v8 147.3.0

Rust bindings to V8
Documentation 
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

#include <benchmark/benchmark.h>

#include <fp16.h>

#include <vector>
#include <random>
#include <chrono>
#include <functional>
#include <algorithm>

#ifdef FP16_COMPARATIVE_BENCHMARKS
	#include <third-party/THHalf.h>
	#include <third-party/npy-halffloat.h>
	#include <third-party/eigen-half.h>
	#include <third-party/float16-compressor.h>
	#include <third-party/half.hpp>
#endif


static void fp16_ieee_to_fp32_bits(benchmark::State& state) {
	const uint_fast32_t seed = std::chrono::system_clock::now().time_since_epoch().count();
	auto rng = std::bind(std::uniform_real_distribution<float>(-1.0f, 1.0f), std::mt19937(seed));

	std::vector<uint16_t> fp16(state.range(0));
	std::vector<uint32_t> fp32(state.range(0));
	std::generate(fp16.begin(), fp16.end(),
		[&rng]{ return fp16_ieee_from_fp32_value(rng()); });

	while (state.KeepRunning()) {
		uint16_t* input = fp16.data();
		benchmark::DoNotOptimize(input);

		uint32_t* output = fp32.data();
		const size_t n = state.range(0);
		for (size_t i = 0; i < n; i++) {
			output[i] = fp16_ieee_to_fp32_bits(input[i]);
		}

		benchmark::DoNotOptimize(output);
	}
	state.SetItemsProcessed(int64_t(state.iterations()) * int64_t(state.range(0)));
}
BENCHMARK(fp16_ieee_to_fp32_bits)->RangeMultiplier(2)->Range(1<<10, 64<<20);

static void fp16_ieee_to_fp32_value(benchmark::State& state) {
	const uint_fast32_t seed = std::chrono::system_clock::now().time_since_epoch().count();
	auto rng = std::bind(std::uniform_real_distribution<float>(-1.0f, 1.0f), std::mt19937(seed));

	std::vector<uint16_t> fp16(state.range(0));
	std::vector<float> fp32(state.range(0));
	std::generate(fp16.begin(), fp16.end(),
		[&rng]{ return fp16_ieee_from_fp32_value(rng()); });

	while (state.KeepRunning()) {
		uint16_t* input = fp16.data();
		benchmark::DoNotOptimize(input);

		float* output = fp32.data();
		const size_t n = state.range(0);
		for (size_t i = 0; i < n; i++) {
			output[i] = fp16_ieee_to_fp32_value(input[i]);
		}

		benchmark::DoNotOptimize(output);
	}
	state.SetItemsProcessed(int64_t(state.iterations()) * int64_t(state.range(0)));
}
BENCHMARK(fp16_ieee_to_fp32_value)->RangeMultiplier(2)->Range(1<<10, 64<<20);

#ifdef FP16_COMPARATIVE_BENCHMARKS
	static void TH_halfbits2float(benchmark::State& state) {
		const uint_fast32_t seed = std::chrono::system_clock::now().time_since_epoch().count();
		auto rng = std::bind(std::uniform_real_distribution<float>(-1.0f, 1.0f), std::mt19937(seed));

		std::vector<uint16_t> fp16(state.range(0));
		std::vector<float> fp32(state.range(0));
		std::generate(fp16.begin(), fp16.end(),
			[&rng]{ return fp16_ieee_from_fp32_value(rng()); });

		while (state.KeepRunning()) {
			uint16_t* input = fp16.data();
			benchmark::DoNotOptimize(input);

			float* output = fp32.data();
			const size_t n = state.range(0);
			for (size_t i = 0; i < n; i++) {
				TH_halfbits2float(&input[i], &output[i]);
			}

			benchmark::DoNotOptimize(output);
		}
		state.SetItemsProcessed(int64_t(state.iterations()) * int64_t(state.range(0)));
	}
	BENCHMARK(TH_halfbits2float)->RangeMultiplier(2)->Range(1<<10, 64<<20);

	static void npy_halfbits_to_floatbits(benchmark::State& state) {
		const uint_fast32_t seed = std::chrono::system_clock::now().time_since_epoch().count();
		auto rng = std::bind(std::uniform_real_distribution<float>(-1.0f, 1.0f), std::mt19937(seed));

		std::vector<uint16_t> fp16(state.range(0));
		std::vector<uint32_t> fp32(state.range(0));
		std::generate(fp16.begin(), fp16.end(),
			[&rng]{ return fp16_ieee_from_fp32_value(rng()); });

		while (state.KeepRunning()) {
			uint16_t* input = fp16.data();
			benchmark::DoNotOptimize(input);

			uint32_t* output = fp32.data();
			const size_t n = state.range(0);
			for (size_t i = 0; i < n; i++) {
				output[i] = npy_halfbits_to_floatbits(input[i]);
			}

