Ramy Elgammal | f26ea2f | 2023-03-24 11:42:03 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2023 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_MATMULFIXTURE |
| 25 | #define TESTS_VALIDATION_FIXTURES_MATMULFIXTURE |
| 26 | |
| 27 | #include "arm_compute/core/Types.h" |
| 28 | #include "tests/framework/Fixture.h" |
| 29 | #include "tests/validation/reference/GEMM.h" |
| 30 | #include "tests/validation/reference/Permute.h" |
| 31 | #include "tests/validation/reference/Permute.h" |
| 32 | #include "tests/validation/reference/ReshapeLayer.h" |
| 33 | #include <random> |
| 34 | namespace arm_compute |
| 35 | { |
| 36 | namespace test |
| 37 | { |
| 38 | namespace validation |
| 39 | { |
| 40 | template <typename TensorType, typename AccessorType, typename FunctionType, typename T> |
| 41 | class MatMulValidationFixture : public framework::Fixture |
| 42 | { |
| 43 | public: |
| 44 | template <typename...> |
| 45 | void setup(TensorShape shape_a, TensorShape shape_b, TensorShape output_shape, bool pretranspose_a, bool pretranspose_b, DataType data_type) |
| 46 | { |
| 47 | // For brevity, the input shapes are assumed to be not-transposed for both Lhs and Rhs matrices. |
| 48 | if(pretranspose_a) |
| 49 | { |
| 50 | permute(shape_a, PermutationVector(1U, 0U)); |
| 51 | } |
| 52 | if(pretranspose_b) |
| 53 | { |
| 54 | permute(shape_b, PermutationVector(1U, 0U)); |
| 55 | } |
| 56 | _target = compute_target(shape_a, shape_b, output_shape, pretranspose_a, pretranspose_b, data_type); |
| 57 | _reference = compute_reference(shape_a, shape_b, output_shape, pretranspose_a, pretranspose_b, data_type); |
| 58 | } |
| 59 | |
| 60 | protected: |
| 61 | template <typename U> |
| 62 | void fill(U &&tensor, int i, float lo = -1.f, float hi = 1.f) |
| 63 | { |
| 64 | switch(tensor.data_type()) |
| 65 | { |
| 66 | case DataType::F16: |
| 67 | { |
| 68 | arm_compute::utils::uniform_real_distribution_16bit<half> distribution{ float(lo), float(hi) }; |
| 69 | library->fill(tensor, distribution, i); |
| 70 | break; |
| 71 | } |
| 72 | case DataType::F32: |
| 73 | { |
| 74 | std::uniform_real_distribution<float> distribution(lo, hi); |
| 75 | library->fill(tensor, distribution, i); |
| 76 | break; |
| 77 | } |
| 78 | default: |
| 79 | library->fill_tensor_uniform(tensor, i); |
| 80 | } |
| 81 | } |
| 82 | TensorType compute_target(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &output_shape, bool pretranspose_a, bool pretranspose_b, DataType data_type) |
| 83 | { |
| 84 | // 1. Create Classes and configure function |
| 85 | // Create tensors |
| 86 | TensorType a = create_tensor<TensorType>(shape_a, data_type, 1); |
| 87 | TensorType b = create_tensor<TensorType>(shape_b, data_type, 1); |
| 88 | TensorType dst = create_tensor<TensorType>(output_shape, data_type, 1); |
| 89 | FunctionType matmul; |
| 90 | // Configure MatMulInfo class |
| 91 | MatMulInfo info; |
| 92 | info.adj_lhs(pretranspose_a); |
| 93 | info.adj_rhs(pretranspose_b); |
| 94 | matmul.configure(&a, &b, &dst, info); |
| 95 | // Assertions |
| 96 | ARM_COMPUTE_ASSERT(a.info()->is_resizable()); |
| 97 | ARM_COMPUTE_ASSERT(b.info()->is_resizable()); |
| 98 | ARM_COMPUTE_ASSERT(dst.info()->is_resizable()); |
| 99 | // Allocate tensors |
| 100 | a.allocator()->allocate(); |
| 101 | b.allocator()->allocate(); |
| 102 | dst.allocator()->allocate(); |
| 103 | ARM_COMPUTE_ASSERT(!a.info()->is_resizable()); |
| 104 | ARM_COMPUTE_ASSERT(!b.info()->is_resizable()); |
| 105 | ARM_COMPUTE_ASSERT(!dst.info()->is_resizable()); |
| 106 | |
