File size: 7,785 Bytes
8b7c501 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 |
// Copyright 2020 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.
#pragma once
#include <gtest/gtest.h>
#include <algorithm>
#include <cassert>
#include <cmath>
#include <cstddef>
#include <cstdlib>
#include <memory>
#include <random>
#include <vector>
#include <fp16/fp16.h>
#include <xnnpack.h>
class AbsOperatorTester {
public:
inline AbsOperatorTester& channels(size_t channels) {
assert(channels != 0);
this->channels_ = channels;
return *this;
}
inline size_t channels() const {
return this->channels_;
}
inline AbsOperatorTester& input_stride(size_t input_stride) {
assert(input_stride != 0);
this->input_stride_ = input_stride;
return *this;
}
inline size_t input_stride() const {
if (this->input_stride_ == 0) {
return this->channels_;
} else {
assert(this->input_stride_ >= this->channels_);
return this->input_stride_;
}
}
inline AbsOperatorTester& output_stride(size_t output_stride) {
assert(output_stride != 0);
this->output_stride_ = output_stride;
return *this;
}
inline size_t output_stride() const {
if (this->output_stride_ == 0) {
return this->channels_;
} else {
assert(this->output_stride_ >= this->channels_);
return this->output_stride_;
}
}
inline AbsOperatorTester& batch_size(size_t batch_size) {
assert(batch_size != 0);
this->batch_size_ = batch_size;
return *this;
}
inline size_t batch_size() const {
return this->batch_size_;
}
inline AbsOperatorTester& iterations(size_t iterations) {
this->iterations_ = iterations;
return *this;
}
inline size_t iterations() const {
return this->iterations_;
}
void TestF16() const {
std::random_device random_device;
auto rng = std::mt19937(random_device());
std::uniform_real_distribution<float> f32dist(-1.0f, 1.0f);
std::vector<uint16_t> input(XNN_EXTRA_BYTES / sizeof(uint16_t) +
(batch_size() - 1) * input_stride() + channels());
std::vector<uint16_t> output((batch_size() - 1) * output_stride() + channels());
std::vector<uint16_t> output_ref(batch_size() * channels());
for (size_t iteration = 0; iteration < iterations(); iteration++) {
std::generate(input.begin(), input.end(), [&]() { return fp16_ieee_from_fp32_value(f32dist(rng)); });
std::fill(output.begin(), output.end(), UINT16_C(0x7E00) /* NaN */);
// Compute reference results.
for (size_t i = 0; i < batch_size(); i++) {
for (size_t c = 0; c < channels(); c++) {
output_ref[i * channels() + c] = input[i * input_stride() + c] & UINT16_C(0x7FFF);
}
}
// Create, setup, run, and destroy Abs operator.
ASSERT_EQ(xnn_status_success, xnn_initialize(nullptr /* allocator */));
xnn_operator_t abs_op = nullptr;
const xnn_status status = xnn_create_abs_nc_f16(
channels(), input_stride(), output_stride(),
0, &abs_op);
if (status == xnn_status_unsupported_hardware) {
GTEST_SKIP();
}
ASSERT_EQ(xnn_status_success, status);
ASSERT_NE(nullptr, abs_op);
// Smart pointer to automatically delete abs_op.
std::unique_ptr<xnn_operator, decltype(&xnn_delete_operator)> auto_abs_op(abs_op, xnn_delete_operator);
ASSERT_EQ(xnn_status_success, xnn_reshape_abs_nc_f16(abs_op, batch_size(), /*threadpool=*/nullptr));
ASSERT_EQ(xnn_status_success, xnn_setup_abs_nc_f16(abs_op, input.data(), output.data()));
ASSERT_EQ(xnn_status_success, xnn_run_operator(abs_op, /*threadpool=*/nullptr));
// Verify results.
