C66x_h264hpvenc cannot run much time

Chandler

Intellectual 440 points

Dear TI Experts,

Following is my working enviorenment:

Board：EVMK2H V3.0

Software:CCS5.4

Example:C66x_h264hpvenc_01_00_03_00_ELF

I tested the stability of the example on the k2h,and the dsp stopped running when I encoded about 10000~30000 frames(the framesToEncode of encoder.cfg was set to 100000 ).

For running fastly , I commented the reading and writing files.

I just encoded the one same frame of YUV 420P data, and tested some more senarios, for example:

1.input:1920x1200 output: high profile, no B frame result: only encoded 18700 frames

2.input:1920x1200 output: high profile, include I P B frame result: only encoded 35000 frames

3.input:352X288 output: high profile, include I P B frame result: only encoded 59500 frames

4.input:1920x1200 output: Baseline profile, include I P B frame result: only encoded 18700 frames

5.input:1920x1200 output: Baseline profile, include I P B frame, there were two testcases and the framesToEncode = 10000

result: the first testcase encoded 10000 frames, and the second testcase encode 8700.(this step I wanted to initialize again, but the same result)

The most result is 18700, and there is a core maybe core7 or core5 abort. Following is my console print:

numFramesEncoded : 18400
numFramesEncoded : 18500
numFramesEncoded : 18600
numFramesEncoded : 18700
[C66xx_7] ?????????????
xdc.runtime.Error.raise: terminating execution

The above is core7 abort.

another one:

numFramesEncoded : 59400
numFramesEncoded : 59500
[C66xx_2] ??

xdc.runtime.Error.raise: terminating execution

The above were core2 abort.I printed every 100 frames encoded,for runing fastly.

The third:

numFramesEncoded : 7700
numFramesEncoded : 7800
numFramesEncoded : 7900
numFramesEncoded : 8000
numFramesEncoded : 8100
numFramesEncoded : 8200
numFramesEncoded : 8300
numFramesEncoded : 8400
numFramesEncoded : 8500
numFramesEncoded : 8600
[C66xx_1] >mmm>mm>mm>mm>mm>mm>mm>mm>m>mmm��?xdc.runtime.Error.raise: terminating execution
[C66xx_0] >mmm>mm>mm>mm>mm>mm>mm>mm>m>mmm��?xdc.runtime.Error.raise: terminating execution
[C66xx_2] >mmm>mm>mm>mm>mm>mm>mm>mm>m>mmm��?[C66xx_3] >mmm>mm>mm>mm>mm>mm>mm>mm>m>mmm��?[C66xx_4] >mmm>mm>mm>mm>mm>mm>mm>mm>m>mmm��?[C66xx_6] >mmm>mm>mm>mm>mm>mm>mm>mm>m>mmm��?[C66xx_7] >mmm>mm>mm>mm>mm>mm>mm>mm>m>mmm��?[C66xx_2] xdc.runtime.Error.raise: terminating execution
[C66xx_3] xdc.runtime.Error.raise: terminating execution
[C66xx_4] xdc.runtime.Error.raise: terminating execution
[C66xx_6] xdc.runtime.Error.raise: terminating execution
[C66xx_7] xdc.runtime.Error.raise: terminating execution

The following is my ccs picture when the dsp aborted:

over 8 years ago

0 Chandler over 8 years ago

Intellectual 440 points

My h264hpvenc_ti_testapp.c is attached below:

h264hpvenc_ti_testapp.zip

0 Chandler over 8 years ago in reply to Chandler

Intellectual 440 points

can anybody help me?

0 Paula Carrillo over 8 years ago in reply to Chandler

TI__Mastermind 39270 points

Hi Chandler, sorry I missed this before, I will send it to the codec developers for ideas
Thank you,

Paula

0 Chandler over 8 years ago in reply to Paula Carrillo

Intellectual 440 points

Hi Paula ,

Thank you for responsing me.

Last week, I have tried another version (C66X_h264hpvenc_01_00_01_03_ELF_Hawking).But the result is the same.

Today I will try to codec with 7 cores(0~6).After that I will tell you how about it.

0 Chandler over 8 years ago in reply to Paula Carrillo

Intellectual 440 points

Hi Paula,

Just now, I have tried to use 0~6 cores to encode, and it could encode 1920x1200 about more than 60000 frames(below 30000 frames before when used 8 cores to encode).So I think there is some memory leaking.

Thank you.

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Chandler

Intellectual 650 points

Hi Chandler,

Are you facing the same issue when you enable file read and file write?

Regards,

Mashkur

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur,

It doesn't matter with the file operations, and there is no file read or file write in the example.

I just loaded picture image to memory (not read picture file),and then encoded without saving the h264 data and counted the encoded frames.

Note:I encoded one same frame all the time.When the frame was encoded, I pointed the input buffer to the beginning of the frame.

