[PATCH] D62262: [PGO][CHR] Speed up following long use-def chains.

[Hiroshi Yamauchi via Phabricator via llvm-commits]

yamauchi added a comment.

In D62262#1512384 <https://reviews.llvm.org/D62262#1512384>, @davidxl wrote:

> was there a quadratic behavior before ? It seems linear before and after the patch -- but just cut the computation by a factor of 2 .  Do you need an option to control the max chain length?


I think at least something more than linear. The downstream nodes got visited many, many times (eg. in the test, imagine how many paths there are from %v301 to %v298, to %v296, then all the way down to %v0.)

We could use a max length option in the future but for now it seems fine.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D62262/new/

https://reviews.llvm.org/D62262