sol2_test/tests/perf_tests.cpp

#include <fmt/core.h>
#include <string>
#include <fuc2/testing.hpp>
#include "stats.hpp"

using namespace fuc2;

namespace perf_tests {
  struct LargeSOA {
    std::vector<int> a01;
    std::vector<int> a02;
    std::vector<int> a03;
    std::vector<int> a04;
    std::vector<int> a05;
    std::vector<int> a06;
    std::vector<int> a07;
    std::vector<int> a08;
    std::vector<int> a09;
    std::vector<int> a10;
    std::vector<int> a11;
    std::vector<int> a12;
    std::vector<int> a13;
    std::vector<int> a14;
    std::vector<int> a15;
    std::vector<int> a16;
    std::vector<int> a17;
    std::vector<int> a18;
    std::vector<int> a19;
    std::vector<int> a20;

    void push_back(int x) {
      a01.push_back(x);
      a02.push_back(x);
      a03.push_back(x);
      a04.push_back(x);
      a05.push_back(x);
      a06.push_back(x);
      a07.push_back(x);
      a08.push_back(x);
      a09.push_back(x);
      a10.push_back(x);
      a11.push_back(x);
      a12.push_back(x);
      a13.push_back(x);
      a14.push_back(x);
      a15.push_back(x);
      a16.push_back(x);
      a17.push_back(x);
      a18.push_back(x);
      a19.push_back(x);
      a20.push_back(x);
    }
  };

  struct LargeStruct {
    int a01;
    int a02;
    int a03;
    int a04;
    int a05;
    int a06;
    int a07;
    int a08;
    int a09;
    int a10;
    int a11;
    int a12;
    int a13;
    int a14;
    int a15;
    int a16;
    int a17;
    int a18;
    int a19;
    int a20;

    LargeStruct(int x) :
      a01(x),
      a02(x),
      a03(x),
      a04(x),
      a05(x),
      a06(x),
      a07(x),
      a08(x),
      a09(x),
      a10(x),
      a11(x),
      a12(x),
      a13(x),
      a14(x),
      a15(x),
      a16(x),
      a17(x),
      a18(x),
      a19(x),
      a20(x) {}
  };


  struct S1 {
    int a01;
    int a02;
    int a03;
    int a04;

    S1(int x) : a01(x), a02(x), a03(x), a04(x) { }
  };

  struct S2 {
    int a05;
    int a06;
    int a07;
    int a08;
    S2(int x) : a05(x), a06(x), a07(x), a08(x) { }
  };

  struct S3 {
    int a09;
    int a10;
    int a11;
    int a12;
    S3(int x) : a09(x), a10(x), a11(x), a12(x) { }
  };

  struct S4 {
    int a13;
    int a14;
    int a15;
    int a16;
    S4(int x) : a13(x), a14(x), a15(x), a16(x) { }
  };

  struct S5 {
    int a17;
    int a18;
    int a19;
    int a20;
    S5(int x) : a17(x), a18(x), a19(x), a20(x) { }
  };

  struct MultipleAOS {
    std::vector<S1> s1;
    std::vector<S2> s2;
    std::vector<S3> s3;
    std::vector<S4> s4;
    std::vector<S5> s5;

    void push_back(int x) {
      s1.push_back({x});
      s2.push_back({x});
      s3.push_back({x});
      s4.push_back({x});
      s5.push_back({x});
    }
  };

  struct Samples {
    std::vector<Stats> soa_big_for;
    std::vector<Stats> soa_many_for;
    std::vector<Stats> ls_big_for;
    std::vector<Stats> ls_many_for;
    std::vector<Stats> maos_big_for;
    std::vector<Stats> maos_many_for;
  };

  void test_large_soa(Samples& samples, int sample_size, int sample_number) {
    auto soa = new LargeSOA();

    for(int i = 0; i < sample_size; i++) {
      soa->push_back(i);
    }

    Stats big_for;
    for(int sample = 0; sample < sample_number; sample++) {
      auto start = big_for.time_start();
      for(int i = 0; i < sample_size; i++) {
        soa->a01[i] += soa->a02[i] + sample;
        soa->a02[i] += soa->a03[i] + sample;
        soa->a03[i] += soa->a04[i] + sample;
        soa->a04[i] += soa->a05[i] + sample;
        soa->a05[i] += soa->a06[i] + sample;
        soa->a06[i] += soa->a07[i] + sample;
        soa->a07[i] += soa->a08[i] + sample;
        soa->a08[i] += soa->a09[i] + sample;
        soa->a09[i] += soa->a10[i] + sample;
        soa->a10[i] += soa->a11[i] + sample;
        soa->a11[i] += soa->a12[i] + sample;
        soa->a12[i] += soa->a13[i] + sample;
        soa->a13[i] += soa->a14[i] + sample;
        soa->a14[i] += soa->a15[i] + sample;
        soa->a15[i] += soa->a16[i] + sample;
        soa->a16[i] += soa->a17[i] + sample;
        soa->a17[i] += soa->a18[i] + sample;
        soa->a18[i] += soa->a19[i] + sample;
        soa->a19[i] += soa->a20[i] + sample;
      }
      big_for.sample_time(start);
    }
    samples.soa_big_for.push_back(big_for);

