AM62A7: Question related to VPAC sub modules

Hi Jingjie,

jingjie.huang said:

they all said LSC should be prior to NF

Yes, LSC is before NSF4.

jingjie.huang said:

In 7.5.3.4.3, the lens shading correction compensation is added because the LSC is implemented prior to NSF.

LSCC is needed because LSC is before NSF4.

jingjie.huang said:

2. How to configure the modules to work in the On-The-Fly mode? I've seen chapters talking about this mode for RAWFE DPC and MSC,etc. 

I am not sure what you are asking exactly.
Could you please explain?

jingjie.huang said:

I'm wondering between LDC, TEE and MSC, which one have optimal performance when duplicating the exact same output?

Do you mean copying an image?

Simply DMA copying would be enough.

I am not sure what you mean by TEE.

Hi Gang,

Please allow me to further explain the question：

1. The first question is just me not understanding why putting LSC in the first is not optimal.

2. The second question is about the OTF and M2M operations. The concept can be found in multiple chapter and apparently the OTF mode has shorter pipeline thus introduces less latency. We would like to know what mode we are using when invoking tiovxisp plugin in gstreamer, and how to configure it so that OTF mode is applied.

3. The third question is about duplicating a video stream inside the gstreamer pipeline. We've tried two plugins: tiovxmultiscaler(MSC based) and tee (software based). The TRM said that LDC can produce dual output and I think it is suitable to achieve the aformentioned dulipcation process. Unfortunately the tiovxldc plugin doesn't seems capable of outputing two streams. Again this question is about finding an optimal way so that latency can be reduced.

I hope the questions are now more clear. If you need more information, please don't hesitate to say.

Regards,

Huang Jingjie

Hi jingjie,

jingjie.huang said:

We would like to know what mode we are using when invoking tiovxisp plugin in gstreamer, and how to configure it so that OTF mode is applied.

OTF mode is supported by CSI-RX and VPAC-VISS H/W.

In OTF mode, VISS can only support a single camera as its input.
We cannot use VISS with any other camera input at the same time.

I don't think gstreamer and tiovx can support OTF mode now.

jingjie.huang said:

The TRM said that LDC can produce dual output and I think it is suitable to achieve the aformentioned dulipcation process.

Yes, LDC H/W can support dual output.

jingjie.huang said:

Unfortunately the tiovxldc plugin doesn't seems capable of outputing two streams.

I am not familiar with gstreamer details.
It is possible that this feature is not implemented by gstreamer.

I cannot tell if dual output from LDC is better than other options.
If LDC causes high input DDR BW, increasing its output DDR BW may slow down LDC further.

Hi Gang,

Thanks for the explaination.

So currently for AM62A, which approach can actually utilize OTF mode? Do you have any plan to add this feature to gstreamer based applications?

Regards,

Huang Jingjie

Thanks for the reply. It's unfortunate this advanced feature is not available for the most common streaming application.

Hi Jingjie,

There is some difficulty on AM62A for OTF mode because sensor driver is on Linux/A53 side and VISS driver is on the RTOS/R5 side.

There is a plan for supporting OTF on TDA4 RTOS devices in 10.2 first and then we may start planning for AM62A.