Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 1 | /* |
Michele Di Giorgio | d9eaf61 | 2020-07-08 11:12:57 +0100 | [diff] [blame] | 2 | * Copyright (c) 2019-2020 Arm Limited. |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [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 | */ |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 24 | #include "arm_compute/core/KernelDescriptors.h" |
| 25 | #include "arm_compute/core/Types.h" |
| 26 | #include "arm_compute/core/utils/misc/ShapeCalculator.h" |
| 27 | #include "arm_compute/runtime/CL/CLTensor.h" |
| 28 | #include "arm_compute/runtime/CL/CLTensorAllocator.h" |
Sang-Hoon Park | bef7fa2 | 2020-10-21 15:58:54 +0100 | [diff] [blame] | 29 | #include "src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h" |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 30 | #include "tests/CL/CLAccessor.h" |
| 31 | #include "tests/CL/Helper.h" |
| 32 | #include "tests/PaddingCalculator.h" |
| 33 | #include "tests/datasets/ShapeDatasets.h" |
| 34 | #include "tests/framework/Asserts.h" |
| 35 | #include "tests/framework/Macros.h" |
| 36 | #include "tests/framework/datasets/Datasets.h" |
| 37 | #include "tests/validation/Validation.h" |
| 38 | #include "tests/validation/fixtures/GEMMFixture.h" |
| 39 | |
| 40 | namespace arm_compute |
| 41 | { |
| 42 | namespace test |
| 43 | { |
| 44 | namespace validation |
| 45 | { |
| 46 | using namespace arm_compute::misc::shape_calculator; |
| 47 | |
| 48 | // Create function for CLGEMMMatrixMultiplyKernel |
| 49 | using CLGEMMMatrixMultiplyNative = CLSynthetizeFunction<CLGEMMMatrixMultiplyKernel>; |
| 50 | |
| 51 | // Fixture for GEMMMatrixMultiplyValidationFixture |
| 52 | template <typename T> |
| 53 | using CLGEMMMatrixMultiplyNativeFixture = GEMMMatrixMultiplyValidationFixture<CLTensor, CLAccessor, T, CLGEMMMatrixMultiplyNative>; |
| 54 | |
| 55 | // Fixture for GEMMMatrixMultiply3DValidationFixture |
| 56 | template <typename T> |
| 57 | using CLGEMMMatrixMultiplyNative3DFixture = GEMMMatrixMultiply3DValidationFixture<CLTensor, CLAccessor, T, CLGEMMMatrixMultiplyNative>; |
| 58 | |
| 59 | namespace |
| 60 | { |
| 61 | // *INDENT-OFF* |
| 62 | // clang-format off |
| 63 | RelativeTolerance<float> rel_tolerance_f32(0.001f); |
| 64 | constexpr float abs_tolerance_f32(0.0001f); |
| 65 | |
| 66 | RelativeTolerance<half> rel_tolerance_f16(half(0.2)); |
| 67 | constexpr float tolerance_num_f16 = 0.02f; |
| 68 | |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 69 | /** Alpha values to test */ |
Gian Marco Iodice | f3622be | 2019-07-29 14:27:16 +0100 | [diff] [blame] | 70 | const auto alpha_values = framework::dataset::make("alpha", {1.0f, -0.75f} ); |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 71 | |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 72 | /** Beta values to test */ |
Gian Marco Iodice | d820db6 | 2019-08-05 14:23:23 +0100 | [diff] [blame] | 73 | const auto beta_values = framework::dataset::make("beta", {-0.35f, 0.0f} ); |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 74 | |
SiCong Li | 7e5b7bf | 2020-11-17 09:41:13 +0000 | [diff] [blame] | 75 | /** M, N combinations to test |
| 76 | * 1: Special 1x1 case |
| 77 | * 2: Special multples of processor size in both dimensions |
| 78 | * 3: Non multiples of processor size in both dimensions |
| 79 | * 4: Special 1x1003 case |
| 80 | */ |
| 81 | const auto m_n_values = zip( |
| 82 | framework::dataset::make("M", {1, 16, 37, 1}), |
