Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 1 | /* |
Sheri Zhang | ac6499a | 2021-02-10 15:32:38 +0000 | [diff] [blame] | 2 | * Copyright (c) 2020-2021 Arm Limited. |
Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 3 | * |
| 4 | * SPDX-License-Identifier: MIT |
| 5 | * |
| 6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 7 | * of this software and associated documentation files (the "Software"), to |
| 8 | * deal in the Software without restriction, including without limitation the |
| 9 | * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| 10 | * sell copies of the Software, and to permit persons to whom the Software is |
| 11 | * furnished to do so, subject to the following conditions: |
| 12 | * |
| 13 | * The above copyright notice and this permission notice shall be included in all |
| 14 | * copies or substantial portions of the Software. |
| 15 | * |
| 16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 19 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 22 | * SOFTWARE. |
| 23 | */ |
Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 24 | #include "arm_compute/core/Types.h" |
| 25 | #include "arm_compute/runtime/Tensor.h" |
| 26 | #include "arm_compute/runtime/TensorAllocator.h" |
Michalis Spyrou | ebcebf1 | 2020-10-21 00:04:14 +0100 | [diff] [blame] | 27 | #include "src/core/NEON/kernels/NEQLSTMLayerNormalizationKernel.h" |
Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 28 | #include "tests/NEON/Accessor.h" |
Sheri Zhang | 45198c8 | 2020-04-14 22:29:36 +0100 | [diff] [blame] | 29 | #include "tests/NEON/Helper.h" |
Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 30 | #include "tests/PaddingCalculator.h" |
| 31 | #include "tests/datasets/ShapeDatasets.h" |
| 32 | #include "tests/framework/Asserts.h" |
| 33 | #include "tests/framework/Macros.h" |
| 34 | #include "tests/framework/datasets/Datasets.h" |
| 35 | #include "tests/validation/Helpers.h" |
| 36 | #include "tests/validation/Validation.h" |
| 37 | #include "tests/validation/fixtures/QLSTMLayerNormalizationFixture.h" |
| 38 | |
| 39 | namespace arm_compute |
| 40 | { |
| 41 | namespace test |
| 42 | { |
| 43 | namespace validation |
| 44 | { |
| 45 | namespace |
| 46 | { |
| 47 | constexpr uint32_t vector_size_byte = 16; |
| 48 | |
| 49 | using test::datasets::ShapeDataset; |
Sheri Zhang | 45198c8 | 2020-04-14 22:29:36 +0100 | [diff] [blame] | 50 | using NEQLSTMLayerNormalization = NESynthetizeFunction<NEQLSTMLayerNormalizationKernel>; |
| 51 | |
Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 52 | template <uint32_t num_elements_per_iter, uint32_t num_batches, uint32_t num_iteration> |
| 53 | class QLSTMLayerNormShapeDataSet : public ShapeDataset |
| 54 | { |
| 55 | static constexpr auto boundary_minus_one = num_elements_per_iter * num_iteration - 1; |
| 56 | static constexpr auto boundary = num_elements_per_iter * num_iteration; |
| 57 | static constexpr auto boundary_plus_one = num_elements_per_iter * num_iteration + 1; |
| 58 | |
| 59 | public: |
| 60 | QLSTMLayerNormShapeDataSet(std::string name) |
| 61 | : ShapeDataset(name, |
| 62 | { |
| 63 | TensorShape{ boundary_minus_one, num_batches }, |
