TDA4VM: Is there any node in TIOVX execute the function as Null link in TDA2 VSDK?

Hi,

May I know is there any update?

Thanks and regards,

Cherry

Hi Cherry,

Cherry Zhou said:

Is there any node in TIOVX execute the function as Null link in TDA2 VSDK? which could be used do something in A72?

No there is no null node in TIOVX.

Cherry Zhou said:

In TDA2 VSDK, there is  Null link which could do some process at A15 core and by callback function when receiving the images from other link while there isn't any node could do the similar thing on TIOVX, so it's difficult to process the image from camera in A72.

Do you mean null-sink link? i dont recollect if null-sink was doing any processing..  It can just receive input some other link and return the input as it is back.. There is not any node like this in TIOVX, but you could have output of node unconnected or can declare it as graph parameter and can get this output in the application. 

The above graph could be realized by opening capture node two times, one for 4 channels and second for 1 channel. 4-channels can be for encode and mosaic and fifth channel for the dms path.

Regards,

Brijesh 

Hello Brijesh

Nice to communicate with you.

I'm the one post the question. and now this issue is fixed and closed by using a select node which created by PyTIOVX tools. But I'm not sure this is the best solution for this certain question.

on the other side, it seems interesting to solve this question by opening capture node twice. could you kindly give the example code ? I wanna know how to implement this.

Thanks and best rgds!

have a good weekend.

Hi henry o,

There is only one example in sdk, where multiple instance of the capture node are created. You could find this example in the file ti-processor-sdk-rtos-j721e-evm-08_02_00_05\tiovx\kernels_j7\hwa\test\test_csitx_csirx.c. This essentially is CSITX -> CSIRX example, where separate node instance is created for each input channel.. 

henry o said:

I'm the one post the question. and now this issue is fixed and closed by using a select node which created by PyTIOVX tools. But I'm not sure this is the best solution for this certain question.

That's interesting. Could you please help me understand what this select node is? If possible, can you share this node, so that it would be helpful?

Regards,

Brijesh

Hello Brijesh,

Thanks for your kindly help and response.

For select node, the input is object array contrains image frames which comes from capture node or something else, the output the another object array consists of images. e.g. there are 5 camera, channel [0,1,2,3,4], every one produce image frames at certain rate. and the first four should send their output into encode node and the fifth one should send to process the image.

then we could create two select node, both of them take the 5 cameras' output as input, and the first select node choose the first four channel frames and copy the frame to output object array. the second select node choose the fifth channel frame and copy it to output object array.

this select node works as the select link in vision sdk (or processor sdk in tda2x).

wish it's helpful.

best rgds.

Hello henry o,

Thanks for the explanation. Just one more question, how do you copy images from input to output object arrays? Are just copying buffer pointer or images using DMA?

Regards,

Brijesh

Hello Brijesh,

since it's a little difficult to use DMA copying the image buffer data, the memcpy is used here to copy the image.

furthermore, I take account of copying buffer pointer, but it seems that it will bring the coupling issue,  e.g. the pipeline could be blocked or slow down when some node processes the buffer pointer slower than expected.

so, at last, the memcpy is chosen.

Regards,

Henry