DM365 capture device setting

I am not really familiar with the IPNC kernel, so it is hard for me to say how similar it is to our DVSDK kernel.  Often, third party solutions such as IPNC start from our software and make necessary changes to support their solution, and of course these solutions is supported by the corresponding third parties at this point.   Things to look out for is what version of our DVSDK did they base their software on (assuming they based it on one of our DVSDKs); grabing the same version or asking them if they have ported their current solution to our latest DVSDK can resolve things...

That said, assuming the kernels (IPNC vs DVSDK) behave very similar, the following part of bootargs "davinci_capture.device_type=4" selects the input capture device.  The current settings are the default and I believe select the component input on our DVEVM.  Also, the encodedecode demo has options you can pass to it (see encodedecode.txt or run encodedecode with -h option) that can allow you to change some of the default input attributes.  You can also try running the encode demo alone which has more options for resolutions (NTSC, 720p...).

Hi,

IPNC Ref Design from Appro/TI does not use the video capture drivers from the LSP (Linux support package) of DVSDK. The VD0 interrupt is not connected to the interrupt handler in the LSP driver and instead is being used in our own implementation of capture driver.

This is the reason IPNC Ref Design comes with its own version of LSP package and DVSDK components.

Please do not use DVSDK demos like encode, encodedecode etc. on the IPNC LSP package.

What exactly is the requirement to use IPNC codebase and run DVSDK encodedecode demo? BTW, IPNC system does not use DMAI also.

Regards,

Anshuman

Hi,

Thank you very much, and I also know a little more about DM365 SDK.

Are there tips to understand capture driver for linux more quickly?

If we need to change the capture device(to TVP5158), what should I know expect reading sprufg8.pdf(VPFE)?

To use DVSDK demos in IPNC codebase is not our purpose, it just is my testing.

I found there are three ways to use Codec Engine.

1.  "av_capture" framework from Appro (or TI's secret weapon ?)

2. DMAI

3. dm365_codecs packages

How do I choose one of them? Why IPNC system don't use DMAI?

Best Regards,

Person

I have encounted the same problem.

I am running the encodedecode demo app:

# ./loadmodules_sd.sh

#./encodedecode -y 2 -v h264 -o

the output is as following:

...

Error: Failed to create capture device
Process one buffer to figure out buffer requirements

My question is what reason can cause this problem ? how to fix it ?

thanks!

Maybe there is something wrong with the capture device driver, such as vpfe driver and decoder device driver, so you can not operate the capture device in the user space. You should read some source code of the DVSDK, and add debug messages to figure out what's going on.

Xing Hong,

Let me get this post back to the original question. The issue was related to not receiving VPFE interrupt using IP Camera reference design kernel, using encodedecode demo.

If the case is same, i and Juan have already explained details about it.

If that's not the case, then we would request you to provide more details. encodedecode demo has been tested as part of DVSDK 2.x and 3.x release on DM365 on TI EVM. If you are using anything different, you need to analyze the error logs and share the setup details with us.

Regards,

Anshuman

I need  help on encodedecode demo to configure capture driver not to do de-interlacing of captured data.  I am using encodedecode demo in passThrough mode without any encoders as I need the info from both fields in a frame.  I am not a Linux guru and am struggling to find where driver is set up to do automatic de-nterlacing.  Is it in resizer/previewer running in chained mode or is it through SDOFST where VPFE doc (sprufg8a) says "This provides a means to deinterlace an interlaced, two-field input and also to invert an input image vertically" ? (p66).  Is this setting set up on initialisation and which source file should I look in to make the change?

Thanks, Jinh T

Jinh,

- As you mentionned de-interlacing is done directly on on-the-fly so the video during capture is ouput to memory via the resizer. This is setup in the Linux Video drivers that is used by the DMAI interface.
Refering to the LSP 2.10 of the DM365 DVSDK 2.10 details on the Linux driver are provided in the following documents:
Driver user's manual - SPRS566A
http://software-dl.ti.com/dsps/dsps_registered_sw/sdo_sb/targetcontent/psp/mv_lsp_2_10/02_10_00_14/exports/docs/DaVinciLinux_DriversDataManual_SPRS566A.pdf

- There are ways to use the video capture so that the de-interlacing  is not done (so the VPFE is not used in continuous mode but in one-shot mode and data does not go though the resizer).
More information are provided in the following document (Refering to the LSP 2.10 of the DM365 DVSDK 2.10):
VPFE Capture Driver UsersGuide - SPRUGP6
http://software-dl.ti.com/dsps/dsps_registered_sw/sdo_sb/targetcontent/psp/mv_lsp_2_10/02_10_00_14/exports/docs/DaVinciLinux_VPFECaptureDriverUsersGuide_SPRUGP6.pdf

Doc should be located in the following folder after installation of the DVSDK:
    dvsdk_2_xx_xx_x/PSP_02_xx_x_xx/docs

It looks like that video data need to use what is called data path 1 in the document. Data path 2 will be used for on the fly de-interlacing. See section 2.1 and 2.2.2.6 of this document.
I think that there is no need to change the V4L2 driver source code. It should be possible to select the data path using some IOCTL. Still I am not sure if the DMAI interface enables to do that diretcly (but I would think so) or if the DMAI src need to be changed.

