cbalint13 commented on PR #12087:
URL: https://github.com/apache/tvm/pull/12087#issuecomment-1188904111
> Thank you for the great contribution for better accelerator supports!
partying_face I enjoyed reading the code and learning the design. Overall,
implementation looks good to me. Here are a few high-level comments:
>
> * Could you add more documentation in general (e.g., explain what
parameters are)?
> * It would be great to have a testing for end-to-end model-level testing.
(e.g., run BERT with UMA on ARM EthosU)
>
> Please see below for more comments. Thanks!
> Thank you for the great contribution for better accelerator supports!
partying_face I enjoyed reading the code and learning the design. Overall,
implementation looks good to me. Here are a few high-level comments:
>
Personal view over the first part (non code related) of @sunggg requests, of
course just IMHO:
> * Could you add more documentation in general (e.g., explain what
parameters are)?
- Current sample (_template) & doc are fine for a **contrib** module.
- It greatly enables **"out-of-tree"** maintenance towards custom HW
accelarator design.
- Explanatory parts **Vanilla**, **Strawberry** & **Chocolate** are nice
levels of introductory guide.
> * It would be great to have a testing for end-to-end model-level testing.
(e.g., run BERT with UMA on ARM EthosU)
- Sounds beyond of scope, same way we could "dream" a fully verilated BERT
running **verilator-contrib** module (!) @tmoreau89 , @vegaluisjose
- As for ARM-Ethos, my understanding is a quite different (not so
**"out-of-tree"** principled) way to connect things inside, but @manupa-arm can
comment more on this.
- Would be possible in the future to decouple ARM-Ethos (or Vitis-AI) to be
more **"out-of-tree"** by porting it over UMA, but this would be quite bit of
another task.
Some personal efforts envisioned similar **out-of-tree** approaches {for
i.e. [OLIMP](https://github.com/cbalint13/OLIMP) /
[e-verest](https://github.com/cbalint13/e-verest) }, and UMA just landed as
super nice **"out-of-tree"** topping. As my last wish, I put forward hope that
we community one day will be able to leverage jointly the complex
hardware-design space within TVM's unique compiler-space.
Beyond code-related requests below, I halt my review here with a @lgtm
claim, and a go for UMA v1.0 contrib !
> Please see below for more comments. Thanks!
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
