Fix Scale CTS failures

- Move back to use template for NEON implementaion to be more mantainable

Resolves: COMPMID-4112

Signed-off-by: Sheri Zhang <sheri.zhang@arm.com>
Change-Id: I27b60674d743e3d7ec5bb065898cb75cb40bcf0e
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4900
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Manuel Bottini <manuel.bottini@arm.com>

8 files changed, 66 insertions, 633 deletions

diff --git a/Android.bp b/Android.bp
index 30ed0acf17..7851eb6cee 100644
--- a/Android.bp
+++ b/Android.bp
@@ -387,9 +387,6 @@ cc_library_static {
         "src/core/NEON/kernels/convolution/winograd/winograd_transforms/weights_4x4_3x3_fp16_fp16_integers.cpp",
         "src/core/NEON/kernels/convolution/winograd/winograd_transforms/weights_4x4_3x3_fp32_fp32_integers.cpp",
         "src/core/NEON/kernels/convolution/winograd/winograd_transforms/weights_6_3_fp32_fp32_integers.cpp",
-        "src/core/NEON/kernels/scale/impl/NEON/fp16.cpp",
-        "src/core/NEON/kernels/scale/impl/NEON/fp32.cpp",
-        "src/core/NEON/kernels/scale/impl/NEON/integer.cpp",
         "src/core/NEON/kernels/scale/impl/NEON/qasymm8.cpp",
         "src/core/NEON/kernels/scale/impl/NEON/qasymm8_signed.cpp",
         "src/core/NEON/kernels/scale/impl/SVE/fp16.cpp",
diff --git a/src/core/NEON/kernels/NEScaleKernel.cpp b/src/core/NEON/kernels/NEScaleKernel.cpp
index 1fbd5fb60a..39ed6317a1 100644
--- a/src/core/NEON/kernels/NEScaleKernel.cpp
+++ b/src/core/NEON/kernels/NEScaleKernel.cpp
@@ -28,7 +28,8 @@
 #include "arm_compute/core/utils/misc/Utility.h"
 #include "src/core/AccessWindowStatic.h"
 #include "src/core/CPP/Validate.h"
-#include "src/core/NEON/kernels/scale/impl/list.h"
+#include "src/core/NEON/kernels/scale/impl/NEON/list.h"
+#include "src/core/NEON/kernels/scale/impl/SVE/list.h"
 #include "src/core/NEON/wrapper/wrapper.h"
 #include "src/core/common/Registrars.h"
 #include "src/core/helpers/AutoConfiguration.h"
@@ -93,15 +94,15 @@ static const ScaleKernel available_kernels[] =
 #else /* !defined(__ARM_FEATURE_SVE) */
 #if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)
     {
-        "fp16_neon_scale",
+        "common_neon_scale",
         [](const ScaleSelectorData & data) { return data.dt == DataType::F16; },
-        REGISTER_FP16_NEON(arm_compute::cpu::fp16_neon_scale)
+        REGISTER_FP16_NEON(arm_compute::cpu::common_neon_scale<float16_t>)
     },
 #endif /* !defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) */
     {
-        "f32_neon_scale",
+        "common_neon_scale",
         [](const ScaleSelectorData & data) { return data.dt == DataType::F32; },
-        REGISTER_FP32_NEON(arm_compute::cpu::fp32_neon_scale)
+        REGISTER_FP32_NEON(arm_compute::cpu::common_neon_scale<float>)
     },
     {
         "qasymm8_neon_scale",
@@ -114,14 +115,14 @@ static const ScaleKernel available_kernels[] =
         REGISTER_QASYMM8_SIGNED_NEON(arm_compute::cpu::qasymm8_signed_neon_scale)
     },
     {
-        "u8_neon_scale",
+        "common_neon_scale",
         [](const ScaleSelectorData & data) { return data.dt == DataType::U8; },
-        REGISTER_INTEGER_NEON(arm_compute::cpu::u8_neon_scale)
+        REGISTER_INTEGER_NEON(arm_compute::cpu::common_neon_scale<uint8_t>)
     },
     {
-        "s16_neon_scale",
+        "common_neon_scale",
         [](const ScaleSelectorData & data) { return data.dt == DataType::S16; },
-        REGISTER_INTEGER_NEON(arm_compute::cpu::s16_neon_scale)
+        REGISTER_INTEGER_NEON(arm_compute::cpu::common_neon_scale<int16_t>)
     },
 #endif /* !defined(__ARM_FEATURE_SVE) */
 };
diff --git a/src/core/NEON/kernels/scale/impl/NEON/fp16.cpp b/src/core/NEON/kernels/scale/impl/NEON/fp16.cpp
deleted file mode 100644
index 326723e06d..0000000000
--- a/src/core/NEON/kernels/scale/impl/NEON/fp16.cpp
+++ /dev/null
@@ -1,175 +0,0 @@
-/*
- * Copyright (c) 2021 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.
- */
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/ITensorPack.h"
-#include "arm_compute/core/Window.h"
-#include "src/core/NEON/NEMath.h"
-#include "src/core/NEON/wrapper/wrapper.h"
-#include "src/core/common/Validate.h"
-#include "src/core/helpers/ScaleHelpers.h"
-#include "src/core/utils/ScaleUtils.h"
-#include "support/Rounding.h"
-
-#include <arm_neon.h>
-#include <cmath>
-#include <cstddef>
-
-#if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS)
-
-namespace arm_compute
-{
-namespace
-{
-void fp16_neon_scale_nearest(const ITensor *src, ITensor *dst, const ITensor *offsets,
-                             float sampling_offset, bool align_corners, const Window &window)
-{
-    const size_t in_stride_c  = src->info()->dimension(0) + src->info()->padding().left + src->info()->padding().right;
-    const size_t in_stride_w  = src->info()->dimension(1) + src->info()->padding().top + src->info()->padding().bottom;
-    const size_t in_stride_wc = in_stride_w * in_stride_c;
-    const size_t in_dim_h     = src->info()->dimension(2);
-
-    // Compute the ratio between source height and destination height
-    const auto hr             = scale_utils::calculate_resize_ratio(in_dim_h, dst->info()->dimension(2), align_corners);
-    const auto window_start_x = static_cast<int32_t>(window.x().start());
-    const auto window_end_x   = static_cast<int32_t>(window.x().end());
-    const int  window_step_x  = 8;
-
-    Window win(window);
-    win.set(Window::DimX, Window::Dimension(0, 1, 1));
-    Iterator out(dst, win);
-
-    const uint8_t     *in_ptr_start        = src->buffer() + src->info()->offset_first_element_in_bytes();
-    const unsigned int in_stride_bytes_hwc = src->info()->strides_in_bytes()[3];
-
-    execute_window_loop(win, [&](const Coordinates & id)
-    {
-        const int32_t    offset     = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z()))) * in_stride_c;
-        const auto       in_hi      = static_cast<int>(align_corners ? utils::rounding::round_half_away_from_zero((id.z() + sampling_offset) * hr) : std::floor((id.z() + sampling_offset) * hr));
-        const int        offset_row = in_hi * in_stride_wc;
-        int32_t          x          = window_start_x;
-        const float16_t *in_ptr     = reinterpret_cast<const float16_t *>(in_ptr_start + in_stride_bytes_hwc * id[3]);
-
-        for(; x <= window_end_x - window_step_x; x += window_step_x)
-        {
-            wrapper::vstore(reinterpret_cast<float16_t *>(out.ptr()) + x,
-                            wrapper::vloadq(in_ptr + offset + offset_row + x));
-        }
-        for(; x < window_end_x; ++x)
-        {
-            *(reinterpret_cast<float16_t *>(out.ptr()) + x) = *(in_ptr + offset + offset_row + x);
-        }
-    },
-    out);
-}
-
-void fp16_neon_scale_bilinear(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
-                              BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
-                              bool align_corners, const Window &window)
-{