| 83 | framework::dataset::make("N", {1, 16, 51, 1003}) |
| 84 | ); |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 85 | |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 86 | /** N values to test */ |
| 87 | const auto n_values = framework::dataset::make("N", {51, 1003}); |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 88 | |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 89 | /** K values to test */ |
| 90 | const auto k_values = framework::dataset::make("K", 23); |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 91 | |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 92 | /** M_W values to test */ |
| 93 | const auto m_w_values = framework::dataset::make("M_W", 5); |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 94 | |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 95 | /** M_H values to test */ |
| 96 | const auto m_h_values = framework::dataset::make("M_H", 7); |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 97 | |
| 98 | /** Batch size values to test */ |
| 99 | const auto b_values = framework::dataset::make("batch_size", 1, 3); |
| 100 | |
| 101 | /** Activation values to test */ |
| 102 | const auto act_values = framework::dataset::make("Activation", |
| 103 | { |
| 104 | ActivationLayerInfo(), |
| 105 | ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 8.f, 2.f), |
| 106 | }); |
| 107 | |
| 108 | /** Broadcast bias from vector to matrix */ |
Gian Marco Iodice | d820db6 | 2019-08-05 14:23:23 +0100 | [diff] [blame] | 109 | const auto broadcast_bias_values = framework::dataset::make("broadcast_bias", { false, true } ); |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 110 | |
| 111 | /** GPU architectures values to test */ |
| 112 | const auto gpu_arch_values = framework::dataset::make("GPUArch", |
| 113 | { |
| 114 | GPUTarget::MIDGARD, |
| 115 | GPUTarget::BIFROST |
| 116 | }); |
| 117 | |
| 118 | /** Data types values to test in the configuration */ |
| 119 | const auto data_type_values = framework::dataset::make("DataType", |
| 120 | { |
| 121 | DataType::F32, |
| 122 | DataType::F16 |
| 123 | }); |
| 124 | |
| 125 | /** M values to test */ |
| 126 | const auto fp16_mixed_precision_values = framework::dataset::make("fp16_mixed_precision", {true, false}); |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 127 | } // namespace |
| 128 | |
| 129 | TEST_SUITE(CL) |
| 130 | TEST_SUITE(GEMMMatrixMultiply) |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 131 | TEST_CASE(Negative, framework::DatasetMode::ALL) |
| 132 | { |
| 133 | // Unsupported QASYMM8 data type |
| 134 | { |
| 135 | const auto lhs = TensorInfo(TensorShape(13U, 12U, 1U, 1U), 1, DataType::QASYMM8); |
| 136 | const auto rhs = TensorInfo(TensorShape(14U, 13U, 1U, 1U), 1, DataType::QASYMM8); |
| 137 | const auto out = TensorInfo(TensorShape(14U, 12U, 1U, 1U), 1, DataType::QASYMM8); |
| 138 | constexpr float alpha = 1.3f; |
| 139 | constexpr float beta = 0.7f; |
| 140 | const bool is_interleaved_transposed = false; |
| 141 | const GEMMReshapeInfo reshape_info = GEMMReshapeInfo(12, 14, 13, 1, 1, 0, false, false); |
| 142 | const GPUTarget gpu_target = GPUTarget::MIDGARD; |
| 143 | const auto status = CLGEMMMatrixMultiplyKernel::validate(&lhs, &rhs, nullptr, &out, alpha, beta, is_interleaved_transposed, reshape_info, gpu_target); |
| 144 | ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS); |
| 145 | } |
| 146 | |
| 147 | // Unsupported SIZE_T data type |
| 148 | { |
