TDA4VM: The graph does not update output when multiple graph running in different threads

Hi,

I have assigned this thread to correct expert.

Please expect response from him/her.

Hi,

qing wang said:

1. If input graph and model 1 graph are combined into one graph which graph run normally, but the dequeue data of node3  in model 2 graph is always the same.

If output of capture node is going to both the graphs, then I would suggest keeping capture node as a separate graph and then feed this buffer to both the graphs.

qing wang said:

.When the pipeline depth and buffer depth of input graph and model 1 graph are set to 4, but the pipeline depth and buffer depth of model 2graph are set to 1, Which buffer of the input graph does the model2 graph use?

Pipeline depth is for the graph. So if you set pipeline depth as 1 for model2 graph, it would take the same

Buffer depth is for the node. Here, in the graph parameter list, you should ensure that the ref list for capture node output, node 1 input and node 3 input are the same. If this is the case, then the buffer depth for all the 3 will be the same.

Regards,

Nikhil

Nikhil Dasan said:

If output of capture node is going to both the graphs, then I would suggest keeping capture node as a separate graph and then feed this buffer to both the graphs

Can you explain the reason in detail?

Nikhil Dasan said:

it would take the same

Don't quite understand.

Nikhil Dasan said:

Buffer depth is for the node. Here, in the graph parameter list, you should ensure that the ref list for capture node output, node 1 input and node 3 input are the same. If this is the case, then the buffer depth for all the 3 will be the same.

I need to make sure three image(capture node、node 1 input and node 3 input) are consistent，but model2 graph is not in the same thread as other graphs and runs at different speeds. In this case, if the buffer depth is 1 and model1 consumes this frame first, will model2 lose data of capture image?

The output of node2 as a graph parameter for model1 graph and buffer depth is 4, is the buffer depth of node1 output the default value of 1? If model1 graph pipeline depth is also 4, will each graph instance use the same buffer of node 1?

Thanks.

Hi,

qing wang said:

If output of capture node is going to both the graphs, then I would suggest keeping capture node as a separate graph and then feed this buffer to both the graphs

Can you explain the reason in detail?

Here your capture node output would be a graph parameter as you would have to dequeue and enqueue to the other graph. 

Hence, now since the application takes care of enqueue and dequeue, it is better to keep the receiver nodes in separate graph so that whenever there is buffer available, the nodes could use it (based on the buffer depth) and does not interfere with each other.

qing wang said:

Don't quite understand.

Pipeline depth is specific to a graph, so if you set the model 2 graph with pipeline depth as 2 then it will of pipeline depth 2

qing wang said:

I need to make sure three image(capture node、node 1 input and node 3 input) are consistent，but model2 graph is not in the same thread as other graphs and runs at different speeds. In this case, if the buffer depth is 1 and model1 consumes this frame first, will model2 lose data of capture image?

The output of node2 as a graph parameter for model1 graph and buffer depth is 4, is the buffer depth of node1 output the default value of 1? If model1 graph pipeline depth is also 4, will each graph instance use the same buffer of node 1?

In order to make the flow consistent, you could do the below.

in the graph parameter list, you should ensure that the ref list for capture node output, node 1 input and node 3 input are the same.

Then, in your application, you dequeue from capture node and enqueue to node 1 and node 3. In this case, your flow will be consistent

Regards,

Nikhil

Nikhil Dasan said:

The output of node2 as a graph parameter for model1 graph and buffer depth is 4, is the buffer depth of node1 output the default value of 1? If model1 graph pipeline depth is also 4, will each graph instance use the same buffer of node 1?

Thank you for your reply.

please help to answer this question

Hi

qing wang said:

The output of node2 as a graph parameter for model1 graph and buffer depth is 4, is the buffer depth of node1 output the default value of 1?

Yes, node 1 will have buffer depth as 1 unless you set it another value.

qing wang said:

If model1 graph pipeline depth is also 4, will each graph instance use the same buffer of node 1?

Yes, But since in each graph instance, as the node runs on the same target, they would run one after the other. And if there is node 2 is accessing this buffer, then this would block node 1. Hence we suggest to have a buffer depth more than 1 in such cases.

Regards,

Nikhil