-    // Compute the ratio between source height and destination height
-    const auto hr = scale_utils::calculate_resize_ratio(src->info()->dimension(2), dst->info()->dimension(2), align_corners);
-
-    Iterator  out(dst, window);
-    const int in_stride_c  = src->info()->dimension(0) + src->info()->padding().left + src->info()->padding().right;
-    const int in_dim_w     = src->info()->dimension(1);
-    const int in_dim_h     = src->info()->dimension(2);
-    const int in_stride_wc = in_stride_c * (in_dim_w + src->info()->padding().top + src->info()->padding().bottom);
-
-    // Don't increment in Y and Z direction for the input tensor
-    // A pointer to the start of this plane is needed as base for the precomputed offsets
-    Window win_in(window);
-    win_in.set(Window::DimY, Window::Dimension(0, 0, 0));
-    win_in.set(Window::DimZ, Window::Dimension(0, 0, 0));
-    Iterator in(src, win_in);
-
-    if(border_mode == BorderMode::CONSTANT)
-    {
-        using ConstType = typename std::conditional<std::is_same<float16_t, float16_t>::value, half, float16_t>::type;
-
-        const float16_t const_border_value = static_cast<float16_t>(constant_border_value.get<ConstType>());
-        execute_window_loop(window, [&](const Coordinates & id)
-        {
-            const auto       offset = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto       dx_val = *reinterpret_cast<const float *>(dx->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto       dy_val = *reinterpret_cast<const float *>(dy->ptr_to_element(Coordinates(id.y(), id.z())));
-            const int32_t    in_hi  = std::floor((id.z() + sampling_offset) * hr - sampling_offset);
-            const float16_t *in_ptr = reinterpret_cast<const float16_t *>(in.ptr()) + offset * in_stride_c + in_hi * in_stride_wc;
-
-            const auto a00 = (0 <= offset && offset < in_dim_w && 0 <= in_hi && in_hi < in_dim_h) ? *in_ptr : const_border_value;
-            const auto a01 = (-1 <= offset && offset < in_dim_w - 1 && 0 <= in_hi && in_hi < in_dim_h) ? *(in_ptr + in_stride_c) : const_border_value;
-            const auto a10 = (0 <= offset && offset < in_dim_w && -1 <= in_hi && in_hi < in_dim_h - 1) ? *(in_ptr + in_stride_wc) : const_border_value;
-            const auto a11 = (-1 <= offset && offset < in_dim_w - 1 && -1 <= in_hi && in_hi < in_dim_h - 1) ? *(in_ptr + in_stride_c + in_stride_wc) : const_border_value;
-
-            *reinterpret_cast<float16_t *>(out.ptr()) = static_cast<float16_t>(scale_helpers::delta_bilinear(a00, a01, a10, a11, dx_val, dy_val));
-        },
-        in, out);
-    }
-    else if(border_mode == BorderMode::REPLICATE)
-    {
-        execute_window_loop(window, [&](const Coordinates & id)
-        {
-            const auto offset = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto dx_val = *reinterpret_cast<const float *>(dx->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto dy_val = *reinterpret_cast<const float *>(dy->ptr_to_element(Coordinates(id.y(), id.z())));
-            const int  in_hi  = std::floor((id.z() + sampling_offset) * hr - sampling_offset);
-
-            auto clamped_w  = utility::clamp<int>(offset, 0, in_dim_w - 1);
-            auto clamped_w1 = utility::clamp<int>(offset + 1, 0, in_dim_w - 1);
-            auto clamped_h  = utility::clamp<int>(in_hi, 0, in_dim_h - 1);
-            auto clamped_h1 = utility::clamp<int>(in_hi + 1, 0, in_dim_h - 1);
-
-            const auto a00 = *(reinterpret_cast<const float16_t *>(in.ptr()) + clamped_w * in_stride_c + clamped_h * in_stride_wc);
-            const auto a01 = *(reinterpret_cast<const float16_t *>(in.ptr()) + clamped_w1 * in_stride_c + clamped_h * in_stride_wc);
-            const auto a10 = *(reinterpret_cast<const float16_t *>(in.ptr()) + clamped_w * in_stride_c + clamped_h1 * in_stride_wc);
-            const auto a11 = *(reinterpret_cast<const float16_t *>(in.ptr()) + clamped_w1 * in_stride_c + clamped_h1 * in_stride_wc);
-
-            *reinterpret_cast<float16_t *>(out.ptr()) = static_cast<float16_t>(scale_helpers::delta_bilinear(a00, a01, a10, a11, dx_val, dy_val));
-        },
-        in, out);
-    }
-    else
-    {
-        ARM_COMPUTE_ERROR("Not implemented");
-    }
-}
-}
-namespace cpu
-{
-void fp16_neon_scale(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
-                     InterpolationPolicy policy, BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
-                     bool align_corners, const Window &window)
-{
-    if(policy == InterpolationPolicy::BILINEAR)
-    {
-        fp16_neon_scale_bilinear(src, dst, offsets, dx, dy, border_mode, constant_border_value, sampling_offset, align_corners, window);
-    }
-    else if(policy == InterpolationPolicy::NEAREST_NEIGHBOR)
-    {
-        fp16_neon_scale_nearest(src, dst, offsets, sampling_offset, align_corners, window);
-    }
-}
-} // namespace cpu
-} // namespace arm_compute
-
-#endif /* defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS) */
\ No newline at end of file
diff --git a/src/core/NEON/kernels/scale/impl/NEON/integer.cpp b/src/core/NEON/kernels/scale/impl/NEON/integer.cpp
deleted file mode 100644
index 7fc8693e95..0000000000
--- a/src/core/NEON/kernels/scale/impl/NEON/integer.cpp
+++ /dev/null
@@ -1,294 +0,0 @@
-/*
- * Copyright (c) 2021 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.
- */
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/ITensorPack.h"
-#include "arm_compute/core/Window.h"
-#include "src/core/NEON/NEMath.h"
-#include "src/core/NEON/wrapper/wrapper.h"
-#include "src/core/common/Validate.h"
-#include "src/core/helpers/ScaleHelpers.h"
-#include "src/core/utils/ScaleUtils.h"
-#include "support/Rounding.h"
-
-#include <arm_neon.h>
-#include <cmath>
-#include <cstddef>
-
-namespace arm_compute
-{
-namespace
-{
-void u8_neon_scale_nearest(const ITensor *src, ITensor *dst, const ITensor *offsets,
-                           float sampling_offset, bool align_corners, const Window &window)
-{
-    const size_t in_stride_c  = src->info()->dimension(0) + src->info()->padding().left + src->info()->padding().right;
-    const size_t in_stride_w  = src->info()->dimension(1) + src->info()->padding().top + src->info()->padding().bottom;
-    const size_t in_stride_wc = in_stride_w * in_stride_c;
-    const size_t in_dim_h     = src->info()->dimension(2);
-
-    // Compute the ratio between source height and destination height
-    const auto hr             = scale_utils::calculate_resize_ratio(in_dim_h, dst->info()->dimension(2), align_corners);
-    const auto window_start_x = static_cast<int32_t>(window.x().start());
-    const auto window_end_x   = static_cast<int32_t>(window.x().end());
-    const int  window_step_x  = 16;
-
-    Window win(window);
-    win.set(Window::DimX, Window::Dimension(0, 1, 1));
-    Iterator out(dst, win);
-
-    const uint8_t     *in_ptr_start        = src->buffer() + src->info()->offset_first_element_in_bytes();
