xref: /aosp_15_r20/external/XNNPACK/bench/s16-vlshift.cc (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
1 // Copyright 2022 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 <numeric>
10 #include <vector>
11 
12 #include "bench/utils.h"
13 #include <benchmark/benchmark.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/vlshift.h>
20 
21 
vlshift(benchmark::State & state,xnn_s16_vlshift_ukernel_function vlshift,benchmark::utils::IsaCheckFunction isa_check=nullptr)22 void vlshift(
23     benchmark::State& state,
24     xnn_s16_vlshift_ukernel_function vlshift,
25     benchmark::utils::IsaCheckFunction isa_check = nullptr)
26 {
27   if (isa_check && !isa_check(state)) {
28     return;
29   }
30   const size_t batch = state.range(0);
31 
32   std::vector<int16_t, AlignedAllocator<int16_t, 64>> input(
33       batch + XNN_EXTRA_BYTES / sizeof(int16_t));
34   std::vector<int16_t, AlignedAllocator<int16_t, 64>> output(batch);
35   std::iota(input.begin(), input.end(), 0);
36   std::iota(output.begin(), output.end(), 1);
37 
38   for (auto _ : state) {
39     vlshift(batch, input.data(), output.data(), 4 /* shift */);
40   }
41 
42   const uint64_t cpu_frequency = benchmark::utils::GetCurrentCpuFrequency();
43   if (cpu_frequency != 0) {
44     state.counters["cpufreq"] = cpu_frequency;
45   }
46 }
47 
BenchmarkKernelSize(benchmark::internal::Benchmark * b)48 static void BenchmarkKernelSize(benchmark::internal::Benchmark* b)
49 {
50   b->ArgNames({"batch"});
51   b->Args({32});
52   b->Args({64});
53   b->Args({117});
54   b->Args({400});
55   b->Args({1000});
56   b->Args({10000});
57 }
58 
59 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
60 BENCHMARK_CAPTURE(vlshift, s16_neon_x8, xnn_s16_vlshift_ukernel__neon_x8)
61     ->Apply(BenchmarkKernelSize)->UseRealTime();
62 BENCHMARK_CAPTURE(vlshift, s16_neon_x16, xnn_s16_vlshift_ukernel__neon_x16)
63     ->Apply(BenchmarkKernelSize)->UseRealTime();
64 BENCHMARK_CAPTURE(vlshift, s16_neon_x24, xnn_s16_vlshift_ukernel__neon_x24)
65     ->Apply(BenchmarkKernelSize)->UseRealTime();
66 BENCHMARK_CAPTURE(vlshift, s16_neon_x32, xnn_s16_vlshift_ukernel__neon_x32)
67     ->Apply(BenchmarkKernelSize)->UseRealTime();
68 #endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
69 
70 BENCHMARK_CAPTURE(vlshift, s16_scalar_x1, xnn_s16_vlshift_ukernel__scalar_x1)
71     ->Apply(BenchmarkKernelSize)->UseRealTime();
72 BENCHMARK_CAPTURE(vlshift, s16_scalar_x2, xnn_s16_vlshift_ukernel__scalar_x2)
73     ->Apply(BenchmarkKernelSize)->UseRealTime();
74 BENCHMARK_CAPTURE(vlshift, s16_scalar_x3, xnn_s16_vlshift_ukernel__scalar_x3)
75     ->Apply(BenchmarkKernelSize)->UseRealTime();
76 BENCHMARK_CAPTURE(vlshift, s16_scalar_x4, xnn_s16_vlshift_ukernel__scalar_x4)
77     ->Apply(BenchmarkKernelSize)->UseRealTime();
78 
79 #ifndef XNNPACK_BENCHMARK_NO_MAIN
80 BENCHMARK_MAIN();
81 #endif
82