    Stats many_for;
    for(int sample = 0; sample < sample_number; sample++) {
      auto start = many_for.time_start();
      for(int i = 0; i < sample_size; i++) soa->a01[i] += soa->a02[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a02[i] += soa->a03[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a03[i] += soa->a04[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a04[i] += soa->a05[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a05[i] += soa->a06[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a06[i] += soa->a07[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a07[i] += soa->a08[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a08[i] += soa->a09[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a09[i] += soa->a10[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a10[i] += soa->a11[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a11[i] += soa->a12[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a12[i] += soa->a13[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a13[i] += soa->a14[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a14[i] += soa->a15[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a15[i] += soa->a16[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a16[i] += soa->a17[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a17[i] += soa->a18[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a18[i] += soa->a19[i] + sample;
      for(int i = 0; i < sample_size; i++) soa->a19[i] += soa->a20[i] + sample;
      many_for.sample_time(start);
    }
    samples.soa_many_for.push_back(many_for);
  }

  void test_large_struct(Samples& samples, int sample_size, int sample_number) {
    auto ls = new std::vector<LargeStruct>;

    for(int i = 0; i < sample_size; i++) {
      ls->push_back({i});
    }

    Stats ls_big_for;
    for(int sample = 0; sample < sample_number; sample++) {
      auto start = ls_big_for.time_start();
      for(auto& el : *ls) {
        el.a01 += el.a02 + sample;
        el.a02 += el.a03 + sample;
        el.a03 += el.a04 + sample;
        el.a04 += el.a05 + sample;
        el.a05 += el.a06 + sample;
        el.a06 += el.a07 + sample;
        el.a07 += el.a08 + sample;
        el.a08 += el.a09 + sample;
        el.a09 += el.a10 + sample;
        el.a10 += el.a11 + sample;
        el.a11 += el.a12 + sample;
        el.a12 += el.a13 + sample;
        el.a13 += el.a14 + sample;
        el.a14 += el.a15 + sample;
        el.a15 += el.a16 + sample;
        el.a16 += el.a17 + sample;
        el.a17 += el.a18 + sample;
        el.a18 += el.a19 + sample;
        el.a19 += el.a20 + sample;
      }
      ls_big_for.sample_time(start);
    }

    samples.ls_big_for.push_back(ls_big_for);

    Stats ls_many_for;
    for(int sample = 0; sample < sample_number; sample++) {
      auto start = ls_many_for.time_start();
      for(auto& el : *ls) el.a01 += el.a02 + sample;
      for(auto& el : *ls) el.a02 += el.a03 + sample;
      for(auto& el : *ls) el.a03 += el.a04 + sample;
      for(auto& el : *ls) el.a04 += el.a05 + sample;
      for(auto& el : *ls) el.a05 += el.a06 + sample;
      for(auto& el : *ls) el.a06 += el.a07 + sample;
      for(auto& el : *ls) el.a07 += el.a08 + sample;
      for(auto& el : *ls) el.a08 += el.a09 + sample;
      for(auto& el : *ls) el.a09 += el.a10 + sample;
      for(auto& el : *ls) el.a10 += el.a11 + sample;
      for(auto& el : *ls) el.a11 += el.a12 + sample;
      for(auto& el : *ls) el.a12 += el.a13 + sample;
      for(auto& el : *ls) el.a13 += el.a14 + sample;
      for(auto& el : *ls) el.a14 += el.a15 + sample;
      for(auto& el : *ls) el.a15 += el.a16 + sample;
      for(auto& el : *ls) el.a16 += el.a17 + sample;
      for(auto& el : *ls) el.a17 += el.a18 + sample;
      for(auto& el : *ls) el.a18 += el.a19 + sample;
      for(auto& el : *ls) el.a19 += el.a20 + sample;
      ls_many_for.sample_time(start);
    }
    samples.ls_many_for.push_back(ls_many_for);
  }


  void test_maos_style(Samples& samples, int sample_size, int sample_number) {
    auto maos = new MultipleAOS();

    for(int i = 0; i < sample_size; i++) {
      maos->push_back(i);
    }

    Stats maos_big_for;
    for(int sample = 0; sample < sample_number; sample++) {
      auto start = maos_big_for.time_start();
      for(int i = 0; i < sample_size; i++) {
        maos->s1[i].a01 += maos->s1[i].a02 + sample;
        maos->s1[i].a02 += maos->s1[i].a03 + sample;
        maos->s1[i].a03 += maos->s1[i].a04 + sample;
        maos->s1[i].a04 += maos->s2[i].a05 + sample;

        maos->s2[i].a05 += maos->s2[i].a06 + sample;
        maos->s2[i].a06 += maos->s2[i].a07 + sample;
        maos->s2[i].a07 += maos->s2[i].a08 + sample;
        maos->s2[i].a08 += maos->s3[i].a09 + sample;

