diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2018-12-03 14:30:05 +0000 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-01-14 17:53:22 +0000 |
commit | 5a5945387e70f62e6e1e95a177fae261d7570443 (patch) | |
tree | ff8bd61c2e071b5a0b923f4a0d1bef72486435e9 /arm_compute/core/NEON/kernels | |
parent | dea2d2d58fe3a742e6f66fe50befbe0044e15ad1 (diff) | |
download | ComputeLibrary-5a5945387e70f62e6e1e95a177fae261d7570443.tar.gz |
COMPMID-1809: Remove padding in NEGEMMConvolutionLayer 64-bit path.
Change-Id: I1806591a2c73a1f057f13d8c6107d7b9796a82c8
Reviewed-on: https://review.mlplatform.org/370
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Diffstat (limited to 'arm_compute/core/NEON/kernels')
3 files changed, 10 insertions, 15 deletions
diff --git a/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h b/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h index 0290e32085..447f4880ee 100644 --- a/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h +++ b/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2018 ARM Limited. + * Copyright (c) 2017-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -26,6 +26,7 @@ #include "arm_compute/core/NEON/INEKernel.h" #include "arm_compute/core/QAsymm8.h" +#include "arm_compute/core/utils/misc/Traits.h" #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC #include <arm_fp16.h> @@ -89,15 +90,8 @@ private: * @param[in] window Region on which to execute the kernel */ template <ActivationLayerInfo::ActivationFunction F, typename T> - typename std::enable_if<std::is_same<T, float>::value, void>::type activation(const Window &window); -#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC - /** Function to apply an activation function on a tensor. - * - * @param[in] window Region on which to execute the kernel - */ - template <ActivationLayerInfo::ActivationFunction F, typename T> - typename std::enable_if<std::is_same<T, float16_t>::value, void>::type activation(const Window &window); -#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ + typename std::enable_if<arm_compute::utils::traits::is_floating_point<T>::value, void>::type + activation(const Window &window); /** Function to apply an activation function on a tensor. * * @param[in] window Region on which to execute the kernel diff --git a/arm_compute/core/NEON/kernels/NEArithmeticAdditionKernel.h b/arm_compute/core/NEON/kernels/NEArithmeticAdditionKernel.h index 73beca6ded..872c3a5b6b 100644 --- a/arm_compute/core/NEON/kernels/NEArithmeticAdditionKernel.h +++ b/arm_compute/core/NEON/kernels/NEArithmeticAdditionKernel.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2018 ARM Limited. + * Copyright (c) 2016-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -84,7 +84,6 @@ public: // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; - BorderSize border_size() const override; private: /** Common signature for all the specialised add functions @@ -92,14 +91,16 @@ private: * @param[in] input1 An input tensor. Data types supported: U8/QASYMM8/S16/F16/F32 * @param[in] input2 An input tensor. Data types supported: U8/QASYMM8/S16/F16/F32 * @param[out] output The output tensor. Data types supported: U8/QASYMM8/S16/F16/F32. + * @param[in] policy Overflow policy. * @param[in] window Region on which to execute the kernel. */ - using AddFunction = void(const ITensor *input1, const ITensor *input2, ITensor *output, const Window &window); + using AddFunction = void(const ITensor *input1, const ITensor *input2, ITensor *output, ConvertPolicy policy, const Window &window); /** Add function to use for the particular tensor types passed to configure() */ AddFunction *_func; const ITensor *_input1; const ITensor *_input2; ITensor *_output; + ConvertPolicy _policy; }; } // namespace arm_compute #endif /*__ARM_COMPUTE_NEARITHMETICADDITIONKERNEL_H__ */ diff --git a/arm_compute/core/NEON/kernels/detail/NEActivationFunctionDetail.h b/arm_compute/core/NEON/kernels/detail/NEActivationFunctionDetail.h index 9344235d09..bde3ac82e7 100644 --- a/arm_compute/core/NEON/kernels/detail/NEActivationFunctionDetail.h +++ b/arm_compute/core/NEON/kernels/detail/NEActivationFunctionDetail.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018 ARM Limited. + * Copyright (c) 2018-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -139,7 +139,7 @@ struct logistic */ void operator()(ExactType &vval) { - vval = wrapper::vinv(wrapper::vadd(vone, wrapper::vexpq(wrapper::vnegq(vval)))); + vval = wrapper::vinv(wrapper::vadd(vone, wrapper::vexpq(wrapper::vneg(vval)))); } /** Vector of ones. */ |