| 64 | TensorShape{ boundary, num_batches }, |
| 65 | TensorShape{ boundary_plus_one, num_batches } |
| 66 | }) |
| 67 | { |
| 68 | } |
| 69 | }; |
| 70 | |
| 71 | template <uint32_t num_elements_per_iter, uint32_t num_batches> |
| 72 | class QLSTMLayerNormShapeDataSet<num_elements_per_iter, num_batches, 0> : public ShapeDataset |
| 73 | { |
| 74 | public: |
| 75 | QLSTMLayerNormShapeDataSet(std::string name) |
| 76 | : ShapeDataset(name, |
| 77 | { |
| 78 | TensorShape{ 1, num_batches }, |
| 79 | TensorShape{ 2, num_batches } |
| 80 | }) |
| 81 | { |
| 82 | } |
| 83 | }; |
| 84 | } // namespace |
| 85 | TEST_SUITE(NEON) |
| 86 | TEST_SUITE(QLSTMLayerNormalization) |
| 87 | |
| 88 | static const TensorShape correct_input_shape{ TensorShape(15U, 2U) }; |
| 89 | static const TensorShape correct_weight_shape{ TensorShape(15U) }; |
| 90 | static const TensorShape correct_bias_shape{ TensorShape(15U) }; |
| 91 | static const TensorShape correct_output_shape{ correct_input_shape }; |
| 92 | static const DataType correct_input_dt{ DataType::QSYMM16 }; |
| 93 | static const DataType correct_weight_dt{ DataType::QSYMM16 }; |
| 94 | static const DataType correct_bias_dt{ DataType::S32 }; |
| 95 | static const DataType correct_output_dt{ correct_input_dt }; |
| 96 | static const uint32_t tensor_num_channel{ 1 }; |
| 97 | |
| 98 | // *INDENT-OFF* |
| 99 | // clang-format off |
| 100 | |
| 101 | DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, |
| 102 | zip(zip(zip( |
| 103 | framework::dataset::make("InputInfo", { |
| 104 | TensorInfo(correct_input_shape, tensor_num_channel, DataType::F16), // input supports only QSYMM16 |
| 105 | TensorInfo(correct_input_shape, tensor_num_channel, correct_input_dt), // weight supports only QSYMM16 |
| 106 | TensorInfo(correct_input_shape, tensor_num_channel, correct_input_dt), // bias supports only S32 |
| 107 | TensorInfo(TensorShape(15U, 2U, 2U), tensor_num_channel, correct_input_dt), // input supports only up to 2D |
| 108 | TensorInfo(correct_input_shape, tensor_num_channel, correct_input_dt), // weight supports only up to 1D |
| 109 | TensorInfo(correct_input_shape, tensor_num_channel, correct_input_dt), // bias supports only up to 1D |
| 110 | TensorInfo(correct_input_shape, tensor_num_channel, correct_input_dt), // input_shape[0] != weight_shape[0] should fail |
| 111 | TensorInfo(correct_input_shape, tensor_num_channel, correct_input_dt), // weight_shape[0] != bias_shape[0] should fail |
| 112 | TensorInfo(correct_input_shape, tensor_num_channel, correct_input_dt), // output shape mismatches with input shape |
| 113 | TensorInfo(correct_input_shape, tensor_num_channel, correct_input_dt), // output data type mismatches with input data type |
| 114 | }), |
| 115 | framework::dataset::make("WeightInfo", { |
| 116 | TensorInfo(correct_weight_shape, tensor_num_channel, correct_weight_dt), |
| 117 | TensorInfo(correct_weight_shape, tensor_num_channel, DataType::F16), |
| 118 | TensorInfo(correct_weight_shape, tensor_num_channel, correct_weight_dt), |
| 119 | TensorInfo(correct_weight_shape, tensor_num_channel, correct_weight_dt), |
| 120 | TensorInfo(TensorShape(15U, 2U), tensor_num_channel, correct_weight_dt), |
