[llvm-dev] Loop Opt WG Meeting Minutes for Aug 28, 2019

[Bardia Mahjour via llvm-dev]

---------------------------
Wed, Aug 28, 2019:
---------------------------

- Bump normalization (to make loop induction variables start from 0 and
  increment by 1)
   - History behind why it was removed: some targets were not able to
      reverse that transformation, and the transformation pessimized the
      input (eg. normalizing a loop that iterates over a pointer inserts
      a new variable for IV but the original pointer is still needed).
      Loop Strength Reduce should be able to clean it up it in theory.

- Delinearization/Multi-dimensional array indexing RFC
 (Siddharth)
  - A nice summary of the points raised so far prior to today's meeting
    are available at:
https://github.com/bollu/llvm-multidim-array-indexing-proposal/blob/master/discussion-first-round.md
  - Another possible idea was raised as a result of discussion in the
    meeting to extend the GEP instruction with operand bundles.
  - The bundle on GEP would be optional (like it is on calls) but it
    would also be "droppable" meaning that an optimization who doesn't
    know about it can drop it without changing the observable
    behaviour of the program. Siddharth to investigate the idea and
    share his findings at the next meeting.


----------------------------
Tentative Agenda for Sept 11
----------------------------

- LICM vs Loop Sink Strategy (Whitney)

- Status Updates
  - Follow up on multi-dimensional array indexing RFC (Siddharth)
  - Data Dependence Graph (https://reviews.llvm.org/D65350) (Bardia)
  - Impact of Loop Rotation on existing passes (Min-Yih)
  - Bugzilla bugs update (Vivek)
  - Others?


Bardia Mahjour
Compiler Optimizations
IBM Toronto Software Lab
bmahjour at ca.ibm.com (905) 413-2336
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20190828/aaef72c9/attachment.html>