[PATCH] D150994: Bump coalescing limit

[Matthias Braun via Phabricator via llvm-commits]

MatzeB added a comment.

FWIW I noticed that fixing this dramatically improves IPC and instructions executed, but does not actually affect benchmark runtimes. For my microbenchmark.

Before:

  5,581,342,454      cycles                    #    3.288 GHz                      (50.02%)
  11,940,550,021      instructions              #    2.14  insn per cycle           (50.05%)

After:

  5,595,396,911      cycles                    #    3.264 GHz                      (50.01%)
  6,809,294,177      instructions              #    1.22  insn per cycle           (50.06%)

so this isn't as critical as I initially thought, but I also don't think it hurt anything and improves code size.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D150994/new/

https://reviews.llvm.org/D150994