TMS320DM8148: VIP Port Latency

Shaun

Using TI_EZSDK_EVM_05_05_02_00 and OMX, I'm creating an application that utilizes both the "OMX.TI.VPSSM3.VFCC" and the "OMX.TI.DUCATI.VIDENC" components to grab frames from the VIP port to encode. I'm working on better understanding the latency incurred from the VIP capture OMX component ("OMX.TI.VPSSM3.VFCC"). To help us better understand the latency, we replace the first pixel of every frame sent to the VIP port with a counter value. We also have the ability to inject a fixed number of frames into the VIP port.

We notice we have to send 6 frames to the VIP port before the first frame is passed to the encoder for encoding. We added code to ProcessPipeCmdETB to print out the first pixel of every frame to see which frame is passed to the encoder. It shows us that when we send the 6th frame to "OMX.TI.VPSSM3.VFCC", "OMX.TI.DUCATI.VIDENC" gets a frame with the 4th frame's counter value. As we keep passing frames, it appears the encoder continually gets frames that are 2 frame counter values behind. Is there anything we can remove this 2 frame latency? I don't see any settings for "OMX.TI.VPSSM3.VFCC" that would help reduce this number.

Thanks,

Shaun

over 8 years ago

0 Margarita Gashova over 8 years ago

TI__Guru* 80190 points

Hello,

Please check this measurements:
processors.wiki.ti.com/.../Latency_Measurement_on_Capture_Encode_Decode_Display_Demo
I would recommend you to check these threads also:
e2e.ti.com/.../708203
e2e.ti.com/.../601811

BR
Margarita

0 Shaun over 8 years ago in reply to Margarita Gashova

Intellectual 330 points

Hello,

Looking at the measurements you sent, I should be looking at 0.5 to 1 frame latency (Buffers in Capture driver latency). I've looked over the attached code in the .zip file provided in the link and there doesn't appear to be any significant differences that would explain the additional frame of latency.

Looking at this link, https://e2e.ti.com/support/dsp/davinci_digital_media_processors/f/717/p/198860/708203, Hardik Shah says that there are 2 frames of latency that cannot be removed. This contradicts the findings in the Latency Measurement on Capture Encode Decode Display Demo (0.5 to 1 frame of latency).

Which of these is correct?

Shaun

0 Brijesh Jadav over 8 years ago in reply to Shaun

TI__Guru**** 499425 points

Shaun,

VIP driver works slightly differently. It has internal queue of size 2 and when you enqueue a frame to the driver, depending on the size of the request queue, the enqueued buffer can either go to the request queue or internal queue. If it goes to the internal queue, then the at max the latency would be around 2 frames, but if it goes to the request queue, it can further increase the latency.

Regards,
Brijesh

0 Shaun over 8 years ago in reply to Brijesh Jadav

Intellectual 330 points

Thanks Brijesh for clearing that up

Processors

Processors forum

TMS320DM8148: VIP Port Latency