Thank you .

Chandler

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Chandler

Intellectual 650 points

Hi Chandler,

We are able reproduce the issue at our end. We will get back to you soon.

Regards,

Mashkur

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

OK! Thank you .

Chandler

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur,

Is there any progress?

Chandler

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Chandler

Intellectual 650 points

Hi Chandler,

We are working on it. We will get back to you soon.

Regards,

Mashkur

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 650 points

Hi Chandler,

Can you please send the cofig that you are using?

Regards,

Mashkur

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur,

You can see my test senaria:

3.input:352X288 output: high profile, include I P B frame result: only encoded 59500 frames

So , you can use the default encoder.cfg,except change "framesToEncode = 100000".

Regards,

Chandler

0 Chandler over 8 years ago in reply to Chandler

Intellectual 440 points

Also you need modify h264hpvenc_testapp.c about TestApp_ReadInputYUVData function . You can refer to my "h264hpvenc_testapp.c " attachment in first page.

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Chandler

Intellectual 650 points

Hi Chandler,

Thanks for replying for config info. The behavior is inconsistent i.e. it is crashing at different places and different cores. We are working on it. We will get back to you.

Regards,

Mashkur

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur ,

It doesn't matter.

Chandler

0 Paula Carrillo over 8 years ago in reply to Chandler

TI__Mastermind 39270 points

Hi Chandler, could you please elaborate more your comment above?

thank you,

Paula

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Chandler

Intellectual 650 points

Hi Chandler,

We have found the issue which is causing the problem. Working on its solution.

Regards,

Mashkur

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur,

Great! I am looking for your responsing again.

Thanks!

Regards,

Chandler

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur,

How is the problem now?

Regards,

Chandler

0 Chandler over 8 years ago in reply to Paula Carrillo

Intellectual 440 points

Hi Paula,
Can you help me ask the codec developers, how is the encoder crash problem?

Regards
Chandler

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur,

Do you find a solution to resovle the corruption ?

Regards,

Chandler

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Chandler

Intellectual 650 points

Hi Chandler,

It looks like IPC issue and we are working on it. Can you please confirm the same issue on some other platform like k2g etc.

Regards,

Mashkur

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur,

I am sorry I cann't help you,because I don't have any other platform except the k2h evm.

If you have any progress with the corruption, please let me know.

Thank you very much.

Regards,

Chandler

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Chandler

Intellectual 650 points

Hi Chandler,

Please uncomment the following at line 1439 of your testapp and run.

Before :

/*
for (core=0; core<IVIDMC_TI_MAXCORES; core++)
{
for(nal=0; nal<nalSizes[core][align]; nal++)
{
memcpy (pOutputData_loc, pNals[core][nal + align],
nalSizes[core][nal+1+align]);
pOutputData_loc += nalSizes[core][nal+1+align];
bytesGenerated_CB += nalSizes[core][nal+1+align];
printf("\n core %d ,nalSizes %d",
core,nalSizes[core][nal+1+align]);
}
nalSizes[core][align] = 0;
}

printf("\n");

//TestApp_WriteOutputData(outputData_loc,
// bytesGenerated_CB,foutFile);
printf("bytesGenerated_CB=%d\n",bytesGenerated_CB);
*/

After :

for (core = 0; core < IVIDMC_TI_MAXCORES; core++)
{
for (nal = 0; nal < nalSizes[core][align]; nal++)
{
memcpy(pOutputData_loc, pNals[core][nal + align],
nalSizes[core][nal + 1 + align]);
pOutputData_loc += nalSizes[core][nal + 1 + align];
bytesGenerated_CB += nalSizes[core][nal + 1 + align];
/*printf("\n core %d ,nalSizes %d",
core, nalSizes[core][nal + 1 + align]);*/
}
nalSizes[core][align] = 0;
}

//printf("\n");

//TestApp_WriteOutputData(outputData_loc,
// bytesGenerated_CB,foutFile);
//printf("bytesGenerated_CB=%d\n", bytesGenerated_CB);

I have tested at my side for 100000 frames successfully with 8 cores. Please let me know if you face any issues.

Regards,

Mashkur

Please mark this as answered if it solves your issue.

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur，

Thank you for your responsing.

I will do the test next week.

But I don't think this is the reason for the corruption.

Beacause,the h264hpvenc_ti_testapp.c which I give you is just a test file.I didn't change main code except comment some lines.

In fact, in my real project, it can't run the codes which you post above, because the foutFile=NULL.

At the beginning, I found the corruption in my real project. So I tested the original project which had the same problem.

Regards,

Chandler

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Chandler

Intellectual 650 points

Hi Chandler,

Actually you commented some code apart from foutFile code. If you see my last reply in that I have uncommented code which was making nalSizes[core][align] to zero(It was commented earlier by you). This was corrupting a memory section due to which code was crashing. This is the solution. Please do the changes as I have mentioned in last reply. I have tested it at my side successfully for 100000 frames and program ended normally without any crash.

