Port NEGEMMLowp Part 1

Details:
Port NEGEMMLowpQuantizeDownInt32ScaleKernel to CpuGemmLowpQuantizeDownInt32ScaleKernel
Port NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel to CpuGemmLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel
Port NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel to CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel
Port NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel to CpuGemmLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
Port NEGEMMLowpOutputStage functions to CpuGemmLowpOutputStage operators

Partially Resolves: COMPMID-4403

Change-Id: I6d5f45e43f35d731d564ed3b5c0e804d2a318fb1
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5833
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

8 files changed, 0 insertions, 1516 deletions

diff --git a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ScaleKernel.cpp b/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ScaleKernel.cpp
deleted file mode 100644
index 84365ba25b..0000000000
--- a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ScaleKernel.cpp
+++ /dev/null
@@ -1,320 +0,0 @@
-/*
- * Copyright (c) 2020-2021 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 "src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ScaleKernel.h"
-
-#include "arm_compute/core/Error.h"
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/ITensor.h"
-#include "arm_compute/core/Types.h"
-#include "arm_compute/core/Utils.h"
-#include "arm_compute/core/Validate.h"
-#include "arm_compute/core/Window.h"
-#include "arm_compute/core/utils/quantization/AsymmHelpers.h"
-#include "src/core/AccessWindowStatic.h"
-#include "src/core/NEON/wrapper/wrapper.h"
-#include "src/core/helpers/AutoConfiguration.h"
-#include "src/core/helpers/WindowHelpers.h"
-
-#include <arm_neon.h>
-#include <cstddef>
-#include <cstdint>
-
-namespace arm_compute
-{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::S32);
-
-    ARM_COMPUTE_RETURN_ERROR_ON(output_stage->gemmlowp_max_bound > std::get<1>(quantization::get_min_max_values_from_quantized_data_type(output_stage->output_data_type)));
-    ARM_COMPUTE_RETURN_ERROR_ON(output_stage->gemmlowp_min_bound < std::get<0>(quantization::get_min_max_values_from_quantized_data_type(output_stage->output_data_type))
-                                || output_stage->gemmlowp_min_bound > output_stage->gemmlowp_max_bound);
-
-    // Check biases if exist
-    if(bias != nullptr)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, bias);
-        ARM_COMPUTE_RETURN_ERROR_ON(bias->num_dimensions() > 1);
-        ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != bias->dimension(0));
-    }
-
-    if(output->total_size() != 0)
-    {
-        if(output->data_type() != output_stage->output_data_type && (output_stage->output_data_type == DataType::QASYMM8 || output_stage->output_data_type == DataType::QASYMM8_SIGNED))
-        {
-            ARM_COMPUTE_RETURN_ERROR_MSG("Mismatching data types");
-        }
-
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
-    }
-
-    return Status{};
-}
-
-inline void scale_input(int32x4x4_t &in_s32, int32x4_t result_offset_s32, int32_t result_mult_int)
-{
-    // Add the offset terms to GEMM's result
-    in_s32.val[0] = vaddq_s32(in_s32.val[0], result_offset_s32);
-    in_s32.val[1] = vaddq_s32(in_s32.val[1], result_offset_s32);
-    in_s32.val[2] = vaddq_s32(in_s32.val[2], result_offset_s32);
-    in_s32.val[3] = vaddq_s32(in_s32.val[3], result_offset_s32);
-
-    // Multiply by result_mult_int
-    in_s32.val[0] = vmulq_n_s32(in_s32.val[0], result_mult_int);
-    in_s32.val[1] = vmulq_n_s32(in_s32.val[1], result_mult_int);
-    in_s32.val[2] = vmulq_n_s32(in_s32.val[2], result_mult_int);
-    in_s32.val[3] = vmulq_n_s32(in_s32.val[3], result_mult_int);
-}
-
-template <typename T>
-inline typename std::enable_if<std::is_same<T, uint8_t>::value,
-       typename wrapper::traits::neon_vector<T, 16>::type>::type
-       convert_to_8bit(const int16x8x2_t in_s16)
-{
-    return wrapper::vcombine(wrapper::vqmovun(in_s16.val[0]), wrapper::vqmovun(in_s16.val[1]));
-}
-
-template <typename T>
-inline typename std::enable_if<std::is_same<T, int8_t>::value,
-       typename wrapper::traits::neon_vector<T, 16>::type>::type
-       convert_to_8bit(const int16x8x2_t in_s16)
-{
-    return wrapper::vcombine(wrapper::vqmovn(in_s16.val[0]), wrapper::vqmovn(in_s16.val[1]));
-}
-
-template <typename T>
-inline typename wrapper::traits::neon_vector<T, 16>::type finalize_quantization(int32x4x4_t &in_s32, int32x4_t result_shift_s32, typename wrapper::traits::neon_vector<T, 16>::type min,
-                                                                                typename wrapper::traits::neon_vector<T, 16>::type max)
-{
-    // Shift final result (negative value shift right)
-    in_s32.val[0] = vshlq_s32(in_s32.val[0], result_shift_s32);
-    in_s32.val[1] = vshlq_s32(in_s32.val[1], result_shift_s32);
-    in_s32.val[2] = vshlq_s32(in_s32.val[2], result_shift_s32);
-    in_s32.val[3] = vshlq_s32(in_s32.val[3], result_shift_s32);
-
-    // Convert S32 to S16
-    const int16x8x2_t in_s16 =
-    {
-        {
-            vcombine_s16(vqmovn_s32(in_s32.val[0]), vqmovn_s32(in_s32.val[1])),
-            vcombine_s16(vqmovn_s32(in_s32.val[2]), vqmovn_s32(in_s32.val[3]))
-        }
-    };
-
-    // Convert S16 to S8 or U8
-    typename wrapper::traits::neon_vector<T, 16>::type out = convert_to_8bit<T>(in_s16);
-
-    out = wrapper::vmax(out, min);
-    out = wrapper::vmin(out, max);
-
-    return out;
-}
-
-class Coordinates;
-
-template <typename T>
-void NEGEMMLowpQuantizeDownInt32ScaleKernel::run(const Window &window)
-{
-    using VectorType = typename wrapper::traits::neon_vector<T, 16>::type;
-
-    const int32x4_t result_offset_s32 = vdupq_n_s32(_output_stage->gemmlowp_offset);
-    const int32x4_t result_shift_s32  = vdupq_n_s32(-_output_stage->gemmlowp_shift);
-    const int       window_step_x     = 16;
-    const auto      window_start_x    = static_cast<int>(window.x().start());
-    const auto      window_end_x      = static_cast<int>(window.x().end());
-
-    const int clamp_min = (_is_bounded_relu) ? _output_stage->gemmlowp_min_bound : std::numeric_limits<T>::lowest();
-    const int clamp_max = (_is_bounded_relu) ? _output_stage->gemmlowp_max_bound : std::numeric_limits<T>::max();
-
-    VectorType min = wrapper::vdup_n(static_cast<T>(clamp_min), wrapper::traits::vector_128_tag{});
-    VectorType max = wrapper::vdup_n(static_cast<T>(clamp_max), wrapper::traits::vector_128_tag{});
-
-    Window win(window);
-    win.set(Window::DimX, Window::Dimension(0, 1, 1));
-
-    Iterator in(_input, win);
-    Iterator out(_output, win);