			benchmark::DoNotOptimize(output);
		}
		state.SetItemsProcessed(int64_t(state.iterations()) * int64_t(state.range(0)));
	}
	BENCHMARK(npy_halfbits_to_floatbits)->RangeMultiplier(2)->Range(1<<10, 64<<20);

	static void Eigen_half_to_float(benchmark::State& state) {
		const uint_fast32_t seed = std::chrono::system_clock::now().time_since_epoch().count();
		auto rng = std::bind(std::uniform_real_distribution<float>(-1.0f, 1.0f), std::mt19937(seed));

		std::vector<uint16_t> fp16(state.range(0));
		std::vector<float> fp32(state.range(0));
		std::generate(fp16.begin(), fp16.end(),
			[&rng]{ return fp16_ieee_from_fp32_value(rng()); });

		while (state.KeepRunning()) {
			uint16_t* input = fp16.data();
			benchmark::DoNotOptimize(input);

			float* output = fp32.data();
			const size_t n = state.range(0);
			for (size_t i = 0; i < n; i++) {
				output[i] =
					Eigen::half_impl::half_to_float(
						Eigen::half_impl::raw_uint16_to_half(input[i]));
			}

			benchmark::DoNotOptimize(output);
		}
		state.SetItemsProcessed(int64_t(state.iterations()) * int64_t(state.range(0)));
	}
	BENCHMARK(Eigen_half_to_float)->RangeMultiplier(2)->Range(1<<10, 64<<20);

	static void Float16Compressor_decompress(benchmark::State& state) {
		const uint_fast32_t seed = std::chrono::system_clock::now().time_since_epoch().count();
		auto rng = std::bind(std::uniform_real_distribution<float>(-1.0f, 1.0f), std::mt19937(seed));

		std::vector<uint16_t> fp16(state.range(0));
		std::vector<float> fp32(state.range(0));
		std::generate(fp16.begin(), fp16.end(),
			[&rng]{ return fp16_ieee_from_fp32_value(rng()); });

		while (state.KeepRunning()) {
			uint16_t* input = fp16.data();
			benchmark::DoNotOptimize(input);

			float* output = fp32.data();
			const size_t n = state.range(0);
			for (size_t i = 0; i < n; i++) {
				output[i] = Float16Compressor::decompress(input[i]);
			}

			benchmark::DoNotOptimize(output);
		}
		state.SetItemsProcessed(int64_t(state.iterations()) * int64_t(state.range(0)));
	}
	BENCHMARK(Float16Compressor_decompress)->RangeMultiplier(2)->Range(1<<10, 64<<20);

	static void half_float_detail_half2float_table(benchmark::State& state) {
		const uint_fast32_t seed = std::chrono::system_clock::now().time_since_epoch().count();
		auto rng = std::bind(std::uniform_real_distribution<float>(-1.0f, 1.0f), std::mt19937(seed));

		std::vector<uint16_t> fp16(state.range(0));
		std::vector<float> fp32(state.range(0));
		std::generate(fp16.begin(), fp16.end(),
			[&rng]{ return fp16_ieee_from_fp32_value(rng()); });

		while (state.KeepRunning()) {
			uint16_t* input = fp16.data();
			benchmark::DoNotOptimize(input);

			float* output = fp32.data();
			const size_t n = state.range(0);
			for (size_t i = 0; i < n; i++) {
				output[i] = half_float::detail::half2float_impl(input[i],
					half_float::detail::true_type());
			}

			benchmark::DoNotOptimize(output);
		}
		state.SetItemsProcessed(int64_t(state.iterations()) * int64_t(state.range(0)));
	}
	BENCHMARK(half_float_detail_half2float_table)->RangeMultiplier(2)->Range(1<<10, 64<<20);

	static void half_float_detail_half2float_branch(benchmark::State& state) {
		const uint_fast32_t seed = std::chrono::system_clock::now().time_since_epoch().count();
		auto rng = std::bind(std::uniform_real_distribution<float>(-1.0f, 1.0f), std::mt19937(seed));

		std::vector<uint16_t> fp16(state.range(0));
		std::vector<float> fp32(state.range(0));
		std::generate(fp16.begin(), fp16.end(),
			[&rng]{ return fp16_ieee_from_fp32_value(rng()); });

		while (state.KeepRunning()) {
			uint16_t* input = fp16.data();
			benchmark::DoNotOptimize(input);

			float* output = fp32.data();
			const size_t n = state.range(0);
			for (size_t i = 0; i < n; i++) {
				output[i] = half_float::detail::half2float_impl(input[i],
					half_float::detail::false_type());
			}

			benchmark::DoNotOptimize(output);
		}
		state.SetItemsProcessed(int64_t(state.iterations()) * int64_t(state.range(0)));
	}
	BENCHMARK(half_float_detail_half2float_branch)->RangeMultiplier(2)->Range(1<<10, 64<<20);
#endif

BENCHMARK_MAIN();