XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 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 | #pragma once |
| 7 | |
| 8 | #include <gtest/gtest.h> |
| 9 | |
| 10 | #include <algorithm> |
| 11 | #include <cassert> |
| 12 | #include <cstddef> |
| 13 | #include <cstdlib> |
| 14 | #include <functional> |
| 15 | #include <random> |
| 16 | #include <vector> |
| 17 | |
Frank Barchard | b196659 | 2020-05-12 13:47:06 -0700 | [diff] [blame] | 18 | #include <fp16.h> |
| 19 | |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 20 | #include <xnnpack.h> |
Marat Dukhan | eeaa7bd | 2019-10-25 17:31:25 -0700 | [diff] [blame] | 21 | #include <xnnpack/params-init.h> |
Frank Barchard | e0601b5 | 2019-10-25 17:43:34 -0700 | [diff] [blame] | 22 | #include <xnnpack/params.h> |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 23 | |
| 24 | |
| 25 | class HSwishMicrokernelTester { |
| 26 | public: |
| 27 | enum class Variant { |
| 28 | Native, |
| 29 | Scalar, |
| 30 | }; |
| 31 | |
Marat Dukhan | 662faa0 | 2019-12-09 22:48:16 -0800 | [diff] [blame] | 32 | inline HSwishMicrokernelTester& batch_size(size_t batch_size) { |
| 33 | assert(batch_size != 0); |
| 34 | this->batch_size_ = batch_size; |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 35 | return *this; |
| 36 | } |
| 37 | |
Marat Dukhan | 662faa0 | 2019-12-09 22:48:16 -0800 | [diff] [blame] | 38 | inline size_t batch_size() const { |
| 39 | return this->batch_size_; |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 40 | } |
| 41 | |
| 42 | inline HSwishMicrokernelTester& inplace(bool inplace) { |
| 43 | this->inplace_ = inplace; |
| 44 | return *this; |
| 45 | } |
| 46 | |
| 47 | inline bool inplace() const { |
| 48 | return this->inplace_; |
| 49 | } |
| 50 | |
| 51 | inline HSwishMicrokernelTester& iterations(size_t iterations) { |
| 52 | this->iterations_ = iterations; |
| 53 | return *this; |
| 54 | } |
| 55 | |
| 56 | inline size_t iterations() const { |
| 57 | return this->iterations_; |
| 58 | } |
| 59 | |
Frank Barchard | b196659 | 2020-05-12 13:47:06 -0700 | [diff] [blame] | 60 | void Test(xnn_f16_hswish_ukernel_function hswish) const { |
| 61 | std::random_device random_device; |
| 62 | auto rng = std::mt19937(random_device()); |
Frank Barchard | aed3278 | 2020-09-30 03:04:35 -0700 | [diff] [blame] | 63 | auto f32rng = std::bind(std::uniform_real_distribution<float>(-4.0f, 4.0f), std::ref(rng)); |
Frank Barchard | b196659 | 2020-05-12 13:47:06 -0700 | [diff] [blame] | 64 | auto f16rng = std::bind(fp16_ieee_from_fp32_value, f32rng); |
| 65 | |
| 66 | std::vector<uint16_t> x(batch_size() + XNN_EXTRA_BYTES / sizeof(uint16_t)); |
| 67 | std::vector<uint16_t> y(batch_size() + (inplace() ? XNN_EXTRA_BYTES / sizeof(uint16_t) : 0)); |
| 68 | std::vector<float> y_ref(batch_size()); |
| 69 | for (size_t iteration = 0; iteration < iterations(); iteration++) { |
| 70 | std::generate(x.begin(), x.end(), std::ref(f16rng)); |
| 71 | if (inplace()) { |
| 72 | std::generate(y.begin(), y.end(), std::ref(f16rng)); |
| 73 | } else { |
| 74 | std::fill(y.begin(), y.end(), UINT16_C(0x7E00) /* NaN */); |
| 75 | } |
| 76 | const uint16_t* x_data = inplace() ? y.data() : x.data(); |
| 77 | |
Frank Barchard | 9f3a843 | 2020-06-02 13:59:35 -0700 | [diff] [blame] | 78 | // Prepare parameters. |
Frank Barchard | b196659 | 2020-05-12 13:47:06 -0700 | [diff] [blame] | 79 | struct xnn_f16_hswish_params params = xnn_init_f16_hswish_params(); |
| 80 | |
| 81 | // Compute reference results. |
| 82 | for (size_t i = 0; i < batch_size(); i++) { |
Marat Dukhan | 201ea0e | 2020-07-10 22:47:19 -0700 | [diff] [blame] | 83 | y_ref[i] = (fp16_ieee_to_fp32_value(x_data[i]) / 6.0f) * std::max(std::min(fp16_ieee_to_fp32_value(x_data[i]) + 3.0f, 6.0f), 0.0f); |
Frank Barchard | b196659 | 2020-05-12 13:47:06 -0700 | [diff] [blame] | 84 | } |
| 85 | |
| 86 | // Call optimized micro-kernel. |
| 87 | hswish(batch_size() * sizeof(uint16_t), x_data, y.data(), ¶ms); |
