Enhancements in toy tutorial of MLIR

MLIR
1

Hi , the toy tutorial is awesome as it develops a complete/self-sufficient end-to-end compiler using MLIR framework.
For this awesome toy tutorial, could there be list of potential modifications also suggested with solutions so that developers going through the tutorial can make code-changes to understand it better –

Following are some of the modifications I am trying on my own end to understand the framework better :

  1. For chap1, there can be support for more statements, data-type.
  2. For chap2, : there can be task like support for subtraction operator
  3. For chap3: there can be task like support for pattern like multiplication of a tensor with 0 , multiplication with 2 replaced by shift operator etc.
  4. For chap4 : An interface for custom prints and so on …

I am a beginner and may be wrong at above points but these lists /code availability from the experts would be really a very important helper for beginners.

If this sounds correct, could we assume something like that in near future?
Or, any other inputs provided on how can I understand the MLIR framework in a better,faster and easier way could be really helpful ? :smiley: :grinning:

2 Likes
2

These look potentially interesting, are you thinking about trying to send PRs for these?

3

Agreed. I would like to suggest one more thing, It would be better if it toy tutorial explains more about ‘type converter’.
I think type converter customization used very frequently when converting between different dialects, but toy tutorial only explains about default type converter.

1 Like
4

Yeah, It’s a nice proposal for learning by doing, Proper exercise really improves one’s understanding of MLIR framework

5

I apologise for the delayed reply. Sure, I can send PR for some of these … :smiley: :smiley: