[PATCH] D125588: [MachineCombiner] Improve MachineCombiner's cost model

[Guozhi Wei via Phabricator via llvm-commits]

Carrot added a comment.
Herald added a subscriber: jsji.

I tried SPEC2006 on a AArch64 machine.  But for some unknown problem I couldn't run 464.h264ref correctly, even gcc compiled code generates same output.

  base
  400.perlbench    9770        306       31.9 *
  401.bzip2        9650        418       23.1 *
  403.gcc          8050        206       39.1 *
  429.mcf          9120        332       27.5 *
  445.gobmk       10490        337       31.1 *
  456.hmmer        9330        275       34.0 *
  458.sjeng       12100        391       30.9 *
  462.libquantum  20720        214       96.8 *
  464.h264ref                                 NR
  471.omnetpp      6250        229       27.3 *
  473.astar        7020        313       22.4 *
  483.xalancbmk    6900        174       39.6 *
   Est. SPECint(R)_base2006              33.6
  
  
  exp
  400.perlbench    9770        306       31.9 *
  401.bzip2        9650        418       23.1 *
  403.gcc          8050        207       38.9 *
  429.mcf          9120        330       27.6 *
  445.gobmk       10490        337       31.1 *
  456.hmmer        9330        280       33.4 *
  458.sjeng       12100        391       31.0 *
  462.libquantum  20720        213       97.5 *
  464.h264ref                                 NR
  471.omnetpp      6250        230       27.2 *
  473.astar        7020        313       22.4 *
  483.xalancbmk    6900        177       39.1 *
   Est. SPECint(R)_base2006              33.5
  
  The biggest difference is from 456.hmmer. I reran it and got same result
  456.hmmer        9330        273       34.1 S
  456.hmmer        9330        274       34.0 *
  456.hmmer        9330        274       34.0 S

So there is no real performance impact of SPEC2006 on AArch64.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D125588/new/

https://reviews.llvm.org/D125588