Issue of increasing HDVICP2 heap memory (DM8168)

Hi Llano,

Could you please look here : http://e2e.ti.com/support/dsp/davinci_digital_media_processors/f/717/p/208122/738965.aspx

There is a step by step guide to steal memory from DSP to allocate to M3 heaps. Does this work for you?

Best Regards,

Pavel

Hi Pavel,

Thanks for your reply.

The guide above instruct how to allocate more memory on shared region(sr0,sr1,sr2) and M3's internal heap (VIDEO_M3_INT_HEAP_CACHED, VPSS_M3_INT_HEAP_CACHED).

The heap I mentioned is not these ones.

As sys_top shows( green underline in the above picture),  this 2MB  is not SR0,1,2 and M3's internal INT_HEAP_CACHED.

I am now stuck on increase this one, for it will be used up when allocate more codecs.

Does this heap is provided within HDVICP2 binary, which means I should get the source of HDVICP2 and recompile it?

Thanks,

Llano

Hi Llano,

The green underlined heap is exactly the Video-M3 (HDVICP2) internal heap size.

Can you double check that following the instructions, you can not increase the size.

Best Regards,

Pavel

Hi Pavel,

  I have the same problem with Llano, the following is the heap size in my board. As Llano said, the 0 Heap Size of HPVICP2 (the first one marked in the picture) was not the Video-M3(HDVICP2) internal heap size. The third one marked in the picture is the VPSS-M3(HDVPSS) internal heap size, just the same as defined in file " memsegdef.c". So I wander if the 0 Heap Size is the "VIDEO_M3_CODE" which size is 1M, and the  0 Heap Size of HDVPSS(the second one marked in the picture) is the "VPSS_M3_CODE" which size is 2M.

Hi Liu,

I use ezsdk5_04, on dm8168. sys_top outputs the same:  hdvicp2's 0 heap:Size 1048576. I just use the pic got from Memory Map's wiki.

Hi Pavel,

The video M3's internal heap(MC_HDVICP2_INT_HEAP_CACHED) size as default is 24MB, and I have increased it to 50MB. So it can create more than four 1080p decoders.

This hdvicp2's 0 heap:Size 1048576 is an 1MB memory, from memory map wiki, I cannot tell  which one it is. Perhaps as Liu said, it is HDVICP's VIDEO_M3_CODE section which matches size,    or VPSS_M3_INT_HEAP_NONCACHED section? I don't know.

Thanks,

Llano

Hi Llano,

     Thanks for replay! Did you modify the  file memsegdef_default.c to increase the MC_HDVICP2_INT_HEAP_CACHED ? Should I rebuild the ezsdk to generate a new firmware_loader and copy it to the EVM8168 board to make it work?

   Another question, I met some problem when create more than 5 decoders and make them work at the same time. Could you give me a favor to see the issue that I posted at http://e2e.ti.com/support/dsp/davinci_digital_media_processors/f/717/t/219454.aspx 

  Thanks very much!

Hi Llano,

The HDVICP2 internal/local heap (VIDEO_M3_INT_HEAP_CACHED) is used for codec memory allocation. And from what I understand, you need exactly this type of heap.

By default this heap size is 25MB, see memsegdef_default.c file:

/* Segment 2 , Temporarily used by VFPC internal buff heap */
  {
   1,                           /* valid */
   "VIDEO_M3_INT_HEAP_CACHED",  /* name */
   0x01800000,                  /* size */
   LDR_SEGMENT_TYPE_DYNAMIC_LOCAL_HEAP,        /* seg_type */
   0,                           /* flags */
   0x9BD00000,                  /* system_addr */
   0x9BD00000,                  /* slave_virtual_addr */
   -1,                          /* master_core_id */
   (1 << LDR_CORE_ID_VM3),      /* core_id_mask */
   1,                           /* cache_enable_mask */
   0,                           /* cache_operation_mask */
   -1                           /* shared_region_id */
  },

In the How_to_increase_Heap_memory_available_to_VPSS_and_VICP_M3_cores_v2.docx is described how you can enlarge this heap size to 52MB.

