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

[Guozhi Wei via Phabricator via llvm-commits]

Carrot added a comment.

I tested SPEC2006 int on my skylake desktop.

  base
  400.perlbench    9770        241       40.6 *                                  
  401.bzip2        9650        380       25.4 *                                  
  403.gcc          8050        198       40.7 *                                  
  429.mcf          9120        204       44.8 *                                  
  445.gobmk       10490        343       30.6 *                                  
  456.hmmer        9330        253       36.9 *                                  
  458.sjeng       12100        355       34.1 *                                  
  462.libquantum  20720        308       67.3 *                                  
  464.h264ref     22130        325       68.1 *                                  
  471.omnetpp      6250        239       26.1 *                                  
  473.astar        7020        295       23.8 *                                  
  483.xalancbmk    6900        144       47.8 *                                  
   Est. SPECint(R)_base2006              38.3
  
  exp
  400.perlbench    9770        241       40.6 *                                  
  401.bzip2        9650        381       25.3 *                                  
  403.gcc          8050        198       40.7 *                                  
  429.mcf          9120        202       45.3 *                                  
  445.gobmk       10490        342       30.7 *                                  
  456.hmmer        9330        253       36.9 *                                  
  458.sjeng       12100        354       34.1 *                                  
  462.libquantum  20720        318       65.2 *                                  
  464.h264ref     22130        324       68.3 *                                  
  471.omnetpp      6250        237       26.3 *                                  
  473.astar        7020        295       23.8 *                                  
  483.xalancbmk    6900        145       47.5 *                                  
   Est. SPECint(R)_base2006              38.2
  
  Only mcf and libquantum have a little larger delta, they are caused by larger run to run variation.
  
  base
  429.mcf          9120        207       44.1 S                                  
  429.mcf          9120        204       44.8 *                                  
  429.mcf          9120        198       46.0 S         
  462.libquantum  20720        302       68.7 S                                  
  462.libquantum  20720        334       62.1 S                                  
  462.libquantum  20720        308       67.3 *                                  
  
  exp
  429.mcf          9120        202       45.3 *                                  
  429.mcf          9120        207       44.0 S                                  
  429.mcf          9120        197       46.2 S                                  
  462.libquantum  20720        325       63.8 S                                  
  462.libquantum  20720        318       65.2 *                                  
  462.libquantum  20720        296       70.1 S                            

So basically no difference.


Repository:
  rG LLVM Github Monorepo

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

https://reviews.llvm.org/D125588