Sang-Hoon Park | 201e0fe | 2021-01-27 13:14:56 +0000 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright (c) 2021 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 ARM_COMPUTE_CL_SOFTMAX_H |
| 25 | #define ARM_COMPUTE_CL_SOFTMAX_H |
| 26 | |
| 27 | #include "arm_compute/runtime/CL/CLTensor.h" |
| 28 | #include "src/core/gpu/cl/ClCompileContext.h" |
| 29 | #include "src/runtime/gpu/cl/IClOperator.h" |
| 30 | |
| 31 | namespace arm_compute |
| 32 | { |
| 33 | struct SoftmaxKernelInfo; |
| 34 | |
| 35 | namespace opencl |
| 36 | { |
| 37 | class ClPermute; |
| 38 | namespace kernels |
| 39 | { |
| 40 | class ClLogits1DMaxShiftExpSumKernel; |
| 41 | class ClLogits1DNormKernel; |
| 42 | } // namespace kernels |
| 43 | class ClSoftmax : public IClOperator |
| 44 | { |
| 45 | public: |
| 46 | /** Constructor */ |
| 47 | ClSoftmax(); |
| 48 | /** Configure the operator |
| 49 | * |
| 50 | * @param[in] compile_context The compile context to be used. |
| 51 | * @param[in] src Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32 for Softmax and F16/F32 for Log Softmax |
| 52 | * @param[out] dst Destination tensor info. Data types supported: same as @p src |
| 53 | * @param[in] info Contains information consumed by kernels for softmax described in @ref SoftmaxKernelInfo. |
| 54 | * |
| 55 | */ |
| 56 | void configure(const CLCompileContext &compile_context, const ITensorInfo &src, ITensorInfo &dst, const SoftmaxKernelInfo &info); |
| 57 | /** Static function to check if the given info will lead to a valid configuration |
| 58 | * |
| 59 | * @param[in] src Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32 for Softmax and F16/F32 for Log Softmax |
| 60 | * @param[out] dst Destination tensor info. Data types supported: same as @p src |
| 61 | * @param[in] info Contains information consumed by kernels for softmax described in @ref SoftmaxKernelInfo. |
| 62 | * |
| 63 | */ |
| 64 | static Status validate(const ITensorInfo &src, const ITensorInfo &dst, const SoftmaxKernelInfo &info); |
| 65 | // Inherited methods overridden: |
| 66 | void run(ITensorPack &tensors) override; |
| 67 | experimental::MemoryRequirements workspace() const override; |
| 68 | |
| 69 | private: |
| 70 | enum class InternalTensorIdx |
| 71 | { |
| 72 | MAX = 0, |
| 73 | SUM, |
| 74 | TMP, |
| 75 | PERMUTED_SRC, |
| 76 | PERMUTED_DST, |
| 77 | COUNT |
| 78 | }; |
| 79 | |
| 80 | /** Create a single internal tensor |
| 81 | * |
| 82 | * @param[in] info The information used to create a tensor |
| 83 | * @param[in] idx The index within the internal array the created tensor will be held |
| 84 | */ |
| 85 | void create_internal_tensor(TensorInfo &info, InternalTensorIdx idx); |
| 86 | /** Create all required internal tensors */ |
| 87 | void create_internal_tensor(); |
| 88 | /** Function to convert from internal tensor index to @ref TensorType used externally */ |
| 89 | TensorType convert_internal_idx_to_tensor_type(InternalTensorIdx idx) const; |
| 90 | /** Function to import workspace memory allocated by the caller into internal tensor instances */ |
| 91 | void import_workspace_memory(ITensorPack &tensors); |
| 92 | /** Function to permute the given source tensor when permutation is required */ |
| 93 | void run_source_permute(const ITensor *src); |
| 94 | /** Function to permute the intemediate tensor to the final destination tensor when permutation is required */ |
| 95 | void run_destination_permute(ITensor *dst); |
| 96 | /** Function to run @ref arm_compute::opencl::kernels::ClLogits1DMaxShiftExpSumKernel */ |
| 97 | void run_max_sum(const ITensor *src); |
| 98 | /** Function to run @ref kernels::ClLogits1DNormKernel */ |
| 99 | void run_norm(ITensor *dst); |
| 100 | |
| 101 | std::unique_ptr<ClPermute> _permute_input; |
| 102 | std::unique_ptr<ClPermute> _permute_output; |
| 103 | std::unique_ptr<kernels::ClLogits1DMaxShiftExpSumKernel> _max_shift_exp_sum_kernel; |
| 104 | std::unique_ptr<kernels::ClLogits1DNormKernel> _norm_kernel; |
| 105 | bool _needs_permute{ false }; |
| 106 | |
| 107 | std::array<TensorInfo, static_cast<uint32_t>(InternalTensorIdx::COUNT)> _internal_info{}; |
| 108 | std::array<std::unique_ptr<CLTensor>, static_cast<uint32_t>(InternalTensorIdx::COUNT)> _internal_tensor{}; |
| 109 | |
| 110 | TensorInfo &_max_info; |
| 111 | TensorInfo &_sum_info; |
| 112 | TensorInfo &_tmp_info; |
| 113 | TensorInfo &_permuted_src_info; |
| 114 | TensorInfo &_permuted_dst_info; |
| 115 | }; |
| 116 | |
| 117 | } // opencl |
| 118 | } // arm_compute |
| 119 | #endif /* ARM_COMPUTE_CL_SOFTMAX_H */ |