forked from ROCm/rocPRIM
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbenchmark_device_binary_search.cpp
280 lines (243 loc) · 11.4 KB
/
benchmark_device_binary_search.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
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
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
// MIT License
//
// Copyright (c) 2019-2024 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 "benchmark_device_binary_search.parallel.hpp"
#include "benchmark_utils.hpp"
// CmdParser
#include "cmdparser.hpp"
// Google Benchmark
#include <benchmark/benchmark.h>
// HIP API
#include <hip/hip_runtime.h>
// rocPRIM
#include <rocprim/device/config_types.hpp>
#include <iostream>
#include <limits>
#include <string>
#include <vector>
#include <cstdio>
#include <cstdlib>
#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, class AlgorithmSelectorTag>
void run_benchmark(benchmark::State& state,
size_t haystack_size,
const managed_seed& seed,
hipStream_t stream,
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);
const auto random_range = limit_random_range<needle_type>(0, haystack_size);
std::vector<needle_type> needles = get_random_data<needle_type>(needles_size,
random_range.first,
random_range.second,
seed.get_0());
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;
auto dispatch_helper = dispatch_binary_search_helper<rocprim::default_config>();
HIP_CHECK(dispatch_helper.dispatch_binary_search(AlgorithmSelectorTag{},
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(dispatch_helper.dispatch_binary_search(AlgorithmSelectorTag{},
d_temporary_storage,
temporary_storage_bytes,
d_haystack,
d_needles,
d_output,
haystack_size,
needles_size,
compare_op,
stream));
}
HIP_CHECK(hipDeviceSynchronize());
// HIP events creation
hipEvent_t start, stop;
HIP_CHECK(hipEventCreate(&start));
HIP_CHECK(hipEventCreate(&stop));
for(auto _ : state)
{
// Record start event
HIP_CHECK(hipEventRecord(start, stream));
for(size_t i = 0; i < batch_size; i++)
{
HIP_CHECK(dispatch_helper.dispatch_binary_search(AlgorithmSelectorTag{},
d_temporary_storage,
temporary_storage_bytes,
d_haystack,
d_needles,
d_output,
haystack_size,
needles_size,
compare_op,
stream));
}
// Record stop event and wait until it completes
HIP_CHECK(hipEventRecord(stop, stream));
HIP_CHECK(hipEventSynchronize(stop));
float elapsed_mseconds;
HIP_CHECK(hipEventElapsedTime(&elapsed_mseconds, start, stop));
state.SetIterationTime(elapsed_mseconds / 1000);
}
// Destroy HIP events
HIP_CHECK(hipEventDestroy(start));
HIP_CHECK(hipEventDestroy(stop));
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_BENCHMARK(T, K, SORTED, ALGO_TAG) \
benchmark::RegisterBenchmark( \
bench_naming::format_name( \
"{lvl:device,algo:" + ALGO_TAG{}.name() + ",key_type:" #T ",subalgo:" #K "_percent_" \
+ std::string(SORTED ? "sorted" : "random") + "_needles,cfg:default_config}") \
.c_str(), \
[=](benchmark::State& state) \
{ run_benchmark<T, ALGO_TAG>(state, size, seed, stream, size * K / 100, SORTED); })
#define BENCHMARK_ALGORITHMS(T, K, SORTED) \
CREATE_BENCHMARK(T, K, SORTED, binary_search_subalgorithm), \
CREATE_BENCHMARK(T, K, SORTED, lower_bound_subalgorithm), \
CREATE_BENCHMARK(T, K, SORTED, upper_bound_subalgorithm)
#define BENCHMARK_TYPE(type) \
BENCHMARK_ALGORITHMS(type, 10, true), BENCHMARK_ALGORITHMS(type, 10, false)
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.set_optional<std::string>("name_format",
"name_format",
"human",
"either: json,human,txt");
parser.set_optional<std::string>("seed", "seed", "random", get_seed_message());
#ifdef BENCHMARK_CONFIG_TUNING
// optionally run an evenly split subset of benchmarks, when making multiple program invocations
parser.set_optional<int>("parallel_instance",
"parallel_instance",
0,
"parallel instance index");
parser.set_optional<int>("parallel_instances",
"parallel_instances",
1,
"total parallel instances");
#endif
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");
bench_naming::set_format(parser.get<std::string>("name_format"));
const std::string seed_type = parser.get<std::string>("seed");
const managed_seed seed(seed_type);
// HIP
hipStream_t stream = 0; // default
// Benchmark info
add_common_benchmark_info();
benchmark::AddCustomContext("size", std::to_string(size));
benchmark::AddCustomContext("seed", seed_type);
using custom_float2 = custom_type<float, float>;
using custom_double2 = custom_type<double, double>;
// Add benchmarks
std::vector<benchmark::internal::Benchmark*> benchmarks;
#ifdef BENCHMARK_CONFIG_TUNING
const int parallel_instance = parser.get<int>("parallel_instance");
const int parallel_instances = parser.get<int>("parallel_instances");
config_autotune_register::register_benchmark_subset(benchmarks,
parallel_instance,
parallel_instances,
size,
seed,
stream);
#else // BENCHMARK_CONFIG_TUNING
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)};
#endif // BENCHMARK_CONFIG_TUNING
// 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;
}