OMTB scripts vs OMX C application for "decode_display" functionality discrepancy between 5.01 and 5.02 SDKs

Hi Ralph,

Thanks for the confirmation.

I'd be curious to hear from someone at TI about the why, but at least

now we know not to bang our heads too much with it for now.

Thanks again.

Juan

Hi Juan,

Can you please share the oms script that has failed on your setup, I would like to reproduce the issue on my setup.

I'm almost certain Juan is referring to the "decode_scale_display.oms" in EZSDK 5.02.01.59:

"<EZSDK>/example-applications/omtb_01_00_01_05/packages/ti/sdo/omtb/scripts/decode_scale_display.oms"

As an aside, and hopefully something TI can take on board; Juan has an excellent point saying that once he knows something doesn't work he can move on to something else. This is something lacking time and again in the EZSDK.

Us engineers don't particularly mind things not working, it is a pre-release SDK after all. The issue is, we really would like to know which components of the SDK are not working, then we can move onto other activities without wasting our time getting things to work that won't.

All the same, the SDK does seem to be getting better with each release which is pleasing to see.

Thanks,
Ralph

Srinivasa Reddy DeviReddy said:

Hi Juan,

Can you please share the oms script that has failed on your setup, I would like to reproduce the issue on my setup.

Hello Srinivasa,

It's the official one that came with the EZSDK. We didn't make any changes.

   <EZSDK>/example-applications/omtb_01_00_01_05/packages/ti/sdo/omtb/scripts/decode_scale_display.oms

We ran it from the command line (i.e.: from the serial console, not from any gui front-end [if it makes any difference]).

Juan.

Where do all your files (kernel modules, .xem3 firmware files etc) come from? Are they all definitely from the latest EZSDK?

Thanks,
Ralph

The .ko files came from /lib/modules except for syslink.ko which I got from /usr/share/ti/c6run-apps

The .xem3 files came from /usr/share/ti/ti-media-controller-utils

They are all on the NFS on the targetfs created by the EZSDK 5.02 installer.  I renamed my 5.01 targetfs to something else, so these are definitely from the new EZSDK.

The only thing I rebuild was the omtb_dm816xbm_a8host.xv5T file

Cheers,

Simon

Many apologies, it turns out I didn't have the H264 file in the right place so the script couldn't find it. The OMTB decode_scale_display script now works fine for me.

Back to Juan's issue.

Looks like there is a problem with notify. My guess at the cause is either:

a) You haven't loaded syslink.ko properly, or it's not built against the right kernel configuration.

b) Perhaps more likely, you don't have this line in your "bootargs" U-Boot variable: "notifyk.vpssm3_sva=0xBF900000"

Any help?

Ralph

Hi Srinivasa,

Been in meetings most of this morning so just catching up. I'll try a rebuild of the OMTB

app when I get a chance. Thanks for looking into this. I'll post a followup after I try it.

Juan

_Ralph_ said:

Many apologies, it turns out I didn't have the H264 file in the right place so the script couldn't find it. The OMTB decode_scale_display script now works fine for me.

Back to Juan's issue.

Looks like there is a problem with notify. My guess at the cause is either:

a) You haven't loaded syslink.ko properly, or it's not built against the right kernel configuration.

b) Perhaps more likely, you don't have this line in your "bootargs" U-Boot variable: "notifyk.vpssm3_sva=0xBF900000"

Any help?

Ralph

Hi Ralph,

The bootargs were the default that came after I created an SD bootable media from the SDK. I did

confirm that the notifyk.vpssm3 parameter was set correctly just to make sure.

Same goes with the syslink.ko. No local changes; straight from the SDK. There were no errors during

it's modprobe. Besides, if the syslink.ko hadn't made it OK I would think that the C app version wouldn't

work either but that's just my theory. I would expect that both OMTB and the C app version (from TI-OMX)

would use the same communication across the cores (b/w the A8 and the M3s).

Thanks for the suggestions though.

I'll try Srinivasa's suggestion of rebuilding the OMTB app when I get a chance and post on update.

Juan

Simon Brown said:

Hi Juan, Ralph and Srinivasa,

Just my tuppence worth:

I am running EZSDK 5.02.01.59 and went through the OMTB v01_00_01_05 release notes.  After much cursing and swearing removing Windows paths from env.mk and various other things I managed to get it to compile, and then found the .oms scripts on the targetfs in /usr/share/ti/ti-omtb.  I found various other .ko files kicking around in the /usr/share and copied all to my /home/root/omtb directory I was working in.

From here, I boot the board as normal, nothing changed there.  Then I stop the matrix gui, kill the graphics , stop pvr-init and unload hdvicp and hdvpss.

Then I go through the steps in 4.1 of the OMTB release notes, namely:

prcm_config_app s

insmod syslink.ko

./firmware_loader 1 dm816x_hdvicp.xem3 start

./firmware_loader 2 dm816x_hdvpss.xem3 start

insmod vpss.ko mode=hdmi:1080p-60,hdcomp:1080p-60 i2c_mode=1

insmod ti81xxhdmi.ko

and finally:

./omtb_dm816xbm_a8host.xv5T decode_scale_display.oms and it plays the video absolutely fine.

Not sure what I'm doing differently to you guys, but thought I'd share this, in that I think it does work.

Now all I want is some pointers as to what the different steps in the oms scripts actually do so I can write my own!

Cheers,

Simon

Hello Simon,

Just to make sure I understand your post, it's now working OK after you built it fresh but it DID NOT work

before then with the pre-built binaries. Is that correct? Or did you just rebuilt everything before you attempted

it for the first time?

Just want to know what the data-point actually is. :-)

Thanks for the reply.

Juan Solis

All,

Went ahead and tried rebuilding the OMTB app based on the instructions on the OMTB release notes.

Still get the error. I also tried removing/stopping and re-executing the steps as Simon and the release

notes/Srinivasa mentioned but still no luck. Also tried using the syslink.ko from c6run-apps instead of

the one from /lib/modules. Didn't work either.

Keep in mind that this is a custom board so there may be some other gremlins that the community

may not be seeing on their EVMs. I wasn't too surprised when we started seeing errors under 5.02 that

we weren't seeing under 5.01. However, the C app version works fine and that's what has got us perplexed.

I would have thought they would be doing their communication across the cores in the same way with

the same pre-conditions being applied!?

In any event, thank you all for your suggestions and being a sounding board.

Juan

Srinivasa Reddy DeviReddy said:

Hi Simon,

Thanks for the update, good news that demo runs at your end.

Please find the omtb user guide in "omtb_01_00_01_05/packages/ti/sdo/omtb/docs" folder, explains about omtb application.

Following are the build steps used to build omtb which are relatively simple.

1) DM816x_C6A816x_AM389x_EZ_Software_Developers_Guide.pdf

  "make omtb" from the toplevel directory

2) Set all the paths as arguments in the make command as mentioned in omtb release notes or omtb user guide, this will avoid modification of env.mk file

To all those that provided input on this, THANKS!

We finally found our issue and it seems to be power related. We now have both flavors (script and app)

working. I put up a post to the community here:

    http://e2e.ti.com/support/dsp/davinci_digital_media_processors/f/717/t/133557.aspx

to hopefully spare other people the pained/perplexed headscratching we've gone through.

Juan Solis.