-
-    if(_bias != nullptr)
-    {
-        Window win_biases;
-        win_biases.set(Window::DimX, Window::Dimension(0, 1, 1));
-        win_biases.set(Window::DimY, Window::Dimension(0, 1, 1));
-
-        Iterator bias(_bias, win_biases);
-        execute_window_loop(win, [&](const Coordinates &)
-        {
-            // Compute 16 elements per iteration
-            int x = window_start_x;
-            for(; x <= (window_end_x - window_step_x); x += window_step_x)
-            {
-                int32x4x4_t in_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 4),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 8),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 12)
-                    }
-                };
-
-                const int32x4x4_t bias_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 4),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 8),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 12)
-                    }
-                };
-
-                // Add the bias to GEMM's result
-                in_s32.val[0] = vaddq_s32(in_s32.val[0], bias_s32.val[0]);
-                in_s32.val[1] = vaddq_s32(in_s32.val[1], bias_s32.val[1]);
-                in_s32.val[2] = vaddq_s32(in_s32.val[2], bias_s32.val[2]);
-                in_s32.val[3] = vaddq_s32(in_s32.val[3], bias_s32.val[3]);
-
-                // Add the offset terms to GEMM's result and multiply by result_mult_int
-                scale_input(in_s32, result_offset_s32, _output_stage->gemmlowp_multiplier);
-
-                wrapper::vstore(reinterpret_cast<T *>(out.ptr() + x), finalize_quantization<T>(in_s32, result_shift_s32, min, max));
-            }
-
-            // Compute left-over elements
-            for(; x < window_end_x; ++x)
-            {
-                const int bias_value = *(reinterpret_cast<const int *>(bias.ptr()) + x);
-                int       in_value   = *(reinterpret_cast<const int *>(in.ptr()) + x);
-
-                // Quantize
-                in_value = ((in_value + bias_value + _output_stage->gemmlowp_offset) * _output_stage->gemmlowp_multiplier) >> _output_stage->gemmlowp_shift;
-
-                // Store the result
-                *(out.ptr() + x) = static_cast<T>(utility::clamp<int>(in_value, clamp_min, clamp_max));
-            }
-        },
-        in, bias, out);
-    }
-    else
-    {
-        execute_window_loop(win, [&](const Coordinates &)
-        {
-            // Compute 16 elements per iteration
-            int x = window_start_x;
-            for(; x <= (window_end_x - window_step_x); x += window_step_x)
-            {
-                int32x4x4_t in_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 4),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 8),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 12)
-                    }
-                };
-
-                // Add the offset terms to GEMM's result and multiply by result_mult_int
-                scale_input(in_s32, result_offset_s32, _output_stage->gemmlowp_multiplier);
-
-                wrapper::vstore(reinterpret_cast<T *>(out.ptr() + x), finalize_quantization<T>(in_s32, result_shift_s32, min, max));
-            }
-
-            // Compute left-over elements
-            for(; x < window_end_x; ++x)
-            {
-                int in_value = *(reinterpret_cast<const int *>(in.ptr()) + x);
-
-                // Quantize
-                in_value = ((in_value + _output_stage->gemmlowp_offset) * _output_stage->gemmlowp_multiplier) >> _output_stage->gemmlowp_shift;
-
-                // Store the result
-                *(out.ptr() + x) = static_cast<T>(utility::clamp<int>(in_value, clamp_min, clamp_max));
-            }
-        },
-        in, out);
-    }
-}
-
-NEGEMMLowpQuantizeDownInt32ScaleKernel::NEGEMMLowpQuantizeDownInt32ScaleKernel()
-    : _func(nullptr), _input(nullptr), _bias(nullptr), _output(nullptr), _output_stage(nullptr), _is_bounded_relu(false)
-{
-}
-
-void NEGEMMLowpQuantizeDownInt32ScaleKernel::configure(const ITensor *input, const ITensor *bias, ITensor *output, const GEMMLowpOutputStageInfo *output_stage)
-{
-    // Perform validate step
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, output_stage);
-
-    // Output auto inizialitation if not yet initialized
-    auto_init_if_empty(*output->info(), input->info()->clone()->set_data_type(output_stage->output_data_type));
-
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(),
-                                                  (bias != nullptr) ? bias->info() : nullptr,
-                                                  output->info(),
-                                                  output_stage));
-
-    _input        = input;
-    _bias         = bias;
-    _output       = output;
-    _output_stage = output_stage;
-
-    // Configure kernel window
-    Window win = calculate_max_window(*input->info(), Steps());
-
-    INEKernel::configure(win);
-
-    // Check if we need to clamp the result using min and max
-    _is_bounded_relu = ((_output_stage->gemmlowp_min_bound != _output_stage->gemmlowp_max_bound)
-                        && !(_output_stage->gemmlowp_min_bound == std::get<0>(quantization::get_min_max_values_from_quantized_data_type(output_stage->output_data_type))
-                             && _output_stage->gemmlowp_max_bound == std::get<1>(quantization::get_min_max_values_from_quantized_data_type(output_stage->output_data_type))));
-    if(_output_stage->output_data_type == DataType::QASYMM8)
-    {
-        _func = &NEGEMMLowpQuantizeDownInt32ScaleKernel::run<uint8_t>;
-    }
-    else if(_output_stage->output_data_type == DataType::QASYMM8_SIGNED)
-    {
-        _func = &NEGEMMLowpQuantizeDownInt32ScaleKernel::run<int8_t>;
-    }
-    else
-    {
-        ARM_COMPUTE_ERROR("Data type not supported");
-    }
-}
-
-Status NEGEMMLowpQuantizeDownInt32ScaleKernel::validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, bias, output, output_stage));
-
-    return Status{};
-}
-
-void NEGEMMLowpQuantizeDownInt32ScaleKernel::run(const Window &window, const ThreadInfo &info)
-{
-    ARM_COMPUTE_UNUSED(info);
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
-
-    (this->*_func)(window);
-}
-} // namespace arm_compute
\ No newline at end of file
diff --git a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ScaleKernel.h b/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ScaleKernel.h
deleted file mode 100644
index 021ff8e2e0..0000000000
--- a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ScaleKernel.h
+++ /dev/null
@@ -1,114 +0,0 @@
-/*
- * Copyright (c) 2020-2021 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.
