From 82833b80b1a28ff32fe3c3b2f4996fa21a08bbbc Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Tue, 30 Jan 2018 12:14:24 +0000 Subject: COMPMID-765 : NEON Wrapper initial traits and overloads Change-Id: Iea4c4732d19e8cf9b245ac2a9f75b2aa70a5839e Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/118149 Reviewed-by: Michele DiGiorgio Reviewed-by: Anthony Barbier Tested-by: Jenkins --- src/core/NEON/kernels/NEBitwiseAndKernel.cpp | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-) (limited to 'src/core/NEON/kernels/NEBitwiseAndKernel.cpp') diff --git a/src/core/NEON/kernels/NEBitwiseAndKernel.cpp b/src/core/NEON/kernels/NEBitwiseAndKernel.cpp index 3888300899..c1e3e1f0bc 100644 --- a/src/core/NEON/kernels/NEBitwiseAndKernel.cpp +++ b/src/core/NEON/kernels/NEBitwiseAndKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016, 2017 ARM Limited. + * Copyright (c) 2016-2018 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -25,6 +25,7 @@ #include "arm_compute/core/Helpers.h" #include "arm_compute/core/ITensor.h" +#include "arm_compute/core/NEON/wrapper/wrapper.h" #include "arm_compute/core/Types.h" #include "arm_compute/core/Validate.h" @@ -32,6 +33,7 @@ #include using namespace arm_compute; +using namespace arm_compute::wrapper; namespace arm_compute { @@ -40,12 +42,14 @@ class Coordinates; namespace { -inline void bitwise_and_U8_U8_U8(const uint8_t *__restrict input1, const uint8_t *__restrict input2, uint8_t *__restrict output) +template +inline void bitwise_and(const T *__restrict input1, const T *__restrict input2, T *__restrict output) { - const uint8x16_t val1 = vld1q_u8(input1); - const uint8x16_t val2 = vld1q_u8(input2); + using type = typename wrapper::traits::neon_vector::type; + const type val1 = vloadq(static_cast(input1)); + const type val2 = vloadq(static_cast(input2)); - vst1q_u8(output, vandq_u8(val1, val2)); + vstore(static_cast(output), vand(val1, val2)); } } // namespace @@ -104,7 +108,7 @@ void NEBitwiseAndKernel::run(const Window &window, const ThreadInfo &info) execute_window_loop(window, [&](const Coordinates & id) { - bitwise_and_U8_U8_U8(input1.ptr(), input2.ptr(), output.ptr()); + bitwise_and(input1.ptr(), input2.ptr(), output.ptr()); }, input1, input2, output); } -- cgit v1.2.1