Gunes Bayir | 1dc6ff1 | 2022-12-06 20:48:31 +0000 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright (c) 2022 Arm Limited. |
| 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 | */ |
| 24 | #ifndef TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_CASTFIXTURE |
| 25 | #define TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_CASTFIXTURE |
| 26 | |
| 27 | #include "arm_compute/core/CL/CLKernelLibrary.h" |
| 28 | #include "arm_compute/core/TensorInfo.h" |
| 29 | #include "arm_compute/core/Types.h" |
| 30 | #include "arm_compute/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.h" |
| 31 | #include "arm_compute/dynamic_fusion/sketch/attributes/CastAttributes.h" |
| 32 | #include "arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h" |
| 33 | |
| 34 | #include "tests/framework/Fixture.h" |
| 35 | #include "tests/validation/reference/DepthConvertLayer.h" |
| 36 | |
| 37 | using namespace arm_compute::experimental::dynamic_fusion; |
| 38 | |
| 39 | namespace arm_compute |
| 40 | { |
| 41 | namespace test |
| 42 | { |
| 43 | namespace validation |
| 44 | { |
| 45 | template <typename TensorType, typename AccessorType, typename FunctionType, typename T1, typename T2> |
| 46 | class DynamicFusionCastValidationFixture : public framework::Fixture |
| 47 | { |
| 48 | public: |
| 49 | template <typename...> |
| 50 | void setup(TensorShape shape, DataType dt_in, DataType dt_out, ConvertPolicy policy) |
| 51 | { |
| 52 | _target = compute_target(shape, dt_in, dt_out, policy); |
| 53 | _reference = compute_reference(shape, dt_in, dt_out, policy); |
| 54 | } |
| 55 | |
| 56 | protected: |
| 57 | template <typename U> |
| 58 | void fill(U &&tensor, int i, DataType dt_in, DataType dt_out) |
| 59 | { |
| 60 | // Restricting range to avoid inf values |
| 61 | if(dt_out == DataType::F16) |
| 62 | { |
| 63 | constexpr int signed_min = -32000; |
| 64 | constexpr int signed_max = 32000; |
| 65 | constexpr int unsigned_min = 0; |
| 66 | constexpr int unsigned_max = 65000; |
| 67 | |
| 68 | switch(dt_in) |
| 69 | { |
| 70 | case DataType::U8: |
| 71 | case DataType::QASYMM8: |
| 72 | case DataType::QASYMM8_SIGNED: |
| 73 | case DataType::S8: |
| 74 | case DataType::F32: |
| 75 | { |
| 76 | library->fill_tensor_uniform(tensor, i); |
| 77 | break; |
| 78 | } |
| 79 | case DataType::U16: |
| 80 | { |
| 81 | library->fill_tensor_uniform(tensor, i, static_cast<uint16_t>(unsigned_min), static_cast<uint16_t>(unsigned_max)); |
| 82 | break; |
| 83 | } |
| 84 | case DataType::S16: |
| 85 | { |
| 86 | library->fill_tensor_uniform(tensor, i, static_cast<int16_t>(signed_min), static_cast<int16_t>(signed_max)); |
| 87 | break; |
| 88 | } |
| 89 | case DataType::U32: |
| 90 | { |
| 91 | library->fill_tensor_uniform(tensor, i, static_cast<uint32_t>(unsigned_min), static_cast<uint32_t>(unsigned_max)); |
| 92 | break; |
| 93 | } |
| 94 | case DataType::S32: |
| 95 | { |
| 96 | library->fill_tensor_uniform(tensor, i, static_cast<int32_t>(signed_min), static_cast<int32_t>(signed_max)); |
| 97 | break; |
| 98 | } |
| 99 | default: |
| 100 | ARM_COMPUTE_ERROR("NOT SUPPORTED!"); |
| 101 | } |
| 102 | } |
| 103 | else |
| 104 | { |
| 105 | library->fill_tensor_uniform(tensor, i); |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | // Given input is in nchw format |
| 110 | TensorType compute_target(const TensorShape &shape, const DataType dt_in, const DataType dt_out, const ConvertPolicy policy) |
| 111 | { |
| 112 | // Create a new workload sketch |
| 113 | auto cl_compile_ctx = CLKernelLibrary::get().get_compile_context(); |
| 114 | auto gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx }; |
| 115 | GpuWorkloadSketch sketch{ &gpu_ctx }; |
| 116 | |
| 117 | // Create sketch tensors |
| 118 | TensorInfo src_info = sketch.create_tensor_info(TensorInfo(shape, 1, dt_in, DataLayout::NCHW)); // layout is not important |
| 119 | TensorInfo dst_info = sketch.create_tensor_info(); |
| 120 | |
| 121 | CastAttributes attributes; |
| 122 | attributes.convert_policy(policy).data_type(dt_out); |
| 123 | |
| 124 | FunctionType::create_op(sketch, &src_info, &dst_info, attributes); |
| 125 | |
| 126 | // Configure runtime |
| 127 | ClWorkloadRuntime runtime; |
| 128 | runtime.configure(sketch); |
| 129 | |
| 130 | // (Important) Allocate auxiliary tensor memory if there are any |
| 131 | for(auto &data : runtime.get_auxiliary_tensors()) |
| 132 | { |
| 133 | auto tensor = data.first; |
| 134 | const auto aux_mem_req = data.second; |
| 135 | tensor->allocator()->init(*data.first->info(), aux_mem_req.alignment); |
| 136 | tensor->allocator()->allocate(); |
| 137 | } |
| 138 | |
| 139 | // Construct user tensors |
| 140 | TensorType t_src{}; |
| 141 | TensorType t_dst{}; |
| 142 | |
| 143 | // Initialize user tensors |
| 144 | t_src.allocator()->init(src_info); |
| 145 | t_dst.allocator()->init(dst_info); |
| 146 | |
| 147 | // Allocate and fill user tensors |
| 148 | t_src.allocator()->allocate(); |
| 149 | t_dst.allocator()->allocate(); |
| 150 | |
| 151 | fill(AccessorType(t_src), 0, dt_in, dt_out); |
| 152 | |
| 153 | // Run runtime |
| 154 | runtime.run({ &t_src, &t_dst }); |
| 155 | return t_dst; |
| 156 | } |
| 157 | |
| 158 | SimpleTensor<T2> compute_reference(const TensorShape &shape, const DataType dt_in, const DataType dt_out, const ConvertPolicy policy) |
| 159 | { |
| 160 | // Create reference |
| 161 | SimpleTensor<T1> src{ shape, dt_in, 1 }; |
| 162 | |
| 163 | // Fill reference |
| 164 | fill(src, 0, dt_in, dt_out); |
| 165 | |
| 166 | return reference::depth_convert<T1, T2>(src, dt_out, policy, 0); |
| 167 | } |
| 168 | |
| 169 | TensorType _target{}; |
| 170 | SimpleTensor<T2> _reference{}; |
| 171 | }; |
| 172 | } // namespace validation |
| 173 | } // namespace test |
| 174 | } // namespace arm_compute |
| 175 | #endif /* TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_CASTFIXTURE */ |