- */
-#ifndef ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32SCALEKERNEL_H
-#define ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32SCALEKERNEL_H
-
-#include "src/core/NEON/INEKernel.h"
-
-namespace arm_compute
-{
-class ITensor;
-
-/** Kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8/QASYMM8_SIGNED
- *
- * This kernel takes a final int32 accumulator value (the output of @ref NEGEMMLowpMatrixMultiplyKernel), and processes it to obtain the final QASYMM8/QASYMM8_SIGNED value.
- * The following computations will be performed by the kernel:
- *
- *  -# Add offset terms to final result
- *  -# Multiply each entry of result by result_mult_int
- *  -# Add bias to final result if bias tensor is not a nullptr
- *  -# Shift the int32 accumulator by result_shift
- *  -# Clamp the value between the specified min and max bounds
- *  -# Clamp the resulting int32 values:
- *  -#  -to the [0..255] range and cast to QASYMM8.
- *  -#  -to the [-128..127] range and cast to QASYMM8_SIGNED.
- *
- */
-class NEGEMMLowpQuantizeDownInt32ScaleKernel : public INEKernel
-{
-public:
-    const char *name() const override
-    {
-        return "NEGEMMLowpQuantizeDownInt32ScaleKernel";
-    }
-    /** Constructor */
-    NEGEMMLowpQuantizeDownInt32ScaleKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers)*/
-    NEGEMMLowpQuantizeDownInt32ScaleKernel(const NEGEMMLowpQuantizeDownInt32ScaleKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers)*/
-    NEGEMMLowpQuantizeDownInt32ScaleKernel &operator=(const NEGEMMLowpQuantizeDownInt32ScaleKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    NEGEMMLowpQuantizeDownInt32ScaleKernel(NEGEMMLowpQuantizeDownInt32ScaleKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    NEGEMMLowpQuantizeDownInt32ScaleKernel &operator=(NEGEMMLowpQuantizeDownInt32ScaleKernel &&) = default;
-    /** Default destructor */
-    ~NEGEMMLowpQuantizeDownInt32ScaleKernel() = default;
-    /** Initialise the kernel's input and output.
-     *
-     * @param[in]  input        Input tensor. Data type supported: S32
-     * @param[in]  bias         Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
-     *                          Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
-     * @param[out] output       Output tensor. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED
-     * @param[out] output_stage GEMMLowp output stage metadata.
-     */
-    void configure(const ITensor *input, const ITensor *bias, ITensor *output, const GEMMLowpOutputStageInfo *output_stage);
-    /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ScaleKernel
-     *
-     * @param[in]  input        Input tensor. Data type supported: S32
-     * @param[in]  bias         Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
-     *                          Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
-     * @param[in]  output       Output tensor. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED
-     * @param[out] output_stage GEMMLowp output stage metadata.
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage);
-
-    // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
-
-private:
-    /** Template function to run the NEGEMMLowpQuantizeDownInt32ScaleKernel
-     *
-     * @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
-     */
-    template <typename T>
-    void run(const Window &window);
-
-    /** Common signature for all the specialised NEGEMMLowpQuantizeDownInt32ScaleKernel functions
-     *
-     * @param[in] window Region on which to execute the kernel.
