TDA4VM: How to adjust graph queue depth

Hi Daigo,

It depends on the graph execution, you have to identify if any node is blocked because of buffer then you can increase the buffer depth.

The node may be blocked if the queue is empty ( which means the same node at previous pipe instance might have acquired the buffer and not released it ).

Increasing the queue doesn't affect the performance but only increases memory consumption, wherein decreasing might reduce the pipelining performance.

You have to analyse your graph execution and set this buffer depth to optimize the memory and performance.

In case you have not gone through the docs, i am sharing the link for it,

https://software-dl.ti.com/jacinto7/esd/processor-sdk-rtos-jacinto7/latest/exports/docs/tiovx/docs/user_guide/TIOVX_PIPELINING.html#TIOVX_PIPELINING_PERFORMANCE_OPTIMIZATION

Regards,
Gokul

Hi Gokul,

We realized that we could use the following visualization tool:
https://software-dl.ti.com/jacinto7/esd/processor-sdk-rtos-jacinto7/latest/exports/docs/tiovx/docs/user_guide/DEBUG_LOG_RT.html

We will use this to analyze and adjust the graphs. If we have any further questions, please support us.

Regards,

Daigo

Yeah sure Daigo. I will wait for your update.

Regards,
Gokul