Even though the sys_top utility does not report the current VIDEO_M3_INT_HEAP_CACHED size, do you have problems when you create more than 10 codecs with 52MB heap?

Best Regards,

Pavel

Hi Pavel,

    where can I find the file "How_to_increase_Heap_memory_available_to_VPSS_and_VICP_M3_cores_v2.docx" ? And could you help me with the issue that I posted at http://e2e.ti.com/support/dsp/davinci_digital_media_processors/f/717/t/219454.aspx 

   Thanks very much! 

Hi Bing,

For the docx file, it is attached here:

http://e2e.ti.com/support/dsp/davinci_digital_media_processors/f/717/p/208122/738965.aspx

For your other questions in your post, our OMX experts will have a look.

Best Regards,

Pavel

Thanks so much ! Look forward for your replay !

Could you give me an email address of your OMX expert ?

Hi Pavel,

* Segment 2 , Temporarily used by VFPC internal buff heap */
  {
   1,                           /* valid */
   "VIDEO_M3_INT_HEAP_CACHED",  /* name */
   0x01800000,                  /* size */
   LDR_SEGMENT_TYPE_DYNAMIC_LOCAL_HEAP,        /* seg_type */
   0,                           /* flags */
   0x9BD00000,                  /* system_addr */
   0x9BD00000,                  /* slave_virtual_addr */
   -1,                          /* master_core_id */
   (1 << LDR_CORE_ID_VM3),      /* core_id_mask */
   1,                           /* cache_enable_mask */
   0,                           /* cache_operation_mask */
   -1                           /* shared_region_id */
  },

As you pointed,  "VIDEO_M3_INT_HEAP_CACHED"'s heap size should enlarge to allocate multiple codecs and I have successfully done that.

Right now, the problem is, when I try to allocate more, not the VIDEO_M3_INT_HEAP_CACHED is insufficient, but the following 1048576(1MB) heap is insufficient.

After allocate 4 codecs, this heap's free size will decrease to about 100KB.  So, what this heap? How can I increse it?

Thanks,

Llano

In this picture #1 is HDVICP2 system heap, #2 is HDVPSS system heap and #3 is HDVPSS internal heap (same as VIDEO_M3_INT_HEAP_CACHED in memsegdef.c)

It is the bug with release not to display internal heap (same as VIDEO_M3_INT_HEAP_CACHED in memsegdef.c) for HDVICP2, this will be fixed in subsequent releases.

  A new section add to http://processors.wiki.ti.com/index.php/EZSDK_sys_top_Linux_Utility#How_to_interpret_sys_top_output to describe these in more details.

Thanks & Regards
Velan

Hi Velan,

    Thanks for your replay! Can we modify the HDVICP2 system heap size as it will be used up when creating several decode or encode components at the same time.

Hi Llano,

     You said that you want to creat more than 10 codecs, I want to know how many codecs did you test  at the same time successfully until now? And could you give me an email address for more convenient discussion.

    Thanks very much !

Hi Velan,

I am clear now. Many thanks to your work on sys_top wiki. I have read that and several related system heap posts.

Could you point me out where can I get a guide to increase this #1 is HDVICP2 system heap size?

Do I have to get HDVICP2's source code and rebuild it?

Llano

Hi Llano,

     I know that you have already enlarged the VIDEO_M3_INT_HEAP_CACHED size successfully. Now I try to do the same thing followed the file  "How_to_increase_Heap_memory_available_to_VPSS_and_VICP_M3_cores_v2.docx". I did just exactly what the file said, but there are some problems when I run the omx examples. It seems that the OMX_Init() function failed. I don't know what's the problem, and  I doubt if the "HOST_CODEGEN_INSTALL_DIR" set in the makefile is the gcc or the arm-none-linux-gcc.

   Thanks, and I appreciate your replay!

Hi Liu,

There is a memory map.  1425.EZSDK_MemCfg.xls

When you modify the heap size, it will tell you the base and virtual address you should use to avoid make adjacent memory segment overlapped.

