Open MLIR Meeting 7/15/2021: From MHLO to Linalg on Tensors

This Thursday (9am California Time, 16:00 UTC ), @MaheshRavishankar will talk about the IREE CodeGen, and in particular the path from MHLO to Linalg on Tensors.

Initially IREE used MHLO dialect as the input dialect for its compilation stack. Over time it became apparent that MHLO is at a higher level abstraction than what is needed within IREEs core compilation stack. As a result, the IREE core compilation stack was moved to use Linalg on Tensors as its input dialect. In this talk, Mahesh will walk through the issues that this move helped us resolve and also do a deep dive into the changes in IREE as well as Linalg on tensors that were needed to bridge several abstraction gaps to make this all work cohesively.

As usual the information to join the meeting:
‪+1 218-301-8485‬ PIN: ‪255 745‬#

I’ll also update this thread with slides and recording after the meeting.


Slides and recording are online!

1 Like