Questions regarding the future development of MicroTVM

[LouisLi]

Dear TVM Development Team,

Greetings! I am an engineer from a DSP software development team. Our company is currently planning to develop a product that integrates a DSP with a neural network accelerator unit, similar to Texas Instruments’ TDA4VM architecture. Our team is primarily responsible for implementing the AI compiler. After conducting thorough research, we have decided to adopt TVM as our starting point due to its strong compatibility and effective optimization capabilities. However, we have some questions we hope you could address, and we would greatly appreciate your assistance.

1. Our hardware setup consists solely of a DSP combined with a neural network accelerator unit running on bare metal, without an MCU. Based on the information provided on your website and our understanding, it appears that the TVM Runtime may not be applicable in this scenario. Could you please clarify whether we can utilize MicroTVM under these conditions? Additionally, is the inclusion of an MCU a necessary requirement for using MicroTVM?
2. From your website, we have observed that the TVM team is actively developing MicroTVM, and the list of supported devices predominantly features Arm’s cortex processors. Could you provide an update on the current development progress of MicroTVM?

We sincerely look forward to your response. Thank you very much for your time and assistance!

Best regards,

[ebraheema]

Hi @LouisLi I saw your question and come up on my mind this discussion…

Phasing out Legacy Components

Over the past year, the community has worked hard to bring in and transition to a more flexible and productive flow for ML compilers. One lesson we learned is that it is hard to build a silver bullet for everything. Additionally, given the amount of time and energy contributed by community volunteers, it is hard to build and maintain a single compiler pipeline that aims to fit all purposes, in our case, all backends and use cases. The engineering complexity inevitably grows as we try to grow the…

As i understand from it , microTVM phased out from TVM main branch and the latest one version that support it was on TVM 18.

Best regards, Ebraheem

[LouisLi]

@ebraheema

As i understand from it , microTVM phased out from TVM main branch and the latest one version that support it was on TVM 18.

Hi Ebraheem, Thank you very much for your enthusiastic response. In fact, shortly after I asked this question, I saw a git commit indicating that the TVM main branch has phased out MicroTVM:)

By the way, are there any plans for future versions of TVM to support bare metal devices, especially those without an operating system and MCU devices? Due to cost considerations, we are not currently planning to integrate an MCU into our hardware devices.

Thanks again and hope you have a nice day!

Best Regards, Louis

[Hzfengsy]

The community is open to bringing microTVM back. The main reason to phase out current microTVM is because:

1. There is little maintenance to the microTVM component.
2. The community is pushing the unity flow based on the relax. However, the current microTVM is coupled with some legacy parts.

Any contribution of leveraging microTVM to the relax flow is welcomed by the community

[LouisLi]

Hello Siyuan,

Thank you very much for your prompt response. I will share this information with my colleagues. Wishing you success with your development.

Best regards, Louis Li

[ebraheema]

Hi @Hzfengsy

Thank you for your response. can you elaborate what legacy parts that microTVM is coupled with? what is the essential parts needed to leverage microTVM to relax flow?

Best regards, Ebraheem

[Hzfengsy]

@ebraheema

can you elaborate what legacy parts that microTVM is coupled with?

To be specific, the legacy part is relay flow

@ebraheema

what is the essential parts needed to leverage microTVM to relax flow?

I’m sorry I’m not 100% sure, as I’m not an expert on microTVM. As far as I know, C runtime and AOT need to be added to the relax flow, as well as the compilation integration.