-    const unsigned int in_stride_bytes_hwc = src->info()->strides_in_bytes()[3];
-
-    execute_window_loop(win, [&](const Coordinates & id)
-    {
-        const int32_t  offset     = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z()))) * in_stride_c;
-        const auto     in_hi      = static_cast<int>(align_corners ? utils::rounding::round_half_away_from_zero((id.z() + sampling_offset) * hr) : std::floor((id.z() + sampling_offset) * hr));
-        const int      offset_row = in_hi * in_stride_wc;
-        int32_t        x          = window_start_x;
-        const uint8_t *in_ptr     = reinterpret_cast<const uint8_t *>(in_ptr_start + in_stride_bytes_hwc * id[3]);
-
-        for(; x <= window_end_x - window_step_x; x += window_step_x)
-        {
-            wrapper::vstore(reinterpret_cast<uint8_t *>(out.ptr()) + x,
-                            wrapper::vloadq(in_ptr + offset + offset_row + x));
-        }
-        for(; x < window_end_x; ++x)
-        {
-            *(reinterpret_cast<uint8_t *>(out.ptr()) + x) = *(in_ptr + offset + offset_row + x);
-        }
-    },
-    out);
-}
-
-void u8_neon_scale_bilinear(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
-                            BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
-                            bool align_corners, const Window &window)
-{
-    // Compute the ratio between source height and destination height
-    const auto hr = scale_utils::calculate_resize_ratio(src->info()->dimension(2), dst->info()->dimension(2), align_corners);
-
-    Iterator  out(dst, window);
-    const int in_stride_c  = src->info()->dimension(0) + src->info()->padding().left + src->info()->padding().right;
-    const int in_dim_w     = src->info()->dimension(1);
-    const int in_dim_h     = src->info()->dimension(2);
-    const int in_stride_wc = in_stride_c * (in_dim_w + src->info()->padding().top + src->info()->padding().bottom);
-
-    // Don't increment in Y and Z direction for the input tensor
-    // A pointer to the start of this plane is needed as base for the precomputed offsets
-    Window win_in(window);
-    win_in.set(Window::DimY, Window::Dimension(0, 0, 0));
-    win_in.set(Window::DimZ, Window::Dimension(0, 0, 0));
-    Iterator in(src, win_in);
-
-    if(border_mode == BorderMode::CONSTANT)
-    {
-        const uint8_t const_border_value = static_cast<uint8_t>(constant_border_value.get<uint8_t>());
-        execute_window_loop(window, [&](const Coordinates & id)
-        {
-            const auto     offset = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto     dx_val = *reinterpret_cast<const float *>(dx->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto     dy_val = *reinterpret_cast<const float *>(dy->ptr_to_element(Coordinates(id.y(), id.z())));
-            const int32_t  in_hi  = std::floor((id.z() + sampling_offset) * hr - sampling_offset);
-            const uint8_t *in_ptr = reinterpret_cast<const uint8_t *>(in.ptr()) + offset * in_stride_c + in_hi * in_stride_wc;
-
-            const auto a00 = (0 <= offset && offset < in_dim_w && 0 <= in_hi && in_hi < in_dim_h) ? *in_ptr : const_border_value;
-            const auto a01 = (-1 <= offset && offset < in_dim_w - 1 && 0 <= in_hi && in_hi < in_dim_h) ? *(in_ptr + in_stride_c) : const_border_value;
-            const auto a10 = (0 <= offset && offset < in_dim_w && -1 <= in_hi && in_hi < in_dim_h - 1) ? *(in_ptr + in_stride_wc) : const_border_value;
-            const auto a11 = (-1 <= offset && offset < in_dim_w - 1 && -1 <= in_hi && in_hi < in_dim_h - 1) ? *(in_ptr + in_stride_c + in_stride_wc) : const_border_value;
-
-            *reinterpret_cast<uint8_t *>(out.ptr()) = static_cast<uint8_t>(scale_helpers::delta_bilinear(a00, a01, a10, a11, dx_val, dy_val));
-        },
-        in, out);
-    }
-    else if(border_mode == BorderMode::REPLICATE)
-    {
-        execute_window_loop(window, [&](const Coordinates & id)
-        {
-            const auto offset = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto dx_val = *reinterpret_cast<const float *>(dx->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto dy_val = *reinterpret_cast<const float *>(dy->ptr_to_element(Coordinates(id.y(), id.z())));
-            const int  in_hi  = std::floor((id.z() + sampling_offset) * hr - sampling_offset);
-
-            auto clamped_w  = utility::clamp<int>(offset, 0, in_dim_w - 1);
-            auto clamped_w1 = utility::clamp<int>(offset + 1, 0, in_dim_w - 1);
-            auto clamped_h  = utility::clamp<int>(in_hi, 0, in_dim_h - 1);
-            auto clamped_h1 = utility::clamp<int>(in_hi + 1, 0, in_dim_h - 1);
-
-            const auto a00 = *(reinterpret_cast<const uint8_t *>(in.ptr()) + clamped_w * in_stride_c + clamped_h * in_stride_wc);
-            const auto a01 = *(reinterpret_cast<const uint8_t *>(in.ptr()) + clamped_w1 * in_stride_c + clamped_h * in_stride_wc);
-            const auto a10 = *(reinterpret_cast<const uint8_t *>(in.ptr()) + clamped_w * in_stride_c + clamped_h1 * in_stride_wc);
-            const auto a11 = *(reinterpret_cast<const uint8_t *>(in.ptr()) + clamped_w1 * in_stride_c + clamped_h1 * in_stride_wc);
-
-            *reinterpret_cast<uint8_t *>(out.ptr()) = static_cast<uint8_t>(scale_helpers::delta_bilinear(a00, a01, a10, a11, dx_val, dy_val));
-        },
-        in, out);
-    }
-    else
-    {
-        ARM_COMPUTE_ERROR("Not implemented");
-    }
-}
-
-void s16_neon_scale_nearest(const ITensor *src, ITensor *dst, const ITensor *offsets,
-                            float sampling_offset, bool align_corners, const Window &window)
-{
-    const size_t in_stride_c  = src->info()->dimension(0) + src->info()->padding().left + src->info()->padding().right;
-    const size_t in_stride_w  = src->info()->dimension(1) + src->info()->padding().top + src->info()->padding().bottom;
-    const size_t in_stride_wc = in_stride_w * in_stride_c;
-    const size_t in_dim_h     = src->info()->dimension(2);
-
-    // Compute the ratio between source height and destination height
-    const auto hr             = scale_utils::calculate_resize_ratio(in_dim_h, dst->info()->dimension(2), align_corners);
-    const auto window_start_x = static_cast<int32_t>(window.x().start());
-    const auto window_end_x   = static_cast<int32_t>(window.x().end());
-    const int  window_step_x  = 8;
-
-    Window win(window);
-    win.set(Window::DimX, Window::Dimension(0, 1, 1));
-    Iterator out(dst, win);
-
-    const uint8_t     *in_ptr_start        = src->buffer() + src->info()->offset_first_element_in_bytes();
-    const unsigned int in_stride_bytes_hwc = src->info()->strides_in_bytes()[3];
-
-    execute_window_loop(win, [&](const Coordinates & id)
-    {
-        const int32_t  offset     = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z()))) * in_stride_c;
-        const auto     in_hi      = static_cast<int>(align_corners ? utils::rounding::round_half_away_from_zero((id.z() + sampling_offset) * hr) : std::floor((id.z() + sampling_offset) * hr));
