TDA4VM: Cross Compiling EdgeAI TIDL Tools and Examples

Hi Isharaka,

1. The model compilation should be done on the PC, and we have Python examples that compiles and generates the model artifact required for delegation to the C7x.

https://github.com/TexasInstruments/edgeai-tidl-tools/blob/54b5524ba91b48411a919c2cd47ffe6cbdfa25ac/README.md#python-examples

The compiled models and artifacts from the Python example can then be copied into the Jacinto board target to run with the CPP examples that have been compiled on target.  Also, the CPP example should have the same build steps for running on PC emulation and on target. 

However, I am not aware of a method to cross compile the CPP example on a PC to run on a Jacinto board target, but I can check with the team. Many people are out due to US and India holiday so the response will be sometime next week though.

2. When running the cpp demos from the git repo, there is no need to build the entire SDK. Please use the SD card image as this would be much easier for development. Running the setup_script.sh included with the SD card image should clone the edgeai-tidl-tools repo.

Sincerely,

Takuma

Hi Isharaka,

I checked with the team, but here is the response:

• We do not have support for cross compilation of CPP on the PC for the Jacinto processor.
• For CPP applications, we provide native compilation on the PC and native compilation on the Jacinto processor.
• For models, we provide compilation only on the PC. However, the compiled model can be ran on both the PC and Jacinto processor

In summary, for CPP Edge AI examples we expect customers to always compile natively. On the other hand, for model + model artifacts we expect customers to always compile on the PC and copy the compiled model to the Jacinto processor.

Regards,

Takuma