Open MLIR Meeting 1/13/2021: One-Shot Function Bufferization of Tensor Programs

This Thursday (9am California Time, 17:00 UTC ), @matthias-springer will talk about bufferization: the process of materializing tensor values into memory.

MLIR currently has two solutions for bufferizations:

  • “Core” bufferization is implemented via multiple passes, each of which bufferizes a part of the input IR (partial bufferization). It conservatively inserts buffer allocations/copies on every memory write and relies on subsequent memref-based analyses/passes to remove unneeded allocations/copies. See the original presentation from 2020-09-24 on this approach (slides - recording)

  • Linalg Comprehensive Bufferize is a new bufferization that bufferizes entire functions in “one shot” (single pass). It analyzes use-def chains of tensor values/ops (as opposed to memrefs) to determine if buffer allocations/copies are necessary (before modifying the IR). It often produces fewer buffer copies than core bufferization, especially when the input IR contains matching tensor.extract_slice/tensor.insert_slice pairs, as is often the case after tiling.

In this talk, @matthias-springer will give an overview of the new one-shot bufferization, how to use it, and how it can be extended to support new ops. Finally, he will also discuss plans for unifying both bufferization solutions.

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.

The attached document contains a more detailed explanation of the design of the bufferization framework. There won’t be enough time to explain bufferization in all details and I will refer to this design document here and there. (This is an updated and extended version of the same document that I shared a few months ago.)

one_shot_bufferization_design.pdf (436.2 KB)


Here are the slides and the recording from this morning!

Can you capture the PDF you attached here as a doc in the repo so that it shows up on the website @matthias-springer ?

Yes, I will do that together with the commit that moves stuff over to the bufferization dialect.

1 Like