Optimize facts

[laurenthuberdeau]

Results:

• Significant improvements on benchmarks with lots of rules.
• No regression on other benchmarks

Using Vector

Benchmark       bench-master  bench-vector    
clique/10       0.322e-3      0.039e-3 -87.94%
clique/25       2.201e-3      0.191e-3 -91.34%
clique/40       0.571e-2      0.051e-2 -91.01%
clique/5        0.910e-4      0.199e-4 -78.17%
line/10         0.369e-3      0.266e-3 -28.06%
line/25         3.156e-3      1.798e-3 -43.03%
line/40         1.056e-2      0.562e-2 -46.80%
line/5          0.883e-4      0.785e-4 -11.15%
loop/100        2.032e-3      1.381e-3 -32.03%
loop/1000       0.877e-1      0.842e-1  -3.93%
loop/50         0.505e-3      0.469e-3  -7.13%
loop/500        2.347e-2      2.144e-2  -8.66%
parse/100       1.269e-2      1.285e-2  +1.24%
parse/1000      1.327e-1      1.325e-1  -0.15%
parse/50        0.604e-2      0.602e-2  -0.33%
parse/500       0.662e-1      0.679e-1  +2.62%
tight/100       0.757e-3      0.729e-3  -3.73%
tight/1000      0.775e-1      0.721e-1  -6.89%
tight/50        2.109e-4      2.075e-4  -1.59%
tight/500       1.699e-2      1.655e-2  -2.61%
Geometric mean  0.358e-2      0.208e-2 -41.85%

Forcing the Vector in compilePred using force:

Benchmark       bench-master  bench-vector-strict
clique/10       0.322e-3      0.043e-3 -86.57%   
clique/25       2.201e-3      0.238e-3 -89.21%   
clique/40       0.571e-2      0.098e-2 -82.81%   
clique/5        0.910e-4      0.232e-4 -74.54%   
line/10         0.369e-3      0.267e-3 -27.58%   
line/25         3.156e-3      1.858e-3 -41.14%   
line/40         1.056e-2      0.583e-2 -44.83%   
line/5          0.883e-4      0.804e-4  -9.00%   
loop/100        2.032e-3      1.394e-3 -31.43%   
loop/1000       0.877e-1      0.852e-1  -2.82%   
loop/50         0.505e-3      0.470e-3  -6.93%   
loop/500        2.347e-2      2.140e-2  -8.82%   
parse/100       1.269e-2      1.286e-2  +1.30%   
parse/1000      1.327e-1      1.349e-1  +1.68%   
parse/50        0.604e-2      0.619e-2  +2.55%   
parse/500       0.662e-1      0.693e-1  +4.81%   
tight/100       0.757e-3      0.731e-3  -3.51%   
tight/1000      0.775e-1      0.736e-1  -5.01%   
tight/50        2.109e-4      2.076e-4  -1.54%   
tight/500       1.699e-2      1.700e-2  +0.03%   
Geometric mean  0.358e-2      0.223e-2 -37.73% 

Using Lists

Benchmark       bench-master  bench-list      
clique/10       0.322e-3      0.039e-3 -87.86%
clique/25       2.201e-3      0.192e-3 -91.29%
clique/40       0.571e-2      0.060e-2 -89.51%
clique/5        0.910e-4      0.200e-4 -78.05%
line/10         0.369e-3      0.275e-3 -25.63%
line/25         3.156e-3      1.974e-3 -37.46%
line/40         1.056e-2      0.580e-2 -45.14%
line/5          0.883e-4      0.781e-4 -11.57%
loop/100        2.032e-3      1.467e-3 -27.83%
loop/1000       0.877e-1      0.868e-1  -0.97%
loop/50         0.505e-3      0.480e-3  -4.91%
loop/500        2.347e-2      2.437e-2  +3.84%
parse/100       1.269e-2      1.291e-2  +1.66%
parse/1000      1.327e-1      1.355e-1  +2.13%
parse/50        0.604e-2      0.616e-2  +1.94%
parse/500       0.662e-1      0.713e-1  +7.71%
tight/100       0.757e-3      0.735e-3  -2.88%
tight/1000      0.775e-1      0.707e-1  -8.71%
tight/50        2.109e-4      2.084e-4  -1.19%
tight/500       1.699e-2      1.651e-2  -2.86%
Geometric mean  0.358e-2      0.215e-2 -39.91%

Using hashset:

Benchmark       bench-master  bench-hashset   
clique/10       0.322e-3      0.048e-3 -84.95%
clique/25       2.201e-3      0.297e-3 -86.52%
clique/40       0.571e-2      0.157e-2 -72.58%
clique/5        0.910e-4      0.214e-4 -76.45%
line/10         0.369e-3      0.259e-3 -29.82%
line/25         3.156e-3      1.776e-3 -43.75%
line/40         1.056e-2      0.554e-2 -47.56%
line/5          0.883e-4      0.769e-4 -12.93%
loop/100        2.032e-3      1.345e-3 -33.83%
loop/1000       0.877e-1      0.819e-1  -6.53%
loop/50         0.505e-3      0.456e-3  -9.66%
loop/500        2.347e-2      2.057e-2 -12.35%
parse/100       1.269e-2      1.230e-2  -3.08%
parse/1000      1.327e-1      1.327e-1  +0.04%
parse/50        0.604e-2      0.619e-2  +2.52%
parse/500       0.662e-1      0.663e-1  +0.29%
tight/100       0.757e-3      0.708e-3  -6.45%
tight/1000      0.775e-1      0.698e-1  -9.84%
tight/50        2.109e-4      2.004e-4  -4.96%
tight/500       1.699e-2      1.648e-2  -3.03%
Geometric mean  0.358e-2      0.225e-2 -37.25%

Data:
bench-hashset.csv
bench-list.csv
bench-master.csv
bench-vector-strict.csv
bench-vector.csv

[laurenthuberdeau]

Benchmarks with inlining:

Benchmark       bench-inline-new  bench-optimizations-with-inline
clique/10       1.387e-4          0.303e-4 -78.18%               
clique/25       1.045e-3          0.183e-3 -82.46%               
clique/40       2.687e-3          0.803e-3 -70.12%               
clique/5        0.411e-4          0.119e-4 -70.94%               
line/10         1.528e-4          1.186e-4 -22.35%               
line/25         1.037e-3          0.631e-3 -39.10%               
line/40         2.973e-3          1.646e-3 -44.63%               
line/5          0.419e-4          0.390e-4  -6.93%               
loop/100        0.550e-3          0.540e-3  -1.88%               
loop/1000       1.968e-2          1.827e-2  -7.14%               
loop/50         2.178e-4          2.155e-4  -1.01%               
loop/500        0.614e-2          0.575e-2  -6.37%               
parse/100       1.292e-2          1.274e-2  -1.38%               
parse/1000      1.378e-1          1.361e-1  -1.22%               
parse/50        0.627e-2          0.616e-2  -1.85%               
parse/500       0.686e-1          0.676e-1  -1.48%               
tight/100       1.647e-4          1.464e-4 -11.10%               
tight/1000      1.122e-2          1.008e-2 -10.20%               
tight/50        0.565e-4          0.529e-4  -6.50%               
tight/500       3.024e-3          2.545e-3 -15.84%              
Geometric mean  1.420e-3          0.961e-3 -32.34%   

bench-inline-new.csv
bench-optimizations-with-inline.csv