diff options
Diffstat (limited to 'tests/validation')
-rw-r--r-- | tests/validation/dynamic_fusion/gpu/cl/Cast.cpp | 235 | ||||
-rw-r--r-- | tests/validation/fixtures/dynamic_fusion/operators/CastFixture.h | 175 |
2 files changed, 410 insertions, 0 deletions
diff --git a/tests/validation/dynamic_fusion/gpu/cl/Cast.cpp b/tests/validation/dynamic_fusion/gpu/cl/Cast.cpp new file mode 100644 index 0000000000..cb6c8c52f6 --- /dev/null +++ b/tests/validation/dynamic_fusion/gpu/cl/Cast.cpp @@ -0,0 +1,235 @@ +/* + * Copyright (c) 2022 Arm Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include "arm_compute/core/Types.h" +#include "arm_compute/dynamic_fusion/sketch/gpu/operators/GpuCast.h" +#include "arm_compute/runtime/CL/CLTensor.h" + +#include "tests/CL/CLAccessor.h" +#include "tests/datasets/ConvertPolicyDataset.h" +#include "tests/datasets/ShapeDatasets.h" +#include "tests/framework/Macros.h" +#include "tests/framework/datasets/Datasets.h" +#include "tests/validation/Validation.h" +#include "tests/validation/fixtures/dynamic_fusion/operators/CastFixture.h" + +namespace arm_compute +{ +namespace test +{ +namespace validation +{ +namespace +{ +// Tolerance +constexpr AbsoluteTolerance<float> one_tolerance(1); +constexpr AbsoluteTolerance<float> zero_tolerance(0); + +/** Input data sets **/ +// QASYMM8 +const auto CastQASYMM8toF32Dataset = combine(framework::dataset::make("DataType", DataType::QASYMM8), framework::dataset::make("DataType", DataType::F32)); + +// U8 +const auto CastU8toS8Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::S8)); +const auto CastU8toU16Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::U16)); +const auto CastU8toS16Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::S16)); +const auto CastU8toU32Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::U32)); +const auto CastU8toS32Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::S32)); +const auto CastU8toF16Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::F16)); +const auto CastU8toF32Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::F32)); + +// S8 +const auto CastS8toU8Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::U8)); +const auto CastS8toU16Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::U16)); +const auto CastS8toS16Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::S16)); +const auto CastS8toU32Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::U32)); +const auto CastS8toS32Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::S32)); +const auto CastS8toF16Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::F16)); +const auto CastS8toF32Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::F32)); + +// U16 +const auto CastU16toU8Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::U8)); +const auto CastU16toS8Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::S8)); +const auto CastU16toS16Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::S16)); +const auto CastU16toU32Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::U32)); +const auto CastU16toS32Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::S32)); +const auto CastU16toF16Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::F16)); +const auto CastU16toF32Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::F32)); + +// S16 +const auto CastS16toU8Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::U8)); +const auto CastS16toS8Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::S8)); +const auto CastS16toU16Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::U16)); +const auto CastS16toU32Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::U32)); +const auto CastS16toS32Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::S32)); +const auto CastS16toF16Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::F16)); +const auto CastS16toF32Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::F32)); + +// U32 +const auto CastU32toU8Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::U8)); +const auto CastU32toS8Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::S8)); +const