Change the file 'memsegdef_default.c' in the Media-Controller and rebuild the media-controller. Then put firmware_loader to the /usr/bin of the board's file system.

Modify the file sources/omx_05_02_00_38/src/ti/omx/memcfg/memtbl_cfg.h accordingly, and rebuild the OMX, put libtiomxcore.so.5.2 to /usr/lib

It should take effect now.

Llano

Hi Llano,

     Thanks so much for your replay! Now I have managed to modify the Memory Map successfully just as what you have said.

     For the  previous question about "0 Heap of HPVICP2", the TI employee doesn't give any forther replay. Do you have some progress in this issue?

Hi Velan,

    In the wiki page, you said that there is a .cfg file which defines the HDVICP2 system heap. I want to know where could I find the file and whether could I modify that to change the heap size. As the sys_top command could show the size of this heap, it must be read from the .cfg file.

    I and Llano met the same problem when creat several channels of decoder or encoder. Could you help us to solve the problem so as we can move on?  We  appreciate your replay!

To increase the #1 HDVICP2 system heap, you need to change the memory section and rebuild the M3 firmwares. This can be done by installing and then using the "EZSDK overlay installer". Please contact your local FAE for more details how to download, install and use the "EZSDK overlay installer".

Regards,

Pavel

Hi Pavel,

    Thanks for your kindly replay! I update my board with the latest ezsdk 5.05, and now the #1 HDVICP2 system heap is 1.5M in default. I modify the memsegdef_default.c to enlarge the VIDEO_M3_INT_HEAP_CACHED to 72M. Now I can manage to creat and run 8 channels of decoder at the same time.The input video format is 1280*720 15P. But when I increase the channels up to 9 or more, the OMX_Send_Cmd() function will release "OMX_ErrorInsufficientResources (0x80001000)". I change the input video to 640*480 15P and 320*240 15P, the result is the same as 1280*720 15P . The sys_top command shows that the  "#1 HDVICP2 system heap" and "VIDEO_M3_INT_HEAP_CACHED" are both sufficient to use. So, I don't know which kind of resource is use-up as  "OMX_ErrorInsufficientResources (0x80001000)" indicates. The following is the usage of the HDVICP2 when the input video format is 320*240 15P, and 8 channels of decoders run at the same time.

I appreciate your replay !

Hello,

OMX_ErrorInsufficientResources can indicate several error conditions like unable to allocate memory, error in creating event, error allocating pAppData, error in allocating resources in main, error in event retrieve.

Can you post the piece of code that logs/reports the OMX_ErrorInsufficientResources error, and in which C file ? You can examine the  ti-ezsdk_dm816x-evm_5_05_01_04/component-sources/omx_05_02_00_46/examples/ti/omx/demos/decode/src/decode_test.c file.

I can not locate OMX_Send_Cmd() function in the EZSDK. Do you mean OMX_SendCommand() instead?

Best Regards,

Pavel

Hi,

    The following is the part of code which reports the OMX_ErrorInsufficientResources after the inbuf and outbuf had been allocated. I am sorry for making mistakes about the name of OMX_SendCommand() function. I post the cmd output when running the program in the end, and you could have a look.

for(j=0;j<Decode_Channel;j++)
{
int flag=0;
eError =
OMX_SendCommand (pAppData->pDecHandle[j], OMX_CommandStateSet,
OMX_StateIdle, NULL);
//printf("send cmd success %d\n",j);
if (eError != OMX_ErrorNone)
{
printf ("Error in SendCommand()-OMX_StateIdle State set : %s \n",
IL_ClientErrorToStr (eError));
goto EXIT;
}
/* Allocate I/O Buffers; component would allocate buffers and would return
the buffer header containing the pointer to buffer */
for (i = 0; i < pAppData->decILComp[j]->inPortParams->nBufferCountActual; i++)
{
eError = OMX_AllocateBuffer (pAppData->pDecHandle[j],
&pAppData->decILComp[j]->inPortParams->pInBuff[i],
OMX_VIDDEC_INPUT_PORT, pAppData,
pAppData->decILComp[j]->inPortParams->nBufferSize);
if (eError != OMX_ErrorNone)
{
printf ("Error in OMX_AllocateBuffer()- Input Port State set : %s \n",
IL_ClientErrorToStr (eError));
goto EXIT;
}
}

printf (" decoder inport buffers allocated \n ");