- It looks like as well that some example V4L2 examples are provided for both one-shot and continuous mode (this are not DMAI example but example using directly V4L2 at the application level). You should look at a folder called like the following:
    dvsdk_2_xx_xx_x/PSP_02_xx_x_xx/examples/dm365

Hope it helps,

A.

Thanks for the help.

Just a comment to start with on the documentation - loads of it - that sometimes the english is not up to standard and a bit confusing.

I managed to skip the de-interlacing by running in one_shot mode but from the documentation it was not really clear when to use what mode as one-shot sounds like something for doing stills and continues soemthing to use for streaming video especially if the documantation (sprufgg8a) gives "Resizer Bypass Mode" and mentions using SDOFST as a means to do de-interlacing but this does not seem to be used in the drivers which has probably eveolved from previous platforms and not use all features of DM365.

Running one-shot mode in encodedecode demo I had to change Capture.c in DMAI to not run capture chain mode, change the bootargs to set up one-shot mode and change the color space to UYVY and after the driver not able to allocate enough contiguous memery for 4 display buffers in new pixel format changed it to 3 and it worked.  Now only my display is offset to the left by some pixels that I need to resolve as well as how Black and White is packed in UYVY format for my edge insertion data.

Thanks, Jinh T.

Hi,

I am facing similar problem, but due to different reasons. (I have been through sprufg8b)

The design is similar to reference design of Camera (using MT9P031), except some hardware changes, by our designers.

1> The C_WE_FIELD is not used. Instead, it is input to some other GPIO100/YIN4.

2> CLKOUT0 is not used, instead it is connected to CLKOUT1.

I understand that these changes will require some changes in the respective modules, and I have been digging for some time.

Any lead or pointer will be a big help and highly appreciated.

TIA,

Ranjeet

High Jihn,

i am trying to modify my application from Oneshot mode to Chain mode.

for continuous capture i changed the following two things.

1. In boot args dm365_imp.oper_mode=0

2. captureChainMode = TRUE in Capture.c present inside dmai_2_xx/..../linux folder.

i am getting some error while running my application

-----------------

davinci_resizer davinci_resizer.2: Operation mode doesn't match with current hw mode

ccdc_enable(): SYNCEN:0x00000002

ccdc_output_to_sdram(): SYNCEN:0x00000000

----------------

do i need any other changes ? 

Hi Sujit, from what I could remember is that your mode must be set up correctly, mode=0 is fine, and captureChainMode must be true (I am using encodedecode demo so everywhere was set to D1_PAL and Composite) and the colorspace and pixelformat must be correct (i think).  I was also not using the previewer and resizer and was disabling it everywhere to the edge of not breaking functionality.  Most of these might not have an influence as I was trying to find the reason why the app didn't start up on a black and white camera which was another feature in the video_davinci_evm driver that I had to change.

Hope it helps, cheers, Jinh T.

Hi Jihn

thank you very much for your reply.

actually we are not using EVM, we are using our custom design camera.

also we need the resizer. 

is there anywhere setting required for chain mode in registers , apart from oper_mode at boot env and in Capture.c Chain Mode,  present inside dmai/linux/ folder ?

Hi Sujit,

sorry but about previewer and resizer I can't tell you more than what you see in Capture.c as I didn't explore it at all. 

Hope you come right with this issue.

Regards, Jinh T.

jhin

what should be the colorspace format for chain mode?

Hi Marko,

you are right.

but i am getting the message from driver "  mode doesn't allow multiple instances "

because my application needs multiple instances to be enabled.

for testing i just made rsz_dev.users =0 in the imp_resizer.c file present inside kernel/linux version../drivers/char/

it boots fine but no profiles are available and so no video for any of the codecs in my camera.

encodedecode demo was running ColorSpace_YUV420SEMI in chain mode but then it was automatically deinterlacing the video which I did not want.  Switching to one-shot mode I changed the colorspace to ColorSpace_UYVY.

Let us know how you got it fixed, Jinh T.

Hi Sujit,

In chain mode only one resize per frame is allowed.
You can find a lot on that topic at :  http://e2e.ti.com/support/dsp/davinci_digital_media_processors/f/100/t/7777.aspx?PageIndex=3   (code in : RSZB_Output_Code.zip)

To get  two  resolutions from the resizer in a chain mode, besides what you have already done,
enable and set dimension(width,height) of output2 in DMAI Resize.c in function  Resizer_continous_config() .
Then you can get both, the first and the second-resized frame in the same buffer, just one after another.
Look at attached files in mentioned post.

Marko.