for (size_t i = 0; i < batch_size(); i++) {
for (size_t c = 0; c < channels(); c++) {
EXPECT_EQ(output_ref[i * channels() + c], output[i * output_stride() + c])
<< "at batch " << i << " / " << batch_size() << ", channel " << c << " / " << channels();
}
}
}
}
void TestF32() const {
std::random_device random_device;
auto rng = std::mt19937(random_device());
std::uniform_real_distribution<float> f32dist(-1.0f, 1.0f);
std::vector<float> input(XNN_EXTRA_BYTES / sizeof(float) +
(batch_size() - 1) * input_stride() + channels());
std::vector<float> output((batch_size() - 1) * output_stride() + channels());
std::vector<float> output_ref(batch_size() * channels());
for (size_t iteration = 0; iteration < iterations(); iteration++) {
std::generate(input.begin(), input.end(), [&]() { return f32dist(rng); });
std::fill(output.begin(), output.end(), std::nanf(""));
// Compute reference results.
for (size_t i = 0; i < batch_size(); i++) {
for (size_t c = 0; c < channels(); c++) {
output_ref[i * channels() + c] = std::fabs(input[i * input_stride() + c]);
}
}
// Create, setup, run, and destroy Abs operator.
ASSERT_EQ(xnn_status_success, xnn_initialize(nullptr /* allocator */));
xnn_operator_t abs_op = nullptr;
ASSERT_EQ(xnn_status_success,
xnn_create_abs_nc_f32(
channels(), input_stride(), output_stride(),
0, &abs_op));
ASSERT_NE(nullptr, abs_op);
// Smart pointer to automatically delete abs_op.
std::unique_ptr<xnn_operator, decltype(&xnn_delete_operator)> auto_abs_op(abs_op, xnn_delete_operator);
ASSERT_EQ(xnn_status_success, xnn_reshape_abs_nc_f32(abs_op, batch_size(), /*threadpool=*/nullptr));
ASSERT_EQ(xnn_status_success, xnn_setup_abs_nc_f32(abs_op, input.data(), output.data()));
ASSERT_EQ(xnn_status_success, xnn_run_operator(abs_op, /*threadpool=*/nullptr));
// Verify results.
for (size_t i = 0; i < batch_size(); i++) {
for (size_t c = 0; c < channels(); c++) {
EXPECT_EQ(output_ref[i * channels() + c], output[i * output_stride() + c])
<< "at batch " << i << " / " << batch_size() << ", channel " << c << " / " << channels();
}
}
}
}
void TestRunF32() const {
std::random_device random_device;
auto rng = std::mt19937(random_device());
std::uniform_real_distribution<float> f32dist(-1.0f, 1.0f);
std::vector<float> input(XNN_EXTRA_BYTES / sizeof(float) +
(batch_size() - 1) * input_stride() + channels());
std::vector<float> output((batch_size() - 1) * output_stride() + channels());
std::vector<float> output_ref(batch_size() * channels());
for (size_t iteration = 0; iteration < iterations(); iteration++) {
std::generate(input.begin(), input.end(), [&]() { return f32dist(rng); });
std::fill(output.begin(), output.end(), std::nanf(""));
// Compute reference results.
for (size_t i = 0; i < batch_size(); i++) {
for (size_t c = 0; c < channels(); c++) {
output_ref[i * channels() + c] = std::fabs(input[i * input_stride() + c]);
}
}
ASSERT_EQ(xnn_status_success, xnn_initialize(nullptr /* allocator */));
ASSERT_EQ(xnn_status_success,
xnn_run_abs_nc_f32(
channels(),
input_stride(),
output_stride(),
batch_size(),
input.data(),
output.data(),
0,
nullptr /* thread pool */));
// Verify results.
for (size_t i = 0; i < batch_size(); i++) {
for (size_t c = 0; c < channels(); c++) {
EXPECT_EQ(output_ref[i * channels() + c], output[i * output_stride() + c])
<< "at batch " << i << " / " << batch_size() << ", channel " << c << " / " << channels();
}
}
}
}
private:
size_t batch_size_{1};
size_t channels_{1};
size_t input_stride_{0};
size_t output_stride_{0};
size_t iterations_{15};
};
|