auto CastU32toU16Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::U16)); +const auto CastU32toS16Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::S16)); +const auto CastU32toS32Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::S32)); +const auto CastU32toF16Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::F16)); +const auto CastU32toF32Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::F32)); + +// S32 +const auto CastS32toU8Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::U8)); +const auto CastS32toS8Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::S8)); +const auto CastS32toU16Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::U16)); +const auto CastS32toS16Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::S16)); +const auto CastS32toU32Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::U32)); +const auto CastS32toF16Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::F16)); +const auto CastS32toF32Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::F32)); + +// F16 +const auto CastF16toU8Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::U8)); +const auto CastF16toS8Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::S8)); +const auto CastF16toU16Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::U16)); +const auto CastF16toS16Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::S16)); +const auto CastF16toU32Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::U32)); +const auto CastF16toS32Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::S32)); +const auto CastF16toF32Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::F32)); + +// F32 +const auto CastF32toU8Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::U8)); +const auto CastF32toS8Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::S8)); +const auto CastF32toU16Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::U16)); +const auto CastF32toS16Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::S16)); +const auto CastF32toU32Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::U32)); +const auto CastF32toS32Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::S32)); +const auto CastF32toF16Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::F16)); +} // namespace + +TEST_SUITE(CL) +TEST_SUITE(DYNAMIC_FUSION) +TEST_SUITE(CAST) + +template <typename T> +using DynamicFusionCLCastToU8Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, uint8_t>; +template <typename T> +using DynamicFusionCLCastToS8Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, int8_t>; +template <typename T> +using DynamicFusionCLCastToU16Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, uint16_t>; +template <typename T> +using DynamicFusionCLCastToS16Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, int16_t>; +template <typename T> +using DynamicFusionCLCastToU32Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, uint32_t>; +template <typename T> +using DynamicFusionCLCastToS32Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, int32_t>; +template <typename T> +using DynamicFusionCLCastToF16Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, half>; +template <typename T> +using DynamicFusionCLCastToF32Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, float>; + +#define CAST_SUITE(NAME, idt, odt, type, dataset, tolerance) \ + TEST_SUITE(NAME) \ + FIXTURE_DATA_TEST_CASE(RunSmall, type, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(), dataset), \ + datasets::ConvertPolicies())) \ + { \ + validate(CLAccessor(_target), _reference, tolerance); \ + } \ + TEST_SUITE_END() + +// QASYMM8 +CAST_SUITE(QASYMM8_to_F32, DataType::QASYMM8, DataType::F32, DynamicFusionCLCastToF32Fixture<uint8_t>, CastQASYMM8toF32Dataset, zero_tolerance) + +// U8 +CAST_SUITE(U8_to_S8, DataType::U8, DataType::S8, DynamicFusionCLCastToS8Fixture<uint8_t>, CastU8toS8Dataset, zero_tolerance) +CAST_SUITE(U8_to_U16, DataType::U8, DataType::U16, DynamicFusionCLCastToU16Fixture<uint8_t>, CastU8toU16Dataset, zero_tolerance) +CAST_SUITE(U8_to_S16, DataType::U8, DataType::S16, DynamicFusionCLCastToS16Fixture<uint8_t>, CastU8toS16Dataset, zero_tolerance) +CAST_SUITE(U8_to_U32, DataType::U8, DataType::U32, DynamicFusionCLCastToU32Fixture<uint8_t>, CastU8toU32Dataset, zero_tolerance) +CAST_SUITE(U8_to_S32, DataType::U8, DataType::S32, DynamicFusionCLCastToS32Fixture<uint8_t>, CastU8toS32Dataset, zero_tolerance) +CAST_SUITE(U8_to_F16, DataType::U8, DataType::F16, DynamicFusionCLCastToF16Fixture<uint8_t>, CastU8toF16Dataset, zero_tolerance) +CAST_SUITE(U8_to_F32, DataType::U8, DataType::F32, DynamicFusionCLCastToF32Fixture<uint8_t>, CastU8toF32Dataset, zero_tolerance) + +// S8 +CAST_SUITE(S8_to_U8, DataType::S8, DataType::U8, DynamicFusionCLCastToU8Fixture<int8_t>, CastS8toU8Dataset, zero_tolerance) +CAST_SUITE(S8_to_U16, DataType::S8, DataType::U16, DynamicFusionCLCastToU16Fixture<int8_t>, CastS8toU16Dataset, zero_tolerance) +CAST_SUITE(S8_to_S16, DataType::S8, DataType::S16, DynamicFusionCLCastToS16Fixture<int8_t>, CastS8toS16Dataset, zero_tolerance) +CAST_SUITE(S8_to_U32, DataType::S8, DataType::U32, DynamicFusionCLCastToU32Fixture<int8_t>, CastS8toU32Dataset, zero_tolerance) +CAST_SUITE(S8_to_S32, DataType::S8, DataType::S32, DynamicFusionCLCastToS32Fixture<int8_t>, CastS8toS32Dataset, zero_tolerance) +CAST_SUITE(S8_to_F16, DataType::S8, DataType::F16, DynamicFusionCLCastToF16Fixture<int8_t>, CastS8toF16Dataset, zero_tolerance) +CAST_SUITE(S8_to_F32, DataType::S8, DataType::F32, DynamicFusionCLCastToF32Fixture<int8_t>, CastS8toF32Dataset, zero_tolerance) + +// U16 +CAST_SUITE(U16_to_U8, DataType::U16, DataType::U8, DynamicFusionCLCastToU8Fixture<uint16_t>, CastU16toU8Dataset, zero_tolerance) +CAST_SUITE(U16_to_S8, DataType::U16, DataType::S8, DynamicFusionCLCastToS8Fixture<uint16_t>, CastU16toS8Dataset, zero_tolerance) +CAST_SUITE(U16_to_S16, DataType::U16, DataType::S16, DynamicFusionCLCastToS16Fixture<uint16_t>, CastU16toS16Dataset, zero_tolerance) +CAST_SUITE(U16_to_U32, DataType::U16, DataType::U32, DynamicFusionCLCastToU32Fixture<uint16_t>, CastU16toU32Dataset, zero_tolerance) +CAST_SUITE(U16_to_S32, DataType::U16, DataType::S32, DynamicFusionCLCastToS32Fixture<uint16_t>, CastU16toS32Dataset, zero_tolerance) +CAST_SUITE(U16_to_F16, DataType::U16, DataType::F16, DynamicFusionCLCastToF16Fixture<uint16_t>, CastU16toF16Dataset, zero_tolerance) +CAST_SUITE(U16_to_F32, DataType::U16, DataType::F32, DynamicFusionCLCastToF32Fixture<uint16_t>, CastU16toF32Dataset, zero_tolerance) + +// S16 +CAST_SUITE(S16_to_U8, DataType::S16, DataType::U8, DynamicFusionCLCastToU8Fixture<int16_t>, CastS16toU8Dataset, zero_tolerance) +CAST_SUITE(S16_to_S8, DataType::S16, DataType::S8, DynamicFusionCLCastToS8Fixture<int16_t>, CastS16toS8Dataset, zero_tolerance) +CAST_SUITE(S16_to_U16, DataType::S16, DataType::U16, DynamicFusionCLCastToU16Fixture<int16_t>, CastS16toU16Dataset, zero_tolerance) +CAST_SUITE(S16_to_U32, DataType::S16, DataType::U32, DynamicFusionCLCastToU32Fixture<int16_t>, CastS16toU32Dataset, zero_tolerance) +CAST_SUITE(S16_to_S32, DataType::S16, DataType::S32, DynamicFusionCLCastToS32Fixture<int16_t>, CastS16toS32Dataset, zero_tolerance) +CAST_SUITE(S16_to_F16, DataType::S16, DataType::F16, DynamicFusionCLCastToF16Fixture<int16_t>, CastS16toF16Dataset, zero_tolerance) +CAST_SUITE(S16_to_F32, DataType::S16, DataType::F32, DynamicFusionCLCastToF32Fixture<int16_t>, CastS16toF32Dataset, zero_tolerance) + +// U32 +CAST_SUITE(U32_to_U8, DataType::U32, DataType::U8, DynamicFusionCLCastToU8Fixture<uint32_t>, CastU32toU8Dataset, zero_tolerance) +CAST_SUITE(U32_to_S8, DataType::U32, DataType::S8, DynamicFusionCLCastToS8Fixture<uint32_t>, CastU32toS8Dataset, zero_tolerance) +CAST_SUITE(U32_to_U16, DataType::U32, DataType::U16, DynamicFusionCLCastToU16Fixture<uint32_t>, CastU32toU16Dataset, zero_tolerance) +CAST_SUITE(U32_to_S16, DataType::U32, DataType::S16, DynamicFusionCLCastToS16Fixture<uint32_t>, CastU32toS16Dataset, zero_tolerance) +CAST_SUITE(U32_to_S32, DataType::U32, DataType::S32, DynamicFusionCLCastToS32Fixture<uint32_t>, CastU32toS32Dataset, zero_tolerance) +CAST_SUITE(U32_to_F16, DataType::U32, DataType::F16, DynamicFusionCLCastToF16Fixture<uint32_t>, CastU32toF16Dataset, zero_tolerance) +CAST_SUITE(U32_to_F32, DataType::U32, DataType::F32, DynamicFusionCLCastToF32Fixture<uint32_t>, CastU32toF32Dataset, zero_tolerance) + +// S32 +CAST_SUITE(S32_to_U8, DataType::S32, DataType::U8, DynamicFusionCLCastToU8Fixture<int32_t>, CastS32toU8Dataset, zero_tolerance) +CAST_SUITE(S32_to_S8, DataType::S32, DataType::S8, DynamicFusionCLCastToS8Fixture<int32_t>, CastS32toS8Dataset, zero_tolerance) +CAST_SUITE(S32_to_U16, DataType::S32, DataType::U16, DynamicFusionCLCastToU16Fixture<int32_t>, CastS32toU16Dataset, zero_tolerance) +CAST_SUITE(S32_to_S16, DataType::S32, DataType::S16, DynamicFusionCLCastToS16Fixture<int32_t>, CastS32toS16Dataset, zero_tolerance) +CAST_SUITE(S32_to_U32, DataType::S32, DataType::U32, DynamicFusionCLCastToU32Fixture<int32_t>, CastS32toU32Dataset, zero_tolerance) +CAST_SUITE(S32_to_F16, DataType::S32, DataType::F16, DynamicFusionCLCastToF16Fixture<int32_t>, CastS32toF16Dataset, zero_tolerance) +CAST_SUITE(S32_to_F32, DataType::S32, DataType::F32, DynamicFusionCLCastToF32Fixture<int32_t>, CastS32toF32Dataset, zero_tolerance) + +// F16 +CAST_SUITE(F16_to_U8, DataType::F16, DataType::U8, DynamicFusionCLCastToU8Fixture<half>, CastF16toU8Dataset, one_tolerance) +CAST_SUITE(F16_to_S8, DataType::F16, DataType::S8, DynamicFusionCLCastToS8Fixture<half>, CastF16toS8Dataset, one_tolerance) +CAST_SUITE(F16_to_U16, DataType::F16, DataType::U16, DynamicFusionCLCastToU16Fixture<half>, CastF16toU16Dataset, one_tolerance) +CAST_SUITE(F16_to_S16, DataType::F16, DataType::S16, DynamicFusionCLCastToS16Fixture<half>, CastF16toS16Dataset, one_tolerance) +CAST_SUITE(F16_to_U32, DataType::F16, DataType::U32, DynamicFusionCLCastToU32Fixture<half>, CastF16toU32Dataset, one_tolerance) +CAST_SUITE(F16_to_S32, DataType::F16, DataType::S32, DynamicFusionCLCastToS32Fixture<half>, CastF16toS32Dataset, one_tolerance) +CAST_SUITE(F16_to_F32, DataType::F16, DataType::F32, DynamicFusionCLCastToF32Fixture<half>, CastF16toF32Dataset, zero_tolerance) + +// F32 +CAST_SUITE(F32_to_U8, DataType::F32, DataType::U8, DynamicFusionCLCastToU8Fixture<float>, CastF32toU8Dataset, one_tolerance) +CAST_SUITE(F32_to_S8, DataType::F32, DataType::S8, DynamicFusionCLCastToS8Fixture<float>, CastF32toS8Dataset, one_tolerance) +CAST_SUITE(F32_to_U16, DataType::F32, DataType::U16, DynamicFusionCLCastToU16Fixture<float>, CastF32toU16Dataset, one_tolerance) +CAST_SUITE(F32_to_S16, DataType::F32, DataType::S16, DynamicFusionCLCastToS16Fixture<float>, CastF32toS16Dataset, one_tolerance) +CAST_SUITE(F32_to_U32, DataType::F32, DataType::U32, DynamicFusionCLCastToU32Fixture<float>, CastF32toU32Dataset, one_tolerance) +CAST_SUITE(F32_to_S32, DataType::F32, DataType::S32, DynamicFusionCLCastToS32Fixture<float>, CastF32toS32Dataset, one_tolerance) +CAST_SUITE(F32_to_F16, DataType::F32, DataType::F16, DynamicFusionCLCastToF16Fixture<float>, CastF32toF16Dataset, zero_tolerance) + +TEST_SUITE_END() // CAST +TEST_SUITE_END() // DYNAMIC_FUSION +TEST_SUITE_END() // CL +} // namespace validation +} // namespace test +} // namespace arm_compute diff --git a/tests/validation/fixtures/dynamic_fusion/operators/CastFixture.h b/tests/validation/fixtures/dynamic_fusion/operators/CastFixture.h new file mode 100644 index 0000000000..8553472fb9 --- /dev/null +++ b/tests/validation/fixtures/dynamic_fusion/operators/CastFixture.h @@ -0,0 +1,175 @@ +/* + * Copyright (c) 2022 Arm Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_CASTFIXTURE +#define TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_CASTFIXTURE + +#include "arm_compute/core/CL/CLKernelLibrary.h" +#include "arm_compute/core/TensorInfo.h" +#include "arm_compute/core/Types.h" +#include "arm_compute/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.h" +#include "arm_compute/dynamic_fusion/sketch/attributes/CastAttributes.h" +#include "arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h" + +#include "tests/framework/Fixture.h" +#include "tests/validation/reference/DepthConvertLayer.h" + +using namespace arm_compute::experimental::dynamic_fusion; + +namespace arm_compute +{ +namespace test +{ +namespace validation +{ +template <typename TensorType, typename AccessorType, typename FunctionType, typename T1, typename T2> +class DynamicFusionCastValidationFixture : public framework::Fixture +{ +public: + template <typename...