| 88 | |
| 89 | // Verify results. |
| 90 | for (size_t i = 0; i < batch_size(); i++) { |
Frank Barchard | 7d2c1f2 | 2020-09-14 16:43:53 -0700 | [diff] [blame] | 91 | ASSERT_NEAR(y_ref[i], fp16_ieee_to_fp32_value(y[i]), std::max(1.0e-3f, std::abs(y_ref[i]) * 1.0e-2f)) |
Frank Barchard | b196659 | 2020-05-12 13:47:06 -0700 | [diff] [blame] | 92 | << "at position " << i << ", batch_size = " << batch_size(); |
| 93 | } |
| 94 | } |
| 95 | } |
| 96 | |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 97 | void Test(xnn_f32_hswish_ukernel_function hswish, Variant variant = Variant::Native) const { |
| 98 | std::random_device random_device; |
| 99 | auto rng = std::mt19937(random_device()); |
Frank Barchard | aed3278 | 2020-09-30 03:04:35 -0700 | [diff] [blame] | 100 | auto f32rng = std::bind(std::uniform_real_distribution<float>(-4.0f, 4.0f), rng); |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 101 | |
Marat Dukhan | 662faa0 | 2019-12-09 22:48:16 -0800 | [diff] [blame] | 102 | std::vector<float> x(batch_size() + XNN_EXTRA_BYTES / sizeof(float)); |
| 103 | std::vector<float> y(batch_size() + (inplace() ? XNN_EXTRA_BYTES / sizeof(float) : 0)); |
| 104 | std::vector<float> y_ref(batch_size()); |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 105 | for (size_t iteration = 0; iteration < iterations(); iteration++) { |
| 106 | std::generate(x.begin(), x.end(), std::ref(f32rng)); |
| 107 | if (inplace()) { |
| 108 | std::generate(y.begin(), y.end(), std::ref(f32rng)); |
| 109 | } else { |
| 110 | std::fill(y.begin(), y.end(), std::nanf("")); |
| 111 | } |
| 112 | const float* x_data = inplace() ? y.data() : x.data(); |
| 113 | |
Frank Barchard | 9f3a843 | 2020-06-02 13:59:35 -0700 | [diff] [blame] | 114 | // Prepare parameters. |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 115 | union xnn_f32_hswish_params params = { }; |
| 116 | switch (variant) { |
| 117 | case Variant::Native: |
Marat Dukhan | eeaa7bd | 2019-10-25 17:31:25 -0700 | [diff] [blame] | 118 | params = xnn_init_f32_hswish_params(); |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 119 | break; |
| 120 | case Variant::Scalar: |
Marat Dukhan | eeaa7bd | 2019-10-25 17:31:25 -0700 | [diff] [blame] | 121 | params = xnn_init_scalar_f32_hswish_params(); |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 122 | break; |
| 123 | } |
| 124 | |
| 125 | // Compute reference results. |
Marat Dukhan | 662faa0 | 2019-12-09 22:48:16 -0800 | [diff] [blame] | 126 | for (size_t i = 0; i < batch_size(); i++) { |
Marat Dukhan | 201ea0e | 2020-07-10 22:47:19 -0700 | [diff] [blame] | 127 | y_ref[i] = (x_data[i] / 6.0f) * std::max(std::min(x_data[i] + 3.0f, 6.0f), 0.0f); |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 128 | } |
| 129 | |
| 130 | // Call optimized micro-kernel. |
Marat Dukhan | 662faa0 | 2019-12-09 22:48:16 -0800 | [diff] [blame] | 131 | hswish(batch_size() * sizeof(float), x_data, y.data(), ¶ms); |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 132 | |
| 133 | // Verify results. |
Marat Dukhan | 662faa0 | 2019-12-09 22:48:16 -0800 | [diff] [blame] | 134 | for (size_t i = 0; i < batch_size(); i++) { |
Frank Barchard | aed3278 | 2020-09-30 03:04:35 -0700 | [diff] [blame] | 135 | ASSERT_NEAR(y_ref[i], y[i], std::max(1.0e-7f, std::abs(y_ref[i]) * 1.0e-6f)) |
Marat Dukhan | 662faa0 | 2019-12-09 22:48:16 -0800 | [diff] [blame] | 136 | << "at position " << i << ", batch_size = " << batch_size(); |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 137 | } |
| 138 | } |
| 139 | } |
| 140 | |
| 141 | private: |
Marat Dukhan | 662faa0 | 2019-12-09 22:48:16 -0800 | [diff] [blame] | 142 | size_t batch_size_{1}; |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 143 | bool inplace_{false}; |
Marat Dukhan | 662faa0 | 2019-12-09 22:48:16 -0800 | [diff] [blame] | 144 | size_t iterations_{5}; |
XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame] | 145 | }; |