TMS320F28388D: Edge AI with TMS320F28388D- Support and Compatibility

Prabhanjan Kopp

Hi,
I am currently trying to explore Edge-AI tool and want to understand if it supports TMS320F28388D. The forums suggest, this is possible but if I select c2000 family in my demo example, I cannot find TMS320F28388D in the list of supported devices.

7 months ago

0 Adithya Thonse 7 months ago

TI__Prodigy 580 points

Hi Prabhanjan, F2838 is not supported. F2837 is supported in the command line toolchain: https://github.com/TexasInstruments/tinyml-tensorlab/tree/main

You can also see the list of supported devices in the commandline toolchain here: https://github.com/TexasInstruments/tinyml-tensorlab/tree/main?tab=readme-ov-file#supported-target-devices

All of these devices (except 2837) will be supported on Edge AI Studio in a month's time as well.

0 Prabhanjan Kopp 7 months ago

Prodigy 20 points

Thank you Aditya. I am still a bit confused.
Does it mean we cannot run the neural compiler for F28388D ? I was referring to your answer to the question : https://e2e.ti.com/support/microcontrollers/c2000-microcontrollers-group/c2000/f/c2000-microcontrollers-forum/1470025/tms320f28388d-deploy-neural-network-model-from-my-own-onnx-file-export-by-pytorch-using-edgeai-studio-for-c2000

you suggest here that we can run the neural compiler for F28388D device using the command line but if I see the list of devices supported, I do not see 28388D listed. what am I missing here? Or is it that we cannot use EDGE-AI but can use the command line toolchain for F28388D?

+1 Adithya Thonse 7 months ago in reply to Prabhanjan Kopp

TI__Prodigy 580 points

So Prabhanjan, let me break it down for you.

TI provides 3 toolchains for AI that are interlinked.

1. TI Neural Network Compiler: Just compiles an onnx model to static library. This is for the most advanced customers who have their own training framework, just need a way to compile onnx model onto TI MCUs. This can be used for F2838 as well - Checkout Section 2.1.1 in software-dl.ti.com/.../index.html

2. TI Tiny ML Tensorlab: Commandline toolchain that I was referring to above. It is for customers who understand basic python and know to work around a terminal. It has all the flows for AI tasks and a variety of example applications. It supports plenty of devices including F2837 (but not F2838, this was a conscious choice as there was no customer demand for F2838 support)

3. Edge AI Studio-Model Composer: A GUI based tool for novice customers focused on no code, ease of use. The current release has very few device supported as you have pointed out (and definitely no F2838)

0 Prabhanjan Kopp 7 months ago in reply to Adithya Thonse

Prodigy 20 points

Hi Aditya,
Sorry , I had a related question on the same topic and hence putting it here:
" The F2838 device also has an ARM cortex M4 , does the compiler support putting an onxx model onto ARM Cortex M4?
Or there are other tools which we need to use to accomplish this.

Thanks,

Prabhanjan

0 Adithya Thonse 7 months ago

TI__Prodigy 580 points

Prabhanjan, for Cortex M4, you could use a compilation setting like: -DARM_CPU_INTRINSICS_EXIST -mcpu=cortex-m4 -mfloat-abi=hard -mfpu=fpv4-sp-d16 -O3 -Wl,-u,_c_int00 -Wno-return-type -march=armv7e-m -mthumb

Since I haven't tried anything on F2838, I may not be able to help much. The command I have given above is for another MCU that has a M4 core

0 Prabhanjan Kopp 6 months ago in reply to Adithya Thonse

Prodigy 20 points

Hi Aditya,
Thank you for that. will give it a spin and come back with an update.

C2000™︎ microcontrollers

C2000 microcontrollers forum

TMS320F28388D: Edge AI with TMS320F28388D- Support and Compatibility