xref: /aosp_15_r20/external/XNNPACK/bench/f32-rmax.cc (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
1 // Copyright 2019 Google LLC
2 //
3 // This source code is licensed under the BSD-style license found in the
4 // LICENSE file in the root directory of this source tree.
5 
6 #include <algorithm>
7 #include <cmath>
8 #include <functional>
9 #include <random>
10 #include <vector>
11 
12 #include <benchmark/benchmark.h>
13 #include "bench/utils.h"
14 
15 #include <xnnpack.h>
16 #include <xnnpack/aligned-allocator.h>
17 #include <xnnpack/common.h>
18 #include <xnnpack/microfnptr.h>
19 #include <xnnpack/rmax.h>
20 
21 
f32_rmax(benchmark::State & state,xnn_f32_rmax_ukernel_function f32_rmax,benchmark::utils::IsaCheckFunction isa_check=nullptr)22 static void f32_rmax(
23   benchmark::State& state,
24   xnn_f32_rmax_ukernel_function f32_rmax,
25   benchmark::utils::IsaCheckFunction isa_check = nullptr)
26 {
27   if (isa_check && !isa_check(state)) {
28     return;
29   }
30 
31   const size_t elements = state.range(0);
32 
33   std::random_device random_device;
34   auto rng = std::mt19937(random_device());
35   auto f32rng = std::bind(std::uniform_real_distribution<float>(-10.0f, 10.0f), std::ref(rng));
36 
37   std::vector<float, AlignedAllocator<float, 64>> x(elements);
38   std::generate(x.begin(), x.end(), std::ref(f32rng));
39 
40   float y;
41   for (auto _ : state) {
42     f32_rmax(elements * sizeof(float), x.data(), &y);
43   }
44 
45   const uint64_t cpu_frequency = benchmark::utils::GetCurrentCpuFrequency();
46   if (cpu_frequency != 0) {
47     state.counters["cpufreq"] = cpu_frequency;
48   }
49 
50   const size_t elements_per_iteration = elements;
51   state.counters["elements"] =
52     benchmark::Counter(uint64_t(state.iterations()) * elements_per_iteration, benchmark::Counter::kIsRate);
53 
54   const size_t bytes_per_iteration = elements * sizeof(float);
55   state.counters["bytes"] =
56     benchmark::Counter(uint64_t(state.iterations()) * bytes_per_iteration, benchmark::Counter::kIsRate);
57 }
58 
59 #if XNN_ARCH_X86 || XNN_ARCH_X86_64
60   BENCHMARK_CAPTURE(f32_rmax, sse, xnn_f32_rmax_ukernel__sse)
61     ->RangeMultiplier(10)
62     ->Range(1000, 100000000)
63     ->UseRealTime();
64 
65   BENCHMARK_CAPTURE(f32_rmax, avx, xnn_f32_rmax_ukernel__avx, benchmark::utils::CheckAVX)
66     ->RangeMultiplier(10)
67     ->Range(1000, 100000000)
68     ->UseRealTime();
69 
70   BENCHMARK_CAPTURE(f32_rmax, avx512f, xnn_f32_rmax_ukernel__avx512f, benchmark::utils::CheckAVX512F)
71     ->RangeMultiplier(10)
72     ->Range(1000, 100000000)
73     ->UseRealTime();
74 #endif  // XNN_ARCH_X86 || XNN_ARCH_X86_64
75 
76 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
77   BENCHMARK_CAPTURE(f32_rmax, neon, xnn_f32_rmax_ukernel__neon, benchmark::utils::CheckNEON)
78     ->RangeMultiplier(10)
79     ->Range(1000, 100000000)
80     ->UseRealTime();
81 #endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
82 
83 #if XNN_ARCH_WASMSIMD || XNN_ARCH_WASMRELAXEDSIMD
84   BENCHMARK_CAPTURE(f32_rmax, wasmsimd_arm, xnn_f32_rmax_ukernel__wasmsimd_arm)
85     ->RangeMultiplier(10)
86     ->Range(1000, 100000000)
87     ->UseRealTime();
88 
89   BENCHMARK_CAPTURE(f32_rmax, wasmsimd_x86, xnn_f32_rmax_ukernel__wasmsimd_x86)
90     ->RangeMultiplier(10)
91     ->Range(1000, 100000000)
92     ->UseRealTime();
93 #endif  // XNN_ARCH_WASMSIMD || XNN_ARCH_WASMRELAXEDSIMD
94 
95 BENCHMARK_CAPTURE(f32_rmax, scalar, xnn_f32_rmax_ukernel__scalar)
96   ->RangeMultiplier(10)
97   ->Range(1000, 100000000)
98   ->UseRealTime();
99 
100 #ifndef XNNPACK_BENCHMARK_NO_MAIN
101 BENCHMARK_MAIN();
102 #endif
103