Get unexpected result from DVO2

longxiao zhang

Hi,

We are using DM8168 as a decoder implementation. However, we get some unexpected results from DVO2.

Because we have FPGA behind DVO2 output, so we have the ability to scope the output of DVO2. The problem is, when we generate some simply ascending samples (like 0,1,2,3 ... ,255), and call Vdis_putFullVideoFrames() to feed to DVO2, then observe the output of DVO2 from FPGA. It's just NOT what we expect. The mismatch is not huge, but does exist .

We also studied TRM of DM816x VPSS, and known DVO2 could work in a 'self test' mode. Under that mode, DVO2 does output a perfect color bar.

Any help will be highly appreciated!

zhang

over 10 years ago

0 Margarita Gashova over 10 years ago

TI__Guru* 80190 points

Hello,

What is the software release that you are using here?

BR
Margarita

0 longxiao zhang over 10 years ago in reply to Margarita Gashova

Intellectual 500 points

Hi Margarita,

We are using:

DVRRDK: 04.00.00.03

Linux Kernel: 2.6.37+

Zhang

0 Brijesh Jadav over 10 years ago in reply to longxiao zhang

TI__Guru**** 477075 points

Hello Zhang,

It really depends on what path you are using. On video pipeline, there are many modules that come into the path and some of these modules cannot be even disabled. On Grpx pipeline, you could get the exact same output as input, but you need to configure grpx in rgb888 mode and also you need to set the output format to be rgb888. There is no other way to get exactly same output.

Regards,
Brijesh

0 longxiao zhang over 10 years ago in reply to Brijesh Jadav

Intellectual 500 points

Hi Brijesh,

Yes, you are right. We have tested rgb mode using FB driver, which will go through GRPX. As you mentioned, the result was exactly what we want.

However, and for some reason, we need to use DVRRDK API (something like Vdis_putFullVideoFrames()), which I guess will go through 422P path. I also noticed that there are some modules in 422P path, like CSC_HD.

We tried to bypass CSC_HD, but we got nothing after that.

Is it possiable to find a way to bypass CSC_HD which will modify the result? If yes, is there anything we missed?

BR,

Zhang

0 Brijesh Jadav over 10 years ago in reply to longxiao zhang

TI__Guru**** 477075 points

On YUV Path, it is not possible, as there is a YUV422 to YUV444 convertor which cannot be bypassed. You can use only GRPX path for this.
Regards,Brijesh

0 longxiao zhang over 10 years ago in reply to Brijesh Jadav

Intellectual 500 points

I can understand that chroma will go through convertor and be changed. Does luminance must be converted also?

BR,
Zhang

0 Brijesh Jadav over 10 years ago in reply to longxiao zhang

TI__Guru**** 477075 points

What is you input format? is it YUV422 or YUV420?
No it will not change luma portion.Rgds,Brijesh

0 longxiao zhang over 10 years ago in reply to Brijesh Jadav

Intellectual 500 points

We both tried YUV422 and YUV420. They will be changed.
So, to confirm, RGB is the only way if we want the output unchanged, right?

BR,
Zhang

0 Brijesh Jadav over 10 years ago in reply to longxiao zhang

TI__Guru**** 477075 points

Yes, it is possible for RGB paths.BR,Brijesh

0 longxiao zhang over 10 years ago in reply to Brijesh Jadav

Intellectual 500 points

One more thing, as you said "it will not change luma portion", but we got luma changed indeed.

How to explain it?

BR,

Zhang

0 Brijesh Jadav over 10 years ago in reply to longxiao zhang

TI__Guru**** 477075 points

Again, which path did you select? There are modules on the path like Vcomp, EDE, Cproc, CSC, SC, DEI, which can change luma portion.
Regards,Brijesh

0 longxiao zhang over 10 years ago in reply to Brijesh Jadav

Intellectual 500 points

The path is marked by green dot line.

BR,

Zhang

0 Brijesh Jadav over 10 years ago in reply to longxiao zhang

TI__Guru**** 477075 points

Have you bypassed CSC in the path and the one in the VENC? Did you make back ground color transparent completely by setting correct blend parameters? Finally, none of the submodules in VENC should be enabled..
Regards,Brijesh

0 longxiao zhang over 10 years ago in reply to Brijesh Jadav

Intellectual 500 points

Hi Brijesh,

Sorry for the late reply. We took some time to check the modules in that path. We had CSC bypassed, turned off submodules in HD_VENC except DVO2. But now the mismatch is even bigger.

Now we are investigating what are missed.

BR,

Zhang

0 Brijesh Jadav over 10 years ago in reply to longxiao zhang

TI__Guru**** 477075 points

Hi Zhang,

What is the DVO2 output format in this case? Are you seetting it to YUV422?

Could you again check the DVO2 registers? There is an CSC even in VENC. Also there are a way to set the gains for input pixels. Please check all of these are set to unity.

Regards,

Brijesh

0 Brijesh Jadav over 10 years ago in reply to Brijesh Jadav

TI__Guru**** 477075 points

Can you take a dump of 25 registers from 0x4810a000 and share it?

Rgds,

Brijesh

0 longxiao zhang over 10 years ago in reply to Brijesh Jadav

Intellectual 500 points

Hi Brijesh,

Yes, we are setting it to YUV422.

This is a dump of VENC registers

0x4810a000:44013073
0x4810a004:003F0275
0x4810a008:1EA500BB
0x4810a00c:1F9901C2
0x4810a010:1FD71E67
0x4810a014:004001C2
0x4810a018:00200200
0x4810a01c:184C0C77
0x4810a020:1C0C0C30
0x4810a024:1C0C0C30
0x4810a028:842EE672
0x4810a02c:281A401C
0x4810a030:28508101
0x4810a034:000002ED
0x4810a038:00038338
0x4810a03c:28500172
0x4810a040:0001D000
0x4810a044:002D0000
0x4810a048:05004000
0x4810a04c:05001001
0x4810a050:0018B18B
0x4810a054:2850016A
0x4810a058:0001D001
0x4810a05c:002D11A4
0x4810a060:05001000
0x4810a064:05006176
0x4810a068:00000000
0x4810a06c:00000000
0x4810a070:00000000
0x4810a074:00000000
0x4810a078:00000000
0x4810a07c:00000000
0x4810a080:00000000
0x4810a084:00000000
0x4810a088:00000000
0x4810a08c:00000000
0x4810a090:00000000
0x4810a094:00000000
0x4810a098:00000000
0x4810a09c:00000000
0x4810a0a0:00000000
0x4810a0a4:00000000
0x4810a0a8:00000000
0x4810a0ac:00000000
0x4810a0b0:00000000
0x4810a0b4:00000000
0x4810a0b8:00000000
0x4810a0bc:00000000
0x4810a0c0:00000000
0x4810a0c4:00000000
0x4810a0c8:00000000
0x4810a0cc:00000000
0x4810a0d0:00000000

0 Brijesh Jadav over 10 years ago in reply to longxiao zhang

TI__Guru**** 477075 points

Can you try setting bits[12:11] to 00b? This will make sure that the CSC is completely bypassed.

Regards,

Brijesh

Processors

Processors forum

Get unexpected result from DVO2