-        const int      offset_row = in_hi * in_stride_wc;
-        int32_t        x          = window_start_x;
-        const int16_t *in_ptr     = reinterpret_cast<const int16_t *>(in_ptr_start + in_stride_bytes_hwc * id[3]);
-
-        for(; x <= window_end_x - window_step_x; x += window_step_x)
-        {
-            wrapper::vstore(reinterpret_cast<int16_t *>(out.ptr()) + x,
-                            wrapper::vloadq(in_ptr + offset + offset_row + x));
-        }
-        for(; x < window_end_x; ++x)
-        {
-            *(reinterpret_cast<int16_t *>(out.ptr()) + x) = *(in_ptr + offset + offset_row + x);
-        }
-    },
-    out);
-}
-
-void s16_neon_scale_bilinear(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
-                             BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
-                             bool align_corners, const Window &window)
-{
-    // Compute the ratio between source height and destination height
-    const auto hr = scale_utils::calculate_resize_ratio(src->info()->dimension(2), dst->info()->dimension(2), align_corners);
-
-    Iterator  out(dst, window);
-    const int in_stride_c  = src->info()->dimension(0) + src->info()->padding().left + src->info()->padding().right;
-    const int in_dim_w     = src->info()->dimension(1);
-    const int in_dim_h     = src->info()->dimension(2);
-    const int in_stride_wc = in_stride_c * (in_dim_w + src->info()->padding().top + src->info()->padding().bottom);
-
-    // Don't increment in Y and Z direction for the input tensor
-    // A pointer to the start of this plane is needed as base for the precomputed offsets
-    Window win_in(window);
-    win_in.set(Window::DimY, Window::Dimension(0, 0, 0));
-    win_in.set(Window::DimZ, Window::Dimension(0, 0, 0));
-    Iterator in(src, win_in);
-
-    if(border_mode == BorderMode::CONSTANT)
-    {
-        const int16_t const_border_value = static_cast<int16_t>(constant_border_value.get<int16_t>());
-        execute_window_loop(window, [&](const Coordinates & id)
-        {
-            const auto     offset = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto     dx_val = *reinterpret_cast<const float *>(dx->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto     dy_val = *reinterpret_cast<const float *>(dy->ptr_to_element(Coordinates(id.y(), id.z())));
-            const int32_t  in_hi  = std::floor((id.z() + sampling_offset) * hr - sampling_offset);
-            const int16_t *in_ptr = reinterpret_cast<const int16_t *>(in.ptr()) + offset * in_stride_c + in_hi * in_stride_wc;
-
-            const auto a00 = (0 <= offset && offset < in_dim_w && 0 <= in_hi && in_hi < in_dim_h) ? *in_ptr : const_border_value;
-            const auto a01 = (-1 <= offset && offset < in_dim_w - 1 && 0 <= in_hi && in_hi < in_dim_h) ? *(in_ptr + in_stride_c) : const_border_value;
-            const auto a10 = (0 <= offset && offset < in_dim_w && -1 <= in_hi && in_hi < in_dim_h - 1) ? *(in_ptr + in_stride_wc) : const_border_value;
-            const auto a11 = (-1 <= offset && offset < in_dim_w - 1 && -1 <= in_hi && in_hi < in_dim_h - 1) ? *(in_ptr + in_stride_c + in_stride_wc) : const_border_value;
-
-            *reinterpret_cast<int16_t *>(out.ptr()) = static_cast<int16_t>(scale_helpers::delta_bilinear(a00, a01, a10, a11, dx_val, dy_val));
-        },
-        in, out);
-    }
-    else if(border_mode == BorderMode::REPLICATE)
-    {
-        execute_window_loop(window, [&](const Coordinates & id)
-        {
-            const auto offset = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto dx_val = *reinterpret_cast<const float *>(dx->ptr_to_element(Coordinates(id.y(), id.z())));
-            const auto dy_val = *reinterpret_cast<const float *>(dy->ptr_to_element(Coordinates(id.y(), id.z())));
-            const int  in_hi  = std::floor((id.z() + sampling_offset) * hr - sampling_offset);
-
-            auto clamped_w  = utility::clamp<int>(offset, 0, in_dim_w - 1);
-            auto clamped_w1 = utility::clamp<int>(offset + 1, 0, in_dim_w - 1);
-            auto clamped_h  = utility::clamp<int>(in_hi, 0, in_dim_h - 1);
-            auto clamped_h1 = utility::clamp<int>(in_hi + 1, 0, in_dim_h - 1);
-
-            const auto a00 = *(reinterpret_cast<const int16_t *>(in.ptr()) + clamped_w * in_stride_c + clamped_h * in_stride_wc);
-            const auto a01 = *(reinterpret_cast<const int16_t *>(in.ptr()) + clamped_w1 * in_stride_c + clamped_h * in_stride_wc);
-            const auto a10 = *(reinterpret_cast<const int16_t *>(in.ptr()) + clamped_w * in_stride_c + clamped_h1 * in_stride_wc);
-            const auto a11 = *(reinterpret_cast<const int16_t *>(in.ptr()) + clamped_w1 * in_stride_c + clamped_h1 * in_stride_wc);
-
-            *reinterpret_cast<int16_t *>(out.ptr()) = static_cast<int16_t>(scale_helpers::delta_bilinear(a00, a01, a10, a11, dx_val, dy_val));
-        },
-        in, out);
-    }
-    else
-    {
-        ARM_COMPUTE_ERROR("Not implemented");
-    }
-}
-}
-namespace cpu
-{
-void u8_neon_scale(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
-                   InterpolationPolicy policy, BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
-                   bool align_corners, const Window &window)
-{
-    if(policy == InterpolationPolicy::BILINEAR)
-    {
-        u8_neon_scale_bilinear(src, dst, offsets, dx, dy, border_mode, constant_border_value, sampling_offset, align_corners, window);
-    }
-    else if(policy == InterpolationPolicy::NEAREST_NEIGHBOR)
-    {
-        u8_neon_scale_nearest(src, dst, offsets, sampling_offset, align_corners, window);
-    }
-}
-
-void s16_neon_scale(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
-                    InterpolationPolicy policy, BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
-                    bool align_corners, const Window &window)
-{
-    if(policy == InterpolationPolicy::BILINEAR)
-    {
-        s16_neon_scale_bilinear(src, dst, offsets, dx, dy, border_mode, constant_border_value, sampling_offset, align_corners, window);
-    }
-    else if(policy == InterpolationPolicy::NEAREST_NEIGHBOR)
-    {
-        s16_neon_scale_nearest(src, dst, offsets, sampling_offset, align_corners, window);
-    }
-}
-} // namespace cpu
-} // namespace arm_compute
\ No newline at end of file
diff --git a/src/core/NEON/kernels/scale/impl/NEON/fp32.cpp b/src/core/NEON/kernels/scale/impl/NEON/list.h
index 2e8a66123d..11cac26071 100644
--- a/src/core/NEON/kernels/scale/impl/NEON/fp32.cpp
+++ b/src/core/NEON/kernels/scale/impl/NEON/list.h
@@ -21,6 +21,9 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#ifndef SRC_CORE_NEON_KERNELS_SCALE_LIST_H
+#define SRC_CORE_NEON_KERNELS_SCALE_LIST_H
+
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensorPack.h"
 #include "arm_compute/core/Window.h"
@@ -31,16 +34,23 @@
 #include "src/core/utils/ScaleUtils.h"
 #include "support/Rounding.h"
 
-#include <arm_neon.h>
-#include <cmath>
-#include <cstddef>
-
 namespace arm_compute
 {
-namespace
+namespace cpu
 {
-void fp32_neon_scale_nearest(const ITensor *src, ITensor *dst, const ITensor *offsets,
-                             float sampling_offset, bool align_corners, const Window &window)
+#define DECLARE_SCALE_KERNEL(func_name)                                                                                         \
+    void func_name(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,              \
+                   InterpolationPolicy policy, BorderMode border_mode, PixelValue constant_border_value, float sampling_offset, \
+                   bool align_corners, const Window &window)
+
+DECLARE_SCALE_KERNEL(qasymm8_neon_scale);
+DECLARE_SCALE_KERNEL(qasymm8_signed_neon_scale);
+
+#undef DECLARE_SCALE_KERNEL
+
+template <typename T>
+void nearest_neon_scale(const ITensor *src, ITensor *dst, const ITensor *offsets, float sampling_offset,
+                        bool align_corners, const Window &window)
 {
     const size_t in_stride_c  = src->info()->dimension(0) + src->info()->padding().left + src->info()->padding().right;
     const size_t in_stride_w  = src->info()->dimension(1) + src->info()->padding().top + src->info()->padding().bottom;
@@ -51,7 +61,7 @@ void fp32_neon_scale_nearest(const ITensor *src, ITensor *dst, const ITensor *of
     const auto hr             = scale_utils::calculate_resize_ratio(in_dim_h, dst->info()->dimension(2), align_corners);
     const auto window_start_x = static_cast<int32_t>(window.x().start());
     const auto window_end_x   = static_cast<int32_t>(window.x().end());
-    const int  window_step_x  = 4;
+    const int  window_step_x  = 16 / sizeof(T);
 
     Window win(window);
     win.set(Window::DimX, Window::Dimension(0, 1, 1));
@@ -66,24 +76,25 @@ void fp32_neon_scale_nearest(const ITensor *src, ITensor *dst, const ITensor *of
         const auto    in_hi      = static_cast<int>(align_corners ? utils::rounding::round_half_away_from_zero((id.z() + sampling_offset) * hr) : std::floor((id.z() + sampling_offset) * hr));
         const int     offset_row = in_hi * in_stride_wc;
         int32_t       x          = window_start_x;
-        const float *in_ptr     = reinterpret_cast<const float *>(in_ptr_start + in_stride_bytes_hwc * id[3]);
+        const T      *in_ptr     = reinterpret_cast<const T *>(in_ptr_start + in_stride_bytes_hwc * id[3]);
 
         for(; x <= window_end_x - window_step_x; x += window_step_x)
         {
-            wrapper::vstore(reinterpret_cast<float *>(out.ptr()) + x,
+            wrapper::vstore(reinterpret_cast<T *>(out.ptr()) + x,
                             wrapper::vloadq(in_ptr + offset + offset_row + x));
         }
         for(; x < window_end_x; ++x)
         {
-            *(reinterpret_cast<float *>(out.ptr()) + x) = *(in_ptr + offset + offset_row + x);
+            *(reinterpret_cast<T *>(out.ptr()) + x) = *(in_ptr + offset + offset_row + x);
         }
     },
     out);
 }
 
-void fp32_neon_scale_bilinear(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
-                              BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
-                              bool align_corners, const Window &window)
+template <typename T>
+void bilinear_neon_scale(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
+                         BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
+                         bool align_corners, const Window &window)
 {
     // Compute the ratio between source height and destination height
     const auto hr = scale_utils::calculate_resize_ratio(src->info()->dimension(2), dst->info()->dimension(2), align_corners);
@@ -103,21 +114,26 @@ void fp32_neon_scale_bilinear(const ITensor *src, ITensor *dst, const ITensor *o
 
     if(border_mode == BorderMode::CONSTANT)
     {
-        const float const_border_value = static_cast<float>(constant_border_value.get<float>());
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+        using ConstType = typename std::conditional<std::is_same<T, float16_t>::value, half, T>::type;
+#else  /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+        using ConstType = T;
+#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+        const T const_border_value = static_cast<T>(constant_border_value.get<ConstType>());
         execute_window_loop(window, [&](const Coordinates & id)
         {
             const auto    offset = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z())));
             const auto    dx_val = *reinterpret_cast<const float *>(dx->ptr_to_element(Coordinates(id.y(), id.z())));
             const auto    dy_val = *reinterpret_cast<const float *>(dy->ptr_to_element(Coordinates(id.y(), id.z())));
             const int32_t in_hi  = std::floor((id.z() + sampling_offset) * hr - sampling_offset);
-            const float *in_ptr = reinterpret_cast<const float *>(in.ptr()) + offset * in_stride_c + in_hi * in_stride_wc;
+            const T      *in_ptr = reinterpret_cast<const T *>(in.ptr()) + offset * in_stride_c + in_hi * in_stride_wc;
 
             const auto a00 = (0 <= offset && offset < in_dim_w && 0 <= in_hi && in_hi < in_dim_h) ? *in_ptr : const_border_value;
             const auto a01 = (-1 <= offset && offset < in_dim_w - 1 && 0 <= in_hi && in_hi < in_dim_h) ? *(in_ptr + in_stride_c) : const_border_value;
             const auto a10 = (0 <= offset && offset < in_dim_w && -1 <= in_hi && in_hi < in_dim_h - 1) ? *(in_ptr + in_stride_wc) : const_border_value;
             const auto a11 = (-1 <= offset && offset < in_dim_w - 1 && -1 <= in_hi && in_hi < in_dim_h - 1) ? *(in_ptr + in_stride_c + in_stride_wc) : const_border_value;
 
-            *reinterpret_cast<float *>(out.ptr()) = static_cast<float>(scale_helpers::delta_bilinear(a00, a01, a10, a11, dx_val, dy_val));
+            *reinterpret_cast<T *>(out.ptr()) = static_cast<T>(scale_helpers::delta_bilinear(a00, a01, a10, a11, dx_val, dy_val));
         },
         in, out);
     }
@@ -135,12 +151,12 @@ void fp32_neon_scale_bilinear(const ITensor *src, ITensor *dst, const ITensor *o
             auto clamped_h  = utility::clamp<int>(in_hi, 0, in_dim_h - 1);
             auto clamped_h1 = utility::clamp<int>(in_hi + 1, 0, in_dim_h - 1);
 
-            const auto a00 = *(reinterpret_cast<const float *>(in.ptr()) + clamped_w * in_stride_c + clamped_h * in_stride_wc);
-            const auto a01 = *(reinterpret_cast<const float *>(in.ptr()) + clamped_w1 * in_stride_c + clamped_h * in_stride_wc);
-            const auto a10 = *(reinterpret_cast<const float *>(in.ptr()) + clamped_w * in_stride_c + clamped_h1 * in_stride_wc);
-            const auto a11 = *(reinterpret_cast<const float *>(in.ptr()) + clamped_w1 * in_stride_c + clamped_h1 * in_stride_wc);
+            const auto a00 = *(reinterpret_cast<const T *>(in.ptr()) + clamped_w * in_stride_c + clamped_h * in_stride_wc);
+            const auto a01 = *(reinterpret_cast<const T *>(in.ptr()) + clamped_w1 * in_stride_c + clamped_h * in_stride_wc);
+            const auto a10 = *(reinterpret_cast<const T *>(in.ptr()) + clamped_w * in_stride_c + clamped_h1 * in_stride_wc);
+            const auto a11 = *(reinterpret_cast<const T *>(in.ptr()) + clamped_w1 * in_stride_c + clamped_h1 * in_stride_wc);
 
-            *reinterpret_cast<float *>(out.ptr()) = static_cast<float>(scale_helpers::delta_bilinear(a00, a01, a10, a11, dx_val, dy_val));
+            *reinterpret_cast<T *>(out.ptr()) = static_cast<T>(scale_helpers::delta_bilinear(a00, a01, a10, a11, dx_val, dy_val));
         },
         in, out);
     }
@@ -149,21 +165,22 @@ void fp32_neon_scale_bilinear(const ITensor *src, ITensor *dst, const ITensor *o
         ARM_COMPUTE_ERROR("Not implemented");
     }
 }
-}
-namespace cpu
-{
-void fp32_neon_scale(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
-                     InterpolationPolicy policy, BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
-                     bool align_corners, const Window &window)
+
+template <typename T>
+void common_neon_scale(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
+                       InterpolationPolicy policy, BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
+                       bool align_corners, const Window &window)
 {
     if(policy == InterpolationPolicy::BILINEAR)
     {
-        fp32_neon_scale_bilinear(src, dst, offsets, dx, dy, border_mode, constant_border_value, sampling_offset, align_corners, window);
+        bilinear_neon_scale<T>(src, dst, offsets, dx, dy, border_mode, constant_border_value, sampling_offset, align_corners, window);
     }
     else if(policy == InterpolationPolicy::NEAREST_NEIGHBOR)
     {
-        fp32_neon_scale_nearest(src, dst, offsets, sampling_offset, align_corners, window);
+        nearest_neon_scale<T>(src, dst, offsets, sampling_offset, align_corners, window);
     }
 }
 } // namespace cpu
-} // namespace arm_compute
\ No newline at end of file
+} // namespace arm_compute
+
+#endif /* SRC_CORE_NEON_KERNELS_SCALE_LIST_H */
diff --git a/src/core/NEON/kernels/scale/impl/NEON/qasymm8.cpp b/src/core/NEON/kernels/scale/impl/NEON/qasymm8.cpp
index 2e5d82a65b..536ad2cd17 100644
--- a/src/core/NEON/kernels/scale/impl/NEON/qasymm8.cpp
+++ b/src/core/NEON/kernels/scale/impl/NEON/qasymm8.cpp
@@ -21,66 +21,12 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/ITensorPack.h"
-#include "arm_compute/core/Window.h"
-#include "src/core/NEON/NEMath.h"
-#include "src/core/NEON/wrapper/wrapper.h"
-#include "src/core/common/Validate.h"
-#include "src/core/helpers/ScaleHelpers.h"
-#include "src/core/utils/ScaleUtils.h"
-#include "support/Rounding.h"
-
-#include <arm_neon.h>
-#include <cmath>
-#include <cstddef>
+#include "src/core/NEON/kernels/scale/impl/NEON/list.h"
 
 namespace arm_compute
 {
 namespace
 {
-void qasymm8_neon_scale_nearest(const ITensor *src, ITensor *dst, const ITensor *offsets,
-                                float sampling_offset, bool align_corners, const Window &window)
-{
-    const size_t in_stride_c  = src->info()->dimension(0) + src->info()->padding().left + src->info()->padding().right;
-    const size_t in_stride_w  = src->info()->dimension(1) + src->info()->padding().top + src->info()->padding().bottom;
-    const size_t in_stride_wc = in_stride_w * in_stride_c;
-    const size_t in_dim_h     = src->info()->dimension(2);
-
-    // Compute the ratio between source height and destination height
-    const auto hr             = scale_utils::calculate_resize_ratio(in_dim_h, dst->info()->dimension(2), align_corners);
-    const auto window_start_x = static_cast<int32_t>(window.x().start());
-    const auto window_end_x   = static_cast<int32_t>(window.x().end());
-    const int  window_step_x  = 16;
-
-    Window win(window);
-    win.set(Window::DimX, Window::Dimension(0, 1, 1));
-    Iterator out(dst, win);
-
-    const uint8_t     *in_ptr_start        = src->buffer() + src->info()->offset_first_element_in_bytes();
-    const unsigned int in_stride_bytes_hwc = src->info()->strides_in_bytes()[3];
-
-    execute_window_loop(win, [&](const Coordinates & id)
-    {
-        const int32_t  offset     = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z()))) * in_stride_c;
-        const auto     in_hi      = static_cast<int>(align_corners ? utils::rounding::round_half_away_from_zero((id.z() + sampling_offset) * hr) : std::floor((id.z() + sampling_offset) * hr));
-        const int      offset_row = in_hi * in_stride_wc;
-        int32_t        x          = window_start_x;
-        const uint8_t *in_ptr     = reinterpret_cast<const uint8_t *>(in_ptr_start + in_stride_bytes_hwc * id[3]);
-
-        for(; x <= window_end_x - window_step_x; x += window_step_x)
-        {
-            wrapper::vstore(reinterpret_cast<uint8_t *>(out.ptr()) + x,
-                            wrapper::vloadq(in_ptr + offset + offset_row + x));
-        }
-        for(; x < window_end_x; ++x)
-        {
-            *(reinterpret_cast<uint8_t *>(out.ptr()) + x) = *(in_ptr + offset + offset_row + x);
-        }
-    },
-    out);
-}
-
 void qasymm8_neon_scale_bilinear(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
                                  BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
                                  bool align_corners, const Window &window)
@@ -196,7 +142,7 @@ void qasymm8_neon_scale(const ITensor *src, ITensor *dst, const ITensor *offsets
     }
     else if(policy == InterpolationPolicy::NEAREST_NEIGHBOR)
     {
-        qasymm8_neon_scale_nearest(src, dst, offsets, sampling_offset, align_corners, window);
+        nearest_neon_scale<uint8_t>(src, dst, offsets, sampling_offset, align_corners, window);
     }
 }
 } // namespace cpu
diff --git a/src/core/NEON/kernels/scale/impl/NEON/qasymm8_signed.cpp b/src/core/NEON/kernels/scale/impl/NEON/qasymm8_signed.cpp
index c9d71ef8df..7b1860802f 100644
--- a/src/core/NEON/kernels/scale/impl/NEON/qasymm8_signed.cpp
+++ b/src/core/NEON/kernels/scale/impl/NEON/qasymm8_signed.cpp
@@ -21,66 +21,13 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/ITensorPack.h"
-#include "arm_compute/core/Window.h"
-#include "src/core/NEON/NEMath.h"
-#include "src/core/NEON/wrapper/wrapper.h"
-#include "src/core/common/Validate.h"
-#include "src/core/helpers/ScaleHelpers.h"
-#include "src/core/utils/ScaleUtils.h"
-#include "support/Rounding.h"
 
-#include <arm_neon.h>
-#include <cmath>
-#include <cstddef>
+#include "src/core/NEON/kernels/scale/impl/NEON/list.h"
 
 namespace arm_compute
 {
 namespace
 {
-void qasymm8_signed_neon_scale_nearest(const ITensor *src, ITensor *dst, const ITensor *offsets,
-                                       float sampling_offset, bool align_corners, const Window &window)
-{
-    const size_t in_stride_c  = src->info()->dimension(0) + src->info()->padding().left + src->info()->padding().right;
-    const size_t in_stride_w  = src->info()->dimension(1) + src->info()->padding().top + src->info()->padding().bottom;
-    const size_t in_stride_wc = in_stride_w * in_stride_c;
-    const size_t in_dim_h     = src->info()->dimension(2);
-
-    // Compute the ratio between source height and destination height
-    const auto hr             = scale_utils::calculate_resize_ratio(in_dim_h, dst->info()->dimension(2), align_corners);
-    const auto window_start_x = static_cast<int32_t>(window.x().start());
-    const auto window_end_x   = static_cast<int32_t>(window.x().end());
-    const int  window_step_x  = 16;
-
-    Window win(window);
-    win.set(Window::DimX, Window::Dimension(0, 1, 1));
-    Iterator out(dst, win);
-
-    const uint8_t     *in_ptr_start        = src->buffer() + src->info()->offset_first_element_in_bytes();
-    const unsigned int in_stride_bytes_hwc = src->info()->strides_in_bytes()[3];
-
-    execute_window_loop(win, [&](const Coordinates & id)
-    {
-        const int32_t offset     = *reinterpret_cast<const int32_t *>(offsets->ptr_to_element(Coordinates(id.y(), id.z()))) * in_stride_c;
-        const auto    in_hi      = static_cast<int>(align_corners ? utils::rounding::round_half_away_from_zero((id.z() + sampling_offset) * hr) : std::floor((id.z() + sampling_offset) * hr));
-        const int     offset_row = in_hi * in_stride_wc;
-        int32_t       x          = window_start_x;
-        const int8_t *in_ptr     = reinterpret_cast<const int8_t *>(in_ptr_start + in_stride_bytes_hwc * id[3]);
-
-        for(; x <= window_end_x - window_step_x; x += window_step_x)
-        {
-            wrapper::vstore(reinterpret_cast<int8_t *>(out.ptr()) + x,
-                            wrapper::vloadq(in_ptr + offset + offset_row + x));
-        }
-        for(; x < window_end_x; ++x)
-        {
-            *(reinterpret_cast<int8_t *>(out.ptr()) + x) = *(in_ptr + offset + offset_row + x);
-        }
-    },
-    out);
-}
-
 void qasymm8_signed_neon_scale_bilinear(const ITensor *src, ITensor *dst, const ITensor *offsets, const ITensor *dx, const ITensor *dy,
                                         BorderMode border_mode, PixelValue constant_border_value, float sampling_offset,
                                         bool align_corners, const Window &window)
@@ -196,7 +143,7 @@ void qasymm8_signed_neon_scale(const ITensor *src, ITensor *dst, const ITensor *
     }
     else if(policy == InterpolationPolicy::NEAREST_NEIGHBOR)
     {
-        qasymm8_signed_neon_scale_nearest(src, dst, offsets, sampling_offset, align_corners, window);
+        nearest_neon_scale<int8_t>(src, dst, offsets, sampling_offset, align_corners, window);
     }
 }
 } // namespace cpu
diff --git a/src/core/NEON/kernels/scale/impl/list.h b/src/core/NEON/kernels/scale/impl/SVE/list.h
index eb3ba8b392..b9c3a10a78 100644
--- a/src/core/NEON/kernels/scale/impl/list.h
+++ b/src/core/NEON/kernels/scale/impl/SVE/list.h
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef SRC_CORE_NEON_KERNELS_SCALE_LIST_H
-#define SRC_CORE_NEON_KERNELS_SCALE_LIST_H
+#ifndef SRC_CORE_SVE_KERNELS_SCALE_LIST_H
+#define SRC_CORE_SVE_KERNELS_SCALE_LIST_H
 
 namespace arm_compute
 {
@@ -33,21 +33,15 @@ namespace cpu
                    InterpolationPolicy policy, BorderMode border_mode, PixelValue constant_border_value, float sampling_offset, \
                    bool align_corners, const Window &window)
 
-DECLARE_SCALE_KERNEL(fp16_neon_scale);
 DECLARE_SCALE_KERNEL(fp16_sve_scale);
-DECLARE_SCALE_KERNEL(fp32_neon_scale);
 DECLARE_SCALE_KERNEL(fp32_sve_scale);
-DECLARE_SCALE_KERNEL(s16_neon_scale);
 DECLARE_SCALE_KERNEL(s16_sve_scale);
-DECLARE_SCALE_KERNEL(u8_neon_scale);
 DECLARE_SCALE_KERNEL(u8_sve_scale);
-DECLARE_SCALE_KERNEL(qasymm8_neon_scale);
 DECLARE_SCALE_KERNEL(qasymm8_sve_scale);
-DECLARE_SCALE_KERNEL(qasymm8_signed_neon_scale);
 DECLARE_SCALE_KERNEL(qasymm8_signed_sve_scale);
 
-#undef DECLARE_ACTIVATION_KERNEL
+#undef DECLARE_SCALE_KERNEL
 } // namespace cpu
 } // namespace arm_compute
 
-#endif /* SRC_CORE_NEON_KERNELS_SCALE_LIST_H */
+#endif /* SRC_CORE_SVE_KERNELS_SCALE_LIST_H */