-     */
-    using QuantizeDownFunctionPtr = void (NEGEMMLowpQuantizeDownInt32ScaleKernel::*)(const Window &window);
-
-    QuantizeDownFunctionPtr        _func;
-    const ITensor                 *_input;
-    const ITensor                 *_bias;
-    ITensor                       *_output;
-    const GEMMLowpOutputStageInfo *_output_stage;
-    bool                           _is_bounded_relu;
-};
-} // namespace arm_compute
-
-#endif /* ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32SCALEKERNEL_H */
diff --git a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.cpp b/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.cpp
deleted file mode 100644
index aa54b80436..0000000000
--- a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.cpp
+++ /dev/null
@@ -1,234 +0,0 @@
-/*
- * Copyright (c) 2019-2021 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 "src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h"
-
-#include "arm_compute/core/Error.h"
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/ITensor.h"
-#include "arm_compute/core/TensorInfo.h"
-#include "arm_compute/core/Types.h"
-#include "arm_compute/core/Utils.h"
-#include "arm_compute/core/Validate.h"
-#include "arm_compute/core/Window.h"
-#include "arm_compute/core/utils/misc/ShapeCalculator.h"
-#include "src/core/NEON/NESymm.h"
-#include "src/core/helpers/AutoConfiguration.h"
-#include "src/core/helpers/WindowHelpers.h"
-
-#include <arm_neon.h>
-#include <cstddef>
-#include <cstdint>
-
-namespace arm_compute
-{
-namespace
-{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min, int max)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::S32);
-    ARM_COMPUTE_RETURN_ERROR_ON(min > max);
-
-    // Check biases if exist
-    if(bias != nullptr)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, bias);
-        ARM_COMPUTE_RETURN_ERROR_ON(bias->num_dimensions() > 1);
-        ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != bias->dimension(0));
-    }
-
-    if(output->total_size() != 0)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::QSYMM16);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, input);
-    }
-
-    return Status{};
-}
-
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
-{
-    // Output auto inizialitation if not yet initialized
-    auto_init_if_empty(*output, input->clone()->set_data_type(DataType::QSYMM16));
-
-    // Configure kernel window
-    Window win = calculate_max_window(*input, Steps());
-
-    // NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel doesn't need padding so update_window_and_padding() can be skipped
-
-    return std::make_pair(Status{}, win);
-}
-} // namespace
-
-class Coordinates;
-
-template <bool is_bounded_relu>
-void NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::run(const Window &window)
-{
-    const int16x8_t min_s16 = vdupq_n_s16(static_cast<int16_t>(_min));
-    const int16x8_t max_s16 = vdupq_n_s16(static_cast<int16_t>(_max));
-
-    ARM_COMPUTE_UNUSED(min_s16);
-    ARM_COMPUTE_UNUSED(max_s16);
-
-    const int  window_step_x  = 8;
-    const auto window_start_x = static_cast<int>(window.x().start());
-    const auto window_end_x   = static_cast<int>(window.x().end());
-
-    Window win_collapsed = window.collapse_if_possible(window, Window::DimZ);
-    win_collapsed.set(Window::DimX, Window::Dimension(0, 1, 1));
-
-    Iterator in(_input, win_collapsed);
-    Iterator out(_output, win_collapsed);
-    if(_bias != nullptr)
-    {
-        Window win_biases;
-        win_biases.set(Window::DimX, Window::Dimension(0, 1, 1));
-        win_biases.set(Window::DimY, Window::Dimension(0, 1, 1));
-
-        Iterator bias(_bias, win_biases);
-        execute_window_loop(win_collapsed, [&](const Coordinates &)
-        {
-            // Compute 16 elements per iteration
-            int x = window_start_x;
-            for(; x <= (window_end_x - window_step_x); x += window_step_x)
-            {
-                int32x4x2_t in_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 4)
-                    }
-                };
-
-                const int32x4x2_t bias_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 4)
-                    }
-                };
-
-                // Add the bias to GEMM's result
-                in_s32.val[0] = vaddq_s32(in_s32.val[0], bias_s32.val[0]);
-                in_s32.val[1] = vaddq_s32(in_s32.val[1], bias_s32.val[1]);
-
-                vst1q_s16(reinterpret_cast<int16_t *>(out.ptr()) + x, finalize_quantization_int16<is_bounded_relu>(in_s32, _result_fixedpoint_multiplier, _result_shift, min_s16, max_s16));
-            }
-
-            // Compute left-over elements
-            for(; x < window_end_x; ++x)
-            {
-                const int32_t bias_value = *(reinterpret_cast<const int32_t *>(bias.ptr()) + x);
-                int32_t       in_value   = *(reinterpret_cast<const int32_t *>(in.ptr()) + x);
-
-                // Add bias
-                in_value += bias_value;
-                // Finalize and store the result
-                *(reinterpret_cast<int16_t *>(out.ptr()) + x) = finalize_quantization_int16<is_bounded_relu>(in_value, _result_fixedpoint_multiplier, _result_shift, static_cast<int16_t>(_min),
-                                                                                                             static_cast<int16_t>(_max));
-            }
-        },
-        in, out, bias);
-    }
-    else
-    {
-        execute_window_loop(win_collapsed, [&](const Coordinates &)
-        {
-            // Compute 16 elements per iteration
-            int x = window_start_x;
-            for(; x <= (window_end_x - window_step_x); x += window_step_x)
-            {
-                int32x4x2_t in_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 4)
-                    }
-                };
-
-                vst1q_s16(reinterpret_cast<int16_t *>(out.ptr()) + x, finalize_quantization_int16<is_bounded_relu>(in_s32, _result_fixedpoint_multiplier, _result_shift, min_s16, max_s16));
-            }
-
-            // Compute left-over elements
-            for(; x < window_end_x; ++x)
-            {
-                const int32_t in_value = *(reinterpret_cast<const int32_t *>(in.ptr()) + x);
-                ARM_COMPUTE_UNUSED(in_value);
-                // Finalize and store the result
-                *(reinterpret_cast<int16_t *>(out.ptr()) + x) = finalize_quantization_int16<is_bounded_relu>(in_value, _result_fixedpoint_multiplier, _result_shift, static_cast<int16_t>(_min),
-                                                                                                             static_cast<int16_t>(_max));
-            }
-        },
-        in, out);
-    }
-}
-
-NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel()
-    : _func(nullptr), _input(nullptr), _bias(nullptr), _output(nullptr), _result_fixedpoint_multiplier(0), _result_shift(0), _min(0), _max(0)
-{
-}
-
-void NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift,
-                                                                          int min, int max)
-{
-    // Perform validate step
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (bias != nullptr) ? bias->info() : nullptr, output->info(), min, max));
-
-    _input                        = input;
-    _bias                         = bias;
-    _output                       = output;
-    _result_fixedpoint_multiplier = result_fixedpoint_multiplier;
-    _result_shift                 = result_shift;
-    _min                          = min;
-    _max                          = max;
-
-    // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), output->info());
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
-    INEKernel::configure(win_config.second);
-
-    // Check if we need to clamp the result using min and max
-    const bool is_bounded_relu = !(min <= -32768 && max >= 32767);
-    _func                      = is_bounded_relu ? &NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::run<true> : &NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::run<false>;
-}
-
-Status NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min, int max)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, bias, output, min, max));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);
-
-    return Status{};
-}
-
-void NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::run(const Window &window, const ThreadInfo &info)
-{
-    ARM_COMPUTE_UNUSED(info);
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
-
-    (this->*_func)(window);
-}
-} // namespace arm_compute
diff --git a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h b/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h
deleted file mode 100644
index b01b204a6f..0000000000
--- a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h
+++ /dev/null
@@ -1,118 +0,0 @@
-/*
- * Copyright (c) 2019-2021 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.
- */
-#ifndef ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32TOINT16SCALEBYFIXEDPOINTKERNEL_H
-#define ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32TOINT16SCALEBYFIXEDPOINTKERNEL_H
-
-#include "src/core/NEON/INEKernel.h"
-
-namespace arm_compute
-{
-class ITensor;
-
-/** Kernel used to quantize down the int32 accumulator values of GEMMLowp to QSYMM16
- *
- * This kernel takes a final int32 accumulator value (the output of @ref NEGEMMLowpMatrixMultiplyKernel), and processes it to obtain the final QSYMM16 value.
- * The following computations will be performed by the kernel:
- *
- *  -# Compute fixed point multiplication between each entry of input by result_fixedpoint_multiplier
- *  -# Add bias to final result if bias tensor is not a nullptr
- *  -# Round to nearest division by a power-of-two using result_shift
- *  -# Clamp the value between the specified min and max bounds
- *  -# Clamp the resulting int32 values to the [-32768, 32767] range and cast to QSYMM16.
- *
- */
-class NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel : public INEKernel
-{
-public:
-    const char *name() const override
-    {
-        return "NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel";
-    }
-    /** Constructor */
-    NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers)*/
-    NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel(const NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers)*/
-    NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &operator=(const NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel(NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &operator=(NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &&) = default;
-    /** Default destructor */
-    ~NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel() = default;
-    /** Initialise the kernel's input and output.
-     *
-     * @param[in]  input                        Input tensor. Data type supported: S32
-     * @param[in]  bias                         Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
-     *                                          Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
-     * @param[out] output                       Output tensor. Data type supported: Data type supported: QSYMM16
-     * @param[in]  result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
-     * @param[in]  result_shift                 Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication
-     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to 0.
-     * @param[in]  max                          (Optional) Max value used to saturate up the output result before converting back to QSYMM16.
-     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to 0.
-     */
-    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift, int min = 0, int max = 0);
-    /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel
-     *
-     * @param[in] input  Input tensor info. Data type supported: S32
-     * @param[in] bias   Biases tensor info. Only shared biases supported and it can be a nullptr if the biases addition is not required.
-     *                   Biases are 1D tensor info with dimensions [OFM]. Data type supported: Same as @p input.
-     * @param[in] output Output tensor info. Data type supported: Data type supported: QSYMM16
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to 0.
-     * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QSYMM16,
-     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to 0.
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
-
-    // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
-
-private:
-    /** Template function to run the NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel
-     *
-     * @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
-     */
-    template <bool is_bounded_relu>
-    void run(const Window &window);
-
-    /** Common signature for all the specialised NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel functions
-     *
-     * @param[in] window Region on which to execute the kernel.
-     */
-    using QuantizeDownFunctionPtr = void (NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::*)(const Window &window);
-
-    QuantizeDownFunctionPtr _func;
-    const ITensor          *_input;
-    const ITensor          *_bias;
-    ITensor                *_output;
-    int                     _result_fixedpoint_multiplier;
-    int                     _result_shift;
-    int                     _min;
-    int                     _max;
-};
-} // namespace arm_compute
-#endif /* ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32TOINT16SCALEBYFIXEDPOINTKERNEL_H */
diff --git a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.cpp b/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.cpp
deleted file mode 100644
index 9ed85e62aa..0000000000
--- a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.cpp
+++ /dev/null
@@ -1,243 +0,0 @@
-/*
- * Copyright (c) 2019-2021 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 "src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h"
-
-#include "arm_compute/core/Error.h"
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/ITensor.h"
-#include "arm_compute/core/TensorInfo.h"
-#include "arm_compute/core/Types.h"
-#include "arm_compute/core/Utils.h"
-#include "arm_compute/core/Validate.h"
-#include "arm_compute/core/Window.h"
-#include "arm_compute/core/utils/misc/ShapeCalculator.h"
-#include "src/core/NEON/NEAsymm.h"
-#include "src/core/helpers/AutoConfiguration.h"
-#include "src/core/helpers/WindowHelpers.h"
-
-#include <arm_neon.h>
-#include <cstddef>
-#include <cstdint>
-
-namespace arm_compute
-{
-namespace
-{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min, int max)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::S32);
-    ARM_COMPUTE_RETURN_ERROR_ON(min > max);
-
-    // Check biases if exist
-    if(bias != nullptr)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, bias);
-        ARM_COMPUTE_RETURN_ERROR_ON(bias->num_dimensions() > 1);
-        ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != bias->dimension(0));
-    }
-
-    if(output->total_size() != 0)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::QASYMM8_SIGNED);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, input);
-    }
-
-    return Status{};
-}
-
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
-{
-    // Output auto initialization if not yet initialized
-    auto_init_if_empty(*output, input->clone()->set_data_type(DataType::QASYMM8_SIGNED));
-
-    // Configure kernel window
-    Window win = calculate_max_window(*input, Steps());
-
-    // NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel doesn't need padding so update_window_and_padding() can be skipped
-
-    return std::make_pair(Status{}, win);
-}
-} // namespace
-
-template <bool is_bounded_relu>
-void NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::run(const Window &window)
-{
-    const int32x4_t result_offset_after_shift_s32 = vdupq_n_s32(_result_offset_after_shift);
-    const int8x16_t min_s8                        = vdupq_n_s8(static_cast<int8_t>(_min));
-    const int8x16_t max_s8                        = vdupq_n_s8(static_cast<int8_t>(_max));
-
-    ARM_COMPUTE_UNUSED(min_s8, max_s8);
-
-    const int  window_step_x  = 16;
-    const auto window_start_x = static_cast<int>(window.x().start());
-    const auto window_end_x   = static_cast<int>(window.x().end());
-
-    Window win_collapsed = window.collapse_if_possible(window, Window::DimZ);
-    win_collapsed.set(Window::DimX, Window::Dimension(0, 1, 1));
-
-    Iterator in(_input, win_collapsed);
-    Iterator out(_output, win_collapsed);
-    if(_bias != nullptr)
-    {
-        Window win_biases;
-        win_biases.set(Window::DimX, Window::Dimension(0, 1, 1));
-        win_biases.set(Window::DimY, Window::Dimension(0, 1, 1));
-
-        Iterator bias(_bias, win_biases);
-        execute_window_loop(win_collapsed, [&](const Coordinates &)
-        {
-            // Compute 16 elements per iteration
-            int x = window_start_x;
-            for(; x <= (window_end_x - window_step_x); x += window_step_x)
-            {
-                int32x4x4_t in_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 4),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 8),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 12)
-                    }
-                };
-
-                const int32x4x4_t bias_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 4),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 8),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 12)
-                    }
-                };
-
-                // Add the bias to GEMM's result
-                in_s32.val[0] = vaddq_s32(in_s32.val[0], bias_s32.val[0]);
-                in_s32.val[1] = vaddq_s32(in_s32.val[1], bias_s32.val[1]);
-                in_s32.val[2] = vaddq_s32(in_s32.val[2], bias_s32.val[2]);
-                in_s32.val[3] = vaddq_s32(in_s32.val[3], bias_s32.val[3]);
-
-                vst1q_s8(reinterpret_cast<int8_t *>(out.ptr() + x),
-                         finalize_quantization(in_s32, _result_fixedpoint_multiplier, _result_shift, result_offset_after_shift_s32, min_s8, max_s8, is_bounded_relu));
-            }
-
-            // Compute left-over elements
-            for(; x < window_end_x; ++x)
-            {
-                const int32_t bias_value = *(reinterpret_cast<const int32_t *>(bias.ptr()) + x);
-                int32_t       in_value   = *(reinterpret_cast<const int32_t *>(in.ptr()) + x);
-
-                // Add bias
-                in_value += bias_value;
-                // Finalize and store the result
-                *reinterpret_cast<int8_t *>(out.ptr() + x) = finalize_quantization(in_value, _result_fixedpoint_multiplier, _result_shift, _result_offset_after_shift,
-                                                                                   static_cast<int8_t>(_min), static_cast<int8_t>(_max), is_bounded_relu);
-            }
-        },
-        in, out, bias);
-    }
-    else
-    {
-        execute_window_loop(win_collapsed, [&](const Coordinates &)
-        {
-            // Compute 16 elements per iteration
-            int x = window_start_x;
-            for(; x <= (window_end_x - window_step_x); x += window_step_x)
-            {
-                int32x4x4_t in_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 4),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 8),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 12)
-                    }
-                };
-
-                vst1q_s8(reinterpret_cast<int8_t *>(out.ptr() + x),
-                         finalize_quantization(in_s32, _result_fixedpoint_multiplier, _result_shift, result_offset_after_shift_s32, min_s8, max_s8, is_bounded_relu));
-            }
-
-            // Compute left-over elements
-            for(; x < window_end_x; ++x)
-            {
-                const int32_t in_value = *(reinterpret_cast<const int32_t *>(in.ptr()) + x);
-
-                // Finalize and store the result
-                *reinterpret_cast<int8_t *>(out.ptr() + x) = finalize_quantization(in_value, _result_fixedpoint_multiplier, _result_shift, _result_offset_after_shift,
-                                                                                   static_cast<int8_t>(_min), static_cast<int8_t>(_max), is_bounded_relu);
-            }
-        },
-        in, out);
-    }
-}
-
-NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel()
-    : _func(nullptr), _input(nullptr), _bias(nullptr), _output(nullptr), _result_fixedpoint_multiplier(0), _result_shift(0), _result_offset_after_shift(0), _min(0), _max(0)
-{
-}
-
-void NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift,
-                                                                         int result_offset_after_shift, int min, int max)
-{
-    // Perform validate step
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (bias != nullptr) ? bias->info() : nullptr, output->info(), min, max));
-
-    _input                        = input;
-    _bias                         = bias;
-    _output                       = output;
-    _result_fixedpoint_multiplier = result_fixedpoint_multiplier;
-    _result_shift                 = result_shift;
-    _result_offset_after_shift    = result_offset_after_shift;
-    _min                          = min;
-    _max                          = max;
-
-    // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), output->info());
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
-    INEKernel::configure(win_config.second);
-
-    // Check if we need to clamp the result using min and max
-    const bool is_bounded_relu = !(min <= -128 && max >= 127);
-    _func                      = is_bounded_relu ? &NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::run<true> : &NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::run<false>;
-}
-
-Status NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min, int max)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, bias, output, min, max));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);
-
-    return Status{};
-}
-
-void NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::run(const Window &window, const ThreadInfo &info)
-{
-    ARM_COMPUTE_UNUSED(info);
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
-
-    (this->*_func)(window);
-}
-} // namespace arm_compute
diff --git a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h b/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h
deleted file mode 100644
index 9e7dc2f599..0000000000
--- a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h
+++ /dev/null
@@ -1,121 +0,0 @@
-/*
- * Copyright (c) 2019-2021 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.
- */
-#ifndef ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32TOINT8SCALEBYFIXEDPOINTKERNEL_H
-#define ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32TOINT8SCALEBYFIXEDPOINTKERNEL_H
-
-#include "src/core/NEON/INEKernel.h"
-
-namespace arm_compute
-{
-class ITensor;
-
-/** Kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8_SIGNED
- *
- * This kernel takes a final int32 accumulator value (the output of @ref NEGEMMLowpMatrixMultiplyKernel), and processes it to obtain the final QASYMM8_SIGNED value.
- * The following computations will be performed by the kernel:
- *
- *  -# Compute fixed point multiplication between each entry of input by result_fixedpoint_multiplier
- *  -# Add bias to final result if bias tensor is not a nullptr
- *  -# Round to nearest division by a power-of-two using result_shift
- *  -# Add offset to each result
- *  -# Clamp the value between the specified min and max bounds
- *  -# Clamp the resulting int32 values to the [-128..127] range and cast to QASYMM8_SIGNED.
- *
- */
-class NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel : public INEKernel
-{
-public:
-    const char *name() const override
-    {
-        return "NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel";
-    }
-    /** Constructor */
-    NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers)*/
-    NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel(const NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers)*/
-    NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &operator=(const NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel(NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &operator=(NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &&) = default;
-    /** Default destructor */
-    ~NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel() = default;
-    /** Initialise the kernel's input and output.