| 121 | TensorInfo(correct_weight_shape, tensor_num_channel, correct_weight_dt), |
| 122 | TensorInfo(TensorShape(14U), tensor_num_channel, correct_weight_dt), |
| 123 | TensorInfo(correct_weight_shape, tensor_num_channel, correct_weight_dt), |
| 124 | TensorInfo(correct_weight_shape, tensor_num_channel, correct_weight_dt), |
| 125 | TensorInfo(correct_weight_shape, tensor_num_channel, correct_weight_dt), |
| 126 | }) |
| 127 | ), |
| 128 | framework::dataset::make("BiasInfo", { |
| 129 | TensorInfo(correct_bias_shape, tensor_num_channel, correct_bias_dt), |
| 130 | TensorInfo(correct_bias_shape, tensor_num_channel, correct_bias_dt), |
| 131 | TensorInfo(correct_bias_shape, tensor_num_channel, DataType::QSYMM16), |
| 132 | TensorInfo(correct_bias_shape, tensor_num_channel, correct_bias_dt), |
| 133 | TensorInfo(correct_bias_shape, tensor_num_channel, correct_bias_dt), |
| 134 | TensorInfo(TensorShape(15U, 2U), tensor_num_channel, correct_bias_dt), |
| 135 | TensorInfo(correct_bias_shape, tensor_num_channel, correct_bias_dt), |
| 136 | TensorInfo(TensorShape(14U), tensor_num_channel, correct_bias_dt), |
| 137 | TensorInfo(correct_bias_shape, tensor_num_channel, correct_bias_dt), |
| 138 | TensorInfo(correct_bias_shape, tensor_num_channel, correct_bias_dt), |
| 139 | }) |
| 140 | ), |
| 141 | framework::dataset::make("OutputInfo", { |
| 142 | TensorInfo(correct_output_shape, tensor_num_channel, correct_output_dt), |
| 143 | TensorInfo(correct_output_shape, tensor_num_channel, correct_output_dt), |
| 144 | TensorInfo(correct_output_shape, tensor_num_channel, correct_output_dt), |
| 145 | TensorInfo(correct_output_shape, tensor_num_channel, correct_output_dt), |
| 146 | TensorInfo(correct_output_shape, tensor_num_channel, correct_output_dt), |
| 147 | TensorInfo(correct_output_shape, tensor_num_channel, correct_output_dt), |
| 148 | TensorInfo(correct_output_shape, tensor_num_channel, correct_output_dt), |
| 149 | TensorInfo(correct_output_shape, tensor_num_channel, correct_output_dt), |
| 150 | TensorInfo(TensorShape(15, 3), tensor_num_channel, correct_output_dt), |
| 151 | TensorInfo(correct_output_shape, tensor_num_channel, DataType::S32), |
| 152 | }) |
| 153 | ), |
| 154 | input_info, weight_info, bias_info, output_info) |
| 155 | { |
Sheri Zhang | 45198c8 | 2020-04-14 22:29:36 +0100 | [diff] [blame] | 156 | const Status s = NEQLSTMLayerNormalization::validate(&input_info, &output_info, &weight_info, &bias_info); |
Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 157 | ARM_COMPUTE_EXPECT(!bool(s), framework::LogLevel::ERRORS); |
| 158 | } |
| 159 | |
| 160 | // clang-format on |
| 161 | // *INDENT-ON* |
| 162 | |
| 163 | template <typename T> |
Sheri Zhang | 45198c8 | 2020-04-14 22:29:36 +0100 | [diff] [blame] | 164 | using NEQLSTMLayerNormalizationFixture = QLSTMLayerNormalizationValidationFixture<Tensor, Accessor, NEQLSTMLayerNormalization, T>; |
Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 165 | |
| 166 | TEST_SUITE(Quantized) |
| 167 | TEST_SUITE(QSYMM16) |
| 168 | |
| 169 | /** Tests will be targetting |
Michele Di Giorgio | 33f41fa | 2021-03-09 14:09:08 +0000 | [diff] [blame] | 170 | * - Comparison between optimized kernel and the exact same but scalar version of reference kernel |
Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 171 | * - Input shapes of 1D and 2D with the first dimension covers boundary values of 128-bit vector size (0~3 iterations) |
| 172 | * - Weight and bias 1D shape that have same size as that of input shapes |
| 173 | * - Quantization scale is greater and smaller than one. |
| 174 | * - Input values will be noted in fixture. |
| 175 | * |
| 176 | * What we can't test |
| 177 | * - Since reference kernel uses the exact the same algorithm in the same quantized domain |
| 178 | * it is hard to fully test whether the algorithm accomplishes what it is supposed to. |
| 179 | * - The algorithm has been sensitive to quantization scale but it is hard to fully test |
| 180 | * the sensitivity due to aforementioned reason. |
| 181 | * - Again, it is hard to fully test corner values due to the exact same algorithm of the |
Michele Di Giorgio | 33f41fa | 2021-03-09 14:09:08 +0000 | [diff] [blame] | 182 | * reference kernel and the optimized kernel. |
Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 183 | */ |
| 184 | |
| 185 | constexpr uint32_t qsymm16_per_vector = vector_size_byte / sizeof(int16_t); |
| 186 | |
| 187 | #define QSYMM16_DATASET_ITER(num_input_batch, num_iter) \ |
| 188 | combine(combine(zip(zip(QLSTMLayerNormShapeDataSet<qsymm16_per_vector, num_input_batch, num_iter>("InputShape"), \ |
| 189 | QLSTMLayerNormShapeDataSet<qsymm16_per_vector, 1, num_iter>("WeightShape")), \ |
| 190 | QLSTMLayerNormShapeDataSet<qsymm16_per_vector, 1, num_iter>("BiasShape")), \ |
| 191 | framework::dataset::make("DataType", DataType::QSYMM16)), \ |
| 192 | framework::dataset::make("WeightQuantizationInfo", { QuantizationInfo(1. / 8192), QuantizationInfo(8192) })) |
| 193 | |
| 194 | #define QSYMM16_DATASET_1D \ |
| 195 | concat(concat(QSYMM16_DATASET_ITER(1, 0), QSYMM16_DATASET_ITER(1, 1)), QSYMM16_DATASET_ITER(1, 2)) |
| 196 | |
| 197 | #define QSYMM16_DATASET_2D \ |
| 198 | concat(concat(QSYMM16_DATASET_ITER(3, 0), QSYMM16_DATASET_ITER(3, 1)), QSYMM16_DATASET_ITER(3, 2)) |
| 199 | |
| 200 | FIXTURE_DATA_TEST_CASE(RandomValue1D, NEQLSTMLayerNormalizationFixture<int16_t>, framework::DatasetMode::ALL, QSYMM16_DATASET_1D) |
| 201 | { |
| 202 | // Validate output |
| 203 | validate(Accessor(_target), _reference); |
| 204 | } |
| 205 | |
| 206 | FIXTURE_DATA_TEST_CASE(RandomValue2D, NEQLSTMLayerNormalizationFixture<int16_t>, framework::DatasetMode::ALL, QSYMM16_DATASET_2D) |
| 207 | { |
| 208 | // Validate output |
| 209 | validate(Accessor(_target), _reference); |
| 210 | } |
| 211 | |
| 212 | #undef QSYMM16_DATASET_ITER |
| 213 | #undef QSYMM16_DATASET_2D |
| 214 | #undef QSYMM16_DATASET_1D |
| 215 | |
| 216 | TEST_SUITE_END() // QSYMM16 |
| 217 | TEST_SUITE_END() // Quantized |
| 218 | TEST_SUITE_END() // QLSTMLayerNormalization |
Sheri Zhang | ac6499a | 2021-02-10 15:32:38 +0000 | [diff] [blame] | 219 | TEST_SUITE_END() // Neon |
Sang-Hoon Park | 0d008f7 | 2020-03-13 14:56:05 +0000 | [diff] [blame] | 220 | |
| 221 | } // namespace validation |
| 222 | } // namespace test |
| 223 | } // namespace arm_compute |