/* buffer alloaction for output port */
for (i = 0; i < pAppData->decILComp[j]->outPortParams->nBufferCountActual; i++)
{
eError = OMX_AllocateBuffer (pAppData->pDecHandle[j],
&pAppData->decILComp[j]->outPortParams->pOutBuff[i],
OMX_VIDDEC_OUTPUT_PORT, pAppData,
pAppData->decILComp[j]->outPortParams->nBufferSize);
if (eError != OMX_ErrorNone)
{
printf ("Error in OMX_AllocateBuffer()-Output Port State set : %s \n",
IL_ClientErrorToStr (eError));
goto EXIT;
}
}
printf (" decoder outport buffers allocated \n ");
/* Wait for Idle stete of component after all buffers are alloacted componet would chnage to idle */

semp_pend (pAppData->decILComp[j]->done_sem);
}

Hi Bing,

This looks like error in creating event (set Idle state) or error in event retrieve. What does the logerSMDump utility report ?

Regards,

Pavel

Hi Bing,

What does  'ipcs -ls' say?

Is it possible semaphores are used up.

Llano

Hi Pavel,

    The following is the logerSMDump report when the error occurs, and you could have a look.

N:Video P:1 #:01523 T:00001f591c8d5c17 M:xdc.runtime.Main S:Entered Function :omxrpc_skel_allocbuffer
N:Video P:1 #:01522 T:00001f591c8c31d5 M:xdc.runtime.Main S:Module<OMX.TI.DUCATI.VIDDEC> @<_OMX_BASE_HandleStateTransition> @line<339> msg<Loaded to Idle Transition>
N:Video P:1 #:01525 T:00001f591c910cff M:xdc.runtime.Main S:Entered Function :omxrpc_skel_allocbuffer
N:Video P:1 #:01524 T:00001f591c8fc5f7 M:xdc.runtime.Main S:OMX_TI_VIDDEC_CommandNotify::Line 3079::VDEC->Loaded to Idle Transition Begin
N:Video P:1 #:01526 T:00001f591c9338ef M:xdc.runtime.Main S:OMX_TI_VIDDEC_CommandNotify::Line 3088::VDEC->Before VIDDEC3_create
N:Video P:1 #:01527 T:00001f591c947ff1 M:xdc.runtime.Main S:Entered Function :omxrpc_skel_allocbuffer
N:Video P:1 #:01529 T:00001f591c97f2f9 M:xdc.runtime.Main S:Entered Function :omxrpc_skel_allocbuffer
N:Video P:1 #:01528 T:00001f591c96fa6d M:xdc.runtime.Main S:OMX_TI_VIDDEC_CommandNotify::Line 3096::VDEC Error->VIDDEC3_create Failed
N:Video P:1 #:01530 T:00001f591c9b88af M:xdc.runtime.Main S:Entered Function :omxrpc_skel_allocbuffer
N:Video P:1 #:01531 T:00001f591c9eedad M:xdc.runtime.Main S:Entered Function :omxrpc_skel_allocbuffer
N:Video P:1 #:01532 T:00001f591ca1420f M:xdc.runtime.Main S: OMX Error in _OMX_BASE_HandleStateTransition :: line 344
N:Video P:1 #:01533 T:00001f591ca1c071 M:xdc.runtime.Main S:Module<OMX.TI.DUCATI.VIDDEC> Leaving<_OMX_BASE_HandleStateTransition> @line<483> with error<-2147479552:ErrorInsufficientResources>
N:Video P:1 #:01534 T:00001f591ca54a7f M:xdc.runtime.Main S:Module<OMX.TI.DUCATI.VIDDEC> Leaving<OMX_BASE_PROCESS_CmdEvent> @line<784> with error<-2147479552:ErrorInsufficientResources>
N:Video P:1 #:01535 T:00001f591ca5fc5b M:xdc.runtime.Main S:Module<OMX.TI.DUCATI.VIDDEC> Leaving<OMX_BASE_CmdEventHandler> @line<466> with error<-2147479552:ErrorInsufficientResources>
N:Video P:1 #:01536 T:00001f591ca6981d M:xdc.runtime.Main S:Module<OMX.TI.DUCATI.VIDDEC> Entering<OMX_BASE_CmdEventHandler> @line<434>
N:Video P:1 #:01537 T:00001f591ca71f93 M:xdc.runtime.Main S:Module<OMX.TI.DUCATI.VIDDEC> Leaving<OMX_BASE_CmdEventHandler> @line<466> with error<0:ErrorNone>