> + void setup(TensorShape shape, DataType dt_in, DataType dt_out, ConvertPolicy policy) + { + _target = compute_target(shape, dt_in, dt_out, policy); + _reference = compute_reference(shape, dt_in, dt_out, policy); + } + +protected: + template <typename U> + void fill(U &&tensor, int i, DataType dt_in, DataType dt_out) + { + // Restricting range to avoid inf values + if(dt_out == DataType::F16) + { + constexpr int signed_min = -32000; + constexpr int signed_max = 32000; + constexpr int unsigned_min = 0; + constexpr int unsigned_max = 65000; + + switch(dt_in) + { + case DataType::U8: + case DataType::QASYMM8: + case DataType::QASYMM8_SIGNED: + case DataType::S8: + case DataType::F32: + { + library->fill_tensor_uniform(tensor, i); + break; + } + case DataType::U16: + { + library->fill_tensor_uniform(tensor, i, static_cast<uint16_t>(unsigned_min), static_cast<uint16_t>(unsigned_max)); + break; + } + case DataType::S16: + { + library->fill_tensor_uniform(tensor, i, static_cast<int16_t>(signed_min), static_cast<int16_t>(signed_max)); + break; + } + case DataType::U32: + { + library->fill_tensor_uniform(tensor, i, static_cast<uint32_t>(unsigned_min), static_cast<uint32_t>(unsigned_max)); + break; + } + case DataType::S32: + { + library->fill_tensor_uniform(tensor, i, static_cast<int32_t>(signed_min), static_cast<int32_t>(signed_max)); + break; + } + default: + ARM_COMPUTE_ERROR("NOT SUPPORTED!"); + } + } + else + { + library->fill_tensor_uniform(tensor, i); + } + } + + // Given input is in nchw format + TensorType compute_target(const TensorShape &shape, const DataType dt_in, const DataType dt_out, const ConvertPolicy policy) + { + // Create a new workload sketch + auto cl_compile_ctx = CLKernelLibrary::get().get_compile_context(); + auto gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx }; + GpuWorkloadSketch sketch{ &gpu_ctx }; + + // Create sketch tensors + TensorInfo src_info = sketch.create_tensor_info(TensorInfo(shape, 1, dt_in, DataLayout::NCHW)); // layout is not important + TensorInfo dst_info = sketch.create_tensor_info(); + + CastAttributes attributes; + attributes.convert_policy(policy).data_type(dt_out); + + FunctionType::create_op(sketch, &src_info, &dst_info, attributes); + + // Configure runtime + ClWorkloadRuntime runtime; + runtime.configure(sketch); + + // (Important) Allocate auxiliary tensor memory if there are any + for(auto &data : runtime.get_auxiliary_tensors()) + { + auto tensor = data.first; + const auto aux_mem_req = data.second; + tensor->allocator()->init(*data.first->info(), aux_mem_req.alignment); + tensor->allocator()->allocate(); + } + + // Construct user tensors + TensorType t_src{}; + TensorType t_dst{}; + + // Initialize user tensors + t_src.allocator()->init(src_info); + t_dst.allocator()->init(dst_info); + + // Allocate and fill user tensors + t_src.allocator()->allocate(); + t_dst.allocator()->allocate(); + + fill(AccessorType(t_src), 0, dt_in, dt_out); + + // Run runtime + runtime.run({ &t_src, &t_dst }); + return t_dst; + } + + SimpleTensor<T2> compute_reference(const TensorShape &shape, const DataType dt_in, const DataType dt_out, const ConvertPolicy policy) + { + // Create reference + SimpleTensor<T1> src{ shape, dt_in, 1 }; + + // Fill reference + fill(src, 0, dt_in, dt_out); + + return reference::depth_convert<T1, T2>(src, dt_out, policy, 0); + } + + TensorType _target{}; + SimpleTensor<T2> _reference{}; +}; +} // namespace validation +} // namespace test +} // namespace arm_compute +#endif /* TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_CASTFIXTURE */ |