[PATCH] D60839: [ScheduleDAGInstrs] Compute topological ordering on demand.

[Tom Stellard via Phabricator via llvm-commits]

tstellar added a comment.

Thanks for working on this.  Here are some unscientific benchmarks from a pathological case on AMDGPU:

before rL358583 <https://reviews.llvm.org/rL358583>:
real    0m2.134s
user    0m2.081s
sys     0m0.053s

rL358583 <https://reviews.llvm.org/rL358583>:
real    0m1.858s
user    0m1.810s
sys     0m0.048s

D60839 <https://reviews.llvm.org/D60839>:
real    0m1.798s
user    0m1.742s
sys     0m0.057s

And prior to rL340346 <https://reviews.llvm.org/rL340346> this program took 7.5s to compile, and it's relatively small (2943 IR instructions).


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D60839/new/

https://reviews.llvm.org/D60839