Hi Bing,

It is not possible to go beyond 8 channels with EZSDK. For multichannel, DVR RDK is recommended, where you can modify the M3 source code and go beyond 8 channels.

Can you provide more details for your use-case?

Regards,

Pavel

Hi Pavel,

     Thanks so much for your replay. We want to use EVM8168 board as a media-server for a video conference. The conference has 24 participants at the maximum, and each participant would send a h264 format video stream to the media-server. The media-server would receive these video streams and decode them all, then the server will mix some of these uncompressed video together and encode the mixed video using h264 format.

    So, the board  could have to do 24 channels of decoding at the worst situation. As the EZSDK supports 8 channels at most, I wonder if we could use one decoder to decode 3 video one after another, I mean decode 1 frame of video1, then 1 frame of video2, and 1 frame of video3, just in turn.

Hi Bing,

24 decodes would take lot of A8 cycles for IPC between A8 and M3, so you would run out of A8 cycles for your application. We recommend you to move to RDK, as it is meant for multichannel apps. Please contact your local FAE for RDK.

Decoder has history associated with frames, so it needs separate instances for each channel, else it would mix up all streams. So passing buffers for 24 channels in 8 instances would not be possible.  Still If you wish to use EZSDK, you can enter into NDA and can modify the code as per your needs, without waiting for EZSDK releases. Please contact/check with your local FAE for NDA.

Regards,

Pavel

But you could do it if you decoded each frame as an I-frame, then there would be no requirement for frames to have a "history" of past/future frames. I've done this in the decode_display demo code.

Of course, you would need to ensure that your H264 stream that you were decoding was entirely composed of I-frames in the first place...

Ralph

Hi Pavel,

    Thanks very much for your replay and suggestion ! We will discuss and decide which way to go.

Hi Ralph,

     Maybe our video streams could not match the condition, but thank you all the same !

Hi Pavel, 

     I want to confirm with your about one thing. As we know, frames in different GOPs(Group of Pictures) are independent with each other. So, can we use just one encoder/decoder to deal with multiple videos, given that we see each GOP as a basic unit? That is to say, if we only create one encoder, can we use it to encode GOP1 of video 1, and then use it to encode GOP1 of video 2, and then GOP2 of video 1, GOP2 of video 2……？

   Appreciate your replay !

Hi Bing,

    It is feasible, but still may not meet your requirement. If the GOP length is large, time slice for a certain encoder is too long and won't suit for real time conference. If GOP length is small,  you have to get more I frames which won't make a relative low bitrate.

   Of course if bandwidth is not your concern, you can try GOP method, dynamic setting I frames.

Llano

Hi Llano,

    Thanks for your replay ! You are right. We have to make the balance of GOP size and time delay. I will try and check out if this method could be used in our project. 

    I have another question about the ti-omx encode example in ezsdk5.05. The output h264 format video could not be played normly. I decode the output file back to yuv format and find that the Y component is normal as well as the U and V components are abnormal. The picturea are below. The fisrt one is the original yuv picture. The second one is the yuv picture decoded from the abnormal encoder output file. The third one is the Y component of the second one. It seems that the size of the U and V components is 1/4 of Y. I try to change the params of the encoder but make no progess. 

    So, could you help me to find the issue? 

Bing

Yes, that's right. The H264 encoder can only encode data of YUV 420 format.

Conversely, the decoder will only output YUV420 output. This is what you are seeing. There is Y data for every pixel however the U component is only provided for 1 in 4 pixels, and also the same applies for the V component. That is the definition of the YUV 420 format.

Ralph

Hi Ralph,

    Thanks for your replay! I know that the encoder takes only YUV420SemiPlanar(NV12) format input. I am sure the input file format is right, but the output h264 file is abnormal. I mean it could not be played.

Hello,

What application are you using to play back the encoded video? I have no problems putting the video into VLC and playing it back.

For an 8 pixel image, the decoded data would look like this:

YYYYYYYYUVUV

By the way, it's "reply", not "replay". :-)

Ralph

Hi Ralph,

    Thanks for your quick reply! I use the gstreamer pipeline "gst-launch -v filesrc location=out.264 ! h264parse ! ffdec_h264 ! ffmpegcolorspace ! xvimagesink" to play the video file. The decoded data is right and the format is NV12(YYYYYYYYUVUV).

   Could you offer me your output h264 file after encoding? My email address is liubing204@gmail.com.

Attached

1682.file.zip

It is not a zip file; you need to rename it ".h264" instead of ".zip" as TI forums won't let me upload h264.

Hi Ralph,

    The h264 file you shared has no problem when playing. I will decode it to yuv format and then encode the yuv file to see if there is still some issues. I will tell you the result in a while.

Hi Ralph,

    I tried that, but the result was just the same. I think maybe the codes of the decode and encode examples in my hand were not right. 

   So, could you do me a favor to share your decode and encode examples with me?

   Thanks very much!

In the latest EZSDK, see ti-ezsdk_dm816x-evm_5_05_01_04/component-sources/omx_05_02_00_46/examples/ti/omx/demos/

and in that directory you have "encode" and "decode_display" examples. Use "make omx" from the root of the EZSDK to build them.

Ralph

It is a bit mysterious. I will re-install the ezsdk5.05 to see if it will be right.

Thanks so much, Ralph! And I will let you know the result.

Hi Ralph,

    I install the ezsdk5.05 again, and make the omx examples. The result is just the same. I don't know why.

   Could you offer me your decode_a8host_debug.xv5T and encode_a8host_debug.xv5T? 

Here:

7635.encode_a8host_debug.zip

5228.decode_a8host_debug.zip

Sorry but I don't think I can help any further.

Thanks so much, Ralph!  

Hi,

     I modify the encode example to test using one encoder to encode multiple channels of input videos. I use two channels of input videos which have the same content. The encoder works but one of the output h264 file lacks the h264 header. I could not find which part of codes in the encode example incharge of writing the h264 header.

Does anyone know when and where does the encoder write the h264 header to the output file?

Thanks very much ! 

h264 headers are generated by the encoder firmware itself when encoding the first frame. They are referred as the SPS (sequence parameter set) and PPS (picture parameter set). You can force the codec to generate new header in the stream at any moment using the following code:

OMX_CONFIG_INTRAREFRESHVOPTYPE tVidEncIntraRefresh;
OMX_INIT_PARAM(&tVidEncIntraRefresh);

tVidEncIntraRefresh.nPortIndex = OMX_DirOutput;
eError = OMX_GetConfig(pEncHandle, OMX_IndexConfigVideoIntraVOPRefresh, &tVidEncIntraRefresh);
if (eError != OMX_ErrorNone)

{
 return -1;
}
tVidEncIntraRefresh.IntraRefreshVOP = 1;

eError = OMX_SetConfig(pEncHandle, OMX_IndexConfigVideoIntraVOPRefresh, &tVidEncIntraRefresh);
if (eError != OMX_ErrorNone)
{
 return -1;
}

Hi Belzile,

     Thanks so much for your reply! I will have a try.