-     *
-     * @param[in]  input                        Input tensor. Data type supported: S32
-     * @param[in]  bias                         Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
-     *                                          Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
-     * @param[out] output                       Output tensor. Data type supported: Data type supported: QASYMM8_SIGNED
-     * @param[in]  result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
-     * @param[in]  result_shift                 Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication
-     * @param[in]  result_offset_after_shift    Offset to be applied to result before converting it back to QASYMM8_SIGNED
-     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED
-     * @param[in]  max                          (Optional) Max value used to saturate up the output result before converting back to QASYMM8_SIGNED,
-     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions
-     */
-    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift, int min = 0, int max = 0);
-    /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
-     *
-     * @param[in] input  Input tensor. Data type supported: S32
-     * @param[in] bias   Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
-     *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
-     * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8_SIGNED
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED
-     * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QASYMM8_SIGNED,
-     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
-
-    // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
-
-private:
-    /** Template function to run the NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel
-     *
-     * @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
-     */
-    template <bool is_bounded_relu>
-    void run(const Window &window);
-
-    /** Common signature for all the specialised NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel functions
-     *
-     * @param[in] window Region on which to execute the kernel.
-     */
-    using QuantizeDownFunctionPtr = void (NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::*)(const Window &window);
-
-    QuantizeDownFunctionPtr _func;
-    const ITensor          *_input;
-    const ITensor          *_bias;
-    ITensor                *_output;
-    int                     _result_fixedpoint_multiplier;
-    int                     _result_shift;
-    int                     _result_offset_after_shift;
-    int                     _min;
-    int                     _max;
-};
-} // namespace arm_compute
-#endif /* ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32TOINT8SCALEBYFIXEDPOINTKERNEL_H */
diff --git a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.cpp b/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.cpp
deleted file mode 100644
index 83ca6f944d..0000000000
--- a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.cpp
+++ /dev/null
@@ -1,245 +0,0 @@
-/*
- * Copyright (c) 2017-2021 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 "src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h"
-
-#include "arm_compute/core/Error.h"
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/ITensor.h"
-#include "arm_compute/core/TensorInfo.h"
-#include "arm_compute/core/Types.h"
-#include "arm_compute/core/Utils.h"
-#include "arm_compute/core/Validate.h"
-#include "arm_compute/core/Window.h"
-#include "arm_compute/core/utils/misc/ShapeCalculator.h"
-#include "src/core/NEON/NEAsymm.h"
-#include "src/core/helpers/AutoConfiguration.h"
-#include "src/core/helpers/WindowHelpers.h"
-
-#include <arm_neon.h>
-#include <cstddef>
-#include <cstdint>
-
-namespace arm_compute
-{
-namespace
-{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min, int max)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::S32);
-    ARM_COMPUTE_RETURN_ERROR_ON(min > max);
-
-    // Check biases if exist
-    if(bias != nullptr)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, bias);
-        ARM_COMPUTE_RETURN_ERROR_ON(bias->num_dimensions() > 1);
-        ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != bias->dimension(0));
-    }
-
-    if(output->total_size() != 0)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::QASYMM8);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, input);
-    }
-
-    return Status{};
-}
-
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
-{
-    // Output auto inizialitation if not yet initialized
-    auto_init_if_empty(*output, input->clone()->set_data_type(DataType::QASYMM8));
-
-    // Configure kernel window
-    Window win = calculate_max_window(*input, Steps());
-
-    // NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel doesn't need padding so update_window_and_padding() can be skipped
-
-    return std::make_pair(Status{}, win);
-}
-} // namespace
-
-namespace arm_compute
-{
-class Coordinates;
-} // namespace arm_compute
-
-template <bool is_bounded_relu>
-void NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::run(const Window &window)
-{
-    const int32x4_t  result_offset_after_shift_s32 = vdupq_n_s32(_result_offset_after_shift);
-    const uint8x16_t min_u8                        = vdupq_n_u8(static_cast<uint8_t>(_min));
-    const uint8x16_t max_u8                        = vdupq_n_u8(static_cast<uint8_t>(_max));
-
-    ARM_COMPUTE_UNUSED(min_u8);
-    ARM_COMPUTE_UNUSED(max_u8);
-
-    const int  window_step_x  = 16;
-    const auto window_start_x = static_cast<int>(window.x().start());
-    const auto window_end_x   = static_cast<int>(window.x().end());
-
-    Window win_collapsed = window.collapse_if_possible(window, Window::DimZ);
-    win_collapsed.set(Window::DimX, Window::Dimension(0, 1, 1));
-
-    Iterator in(_input, win_collapsed);
-    Iterator out(_output, win_collapsed);
-    if(_bias != nullptr)
-    {
-        Window win_biases;
-        win_biases.set(Window::DimX, Window::Dimension(0, 1, 1));
-        win_biases.set(Window::DimY, Window::Dimension(0, 1, 1));
-
-        Iterator bias(_bias, win_biases);
-        execute_window_loop(win_collapsed, [&](const Coordinates &)
-        {
-            // Compute 16 elements per iteration
-            int x = window_start_x;
-            for(; x <= (window_end_x - window_step_x); x += window_step_x)
-            {
-                int32x4x4_t in_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 4),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 8),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 12)
-                    }
-                };
-
-                const int32x4x4_t bias_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 4),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 8),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(bias.ptr()) + x + 12)
-                    }
-                };
-
-                // Add the bias to GEMM's result
-                in_s32.val[0] = vaddq_s32(in_s32.val[0], bias_s32.val[0]);
-                in_s32.val[1] = vaddq_s32(in_s32.val[1], bias_s32.val[1]);
-                in_s32.val[2] = vaddq_s32(in_s32.val[2], bias_s32.val[2]);
-                in_s32.val[3] = vaddq_s32(in_s32.val[3], bias_s32.val[3]);
-
-                vst1q_u8(out.ptr() + x, finalize_quantization(in_s32, _result_fixedpoint_multiplier, _result_shift, result_offset_after_shift_s32, min_u8, max_u8, is_bounded_relu));
-            }
-
-            // Compute left-over elements
-            for(; x < window_end_x; ++x)
-            {
-                const int32_t bias_value = *(reinterpret_cast<const int32_t *>(bias.ptr()) + x);
-                int32_t       in_value   = *(reinterpret_cast<const int32_t *>(in.ptr()) + x);
-
-                // Add bias
-                in_value += bias_value;
-                // Finalize and store the result
-                *(out.ptr() + x) = finalize_quantization(in_value, _result_fixedpoint_multiplier, _result_shift, _result_offset_after_shift, static_cast<uint8_t>(_min), static_cast<uint8_t>(_max), is_bounded_relu);
-            }
-        },
