diff options
author | giuros01 <giuseppe.rossini@arm.com> | 2019-05-14 16:12:53 +0100 |
---|---|---|
committer | Giuseppe Rossini <giuseppe.rossini@arm.com> | 2019-06-11 10:38:21 +0000 |
commit | d5134364fc4ca40ea65635192e7959327d690a01 (patch) | |
tree | d6781cc0319e54e538ea2b02ea59e842acfd6e49 /arm_compute | |
parent | e7510622419a63315e5ad5ed7de61a2ce4bd0b49 (diff) | |
download | ComputeLibrary-d5134364fc4ca40ea65635192e7959327d690a01.tar.gz |
COMPMID-2321: PRELU support in NEActivationLayer
Change-Id: Ib320ee7772492cd1b86eba624438da826d47b984
Signed-off-by: giuros01 <giuseppe.rossini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1224
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Diffstat (limited to 'arm_compute')
-rw-r--r-- | arm_compute/core/NEON/wrapper/traits.h | 42 | ||||
-rw-r--r-- | arm_compute/runtime/NEON/NEFunctions.h | 1 | ||||
-rw-r--r-- | arm_compute/runtime/NEON/functions/NEPReluLayer.h | 59 |
3 files changed, 81 insertions, 21 deletions
diff --git a/arm_compute/core/NEON/wrapper/traits.h b/arm_compute/core/NEON/wrapper/traits.h index 0dbd90ddf8..cc22597c29 100644 --- a/arm_compute/core/NEON/wrapper/traits.h +++ b/arm_compute/core/NEON/wrapper/traits.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018 ARM Limited. + * Copyright (c) 2018-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -44,27 +44,27 @@ struct vector_128_tag {}; template <typename T, int S> struct neon_vector; // Specializations #ifndef DOXYGEN_SKIP_THIS -template <> struct neon_vector<uint8_t, 8>{ using type = uint8x8_t; using tag_type = vector_64_tag; }; -template <> struct neon_vector<int8_t, 8>{ using type = int8x8_t; using tag_type = vector_64_tag; }; -template <> struct neon_vector<uint8_t, 16>{ using type = uint8x16_t; using tag_type = vector_128_tag; }; -template <> struct neon_vector<int8_t, 16>{ using type = int8x16_t; using tag_type = vector_128_tag; }; -template <> struct neon_vector<uint16_t, 4>{ using type = uint16x4_t; using tag_type = vector_64_tag; }; -template <> struct neon_vector<int16_t, 4>{ using type = int16x4_t; using tag_type = vector_64_tag; }; -template <> struct neon_vector<uint16_t, 8>{ using type = uint16x8_t; using tag_type = vector_128_tag; }; -template <> struct neon_vector<int16_t, 8>{ using type = int16x8_t; using tag_type = vector_128_tag; }; -template <> struct neon_vector<uint32_t, 2>{ using type = uint32x2_t; using tag_type = vector_64_tag; }; -template <> struct neon_vector<int32_t, 2>{ using type = int32x2_t; using tag_type = vector_64_tag; }; -template <> struct neon_vector<uint32_t, 4>{ using type = uint32x4_t; using tag_type = vector_128_tag; }; -template <> struct neon_vector<int32_t, 4>{ using type = int32x4_t; using tag_type = vector_128_tag; }; -template <> struct neon_vector<uint64_t, 1>{ using type = uint64x1_t; using tag_type = vector_64_tag; }; -template <> struct neon_vector<int64_t, 1>{ using type = int64x1_t; using tag_type = vector_64_tag; }; -template <> struct neon_vector<uint64_t, 2>{ using type = uint64x2_t; using tag_type = vector_128_tag; }; -template <> struct neon_vector<int64_t, 2>{ using type = int64x2_t; using tag_type = vector_128_tag; }; -template <> struct neon_vector<float_t, 2>{ using type = float32x2_t; using tag_type = vector_64_tag; }; -template <> struct neon_vector<float_t, 4>{ using type = float32x4_t; using tag_type = vector_128_tag; }; +template <> struct neon_vector<uint8_t, 8>{ using scalar_type = uint8_t; using type = uint8x8_t; using tag_type = vector_64_tag; }; +template <> struct neon_vector<int8_t, 8>{ using scalar_type = int8_t; using type = int8x8_t; using tag_type = vector_64_tag; }; +template <> struct neon_vector<uint8_t, 16>{ using scalar_type = uint8_t; using type = uint8x16_t; using tag_type = vector_128_tag; }; +template <> struct neon_vector<int8_t, 16>{ using scalar_type = int8_t; using type = int8x16_t; using tag_type = vector_128_tag; }; +template <> struct neon_vector<uint16_t, 4>{ using scalar_type = uint16_t; using type = uint16x4_t; using tag_type = vector_64_tag; }; +template <> struct neon_vector<int16_t, 4>{ using scalar_type = int16_t; using type = int16x4_t; using tag_type = vector_64_tag; }; +template <> struct neon_vector<uint16_t, 8>{ using scalar_type = uint16_t; using type = uint16x8_t; using tag_type = vector_128_tag; }; +template <> struct neon_vector<int16_t, 8>{ using scalar_type = int16_t; using type = int16x8_t; using tag_type = vector_128_tag; }; +template <> struct neon_vector<uint32_t, 2>{ using scalar_type = uint32_t; using type = uint32x2_t; using tag_type = vector_64_tag; }; +template <> struct neon_vector<int32_t, 2>{ using scalar_type = int32_t; using type = int32x2_t; using tag_type = vector_64_tag; }; +template <> struct neon_vector<uint32_t, 4>{ using scalar_type = uint32_t; using type = uint32x4_t; using tag_type = vector_128_tag; }; +template <> struct neon_vector<int32_t, 4>{ using scalar_type = int32_t; using type = int32x4_t; using tag_type = vector_128_tag; }; +template <> struct neon_vector<uint64_t, 1>{ using scalar_type = uint64_t;using type = uint64x1_t; using tag_type = vector_64_tag; }; +template <> struct neon_vector<int64_t, 1>{ using scalar_type = int64_t; using type = int64x1_t; using tag_type = vector_64_tag; }; +template <> struct neon_vector<uint64_t, 2>{ using scalar_type = uint64_t; using type = uint64x2_t; using tag_type = vector_128_tag; }; +template <> struct neon_vector<int64_t, 2>{ using scalar_type = int64_t; using type = int64x2_t; using tag_type = vector_128_tag; }; +template <> struct neon_vector<float_t, 2>{ using scalar_type = float_t; using type = float32x2_t; using tag_type = vector_64_tag; }; +template <> struct neon_vector<float_t, 4>{ using scalar_type = float_t; using type = float32x4_t; using tag_type = vector_128_tag; }; #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC -template <> struct neon_vector<float16_t, 4>{ using type = float16x4_t; using tag_type = vector_64_tag; }; -template <> struct neon_vector<float16_t, 8>{ using type = float16x8_t; using tag_type = vector_128_tag; }; +template <> struct neon_vector<float16_t, 4>{ using scalar_type = float16_t; using type = float16x4_t; using tag_type = vector_64_tag; }; +template <> struct neon_vector<float16_t, 8>{ using scalar_type = float16_t; using type = float16x8_t; using tag_type = vector_128_tag; }; #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC #endif /* DOXYGEN_SKIP_THIS */ diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h index 4e0cdd7a0a..94607364b3 100644 --- a/arm_compute/runtime/NEON/NEFunctions.h +++ b/arm_compute/runtime/NEON/NEFunctions.h @@ -106,6 +106,7 @@ #include "arm_compute/runtime/NEON/functions/NENonMaximaSuppression3x3.h" #include "arm_compute/runtime/NEON/functions/NENormalizationLayer.h" #include "arm_compute/runtime/NEON/functions/NEOpticalFlow.h" +#include "arm_compute/runtime/NEON/functions/NEPReluLayer.h" #include "arm_compute/runtime/NEON/functions/NEPadLayer.h" #include "arm_compute/runtime/NEON/functions/NEPermute.h" #include "arm_compute/runtime/NEON/functions/NEPhase.h" diff --git a/arm_compute/runtime/NEON/functions/NEPReluLayer.h b/arm_compute/runtime/NEON/functions/NEPReluLayer.h new file mode 100644 index 0000000000..52db4279cd --- /dev/null +++ b/arm_compute/runtime/NEON/functions/NEPReluLayer.h @@ -0,0 +1,59 @@ +/* + * Copyright (c) 2019 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 __ARM_COMPUTE_NEPRELULAYER_H__ +#define __ARM_COMPUTE_NEPRELULAYER_H__ + +#include "arm_compute/core/Types.h" +#include "arm_compute/runtime/NEON/INESimpleFunction.h" + +namespace arm_compute +{ +class ITensor; + +/** Basic function to run @ref NEArithmeticOperationKernel for PRELU + * + * @note The function implements an activation layer with the PRELU activation function. + */ +class NEPReluLayer : public INESimpleFunction +{ +public: + /** Set the input and output tensor. + * + * @param[in] input Source tensor. Data types supported: QASYMM8/F16/F32. + * @param[in] alpha Source alpha tensor. Data types supported: same of @p input. + * @param[out] output Destination tensor. Data type supported: same as @p input + */ + void configure(const ITensor *input, const ITensor *alpha, ITensor *output); + /** Static function to check if given info will lead to a valid configuration of @ref NEPReluLayer + * + * @param[in] input Source tensor info. Data types supported: QASYMM8/F16/F32. + * @param[in] alpha Source alpha tensor info. Data types supported: same of @p input. + * @param[in] output Destination tensor info. Data type supported: same as @p input + * + * @return a status + */ + static Status validate(const ITensorInfo *input, const ITensorInfo *alpha, const ITensorInfo *output); +}; +} // namespace arm_compute +#endif /* __ARM_COMPUTE_NEPRELULAYER_H__ */ |