TDA4VM-Q1: if there are other more energy-efficient ways to transmit image data between processes？

Hi,

But why do you require to do memcpy to transfer the images? Why don't you use the same buffer in the next process like scalar? 

Regards,

Brijesh

How can i use the same buffer?

Hi Damon,

Why don't you share the buffer and just pass the pointer of the buffer around the modules, just like OpenVX does?

Regards,

Brijesh

i dont know how to share the Image buffer btween two process..

What are the two processes (modules) that you are using? Are these modules OpenVX based? 

Regards,

Brijesh

it's like this:

/resized-image/__size/920x480/__key/communityserver-discussions-components-files/791/1736218983191.jpg

Assumption: I send the output image of displayM2M node in GraphA to Graph B and Graph C, then 

1.In Graph B/C, can I directly use the displayM2M_output image to perform Enqueue operations?

2.Does Graph A have to wait for Graph B/C finish using the image and return it before continuing to execute the next frame?

Yes, it is possible to share images in this case between GraphA and GraphB&C. Essentially you need to create graph parameter for the output of GraphA and input of GraphB and C. and since all three graphs are running in pipeline mode, there will be multiple images allocated/shared among them, so when graphA is sharing a buffer with GraphB and C, it will be working on another buffer, without block. It will be blocked only if GraphB and C does not return buffers in time and most of the intermediate buffers are input to the GraphB and C. 

Regards,

Brijesh

Is my understanding correct？

/resized-image/__size/750x240/__key/communityserver-discussions-components-files/791/Enqueue_2D00_Dequeue.jpg

And is there some demos that I can refer to?

Regards,

Damon

Yes, it looks correct. Well there are many examples of graph parameter in the SDK. But there isn't exact usecase in the SDK.

Regards,

Brijesh

Thank you, I am familiar with Graph parameters.But confused about how to use the same image between three graphs.

And will this cause synchrinuzation issues between graphs?

I'll try your suggestion and keep you updated on the issue in the future.

Regards,

Damon

Now the image buffer after the Dequeue of process A is given to process B for use. During the use of process B, process A cannot use this buffer (otherwise it will cause buffer synchronization problems), which blocks process A's graph. I cannot increase the depth of the graph and the number of buffers anymore. Is there any other way?

Damon@INVO said:

Now the image buffer after the Dequeue of process A is given to process B for use. During the use of process B, process A cannot use this buffer

This is then not related to two processes, this can happen between two nodes also and in order to run them in parallel, at a full speed, we need to increase pipeline depth between them. 

Btw, in this case, as soon as the first node in processB is completed processing this frame, the frame would be released from processB and would be available to processA for consumption. 

Regards,

Brijesh