| 149 | const auto lhs = TensorInfo(TensorShape(13U, 12U, 1U, 1U), 1, DataType::SIZET); |
| 150 | const auto rhs = TensorInfo(TensorShape(14U, 13U, 1U, 1U), 1, DataType::SIZET); |
| 151 | const auto out = TensorInfo(TensorShape(14U, 12U, 1U, 1U), 1, DataType::SIZET); |
| 152 | constexpr float alpha = 1.3f; |
| 153 | constexpr float beta = 0.7f; |
| 154 | const bool is_interleaved_transposed = false; |
| 155 | const GEMMReshapeInfo reshape_info = GEMMReshapeInfo(12, 14, 13, 1, 1, 0, false, false); |
| 156 | const GPUTarget gpu_target = GPUTarget::MIDGARD; |
| 157 | const auto status = CLGEMMMatrixMultiplyKernel::validate(&lhs, &rhs, nullptr, &out, alpha, beta, is_interleaved_transposed, reshape_info, gpu_target); |
| 158 | ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS); |
| 159 | } |
| 160 | |
| 161 | // Mixed precision with F32 |
| 162 | { |
| 163 | const auto lhs = TensorInfo(TensorShape(13U, 12U, 1U, 1U), 1, DataType::F32); |
| 164 | const auto rhs = TensorInfo(TensorShape(14U, 13U, 1U, 1U), 1, DataType::F32); |
| 165 | const auto out = TensorInfo(TensorShape(14U, 12U, 1U, 1U), 1, DataType::F32); |
| 166 | constexpr float alpha = 1.3f; |
| 167 | constexpr float beta = 0.7f; |
| 168 | const bool is_interleaved_transposed = false; |
| 169 | const GEMMReshapeInfo reshape_info = GEMMReshapeInfo(12, 14, 13, 1, 1, 0, false, false); |
| 170 | const GPUTarget gpu_target = GPUTarget::MIDGARD; |
| 171 | const bool fp_mixed_precision = true; |
| 172 | const auto status = CLGEMMMatrixMultiplyKernel::validate(&lhs, &rhs, nullptr, &out, alpha, beta, is_interleaved_transposed, reshape_info, gpu_target, fp_mixed_precision); |
| 173 | ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS); |
| 174 | } |
| 175 | |
| 176 | // Max number of dimensions LHS matrix |
| 177 | { |
| 178 | const auto lhs = TensorInfo(TensorShape(13U, 12U, 1U, 1U, 4U), 1, DataType::F32); |
| 179 | const auto rhs = TensorInfo(TensorShape(14U, 13U, 1U, 1U), 1, DataType::F32); |
| 180 | const auto out = TensorInfo(TensorShape(14U, 12U, 1U, 1U), 1, DataType::F32); |
| 181 | constexpr float alpha = 1.3f; |
| 182 | constexpr float beta = 0.7f; |
| 183 | const bool is_interleaved_transposed = false; |
| 184 | const GEMMReshapeInfo reshape_info = GEMMReshapeInfo(12, 14, 13, 1, 1, 0, false, false); |
| 185 | const GPUTarget gpu_target = GPUTarget::MIDGARD; |
| 186 | const auto status = CLGEMMMatrixMultiplyKernel::validate(&lhs, &rhs, nullptr, &out, alpha, beta, is_interleaved_transposed, reshape_info, gpu_target); |
| 187 | ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS); |
| 188 | } |
| 189 | |
| 190 | // Max number of dimensions RHS matrix |
| 191 | { |
| 192 | const auto lhs = TensorInfo(TensorShape(13U, 12U, 1U, 4U), 1, DataType::F32); |
| 193 | const auto rhs = TensorInfo(TensorShape(14U, 13U, 1U, 4U), 1, DataType::F32); |
| 194 | const auto out = TensorInfo(TensorShape(14U, 12U, 1U, 4U), 1, DataType::F32); |
| 195 | constexpr float alpha = 1.3f; |
| 196 | constexpr float beta = 0.7f; |
| 197 | const bool is_interleaved_transposed = false; |
| 198 | const GEMMReshapeInfo reshape_info = GEMMReshapeInfo(12, 14, 13, 1, 1, 0, false, false); |
| 199 | const GPUTarget gpu_target = GPUTarget::MIDGARD; |
| 200 | const auto status = CLGEMMMatrixMultiplyKernel::validate(&lhs, &rhs, nullptr, &out, alpha, beta, is_interleaved_transposed, reshape_info, gpu_target); |
| 201 | ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS); |
| 202 | } |
| 203 | |
