Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +0100 | [diff] [blame] | 1 | /* |
Usama Arif | e73686a | 2019-04-08 17:30:48 +0100 | [diff] [blame] | 2 | * Copyright (c) 2018-2019 ARM Limited. |
Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +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 | */ |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 24 | #ifndef ARM_COMPUTE_CLDEPTHWISECONVOLUTIONNCHWKERNEL3x3_H |
| 25 | #define ARM_COMPUTE_CLDEPTHWISECONVOLUTIONNCHWKERNEL3x3_H |
Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +0100 | [diff] [blame] | 26 | |
Giorgio Arena | dfca60b | 2018-01-31 10:30:59 +0000 | [diff] [blame] | 27 | #include "arm_compute/core/CL/kernels/ICLDepthwiseConvolutionLayer3x3Kernel.h" |
Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +0100 | [diff] [blame] | 28 | |
| 29 | namespace arm_compute |
| 30 | { |
| 31 | class ICLTensor; |
| 32 | |
Giorgio Arena | dfca60b | 2018-01-31 10:30:59 +0000 | [diff] [blame] | 33 | /** Interface for the kernel to run a 3x3 depthwise convolution on a tensor when the data layout is NCHW. |
Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +0100 | [diff] [blame] | 34 | */ |
Giorgio Arena | dfca60b | 2018-01-31 10:30:59 +0000 | [diff] [blame] | 35 | class CLDepthwiseConvolutionLayer3x3NCHWKernel : public ICLDepthwiseConvolutionLayer3x3Kernel |
Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +0100 | [diff] [blame] | 36 | { |
| 37 | public: |
| 38 | /** Default constructor */ |
Giorgio Arena | dfca60b | 2018-01-31 10:30:59 +0000 | [diff] [blame] | 39 | CLDepthwiseConvolutionLayer3x3NCHWKernel(); |
Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +0100 | [diff] [blame] | 40 | /** Initialize the function's source, destination, conv and border_size. |
| 41 | * |
Michele Di Giorgio | 37cbc57 | 2019-12-19 16:41:49 +0000 | [diff] [blame^] | 42 | * @param[in] input Source tensor. DataType supported: QASYMM8/QASYMM8_SIGNED/F16/F32. |
Michele Di Giorgio | df4cf57 | 2019-10-09 15:32:39 +0100 | [diff] [blame] | 43 | * @param[in] weights Weights tensor. A 3D tensor with dimensions [3, 3, IFM]. |
Michele Di Giorgio | 37cbc57 | 2019-12-19 16:41:49 +0000 | [diff] [blame^] | 44 | * Data type supported: Same as @p input or QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL when @p input is QASYMM8/QASYMM8_SIGNED. |
Michele Di Giorgio | df4cf57 | 2019-10-09 15:32:39 +0100 | [diff] [blame] | 45 | * @param[in] biases Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. |
Michele Di Giorgio | 37cbc57 | 2019-12-19 16:41:49 +0000 | [diff] [blame^] | 46 | * Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED. |
Michele Di Giorgio | df4cf57 | 2019-10-09 15:32:39 +0100 | [diff] [blame] | 47 | * @param[out] output Destination tensor. Data type supported: Same as @p input. |
| 48 | * @param[in] conv_info Padding and stride information to use for the convolution. |
| 49 | * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. |
| 50 | * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU for QASYMM8 supported. |
| 51 | * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). |
| 52 | * @param[in] output_multipliers (Optional) Output multipliers tensor for quantized computations. In case of per-channel quantization, |
| 53 | * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 |
| 54 | * @param[in] output_shifts (Optional) Output shifts tensor for quantized computations. In case of per-channel quantization, |
| 55 | * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 |
Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +0100 | [diff] [blame] | 56 | */ |
Michele Di Giorgio | df4cf57 | 2019-10-09 15:32:39 +0100 | [diff] [blame] | 57 | void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, |
| 58 | unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U), |
| 59 | const ICLTensor *output_multipliers = nullptr, const ICLTensor *output_shifts = nullptr) override; |
Giorgio Arena | ad0c738 | 2018-04-23 16:16:21 +0100 | [diff] [blame] | 60 | /** Static function to check if given info will lead to a valid configuration of @ref CLDepthwiseConvolutionLayer3x3NCHWKernel |
| 61 | * |
Michele Di Giorgio | 37cbc57 | 2019-12-19 16:41:49 +0000 | [diff] [blame^] | 62 | * @param[in] input Source tensor info. DataType supported: QASYMM8/QASYMM8_SIGNED/F16/F32. |
Michele Di Giorgio | df4cf57 | 2019-10-09 15:32:39 +0100 | [diff] [blame] | 63 | * @param[in] weights Weights tensor info. A 3D tensor with dimensions [3, 3, IFM]. |
Michele Di Giorgio | 37cbc57 | 2019-12-19 16:41:49 +0000 | [diff] [blame^] | 64 | * Data type supported: Same as @p input or QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL when @p input is QASYMM8/QASYMM8_SIGNED. |
Michele Di Giorgio | df4cf57 | 2019-10-09 15:32:39 +0100 | [diff] [blame] | 65 | * @param[in] biases Biases tensor info. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. |
Michele Di Giorgio | 37cbc57 | 2019-12-19 16:41:49 +0000 | [diff] [blame^] | 66 | * Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED. |
Michele Di Giorgio | df4cf57 | 2019-10-09 15:32:39 +0100 | [diff] [blame] | 67 | * @param[in] output Destination tensor. Data type supported: Same as @p input. |
| 68 | * @param[in] conv_info Padding and stride information to use for the convolution. |
| 69 | * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. |
| 70 | * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported. |
| 71 | * @param[in] gpu_target (Optional) GPU target to validate the kernel for. Defaults to midgard. |
| 72 | * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). |
| 73 | * @param[in] output_multipliers (Optional) Output multipliers tensor info for quantized computations. In case of per-channel quantization, |
| 74 | * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 |
| 75 | * @param[in] output_shifts (Optional) Output shifts tensor for quantized computations. In case of per-channel quantization, |
| 76 | * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 |
Giorgio Arena | ad0c738 | 2018-04-23 16:16:21 +0100 | [diff] [blame] | 77 | * |
| 78 | * @return a status |
| 79 | */ |
Michele Di Giorgio | df4cf57 | 2019-10-09 15:32:39 +0100 | [diff] [blame] | 80 | static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, |
| 81 | unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), GPUTarget gpu_target = GPUTarget::MIDGARD, |
| 82 | const Size2D &dilation = Size2D(1U, 1U), const ITensorInfo *output_multipliers = nullptr, const ITensorInfo *output_shifts = nullptr); |
Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +0100 | [diff] [blame] | 83 | |
Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +0100 | [diff] [blame] | 84 | void run(const Window &window, cl::CommandQueue &queue) override; |
| 85 | BorderSize border_size() const override; |
| 86 | |
| 87 | private: |
Giorgio Arena | dfca60b | 2018-01-31 10:30:59 +0000 | [diff] [blame] | 88 | unsigned int _conv_stride_x; |
| 89 | unsigned int _conv_pad_top; |
Giorgio Arena | fa23f11 | 2018-06-19 11:27:38 +0100 | [diff] [blame] | 90 | unsigned int _conv_pad_left; |
Giorgio Arena | 93a690e | 2017-08-01 16:09:33 +0100 | [diff] [blame] | 91 | }; |
Gian Marco Iodice | f670a0a | 2017-09-18 12:20:45 +0100 | [diff] [blame] | 92 | } // namespace arm_compute |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 93 | #endif /*ARM_COMPUTE_CLDEPTHWISECONVOLUTIONNCHWKERNEL3x3_H */ |