        maos->s3[i].a09 += maos->s3[i].a10 + sample;
        maos->s3[i].a10 += maos->s3[i].a11 + sample;
        maos->s3[i].a11 += maos->s3[i].a12 + sample;
        maos->s3[i].a12 += maos->s4[i].a13 + sample;

        maos->s4[i].a13 += maos->s4[i].a14 + sample;
        maos->s4[i].a14 += maos->s4[i].a15 + sample;
        maos->s4[i].a15 += maos->s4[i].a16 + sample;
        maos->s4[i].a16 += maos->s5[i].a17 + sample;

        maos->s5[i].a17 += maos->s5[i].a18 + sample;
        maos->s5[i].a18 += maos->s5[i].a19 + sample;
        maos->s5[i].a19 += maos->s5[i].a20 + sample;
      }
      maos_big_for.sample_time(start);
    }
    samples.maos_big_for.push_back(maos_big_for);

    Stats maos_many_for;
    for(int sample = 0; sample < sample_number; sample++) {
      auto start = maos_many_for.time_start();

      for(auto& el : maos->s1) {
        el.a01 += el.a02 + sample;
        el.a02 += el.a03 + sample;
        el.a03 += el.a04 + sample;
      }

      for(int i = 0; i < sample_size; i++) {
        maos->s1[i].a04 += maos->s2[i].a05 + sample;
      }

      for(auto& el : maos->s2) {
        el.a05 += el.a06 + sample;
        el.a06 += el.a07 + sample;
        el.a07 += el.a08 + sample;
      }

      for(int i = 0; i < sample_size; i++) {
        maos->s2[i].a08 += maos->s3[i].a09 + i;
      }

      for(auto& el : maos->s3) {
        el.a09 += el.a10 + sample;
        el.a10 += el.a11 + sample;
        el.a11 += el.a12 + sample;
      }

      for(int i = 0; i < sample_size; i++) {
        maos->s3[i].a12 += maos->s4[i].a13 + sample;
      }

      for(auto& el : maos->s4) {
        el.a13 += el.a14 + sample;
        el.a14 += el.a15 + sample;
        el.a15 += el.a16 + sample;
      }

      for(int i = 0; i < sample_size; i++) {
        maos->s4[i].a16 += maos->s5[i].a17 + sample;
      }

      for(auto& el : maos->s5) {
        el.a17 += el.a18 + sample;
        el.a18 += el.a19 + sample;
        el.a19 += el.a20 + sample;
      }

      maos_many_for.sample_time(start);
    }

    samples.maos_many_for.push_back(maos_many_for);
  }

  void test_all_performance() {
    Samples samples;

    int sample_size = 100;
    int sample_count = 100;

    for(int i = 0; i < 100; i++) {
      // quick and dirty "randomness"
      if(i % 2 == 0) {
        test_large_soa(samples, sample_size, sample_count);
        test_large_struct(samples, sample_size, sample_count);
        test_maos_style(samples, sample_count, sample_count);
      } else if(i % 3 == 0) {
        test_large_struct(samples, sample_size, sample_count);
        test_large_soa(samples, sample_size, sample_count);
        test_maos_style(samples, sample_count, sample_count);
      } else {
        test_maos_style(samples, sample_count, sample_count);
        test_large_struct(samples, sample_size, sample_count);
        test_large_soa(samples, sample_size, sample_count);
      }
    }

    fmt::println("soa_bf soa_mf ls_bf ls_mf maos_bf maos_mf");

    Samples mofm;

    Stats soa_big_for;
    Stats soa_many_for;
    Stats ls_big_for;
    Stats ls_many_for;
    Stats maos_big_for;
    Stats maos_many_for;

    for(size_t i = 0; i < samples.soa_big_for.size(); i++) {
      fmt::println("{:.2f} {:.2f} {:.2f} {:.2f} {:.2f} {:.2f}",
          samples.soa_big_for[i].mean(),
          samples.soa_many_for[i].mean(),
          samples.ls_big_for[i].mean(),
          samples.ls_many_for[i].mean(),
          samples.maos_big_for[i].mean(),
          samples.maos_many_for[i].mean());

      // this is potentially total bullshit
      soa_big_for.sample(samples.soa_big_for[i].mean());
      soa_many_for.sample(samples.soa_many_for[i].mean());
      ls_big_for.sample(samples.ls_big_for[i].mean());
      ls_many_for.sample(samples.ls_many_for[i].mean());
      maos_big_for.sample(samples.maos_big_for[i].mean());
      maos_many_for.sample(samples.maos_many_for[i].mean());
    }

    // totally not sure if this t-test is legit, must study more
    soa_big_for.t_test(ls_big_for);
    soa_big_for.t_test(maos_big_for);

    soa_many_for.t_test(ls_many_for);
    soa_many_for.t_test(maos_many_for);

    ls_many_for.t_test(maos_many_for);
    ls_big_for.t_test(maos_big_for);

    // the most pathologic case
    fmt::println("---- {}", soa_big_for.mean() - ls_many_for.mean());
    soa_big_for.t_test(ls_many_for);
  }

  fuc2::Set TESTS{
    .name="perf",
      .tests={
        TEST(test_all_performance),
      }
  };
}