Regards,

Mashkur

Please mark this as answered if it solves your issue

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur,

By the way,in my current project, we don't need to write the Output Bitstream to a file,so we wouldn't create a "foutFile".

My current project won't run into following condition:

if((loutArgs.videnc2OutArgs.bytesGenerated ) &&(foutFile != NULL))

{

}

comment or uncomment foutFile code,I don't think it is critical.

But it is still crashing.

Any way,I will check my code next monday and do some tests.I will tell you the results.

Thank you very much.

Regards,

Chandler

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Chandler

Intellectual 650 points

Hi Chandler,

If you are not creating fout file and foutFile = NULL, then please do following change in if condition.

Before:

if((loutArgs.videnc2OutArgs.bytesGenerated ) &&(foutFile != NULL))

{

if (lDynamicParams.videnc2DynamicParams.putDataFxn == NULL)

{

/* code */

}

else

{

/* code */

}

After:

if(loutArgs.videnc2OutArgs.bytesGenerated )

{

if (lDynamicParams.videnc2DynamicParams.putDataFxn == NULL)

{

if(foutFile != NULL)

{

/* code */

}

else

{

/* code */

}

Under that if condition, apart from fout code, there is some other code in else part which you have commented(which is not related to fout) and corrupting memory. It is as follows.

if((loutArgs.videnc2OutArgs.bytesGenerated ) &&(foutFile != NULL))

{

if ((lDynamicParams.videnc2DynamicParams.putDataFxn == NULL)) /* putDataFxn is not NULL, therefore code will execute else part */

{

if(foutFile != NULL)

TestApp_WriteOutputData to fout;

}

else /* You commented this part wheres nalSizes[ ][ ] is getting reset because of which memory is getting corrupted*/

{

Copy Nals;

nalSizes[core][align] = 0;

}

For your reference I have attached the modified testapp file. Please run encoder with this testapp.

2251.h264hpvenc_ti_testapp.c

Regards,

Mashkur

Please mark this as answered if it resolves your issue

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 650 points

Hi Chandler,

If you are not creating fout file and foutFile = NULL, then please do following change in if condition.

Before:

if((loutArgs.videnc2OutArgs.bytesGenerated ) &&(foutFile != NULL))

{

if (lDynamicParams.videnc2DynamicParams.putDataFxn == NULL)

{

/* code */

}

else

{

/* code */

}

After:

if(loutArgs.videnc2OutArgs.bytesGenerated )

{

if (lDynamicParams.videnc2DynamicParams.putDataFxn == NULL)

{

if(foutFile != NULL)

{

/* code */

}

else

{

/* code */

}

Under that if condition, apart from fout code, there is some other code in else part which you have commented(which is not related to fout) and corrupting memory. It is as follows.

if((loutArgs.videnc2OutArgs.bytesGenerated ) &&(foutFile != NULL))

{

if ((lDynamicParams.videnc2DynamicParams.putDataFxn == NULL)) /* putDataFxn is not NULL, therefore code will execute else part */

{

if(foutFile != NULL)

TestApp_WriteOutputData to fout;

}

else /* You commented this part wheres nalSizes[ ][ ] is getting reset because of which memory is getting corrupted*/

{

Copy Nals;

nalSizes[core][align] = 0;

}

For your reference I have attached the modified testapp file. Please run encoder with this testapp file attached.

6281.h264hpvenc_ti_testapp.c

Regards,

Mashkur

Please mark this as answered if it resolves your issue

0 Chandler over 8 years ago in reply to Mashkur Ahmed Badebade74

Intellectual 440 points

Hi Mashkur ,

Exellent!!!!!!!!!!!!!!!

You are right. nalSizes[core][align] = 0 is very important.

What you said above is right, it's a good way to sovle the "fout = NULL".

But I have used another way to test it more than 14 hours about 4,600,000 frames,the code was still running last sunday.

In h264hpvenc_ti_config.c, I modified one line:

before:

dynamicParams->videnc2DynamicParams.putDataFxn = (XDM_DataSyncPutFxn)shipOutNal/*NULL*/;

after:

dynamicParams->videnc2DynamicParams.putDataFxn = NULL;

Then the code will not use the pNals section, and there is no corrupting memory.

I think this is a same function with what you said above.

You've been working very hard！Thank you very much.

Regards，

Chanlder

0 Mashkur Ahmed Badebade74 over 8 years ago in reply to Chandler

Intellectual 650 points

Hi Chandler,

Good to know that issue is resolved. Can you please mark this thread as answered.

Regards,

Mashkur

Processors

Processors forum

C66x_h264hpvenc cannot run much time