[PATCH] D67383: Add new optimization pass of Tree-Height-Reduction

[Florian Hahn via Phabricator via llvm-commits]

fhahn added a comment.

@masakazu.ueno / @fsaintjacques are you still interested in pushing this forward? We recently had a couple of user reports where `-reassoicate` causes large regressions in hand-tuned code written with intrinsics, because it linearizes expressions that have been carefully written to exploit maximal parallelism, e.g. see https://godbolt.org/z/e6fWcbdjn

A general pass to reduce the height of compute expressions would certainly by very valuable for such cases.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D67383/new/

https://reviews.llvm.org/D67383