| 107 | // 2. Fill tensors and run once |
| 108 | // Fill tensors |
| 109 | fill(AccessorType(a), 0); |
| 110 | fill(AccessorType(b), 1); |
| 111 | matmul.run(); // First run |
| 112 | |
| 113 | return dst; |
| 114 | } |
| 115 | SimpleTensor<T> compute_reference(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &output_shape, bool pretranspose_a, bool pretranspose_b, DataType data_type) |
| 116 | { |
| 117 | // We collapse dimensions > 3 onto dimension 3, i.e. 5D+ tensors will look like 4D |
| 118 | // This is necessary unless we choose to extend gemm reference for 5D+ tensors |
| 119 | TensorShape output_shape_collapsed = output_shape.collapsed_from(Window::DimW); |
| 120 | TensorShape a_shape_collapsed = shape_a.collapsed_from(Window::DimW); |
| 121 | TensorShape b_shape_collapsed = shape_b.collapsed_from(Window::DimW); |
| 122 | |
| 123 | // Create reference |
| 124 | SimpleTensor<T> a{ a_shape_collapsed, data_type, 1 }; |
| 125 | SimpleTensor<T> b{ b_shape_collapsed, data_type, 1 }; |
| 126 | SimpleTensor<T> c{ output_shape_collapsed, data_type, 1 }; |
| 127 | |
| 128 | // Fill reference |
| 129 | fill(a, 0); |
| 130 | fill(b, 1); |
| 131 | |
| 132 | /* Note: Assuming the usual batch matmul dimensions A = (B x M x K), B = (B x K x N), if pretranspose_a is set to true, then A is assumed to be (B x K x M), |
| 133 | therefore, A must be pre-transposed before passing it to the fixture. And, we transpose A again in the fixture to make it (B x M x K) |
| 134 | in order to be able to call reference implementation that works with (B x M x K) input. |
| 135 | Similarly, if pretranspose_b is set to true, then B is assumed to be (B x N x K), B must be pre-transposed before passing it to the fixture. */ |
| 136 | |
| 137 | // Define transposed shapes |
| 138 | TensorShape a_transposed_shape(a.shape()); |
| 139 | a_transposed_shape.set(0, a.shape().y()); |
| 140 | a_transposed_shape.set(1, a.shape().x()); |
| 141 | TensorShape b_transposed_shape(b.shape()); |
| 142 | b_transposed_shape.set(0, b.shape().y()); |
| 143 | b_transposed_shape.set(1, b.shape().x()); |
| 144 | |
| 145 | // Define transposed tensors |
| 146 | SimpleTensor<T> a_transposed{ a_transposed_shape, data_type }; |
| 147 | SimpleTensor<T> b_transposed{ b_transposed_shape, data_type }; |
| 148 | |
| 149 | // pretranspose a if necessary |
| 150 | if(pretranspose_a) |
| 151 | { |
| 152 | a_transposed = reference::permute<T>(a, PermutationVector(1U, 0U)); |
| 153 | } |
| 154 | |
| 155 | // pretranspose b if necessary |
| 156 | if(pretranspose_b) |
| 157 | { |
| 158 | b_transposed = reference::permute<T>(b, PermutationVector(1U, 0U)); |
| 159 | } |
| 160 | |
| 161 | // Setting beta to 0 will effectively disable C for the |
| 162 | // computation of the reference: alpha * A * B + 0 * C |
| 163 | // Use transposed tensors if boolean enabled else use original tensors |
| 164 | SimpleTensor<T> result = reference::gemm<T>((pretranspose_a) ? a_transposed : a, (pretranspose_b) ? b_transposed : b, c, 1.0f, 0.f); |
| 165 | |
| 166 | // We reshape the gemm output back if the tensor is high dimensional |
| 167 | if(output_shape_collapsed != output_shape) |
| 168 | { |
| 169 | result = reference::reshape_layer(result, output_shape); |
| 170 | } |
| 171 | |
| 172 | return result; |
| 173 | } |
| 174 | TensorType _target{}; |
| 175 | SimpleTensor<T> _reference{}; |
| 176 | }; |
| 177 | } // namespace validation |
| 178 | } // namespace test |
| 179 | } // namespace arm_compute |
| 180 | #endif /* TESTS_VALIDATION_FIXTURES_MATMULFIXTURE */ |