[PATCH] D56082: [X86][SLP] Enable SLP vectorization for 128-bit horizontal X86 instructions (add, sub)

Fri Feb 8 03:39:20 PST 2019

anton-afanasyev added a comment.

Here are `size..text` results for test-suite compiled with `Os`:

  > ~/llvm/test-suite/utils/compare.py  --filter-short -m size..text results_rel_base_Os.json results_rel_exp_Os.json
  Tests: 1160
  Short Running: 243 (filtered out)
  Remaining: 917
  Metric: size..text

  Program                                         results_rel_base_Os  results_rel_exp_Os diff  

  SingleSour...Regression-C++-pointer_method2   610.00               482.00              -21.0%
  MultiSourc...rks/Prolangs-C++/shapes/shapes   2674.00              2802.00              4.8% 
  MultiSource/Benchmarks/Bullet/bullet          586082.00            569010.00           -2.9% 
  MultiSourc.../Benchmarks/McCat/15-trie/trie   1218.00              1250.00              2.6% 
  MultiSourc...rsaBench/beamformer/beamformer   3826.00              3730.00             -2.5% 
  SingleSour...e/UnitTests/2003-05-07-VarArgs   1346.00              1378.00              2.4% 
  SingleSour.../Vector/SSE/Vector-sse.stepfft   2114.00              2066.00             -2.3% 
  SingleSource/UnitTests/initp1                 1570.00              1538.00             -2.0% 
  SingleSour...arks/Shootout/Shootout-objinst   802.00               818.00               2.0% 
  SingleSour...e/UnitTests/C++11/stdthreadbug   802.00               786.00              -2.0% 
  SingleSour...rks/Shootout/Shootout-methcall   818.00               834.00               2.0% 
  SingleSour...UnitTests/2003-08-11-VaListArg   1682.00              1714.00              1.9% 
  MultiSourc...s/Prolangs-C++/objects/objects   1762.00              1794.00              1.8% 
  SingleSour...tout-C++/Shootout-C++-methcall   898.00               914.00               1.8% 
  SingleSource/Benchmarks/Dhrystone/dry         1010.00              1026.00              1.6%
         results_rel_base_Os  results_rel_exp_Os        diff
  count  9.170000e+02         9.170000e+02        917.000000
  mean   3.171576e+04         3.172267e+04        0.000862  
  std    2.406645e+05         2.405892e+05        0.007775  
  min    3.700000e+02         3.700000e+02       -0.209836  
  25%    1.090000e+03         1.090000e+03        0.000000  
  50%    2.697800e+04         2.704200e+04        0.000000  
  75%    2.713800e+04         2.720200e+04        0.002360  
  max    7.149522e+06         7.148562e+06        0.047868  

and `exec_time` results for test-suite compiled with `O3`:

  > ~/llvm/test-suite/utils/compare.py  --filter-short -m exec_time results_rel_base.json results_rel_exp.json
  Tests: 1160
  Short Running: 762 (filtered out)
  Remaining: 398
  Metric: exec_time

  Program                                         results_rel_base  results_rel_exp diff  

  MicroBench...XRayFDRMultiThreaded/threads:2   524.56             411.88            -21.5%
  MicroBench...RayFDRMultiThreaded/threads:16   1057.94            875.81            -17.2%
  MicroBench...t:BM_PRESSURE_CALC_LAMBDA/5001    13.03              15.23            16.9% 
  MicroBench...XRayFDRMultiThreaded/threads:8   1087.77            912.91            -16.1%
  MultiSourc...nch/beamformer/beamformer.test     0.98               0.86            -11.9%
  MicroBench...mbda.test:BM_PIC_2D_LAMBDA/171     1.95               2.14             9.7% 
  MultiSource/Applications/siod/siod.test         1.80               1.64            -8.7% 
  MicroBench...XRayFDRMultiThreaded/threads:4   669.22             612.08            -8.5% 
  MicroBench...da.test:BM_PIC_1D_LAMBDA/44217   681.44             731.22             7.3% 
  SingleSour...ut-C++/Shootout-C++-hash2.test     2.41               2.58             7.0% 
  SingleSour...bra/kernels/gemver/gemver.test     0.84               0.78            -6.7% 
  MultiSourc...mbolics-dbl/Symbolics-dbl.test     2.99               2.80            -6.6% 
  MicroBench...ambda.test:BM_EOS_LAMBDA/44217    95.19              89.23            -6.3% 
  MicroBench...da.test:BM_HYDRO_2D_LAMBDA/171    16.66              17.69             6.2% 
  MicroBench...RayFDRMultiThreaded/threads:32   1022.42            1084.72            6.1%
         results_rel_base  results_rel_exp        diff
  count  397.000000         398.000000        397.000000
  mean   2536.523923        2523.193837      -0.001760  
  std    17779.494744       17680.005983      0.025555  
  min    0.600000           0.600000         -0.214805  
  25%    2.064000           2.065000         -0.003161    
  50%    6.472000           6.296000          0.000000    
  75%    148.025000         148.279000        0.001152    
  max    227023.000000      226291.000000     0.169066

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D56082/new/

https://reviews.llvm.org/D56082