File: benchmark_device_binary_search.cpp

package info (click to toggle)
rocprim 5.3.3-4
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 4,656 kB
  • sloc: cpp: 60,198; python: 624; sh: 203; xml: 200; makefile: 109
file content (230 lines) | stat: -rw-r--r-- 7,321 bytes parent folder | download
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
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
// MIT License
//
// Copyright (c) 2019 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in all
// copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
// SOFTWARE.

#include <iostream>
#include <chrono>
#include <vector>
#include <limits>
#include <string>
#include <cstdio>
#include <cstdlib>

// Google Benchmark
#include "benchmark/benchmark.h"
// CmdParser
#include "cmdparser.hpp"
#include "benchmark_utils.hpp"

// HIP API
#include <hip/hip_runtime.h>

// rocPRIM
#include <rocprim/rocprim.hpp>

#define HIP_CHECK(condition)         \
  {                                  \
    hipError_t error = condition;    \
    if(error != hipSuccess){         \
        std::cout << "HIP error: " << error << " line: " << __LINE__ << std::endl; \
        exit(error); \
    } \
  }

#ifndef DEFAULT_N
const size_t DEFAULT_N = 1024 * 1024 * 32;
#endif

const unsigned int batch_size = 10;
const unsigned int warmup_size = 5;

template<class T>
void run_lower_bound_benchmark(benchmark::State& state, hipStream_t stream,
                               size_t haystack_size, size_t needles_size,
                               bool sorted_needles)
{
    using haystack_type = T;
    using needle_type = T;
    using output_type = size_t;
    using compare_op_type = typename std::conditional<std::is_same<needle_type, rocprim::half>::value, half_less, rocprim::less<needle_type>>::type;

    compare_op_type compare_op;
    // Generate data
    std::vector<haystack_type> haystack(haystack_size);
    std::iota(haystack.begin(), haystack.end(), 0);

    std::vector<needle_type> needles = get_random_data<needle_type>(
        needles_size, needle_type(0), needle_type(haystack_size)
    );
    if(sorted_needles)
    {
        std::sort(needles.begin(), needles.end(), compare_op);
    }

    haystack_type * d_haystack;
    needle_type * d_needles;
    output_type * d_output;
    HIP_CHECK(hipMalloc(reinterpret_cast<void**>(&d_haystack), haystack_size * sizeof(haystack_type)));
    HIP_CHECK(hipMalloc(reinterpret_cast<void**>(&d_needles), needles_size * sizeof(needle_type)));
    HIP_CHECK(hipMalloc(reinterpret_cast<void**>(&d_output), needles_size * sizeof(output_type)));
    HIP_CHECK(
        hipMemcpy(
            d_haystack, haystack.data(),
            haystack_size * sizeof(haystack_type),
            hipMemcpyHostToDevice
        )
    );
    HIP_CHECK(
        hipMemcpy(
            d_needles, needles.data(),
            needles_size * sizeof(needle_type),
            hipMemcpyHostToDevice
        )
    );

    void * d_temporary_storage = nullptr;
    size_t temporary_storage_bytes;
    HIP_CHECK(
        rocprim::lower_bound(
            d_temporary_storage, temporary_storage_bytes,
            d_haystack, d_needles, d_output,
            haystack_size, needles_size,
            compare_op,
            stream
        )
    );

    HIP_CHECK(hipMalloc(&d_temporary_storage, temporary_storage_bytes));

    // Warm-up
    for(size_t i = 0; i < warmup_size; i++)
    {
        HIP_CHECK(
            rocprim::lower_bound(
                d_temporary_storage, temporary_storage_bytes,
                d_haystack, d_needles, d_output,
                haystack_size, needles_size,
                compare_op,
                stream
            )
        );
    }
    HIP_CHECK(hipDeviceSynchronize());

    for(auto _ : state)
    {
        auto start = std::chrono::high_resolution_clock::now();

        for(size_t i = 0; i < batch_size; i++)
        {
            HIP_CHECK(
                rocprim::lower_bound(
                    d_temporary_storage, temporary_storage_bytes,
                    d_haystack, d_needles, d_output,
                    haystack_size, needles_size,
                    compare_op,
                    stream
                )
            );
        }
        HIP_CHECK(hipDeviceSynchronize());

        auto end = std::chrono::high_resolution_clock::now();
        auto elapsed_seconds =
            std::chrono::duration_cast<std::chrono::duration<double>>(end - start);
        state.SetIterationTime(elapsed_seconds.count());
    }
    state.SetBytesProcessed(state.iterations() * batch_size * needles_size * sizeof(needle_type));
    state.SetItemsProcessed(state.iterations() * batch_size * needles_size);

    HIP_CHECK(hipFree(d_temporary_storage));
    HIP_CHECK(hipFree(d_haystack));
    HIP_CHECK(hipFree(d_needles));
    HIP_CHECK(hipFree(d_output));
}

#define CREATE_LOWER_BOUND_BENCHMARK(T, K, SORTED) \
benchmark::RegisterBenchmark( \
    ( \
        std::string("lower_bound") + "<" #T ">(" #K "\% " + \
        (SORTED ? "sorted" : "random") + " needles)" \
    ).c_str(), \
    [=](benchmark::State& state) { run_lower_bound_benchmark<T>(state, stream, size, size * K / 100, SORTED); } \
)

#define BENCHMARK_TYPE(type) \
    CREATE_LOWER_BOUND_BENCHMARK(type, 10, false), \
    CREATE_LOWER_BOUND_BENCHMARK(type, 10, true)

int main(int argc, char *argv[])
{
    cli::Parser parser(argc, argv);
    parser.set_optional<size_t>("size", "size", DEFAULT_N, "number of values");
    parser.set_optional<int>("trials", "trials", -1, "number of iterations");
    parser.run_and_exit_if_error();

    // Parse argv
    benchmark::Initialize(&argc, argv);
    const size_t size = parser.get<size_t>("size");
    const int trials = parser.get<int>("trials");

    // HIP
    hipStream_t stream = 0; // default

    // Benchmark info
    add_common_benchmark_info();
    benchmark::AddCustomContext("size", std::to_string(size));

    using custom_float2 = custom_type<float, float>;
    using custom_double2 = custom_type<double, double>;

    // Add benchmarks
    std::vector<benchmark::internal::Benchmark*> benchmarks =
    {
        BENCHMARK_TYPE(float),
        BENCHMARK_TYPE(double),
        BENCHMARK_TYPE(int8_t),
        BENCHMARK_TYPE(uint8_t),
        BENCHMARK_TYPE(rocprim::half),
        BENCHMARK_TYPE(custom_float2),
        BENCHMARK_TYPE(custom_double2)
    };

    // Use manual timing
    for(auto& b : benchmarks)
    {
        b->UseManualTime();
        b->Unit(benchmark::kMillisecond);
    }

    // Force number of iterations
    if(trials > 0)
    {
        for(auto& b : benchmarks)
        {
            b->Iterations(trials);
        }
    }

    // Run benchmarks
    benchmark::RunSpecifiedBenchmarks();
    return 0;
}