• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 <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/transpose.h>
20 
21 
transpose(benchmark::State & state,xnn_transposev_ukernel_function transpose,benchmark::utils::IsaCheckFunction isa_check=nullptr)22 void transpose(
23     benchmark::State& state,
24     xnn_transposev_ukernel_function transpose,
25     benchmark::utils::IsaCheckFunction isa_check = nullptr)
26 {
27   if (isa_check && !isa_check(state)) {
28     return;
29   }
30   const size_t height = state.range(0);
31   const size_t width = state.range(1);
32   const size_t element_size = state.range(2);
33   const size_t tile_hbytes = height * 3;
34   const size_t tile_wbytes = width * 3;
35 
36   std::vector<uint8_t, AlignedAllocator<uint8_t, 64>> x(
37       height * width * element_size + XNN_EXTRA_BYTES / sizeof(uint8_t));
38   std::vector<uint8_t, AlignedAllocator<uint8_t, 64>> y(
39       height * width * element_size + XNN_EXTRA_BYTES / sizeof(uint8_t));
40   std::iota(x.begin(), x.end(), 0);
41   std::fill(y.begin(), y.end(), 0);
42 
43   for (auto _ : state) {
44     transpose(x.data(), y.data(), tile_wbytes, tile_hbytes, element_size,
45               element_size, element_size, width, height);
46   }
47 
48   const uint64_t cpu_frequency = benchmark::utils::GetCurrentCpuFrequency();
49   if (cpu_frequency != 0) {
50     state.counters["cpufreq"] = cpu_frequency;
51   }
52 }
53 
BenchmarkKernelSize(benchmark::internal::Benchmark * b)54 static void BenchmarkKernelSize(benchmark::internal::Benchmark* b)
55 {
56   b->ArgNames({"height", "width", "element_size"});
57   //b->Args({32, 32, 3});
58   //b->Args({64, 64, 3});
59   //b->Args({117, 117, 3});
60   //b->Args({1024, 1024, 3});
61   b->Args({49153, 8, 128});
62 }
63 
64 BENCHMARK_CAPTURE(transpose, 1x1_mmemcpy, xnn_xx_transposev_ukernel__1x1_memcpy)
65     ->Apply(BenchmarkKernelSize)->UseRealTime();
66 
67 #ifndef XNNPACK_BENCHMARK_NO_MAIN
68 BENCHMARK_MAIN();
69 #endif
70