Re: vivi kernel driver

2013-04-09 Thread Michal Lazo
memcpy do all cpu load And it is linear with copied memory size On Tue, Apr 9, 2013 at 3:26 PM, Hans Verkuil wrote: > On Tue 9 April 2013 15:05:37 Michal Lazo wrote: >> And can you explain where is problem with memcpy in kernel? >> is it really that big problem to copy 720*576*3 bytes from kernel

Re: vivi kernel driver

2013-04-09 Thread Hans Verkuil
On Tue 9 April 2013 15:05:37 Michal Lazo wrote: > And can you explain where is problem with memcpy in kernel? > is it really that big problem to copy 720*576*3 bytes from kernel > space to user space ? > what all usb camera drivers do ? some dma access directly to destination > buffer? Try this:

Re: vivi kernel driver

2013-04-09 Thread Michal Lazo
And can you explain where is problem with memcpy in kernel? is it really that big problem to copy 720*576*3 bytes from kernel space to user space ? what all usb camera drivers do ? some dma access directly to destination buffer? On Tue, Apr 9, 2013 at 12:11 PM, Hans Verkuil wrote: > On Tue 9 Apr

Re: vivi kernel driver

2013-04-09 Thread Hans Verkuil
On Tue 9 April 2013 11:57:51 Michal Lazo wrote: > Amlogic provide only 2.6.34 driver > can I expect big performance loss ? Why do so many companies stick to ancient kernels? I can never understand that. You have three options: 1) update their driver to a recent kernel 2) create a v4l2 driver for

Re: vivi kernel driver

2013-04-09 Thread Michal Lazo
Amlogic provide only 2.6.34 driver can I expect big performance loss ? On Tue, Apr 9, 2013 at 11:09 AM, Hans Verkuil wrote: > On Tue 9 April 2013 10:52:41 Michal Lazo wrote: >> As we have all source codes >> what do you advice ? > > Write your own driver. Good templates to look at are drivers/med

Re: vivi kernel driver

2013-04-09 Thread Hans Verkuil
On Tue 9 April 2013 10:52:41 Michal Lazo wrote: > As we have all source codes > what do you advice ? Write your own driver. Good templates to look at are drivers/media/pci/sta2x11 and drivers/media/platform/blackfin. The first is functionality-wise probably closer to what you need, but it's a pci

Re: vivi kernel driver

2013-04-09 Thread Michal Lazo
As we have all source codes what do you advice ? On Tue, Apr 9, 2013 at 10:38 AM, Hans Verkuil wrote: > On Tue 9 April 2013 09:58:33 Michal Lazo wrote: >> I want to make API that will provide hw video decoder on Amlogic SOC >> it is ARM cortex 9 >> >> with some proprietary video decoder >> >> am

Re: vivi kernel driver

2013-04-09 Thread Hans Verkuil
On Tue 9 April 2013 09:58:33 Michal Lazo wrote: > I want to make API that will provide hw video decoder on Amlogic SOC > it is ARM cortex 9 > > with some proprietary video decoder > > amlogic provide me with working example that generate output frames in > amlvideo driver. > It is v4l2 driver > a

Re: vivi kernel driver

2013-04-09 Thread Peter Senna Tschudin
Check some patches from Kirill Smelkov like: https://patchwork.kernel.org/patch/1688591/ On Tue, Apr 9, 2013 at 9:58 AM, Michal Lazo wrote: > I want to make API that will provide hw video decoder on Amlogic SOC > it is ARM cortex 9 > > with some proprietary video decoder > > amlogic provide me w

Re: vivi kernel driver

2013-04-09 Thread Michal Lazo
I want to make API that will provide hw video decoder on Amlogic SOC it is ARM cortex 9 with some proprietary video decoder amlogic provide me with working example that generate output frames in amlvideo driver. It is v4l2 driver and it did memcpy to mmap userspace memory buffer_y_start=iore

Re: vivi kernel driver

2013-04-08 Thread Hans Verkuil
On Mon April 8 2013 14:42:32 Michal Lazo wrote: > Hi > 720x576 RGB 25, 30 fps and it take > > 25% cpu load on raspberry pi(ARM 700Mhz linux 3.6.11) or 8% on x86(AMD > 2GHz linux 3.2.0-39) > > it is simply too much No, that's what I would expect. Note that vivi was substantially improved recentl

Re: vivi kernel driver

2013-04-08 Thread Michal Lazo
Hi 720x576 RGB 25, 30 fps and it take 25% cpu load on raspberry pi(ARM 700Mhz linux 3.6.11) or 8% on x86(AMD 2GHz linux 3.2.0-39) it is simply too much On Mon, Apr 8, 2013 at 9:42 AM, Peter Senna Tschudin wrote: > Dear Michal, > > The CPU intensive part of the vivi driver is the image genera

Re: vivi kernel driver

2013-04-08 Thread Peter Senna Tschudin
Dear Michal, The CPU intensive part of the vivi driver is the image generation. This is not an issue for real drivers. Regards, Peter On Sun, Apr 7, 2013 at 9:32 PM, Michal Lazo wrote: > Hi > V4L2 driver vivi > generate 25% cpu load on raspberry pi(linux 3.6.11) or 8% on x86(linux > 3.2.0-39)

vivi kernel driver

2013-04-07 Thread Michal Lazo
Hi V4L2 driver vivi generate 25% cpu load on raspberry pi(linux 3.6.11) or 8% on x86(linux 3.2.0-39) player GST_DEBUG="*:3,v4l2src:3,v4l2:3" gst-launch-0.10 v4l2src device="/dev/video0" norm=255 ! video/x-raw-rgb, width=720, height=576, framerate=3/1001 ! fakesink sync=false Anybody can answe