| 204 | // Broadcast bias |
| 205 | { |
| 206 | const auto lhs = TensorInfo(TensorShape(13U, 12U, 1U, 1U), 1, DataType::F16); |
| 207 | const auto rhs = TensorInfo(TensorShape(14U, 13U, 1U, 1U), 1, DataType::F16); |
| 208 | // The correct shape should be bias = TensorInfo(TensorShape(14U, 1U, 1U, 1U), 1, DataType::F32); |
| 209 | const auto bias = TensorInfo(TensorShape(14U, 12U, 1U, 1U), 1, DataType::F16); |
| 210 | const auto out = TensorInfo(TensorShape(14U, 12U, 1U, 1U), 1, DataType::F16); |
| 211 | constexpr float alpha = 1.3f; |
| 212 | constexpr float beta = 0.7f; |
| 213 | const bool is_interleaved_transposed = false; |
| 214 | const GEMMReshapeInfo reshape_info = GEMMReshapeInfo(12, 14, 13, 1, 1, 0, false, true); |
| 215 | const GPUTarget gpu_target = GPUTarget::MIDGARD; |
| 216 | const bool fp_mixed_precision = false; |
| 217 | const auto status = CLGEMMMatrixMultiplyKernel::validate(&lhs, &rhs, &bias, &out, alpha, beta, is_interleaved_transposed, reshape_info, gpu_target, fp_mixed_precision); |
| 218 | ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS); |
| 219 | } |
| 220 | |
| 221 | // Invalid dimensions for the bias |
| 222 | { |
| 223 | const auto lhs = TensorInfo(TensorShape(13U, 12U, 1U, 1U), 1, DataType::F32); |
| 224 | const auto rhs = TensorInfo(TensorShape(14U, 13U, 1U, 1U), 1, DataType::F32); |
| 225 | // The correct shape should be bias = TensorInfo(TensorShape(14U, 12U, 1U, 1U), 1, DataType::F32); |
| 226 | const auto bias = TensorInfo(TensorShape(14U, 8U, 1U, 1U), 1, DataType::F32); |
| 227 | const auto out = TensorInfo(TensorShape(14U, 12U, 1U, 1U), 1, DataType::F32); |
| 228 | constexpr float alpha = 1.3f; |
| 229 | constexpr float beta = 0.7f; |
| 230 | const bool is_interleaved_transposed = false; |
| 231 | const GEMMReshapeInfo reshape_info = GEMMReshapeInfo(12, 14, 13, 1, 1, 0, false, false); |
| 232 | const GPUTarget gpu_target = GPUTarget::MIDGARD; |
| 233 | const bool fp_mixed_precision = false; |
| 234 | const auto status = CLGEMMMatrixMultiplyKernel::validate(&lhs, &rhs, &bias, &out, alpha, beta, is_interleaved_transposed, reshape_info, gpu_target, fp_mixed_precision); |
| 235 | ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS); |
| 236 | } |
| 237 | |
| 238 | // Invalid dimensions for the output |
| 239 | { |
| 240 | const auto lhs = TensorInfo(TensorShape(13U, 12U, 1U, 1U), 1, DataType::F32); |
| 241 | const auto rhs = TensorInfo(TensorShape(14U, 13U, 1U, 1U), 1, DataType::F32); |
| 242 | // The correct shape should be out = TensorInfo(TensorShape(14U, 12U, 1U, 1U), 1, DataType::F32); |
| 243 | const auto out = TensorInfo(TensorShape(14U, 7U, 1U, 1U), 1, DataType::F32); |
| 244 | constexpr float alpha = 1.3f; |
| 245 | constexpr float beta = 0.7f; |
| 246 | const bool is_interleaved_transposed = false; |
| 247 | const GEMMReshapeInfo reshape_info = GEMMReshapeInfo(12, 14, 13, 1, 1, 0, false, false); |
| 248 | const GPUTarget gpu_target = GPUTarget::MIDGARD; |
| 249 | const auto status = CLGEMMMatrixMultiplyKernel::validate(&lhs, &rhs, nullptr, &out, alpha, beta, is_interleaved_transposed, reshape_info, gpu_target); |
| 250 | ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS); |
| 251 | } |
| 252 | } |
| 253 | |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 254 | TEST_SUITE(Float) |
| 255 | TEST_SUITE(FP32) |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 256 | |
| 257 | FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyNativeFixture<float>, framework::DatasetMode::ALL, |
SiCong Li | 7e5b7bf | 2020-11-17 09:41:13 +0000 | [diff] [blame] | 258 | combine(combine(combine(combine(combine(combine(combine(combine(combine( |
| 259 | m_n_values, |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 260 | k_values), |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 261 | b_values), |
| 262 | alpha_values), |
| 263 | beta_values), |
| 264 | broadcast_bias_values), |
| 265 | framework::dataset::make("fp16_mixed_precision", false)), |
| 266 | act_values), |
| 267 | framework::dataset::make("DataType", DataType::F32)), |
| 268 | gpu_arch_values)) |
| 269 | { |
| 270 | // Validate output |
| 271 | validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); |
| 272 | } |
| 273 | |
| 274 | FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyNative3DFixture<float>, framework::DatasetMode::ALL, |
| 275 | combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 276 | m_w_values, |
| 277 | m_h_values), |
| 278 | n_values), |
| 279 | k_values), |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 280 | b_values), |
| 281 | alpha_values), |
| 282 | beta_values), |
| 283 | broadcast_bias_values), |
| 284 | framework::dataset::make("fp16_mixed_precision", false)), |
| 285 | act_values), |
| 286 | framework::dataset::make("DataType", DataType::F32)), |
| 287 | gpu_arch_values)) |
| 288 | { |
| 289 | // Validate output |
| 290 | validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); |
| 291 | } |
| 292 | |
| 293 | TEST_SUITE_END() // FP32 |
| 294 | |
| 295 | TEST_SUITE(FP16) |
| 296 | FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyNativeFixture<half>, framework::DatasetMode::ALL, |
SiCong Li | 7e5b7bf | 2020-11-17 09:41:13 +0000 | [diff] [blame] | 297 | combine(combine(combine(combine(combine(combine(combine(combine(combine( |
| 298 | m_n_values, |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 299 | k_values), |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 300 | b_values), |
| 301 | alpha_values), |
| 302 | beta_values), |
| 303 | broadcast_bias_values), |
| 304 | fp16_mixed_precision_values), |
| 305 | act_values), |
| 306 | framework::dataset::make("DataType", DataType::F16)), |
| 307 | gpu_arch_values)) |
| 308 | { |
| 309 | // Validate output |
| 310 | validate(CLAccessor(_target), _reference, rel_tolerance_f16, tolerance_num_f16); |
| 311 | } |
| 312 | |
| 313 | FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyNative3DFixture<half>, framework::DatasetMode::ALL, |
| 314 | combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( |
Gian Marco Iodice | c630e94 | 2020-05-11 12:15:54 +0100 | [diff] [blame] | 315 | m_w_values, |
| 316 | m_h_values), |
| 317 | n_values), |
| 318 | k_values), |
Gian Marco Iodice | d1f5476 | 2019-07-19 09:54:47 +0100 | [diff] [blame] | 319 | b_values), |
| 320 | alpha_values), |
| 321 | beta_values), |
| 322 | broadcast_bias_values), |
| 323 | fp16_mixed_precision_values), |
| 324 | act_values), |
| 325 | framework::dataset::make("DataType", DataType::F16)), |
| 326 | gpu_arch_values)) |
| 327 | { |
| 328 | // Validate output |
| 329 | validate(CLAccessor(_target), _reference, rel_tolerance_f16, tolerance_num_f16); |
| 330 | } |
| 331 | |
| 332 | TEST_SUITE_END() // FP16 |
| 333 | TEST_SUITE_END() // Float |
| 334 | TEST_SUITE_END() // GEMMMatrixMuliplty |
| 335 | TEST_SUITE_END() // CL |
| 336 | } // namespace validation |
| 337 | } // namespace test |
| 338 | } // namespace arm_compute |