan 22, 2010 at 11:48 PM, Gibro Vacco <[hidden email]> wrote: >> >> Hi, >> >> would it ber possible to mmap the memory are from the driver so that >> v4l2src passes that very buffer to the remaining of the pipeline? >> >> The only problem would be that, in case of a circular buffer in the >> driver, each memory area will remain locked until the pipeline has >> ended processing it. >> >> On Thu, Jan 21, 2010 at 12:57 AM, yiliang <[hidden email]> wrote: >> > >> > Hi, >> > >> > I am optimizing a V4L2 driver for video capture device. In the v4l2 >> > driver, >> > the DMA is performed based on frames using contingent memory buffers. >> > After >> > a frame is captured, the frame data is copied to a application buffer in >> > v4l2src. >> > >> > gst_v4l2src_grab_frame() >> > -> *buf = gst_buffer_copy (pool_buffer); >> > >> > However, the time spent on copying the buffer is very costly, about 0.01 >> > seconds, which is very large considering frame time is only 0.033 >> > seconds. >> > >> > I do not know whether this is the right forum. Any suggestions will be >> > appreciated. >> > >> > Thanks, >> > Yiliang >> > >> > -- >> > View this message in context: >> > http://n4.nabble.com/Help-needed-to-reduce-unusually-high-CPU-utilization-in-v4l2src-caused-by-memory-copy-tp1019508p1019508.html >> > Sent from the GStreamer-devel mailing list archive at Nabble.com. >> > >> > >> > ------------------------------------------------------------------------------ >> > Throughout its 18-year history, RSA Conference consistently attracts the >> > world's best and brightest in the field, creating opportunities for >> > Conference >> > attendees to learn about information security's most important issues >> > through >> > interactions with peers, luminaries and emerging and established >> > companies. >> > http://p.sf.net/sfu/rsaconf-dev2dev >> > _______________________________________________ >> > gstreamer-devel mailing list >> > [hidden email] >> > https://lists.sourceforge.net/lists/listinfo/gstreamer-devel >> > >> >> >> ------------------------------------------------------------------------------ >> Throughout its 18-year history, RSA Conference consistently attracts the >> world's best and brightest in the field, creating opportunities for >> Conference >> attendees to learn about information security's most important issues >> through >> interactions with peers, luminaries and emerging and established >> companies. >> http://p.sf.net/sfu/rsaconf-dev2dev >> _______________________________________________ >> gstreamer-devel mailing list >> [hidden email] >> https://lists.sourceforge.net/lists/listinfo/gstreamer-devel > > > ------------------------------------------------------------------------------ > Throughout its 18-year history, RSA Conference consistently attracts the > world's best and brightest in the field, creating opportunities for > Conference > attendees to learn about information security's most important issues > through > interactions with peers, luminaries and emerging and established companies. > http://p.sf.net/sfu/rsaconf-dev2dev > _______________________________________________ > gstreamer-devel mailing list > [hidden email] > https://lists.sourceforge.net/lists/listinfo/gstreamer-devel > > -- http://codebad.com/ ------------------------------------------------------------------------------ Throughout its 18-year history, RSA Conference consistently attracts the world's best and brightest in the field, creating opportunities for Conference attendees to learn about information security's most important issues through interactions with peers, luminaries and emerging and established companies. http://p.sf.net/sfu/rsaconf-dev2dev _______________________________________________ gstreamer-devel mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/gstreamer-devel |
«
Return to GStreamer-devel
|
1 view|%1 views
| Free forum by Nabble | Edit this page |