library/cpp/accurate_accumulate/benchmark/main.cpp


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97

#include <library/cpp/accurate_accumulate/accurate_accumulate.h>
#include <library/cpp/testing/benchmark/bench.h>

#include <util/generic/algorithm.h>
#include <util/generic/singleton.h>
#include <util/generic/vector.h>
#include <util/generic/xrange.h>
#include <util/random/fast.h>

namespace {
    template <typename T, size_t N>
    struct TNormalizedExamplesHolder {
        TVector<T> Examples;
        TNormalizedExamplesHolder()
            : Examples(N)
        {
            TFastRng<ui64> prng{sizeof(T) * N * 42u};
            for (auto& x : Examples) {
                x = prng.GenRandReal4();
            }
        }
    };

    template <typename T, size_t N>
    struct TExamplesHolder {
        TVector<T> Examples;
        TExamplesHolder()
            : Examples(N)
        {
            TFastRng<ui64> prng{sizeof(T) * N * 42u + 100500u};
            for (auto& x : Examples) {
                // operations with non-normalized floating point numbers are rumored to work slower
                x = prng.GenRandReal4() + prng.Uniform(1024u);
            }
        }
    };
}

#define DEFINE_BENCHMARK(type, count)                                                                     \
    Y_CPU_BENCHMARK(SimpleNorm_##type##_##count, iface) {                                                 \
        const auto& examples = Default<TNormalizedExamplesHolder<type, count>>().Examples;                \
        for (const auto i : xrange(iface.Iterations())) {                                                 \
            Y_UNUSED(i);                                                                                  \
            Y_DO_NOT_OPTIMIZE_AWAY(                                                                       \
                (type)Accumulate(std::cbegin(examples), std::cend(examples), type{}));                    \
        }                                                                                                 \
    }                                                                                                     \
                                                                                                          \
    Y_CPU_BENCHMARK(KahanNorm_##type##_##count, iface) {                                                  \
        const auto& examples = Default<TNormalizedExamplesHolder<type, count>>().Examples;                \
        for (const auto i : xrange(iface.Iterations())) {                                                 \
            Y_UNUSED(i);                                                                                  \
            Y_DO_NOT_OPTIMIZE_AWAY(                                                                       \
                (type)Accumulate(std::cbegin(examples), std::cend(examples), TKahanAccumulator<type>{})); \
        }                                                                                                 \
    }                                                                                                     \
                                                                                                          \
    Y_CPU_BENCHMARK(Simple_##type##_##count, iface) {                                                     \
        const auto& examples = Default<TExamplesHolder<type, count>>().Examples;                          \
        for (const auto i : xrange(iface.Iterations())) {                                                 \
            Y_UNUSED(i);                                                                                  \
            Y_DO_NOT_OPTIMIZE_AWAY(                                                                       \
                (type)Accumulate(std::cbegin(examples), std::cend(examples), type{}));                    \
        }                                                                                                 \
    }                                                                                                     \
                                                                                                          \
    Y_CPU_BENCHMARK(Kahan_##type##_##count, iface) {                                                      \
        const auto& examples = Default<TExamplesHolder<type, count>>().Examples;                          \
        for (const auto i : xrange(iface.Iterations())) {                                                 \
            Y_UNUSED(i);                                                                                  \
            Y_DO_NOT_OPTIMIZE_AWAY(                                                                       \
                (type)Accumulate(std::cbegin(examples), std::cend(examples), TKahanAccumulator<type>{})); \
        }                                                                                                 \
    }

DEFINE_BENCHMARK(float, 2)
DEFINE_BENCHMARK(float, 4)
DEFINE_BENCHMARK(float, 8)
DEFINE_BENCHMARK(float, 16)
DEFINE_BENCHMARK(float, 32)
DEFINE_BENCHMARK(float, 64)
DEFINE_BENCHMARK(float, 128)
DEFINE_BENCHMARK(float, 256)
DEFINE_BENCHMARK(float, 512)
DEFINE_BENCHMARK(float, 1024)
DEFINE_BENCHMARK(double, 2)
DEFINE_BENCHMARK(double, 4)
DEFINE_BENCHMARK(double, 8)
DEFINE_BENCHMARK(double, 16)
DEFINE_BENCHMARK(double, 32)
DEFINE_BENCHMARK(double, 64)
DEFINE_BENCHMARK(double, 128)
DEFINE_BENCHMARK(double, 256)
DEFINE_BENCHMARK(double, 512)
DEFINE_BENCHMARK(double, 1024)

#undef DEFINE_BENCHMARK