-        in, out, bias);
-    }
-    else
-    {
-        execute_window_loop(win_collapsed, [&](const Coordinates &)
-        {
-            // Compute 16 elements per iteration
-            int x = window_start_x;
-            for(; x <= (window_end_x - window_step_x); x += window_step_x)
-            {
-                int32x4x4_t in_s32 =
-                {
-                    {
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 0),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 4),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 8),
-                        vld1q_s32(reinterpret_cast<const int32_t *>(in.ptr()) + x + 12)
-                    }
-                };
-
-                vst1q_u8(out.ptr() + x, finalize_quantization(in_s32, _result_fixedpoint_multiplier, _result_shift, result_offset_after_shift_s32, min_u8, max_u8, is_bounded_relu));
-            }
-
-            // Compute left-over elements
-            for(; x < window_end_x; ++x)
-            {
-                const int32_t in_value = *(reinterpret_cast<const int32_t *>(in.ptr()) + x);
-
-                // Finalize and store the result
-                *(out.ptr() + x) = finalize_quantization(in_value, _result_fixedpoint_multiplier, _result_shift, _result_offset_after_shift, static_cast<uint8_t>(_min), static_cast<uint8_t>(_max), is_bounded_relu);
-            }
-        },
-        in, out);
-    }
-}
-
-NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel()
-    : _func(nullptr), _input(nullptr), _bias(nullptr), _output(nullptr), _result_fixedpoint_multiplier(0), _result_shift(0), _result_offset_after_shift(0), _min(0), _max(0)
-{
-}
-
-void NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift,
-                                                                          int result_offset_after_shift, int min, int max)
-{
-    // Perform validate step
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (bias != nullptr) ? bias->info() : nullptr, output->info(), min, max));
-
-    _input                        = input;
-    _bias                         = bias;
-    _output                       = output;
-    _result_fixedpoint_multiplier = result_fixedpoint_multiplier;
-    _result_shift                 = result_shift;
-    _result_offset_after_shift    = result_offset_after_shift;
-    _min                          = min;
-    _max                          = max;
-
-    // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), output->info());
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
-    INEKernel::configure(win_config.second);
-
-    // Check if we need to clamp the result using min and max
-    const bool is_bounded_relu = !(min <= 0 && max >= 255);
-    _func                      = is_bounded_relu ? &NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::run<true> : &NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::run<false>;
-}
-
-Status NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min, int max)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, bias, output, min, max));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);
-
-    return Status{};
-}
-
-void NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::run(const Window &window, const ThreadInfo &info)
-{
-    ARM_COMPUTE_UNUSED(info);
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
-
-    (this->*_func)(window);
-}
-} // namespace arm_compute
diff --git a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h b/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h
deleted file mode 100644
index def0573967..0000000000
--- a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h
+++ /dev/null
@@ -1,121 +0,0 @@
-/*
- * Copyright (c) 2017-2021 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.
- */
-#ifndef ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEBYFIXEDPOINTKERNEL_H
-#define ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEBYFIXEDPOINTKERNEL_H
-
-#include "src/core/NEON/INEKernel.h"
-
-namespace arm_compute
-{
-class ITensor;
-
-/** Kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8
- *
- * This kernel takes a final int32 accumulator value (the output of @ref NEGEMMLowpMatrixMultiplyKernel), and processes it to obtain the final QASYMM8 value.
- * The following computations will be performed by the kernel:
- *
- *  -# Compute fixed point multiplication between each entry of input by result_fixedpoint_multiplier
- *  -# Add bias to final result if bias tensor is not a nullptr
- *  -# Round to nearest division by a power-of-two using result_shift
- *  -# Add offset to each result
- *  -# Clamp the value between the specified min and max bounds
- *  -# Clamp the resulting int32 values to the [0..255] range and cast to QASYMM8.
- *
- */
-class NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel : public INEKernel
-{
-public:
-    const char *name() const override
-    {
-        return "NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel";
-    }
-    /** Constructor */
-    NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers)*/
-    NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel(const NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers)*/
-    NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &operator=(const NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel(NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &operator=(NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &&) = default;
-    /** Default destructor */
-    ~NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel() = default;
-    /** Initialise the kernel's input and output.
-     *
-     * @param[in]  input                        Input tensor. Data type supported: S32
-     * @param[in]  bias                         Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
-     *                                          Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
-     * @param[out] output                       Output tensor. Data type supported: Data type supported: QASYMM8
-     * @param[in]  result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
-     * @param[in]  result_shift                 Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication
-     * @param[in]  result_offset_after_shift    Offset to be applied to result before converting it back to QASYMM8
-     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QASYMM8
-     * @param[in]  max                          (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions
-     */
-    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift, int min = 0, int max = 0);
-    /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
-     *
-     * @param[in] input  Input tensor. Data type supported: S32
-     * @param[in] bias   Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
-     *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
-     * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8
-     * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
-
-    // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
-
-private:
-    /** Template function to run the NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
-     *
-     * @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
-     */
-    template <bool is_bounded_relu>
-    void run(const Window &window);
-
-    /** Common signature for all the specialised NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel functions
-     *
-     * @param[in] window Region on which to execute the kernel.
-     */
-    using QuantizeDownFunctionPtr = void (NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::*)(const Window &window);
-
-    QuantizeDownFunctionPtr _func;
-    const ITensor          *_input;
-    const ITensor          *_bias;
-    ITensor                *_output;
-    int                     _result_fixedpoint_multiplier;
-    int                     _result_shift;
-    int                     _result_offset_after_shift;
-    int                     _min;
-    int                     _max;
-};
-} // namespace arm_compute
-#endif /* ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEBYFIXEDPOINTKERNEL_H */