Steps for integrating a new accelerator to TVM

[slai-nick]

Hi,

I am planning to integrate a new accelerator to TVM. I have been following the the BYOC doc and blog post thinking this was sufficient to integrate a new hardware but the device/target interactions section of the design and architecture part of the doc is making me confused now and I am thinking that BYOC might only be the tip of the iceberg.

Could anyone list clearly the steps needed for integrating a new hardware accelerator to TVM?

Cheers

[MJKlaiber]

Hi @slai-nick,

not sure if such a step-by-step tutorial exists yet. We are currently working on an interface for integrating hardware accelerators easily into TVM. It’s still work in progress. Feel free to provide input if it is helpful for you:

[RFC] UMA: Universal Modular Accelerator Interface

UMA: Universal Modular Accelerator Interface Feature Name: Universal Modular Accelerator Interface (UMA) Start Date: 2022 February Authors: Paul Palomero Bernardo @paulpb, Christoph Gerum @cgerum - University of Tübingen Michael J. Klaiber @mjklaiber, Ingo Feldner - Bosch Research Philipp van Kempen @philippvk, Rafael Stahl @r.stahl, Daniel Müller-Gritschneder - Technical University of Munich Johannes Partzsch - TU Dresden Andrew Stevens - Infineon Technologies RFC PR: https://github…

[ant_dev]

Hey @MJKlaiber, I would like to check whether this topic if relevant, as so this topic is 3 years old, would like to check whether currently there is available step by step a-z guide.

Thanks

[pkp124]

Hi all,

I am guessing the interface has now been established, I am also looking to do this exact same exercise. Can anyone please provide some pointers, any tutorials or docs for this would be greatly appreciated. I am still a newbie to the TVM stack, any pointers are helpful.

Thanks