DM8127 camera and isp link

Because I want to use FPGA to implement our own alg such as H3A, WDR, and etc. then FPGA feeds the yuv data to DM8127, I do not need the glbce and noise filter.

1. you said:"This is not possible in ISIF->RSZ->DDR mode."  but how could I implement the YUV resizing?

2. If “ ISIF->DDR->RSZ->DDR mode” is possible, How could I do for it?

3. When the capture mode is one shot mode, the frames that need to be encoded sometimes are skipped. What's the reason for it? Moreover, in continuous capture mode, the skpped frames are less.

But there is another problem coming out---I need four channel different resolution YUV streams.

The camera link only can output two channels. So how could  other  two channels YUV be generated?

Hi, Brijesh. Thanks for your timely replying!

I do not understand that you said "Even in one-shot mode, i think JPEG input is generated by duplicating the 1080p output, so from resizer, there are only two output streams." The three stream are not used dup link before isp link, and not used scaler link for the next operation. They are all generated by isp link in different size. The codes are as follows:

CameraLink_CreateParams_Init(&cameraPrm);
  cameraPrm.captureMode =  CAMERA_LINK_CAPMODE_DDR;
  cameraPrm.outQueParams[0].nextLink = ispId;
  cameraPrm.ispLinkId = ispId;

  /* This is for TVP5158 Audio Channels - Change it to 16 if there are 16
   * audio channels connected in cascade */
  cameraPrm.numAudioChannels = 0;
  cameraPrm.numVipInst = 1;
  cameraPrm.tilerEnable = TRUE;

  pCameraInstPrm = &cameraPrm.vipInst[0];
  pCameraInstPrm->vipInstId = SYSTEM_CAMERA_INST_VP_CSI2;
  pCameraInstPrm->videoDecoderId = MultiCh_getSensorId(gUI_mcfw_config.sensorId);
  pCameraInstPrm->inDataFormat = SYSTEM_DF_YUV420SP_UV;
  pCameraInstPrm->sensorOutWidth  = 1920;
  pCameraInstPrm->sensorOutHeight = 1080;
   pCameraInstPrm->standard = SYSTEM_STD_1080P_30;
  pCameraInstPrm->numOutput = 1;
  pCameraOutPrm = &pCameraInstPrm->outParams[0]; 
  pCameraOutPrm->dataFormat = SYSTEM_DF_BAYER_RAW; 
  pCameraOutPrm->scEnable = FALSE;
  pCameraOutPrm->scOutWidth = 1920;
  pCameraOutPrm->scOutHeight = 1080;
  pCameraOutPrm->outQueId = 0;
  /* 2A config */
  cameraPrm.t2aConfig.n2A_vendor = gUI_mcfw_config.n2A_vendor;
  cameraPrm.t2aConfig.n2A_mode = gUI_mcfw_config.n2A_mode;
  cameraPrm.vnfFullResolution = FALSE;

 /* isp link params */
 ispPrm.inQueParams.prevLinkId    = gVcamModuleContext.cameraId;
 ispPrm.inQueParams.prevLinkQueId = 0;
 ispPrm.numOutQueue               = 3;
 ispPrm.outQueInfo[0].nextLink    = mergeId;
 ispPrm.outQueInfo[1].nextLink    = mergeId;
 ispPrm.outQueInfo[2].nextLink    = dupId;
 ispPrm.clkDivM    = 10;
 ispPrm.clkDivN          = 20;
 ispPrm.vsEnable = gUI_mcfw_config.vsEnable;
 ispPrm.outQueuePrm[0].dataFormat = SYSTEM_DF_YUV420SP_VU;
 ispPrm.outQueuePrm[0].width      = 1920;
 ispPrm.outQueuePrm[0].height     = 1080;
 ispPrm.outQueuePrm[1].dataFormat = SYSTEM_DF_YUV420SP_VU;
 ispPrm.outQueuePrm[1].width      = 720;
 ispPrm.outQueuePrm[1].height     = 576;
 ispPrm.outQueuePrm[2].dataFormat = SYSTEM_DF_YUV420SP_VU;
 ispPrm.outQueuePrm[2].width      = 1920;
 ispPrm.outQueuePrm[2].height     = 1536;
 ispPrm.vnfFullResolution         = FALSE;

The result is OK, and also I could get two H.264 stream and one JPEG(1920*1536) stream.

As you said, the performance is limited.  There are many frames skipped in the stream, especially when there are some motions in the scene.

I have ever tried to increase the output number of camera at continuous mode, but there are some errors of time out.

If I use  duplicate link and then  scalar link(not resize link), would the performance of scaler be limited? Because I have to use two scaler utilizations at least.

Thanks for your help.

But if I want to use camera link for 4 output stream at continuous mode, What should I modify?

I have only implement it at one-shot mode. As I have said previously that I tried to increase the output number of camera at continuous mode, but there were some errors of time out.

Thanks for your useful help.

Which is better for the image capturing in processing  efficiency,   Continuous mode or one shot mode?

In our demo, The stream generated from Continuous mode plays more fluently than  that  generated from one shot mode. Does one shot mode support  processing 1080P YUV@60fps  in  real time?