COMPMID-1047 Extract Flatten function from Im2Col for NEON

Change-Id: I80f3aaadc8cae8c9ca1a5a239e79bda302b89bd8
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/144813
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>

5 files changed, 216 insertions, 143 deletions

diff --git a/src/core/NEON/NEFlattenLayerKernel.cpp b/src/core/NEON/NEFlattenLayerKernel.cpp
new file mode 100644
index 0000000000..b8452fb5fc
--- /dev/null
+++ b/src/core/NEON/NEFlattenLayerKernel.cpp
@@ -0,0 +1,137 @@
+/*
+ * Copyright (c) 2018 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/NEON/kernels/NEFlattenLayerKernel.h"
+
+#include "arm_compute/core/CPP/Validate.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/Validate.h"
+
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
+#include <arm_neon.h>
+
+using namespace arm_compute;
+using namespace misc::shape_calculator;
+
+namespace
+{
+Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QASYMM8,
+                                                         DataType::U16, DataType::S16,
+                                                         DataType::U32, DataType::S32,
+                                                         DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
+
+    // Checks performed when output is configured
+    if(output->total_size() != 0)
+    {
+        const TensorInfo tensor_info_output = input->clone()->set_tensor_shape(compute_flatten_shape(input));
+
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &tensor_info_output);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    }
+
+    return Status{};
+}
+
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
+{
+    // Output tensor auto initialization if not yet initialized
+    auto_init_if_empty(*output, input->clone()->set_tensor_shape(compute_flatten_shape(input)));
+
+    Window win = calculate_max_window(*input, Steps()); // Flatten does not need paddings
+
+    output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
+
+    return std::make_pair(Status{}, win);
+}
+} // namespace
+
+NEFlattenLayerKernel::NEFlattenLayerKernel()
+    : _input(nullptr), _output(nullptr)
+{
+}
+
+void NEFlattenLayerKernel::configure(const ITensor *input, ITensor *output)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info()));
+
+    _input  = input;
+    _output = output;
+
+    // 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);
+}
+
+Status NEFlattenLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);
+    return Status{};
+}
+
+void NEFlattenLayerKernel::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);
+
+    const size_t in_width   = _input->info()->dimension(0);
+    const size_t in_height  = _input->info()->dimension(1);
+    const size_t out_step_x = in_width * _input->info()->element_size();
+    const size_t out_step_y = out_step_x * in_height;
+
+    Window in_window(window);
+    in_window.set(Window::DimX, Window::Dimension(0, 1, 1));
+
+    Window out_window;
+    out_window.use_tensor_dimensions(_output->info()->tensor_shape());
+    out_window.set(Window::DimX, Window::Dimension(out_window.x().start(), out_window.x().end(), in_width));
+
+    Window in_slice  = in_window.first_slice_window_3D();
+    Window out_slice = out_window.first_slice_window_1D();
+
+    do
+    {
+        Iterator in(_input, in_slice);
+        Iterator out(_output, out_slice);
+
+        uint8_t *out_ptr = out.ptr();
+
+        execute_window_loop(in_slice, [&](const Coordinates & id)
+        {
+            memcpy(out_ptr + id.y() * out_step_x + id.z() * out_step_y, in.ptr(), out_step_x);
+        },
+        in);
+    }
+    while(in_window.slide_window_slice_3D(in_slice) && out_window.slide_window_slice_1D(out_slice));
+}
diff --git a/src/core/NEON/kernels/NEIm2ColKernel.cpp b/src/core/NEON/kernels/NEIm2ColKernel.cpp
index 98b1488a9d..e5d31289a4 100644
--- a/src/core/NEON/kernels/NEIm2ColKernel.cpp
+++ b/src/core/NEON/kernels/NEIm2ColKernel.cpp
@@ -41,11 +41,12 @@
 #include <tuple>
 
 using namespace arm_compute;
+using namespace misc::shape_calculator;
 
 namespace
 {
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info,
-                          bool has_bias, const Size2D &dilation, unsigned int num_groups, bool is_fully_connected, bool is_flatten)
+                          bool has_bias, const Size2D &dilation, unsigned int num_groups)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
@@ -55,18 +56,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, c
 
     if(output->total_size() > 0)
     {
-        TensorShape expected_output_shape;
-
-        if(is_flatten || is_fully_connected)
-        {
-            expected_output_shape = misc::shape_calculator::compute_flatten_shape(input);
-        }
-        else
-        {
-            expected_output_shape = misc::shape_calculator::compute_im2col_conv_shape(input, kernel_dims, conv_info, has_bias, dilation, false);
-        }
-
-        TensorInfo expected_output = output->clone()->set_tensor_shape(expected_output_shape);
+        TensorInfo expected_output = output->clone()->set_tensor_shape(compute_im2col_conv_shape(input, kernel_dims, conv_info, has_bias, dilation, false));
         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(&expected_output, output);
         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
     }
@@ -74,6 +64,31 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, c
     return Status{};
 }
 
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info,
+                                                        bool has_bias, const Size2D &dilation)
+{
+    const unsigned int width_idx   = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::WIDTH);
+    const unsigned int height_idx  = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::HEIGHT);
+    const unsigned int channel_idx = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::CHANNEL);
+
+    std::pair<unsigned int, unsigned int> convolved_dims = scaled_dimensions(input->dimension(width_idx), input->dimension(height_idx),
+                                                                             kernel_dims.width, kernel_dims.height,
+                                                                             conv_info, dilation);
+
+    // Output tensor auto initialization if not yet initialized
+    auto_init_if_empty(*output, input->clone()->set_tensor_shape(compute_im2col_conv_shape(input, kernel_dims, conv_info, has_bias, dilation, false)));
+
+    Window win = calculate_max_window(*input, Steps());
+    win.set(width_idx, Window::Dimension(0, convolved_dims.first, 1));
+    win.set(height_idx, Window::Dimension(0, convolved_dims.second, 1));
+    win.set(channel_idx, Window::Dimension(0, 1, 1));
+
+    // The NEIm2ColKernel doesn't need padding so update_window_and_padding() can be skipped
+    output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
+
+    return std::make_pair(Status{}, win);
+}
+
 template <typename T, bool has_pads>
 inline void linearize_volume(const uint8_t *const in_ptr,
                              T                   *out_ptr,
@@ -174,7 +189,7 @@ inline void linearize_volume(const uint8_t *const in_ptr,
 } // namespace
 
 template <typename T, bool has_pads>
-void NEIm2ColKernel::run_generic(const Window &window)
+void NEIm2ColKernel::run_im2col(const Window &window)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
@@ -244,66 +259,21 @@ void NEIm2ColKernel::run_generic(const Window &window)
     in, out);
 }
 
-template <typename T>
-void NEIm2ColKernel::run_reduced(const Window &window)
-{
-    const size_t in_width   = _input->info()->dimension(0);
-    const size_t in_height  = _input->info()->dimension(1);
-    const size_t out_step_x = in_width * _input->info()->element_size();
-    const size_t out_step_y = out_step_x * in_height;
-    const size_t out_width  = _output->info()->dimension(0);
-
-    Window in_window(window);
-    in_window.set(Window::DimX, Window::Dimension(0, 1, 1));
-
-    Window out_window;
-    out_window.use_tensor_dimensions(_output->info()->tensor_shape());
-    out_window.set(Window::DimX, Window::Dimension(out_window.x().start(), out_window.x().end(), in_width));
-
-    Window in_slice  = in_window.first_slice_window_3D();
-    Window out_slice = out_window.first_slice_window_1D();
-
-    do
-    {
-        Iterator in(_input, in_slice);
-        Iterator out(_output, out_slice);
-
-        uint8_t *out_ptr = out.ptr();
-
-        execute_window_loop(in_slice, [&](const Coordinates & id)
-        {
-            memcpy(out_ptr + id.y() * out_step_x + id.z() * out_step_y, in.ptr(), out_step_x);
-        },
-        in);
-
-        // Add bias
-        if(_has_bias)
-        {
-            *(reinterpret_cast<T *>(out_ptr) + out_width - 1) = static_cast<T>(1);
-        }
-    }
-    while(in_window.slide_window_slice_3D(in_slice) && out_window.slide_window_slice_1D(out_slice));
-}
-
 NEIm2ColKernel::NEIm2ColKernel()
     : _func(), _input(nullptr), _output(nullptr), _convolved_dims(), _conv_info(), _kernel_width(0), _kernel_height(0), _has_bias(false), _dilation(1U, 1U)
 {
 }
 
 void NEIm2ColKernel::configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info,
-                               bool has_bias, const Size2D &dilation, unsigned int num_groups, bool is_fully_connected, bool is_flatten)
+                               bool has_bias, const Size2D &dilation, unsigned int num_groups)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-
-    // Perform validation step
-    ARM_COMPUTE_UNUSED(is_fully_connected, is_flatten);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), kernel_dims, conv_info, has_bias, dilation, num_groups));
     ARM_COMPUTE_UNUSED(num_groups);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), kernel_dims, conv_info, has_bias, dilation, num_groups, is_fully_connected, is_flatten));
 
     const DataLayout   data_layout = input->info()->data_layout();
     const unsigned int width_idx   = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
     const unsigned int height_idx  = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
-    const unsigned int channel_idx = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
 
     _input          = input;
     _output         = output;
@@ -316,73 +286,35 @@ void NEIm2ColKernel::configure(const ITensor *input, ITensor *output, const Size
                                         _conv_info, _dilation);
     _has_bias = has_bias;
 
-    unsigned int stride_x = 0;
-    unsigned int stride_y = 0;
-    std::tie(stride_x, stride_y) = conv_info.stride();
-
-    bool run_img2col_reduced = (output->info()->dimension(0) == (input->info()->dimension(0) * input->info()->dimension(1) * input->info()->dimension(2))) && (TensorShape::num_max_dimensions >= 4)
-                               && (std::equal(input->info()->tensor_shape().cbegin() + 3,
-                                              input->info()->tensor_shape().cend(),
-                                              output->info()->tensor_shape().cbegin() + 1))
-                               && ((stride_x == 1) && (stride_y == 1) && !conv_info.has_padding())
-                               && ((dilation.x() == 1) && (dilation.y() == 1));
-
-    Window window = calculate_max_window(*input->info(), Steps());
-
-    if(run_img2col_reduced)
+    switch(_input->info()->data_type())
     {
-        switch(_input->info()->data_type())
-        {
-            case DataType::F32:
-                _func = &NEIm2ColKernel::run_reduced<float>;
-                break;
+        case DataType::F32:
+            _func = (!conv_info.has_padding()) ? &NEIm2ColKernel::run_im2col<float, false> : &NEIm2ColKernel::run_im2col<float, true>;
+            break;
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-            case DataType::F16:
-                _func = &NEIm2ColKernel::run_reduced<float16_t>;
-                break;
+        case DataType::F16:
+            _func = (!conv_info.has_padding()) ? &NEIm2ColKernel::run_im2col<float16_t, false> : &NEIm2ColKernel::run_im2col<float16_t, true>;
+            break;
 #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-            case DataType::QASYMM8:
-                _func = &NEIm2ColKernel::run_reduced<qasymm8_t>;
-                break;
-            default:
-                ARM_COMPUTE_ERROR("Data type not supported");
-                break;
-        }
+        case DataType::QASYMM8:
+            _func = (!conv_info.has_padding()) ? &NEIm2ColKernel::run_im2col<qasymm8_t, false> : &NEIm2ColKernel::run_im2col<qasymm8_t, true>;
+            break;
+        default:
+            ARM_COMPUTE_ERROR("Data type not supported");
+            break;
     }
-    else
-    {
-        switch(_input->info()->data_type())
-        {
-            case DataType::F32:
-                _func = (!conv_info.has_padding()) ? &NEIm2ColKernel::run_generic<float, false> : &NEIm2ColKernel::run_generic<float, true>;
-                break;
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-            case DataType::F16:
-                _func = (!conv_info.has_padding()) ? &NEIm2ColKernel::run_generic<float16_t, false> : &NEIm2ColKernel::run_generic<float16_t, true>;
-                break;
-#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-            case DataType::QASYMM8:
-                _func = (!conv_info.has_padding()) ? &NEIm2ColKernel::run_generic<qasymm8_t, false> : &NEIm2ColKernel::run_generic<qasymm8_t, true>;
-                break;
-            default:
-                ARM_COMPUTE_ERROR("Data type not supported");
-                break;
-        }
-        window.set(width_idx, Window::Dimension(0, _convolved_dims.first, 1));
-        window.set(height_idx, Window::Dimension(0, _convolved_dims.second, 1));
-        window.set(channel_idx, Window::Dimension(0, 1, 1));
-    }
-
-    // The NEIm2ColKernel doesn't need padding so update_window_and_padding() can be skipped
-    output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
 
-    IKernel::configure(window);
+    // Configure kernel window
+    auto win_config = validate_and_configure_window(input->info(), output->info(), kernel_dims, conv_info, has_bias, dilation);
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    INEKernel::configure(win_config.second);
 }
 
 Status NEIm2ColKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info,
-                                bool has_bias, const Size2D &dilation, unsigned int num_groups, bool is_fully_connected, bool is_flatten)
+                                bool has_bias, const Size2D &dilation, unsigned int num_groups)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, kernel_dims, conv_info, has_bias, dilation, num_groups, is_fully_connected, is_flatten));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, kernel_dims, conv_info, has_bias, dilation, num_groups));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), kernel_dims, conv_info, has_bias, dilation).first);
     return Status{};
 }
 
diff --git a/src/runtime/NEON/functions/NEFlattenLayer.cpp b/src/runtime/NEON/functions/NEFlattenLayer.cpp
index 1814d61e2f..57bef2b933 100644
--- a/src/runtime/NEON/functions/NEFlattenLayer.cpp
+++ b/src/runtime/NEON/functions/NEFlattenLayer.cpp
@@ -23,7 +23,7 @@
  */
 #include "arm_compute/runtime/NEON/functions/NEFlattenLayer.h"
 
-#include "arm_compute/core/NEON/kernels/NEIm2ColKernel.h"
+#include "arm_compute/core/NEON/kernels/NEFlattenLayerKernel.h"
 #include "arm_compute/core/Size2D.h"
 #include "support/ToolchainSupport.h"
 
@@ -31,7 +31,12 @@ using namespace arm_compute;
 
 void NEFlattenLayer::configure(const ITensor *input, ITensor *output)
 {
-    auto k = arm_compute::support::cpp14::make_unique<NEIm2ColKernel>();
-    k->configure(input, output, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false, Size2D(1U, 1U), 1, false, true);
+    auto k = arm_compute::support::cpp14::make_unique<NEFlattenLayerKernel>();
+    k->configure(input, output);
     _kernel = std::move(k);
+}
+
+Status NEFlattenLayer::validate(const ITensorInfo *input, const ITensorInfo *output)
+{
+    return NEFlattenLayerKernel::validate(input, output);
 }
\ No newline at end of file
diff --git a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
index f1606aa93e..60f6294394 100644
--- a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
+++ b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
@@ -74,8 +74,8 @@ Status NEFullyConnectedLayerReshapeWeights::validate(const ITensorInfo *input, c
 }
 
 NEFullyConnectedLayer::NEFullyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager)
-    : _memory_group(std::move(memory_manager)), _im2col_kernel(), _convert_weights(), _reshape_weights_function(), _mm_gemm(), _mm_gemmlowp(), _gemmlowp_output_stage(), _accumulate_biases_kernel(),
-      _im2col_output(), _gemmlowp_output(), _converted_weights_output(), _reshape_weights_output(), _original_weights(nullptr), _are_weights_converted(true), _are_weights_reshaped(false),
+    : _memory_group(std::move(memory_manager)), _flatten_kernel(), _convert_weights(), _reshape_weights_function(), _mm_gemm(), _mm_gemmlowp(), _gemmlowp_output_stage(), _accumulate_biases_kernel(),
+      _flatten_output(), _gemmlowp_output(), _converted_weights_output(), _reshape_weights_output(), _original_weights(nullptr), _are_weights_converted(true), _are_weights_reshaped(false),
       _is_fc_after_conv(false), _accumulate_biases(false), _is_quantized(false), _is_prepared(false)
 {
 }
@@ -112,19 +112,19 @@ void NEFullyConnectedLayer::configure_conv_fc(const ITensor *input, const ITenso
 
     // If the fully connected layer is called after a convolution layer, the input tensor must be linearized
 
-    // Initialize output tensor for im2col
-    TensorShape shape_im2col = compute_flatten_shape(input->info());
-    _im2col_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_im2col));
+    // Initialize output tensor for flatten
+    TensorShape shape_flatten = compute_flatten_shape(input->info());
+    _flatten_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_flatten));
 
-    // Configure im2col kernel
-    _memory_group.manage(&_im2col_output);
-    _im2col_kernel.configure(input, &_im2col_output, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false, Size2D(1U, 1U), 1, true);
+    // Configure flatten kernel
+    _memory_group.manage(&_flatten_output);
+    _flatten_kernel.configure(input, &_flatten_output);
 
     // Configure matrix multiply kernel
-    configure_mm(&_im2col_output, weights, output);
+    configure_mm(&_flatten_output, weights, output);
 
-    // Allocate the output tensor for im2col once all the configure methods have been called
-    _im2col_output.allocator()->allocate();
+    // Allocate the output tensor for flatten once all the configure methods have been called
+    _flatten_output.allocator()->allocate();
 }
 
 void NEFullyConnectedLayer::configure_fc_fc(const ITensor *input, const ITensor *weights, ITensor *output)
@@ -249,7 +249,7 @@ Status NEFullyConnectedLayer::validate(const ITensorInfo *input, const ITensorIn
     bool is_fc_after_conv = true;
     bool is_quantized     = is_data_type_quantized_asymmetric(input->data_type());
 
-    const ITensorInfo &im2col_input      = TensorInfo(input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_flatten_shape(input)));
+    const ITensorInfo &flatten_input     = TensorInfo(input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_flatten_shape(input)));
     const ITensorInfo &reshaped_weights  = TensorInfo(weights->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_transposed_shape(*weights)));
     const ITensorInfo &converted_weights = weights_reshaped ? TensorInfo(weights->clone()->set_is_resizable(true).reset_padding()) : TensorInfo(*reshaped_weights.clone());
     const ITensorInfo &gemmlowp_output   = TensorInfo(output->clone()->set_is_resizable(true).reset_padding().set_data_type(DataType::S32));
@@ -307,9 +307,9 @@ Status NEFullyConnectedLayer::validate(const ITensorInfo *input, const ITensorIn
         // Fully Connected layer after a Convolution Layer without batches
         ARM_COMPUTE_RETURN_ERROR_ON((weights_to_use->dimension(1) != (input->dimension(0) * input->dimension(1) * input->dimension(2))));
 
-        // Validate im2col kernel
-        ARM_COMPUTE_RETURN_ON_ERROR(NEIm2ColKernel::validate(input, &im2col_input, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false, Size2D(1U, 1U), 1, true));
-        input_to_use = &im2col_input;
+        // Validate flatten kernel
+        ARM_COMPUTE_RETURN_ON_ERROR(NEFlattenLayerKernel::validate(input, &flatten_input));
+        input_to_use = &flatten_input;
     }
     else
     {
@@ -337,7 +337,7 @@ void NEFullyConnectedLayer::run()
     // Linearize input if it comes from a convolutional layer
     if(_is_fc_after_conv)
     {
-        NEScheduler::get().schedule(&_im2col_kernel, Window::DimY);
+        NEScheduler::get().schedule(&_flatten_kernel, Window::DimY);
     }
 
     // Run matrix multiply
diff --git a/src/runtime/NEON/functions/NEIm2Col.cpp b/src/runtime/NEON/functions/NEIm2Col.cpp
index 4245b650e2..9102fca7f6 100644
--- a/src/runtime/NEON/functions/NEIm2Col.cpp
+++ b/src/runtime/NEON/functions/NEIm2Col.cpp
@@ -34,18 +34,17 @@ NEIm2Col::NEIm2Col()
 {
 }
 
-void NEIm2Col::configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, const Size2D &dilation, unsigned int num_groups,
-                         bool is_fully_connected, bool is_flatten)
+void NEIm2Col::configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, const Size2D &dilation, unsigned int num_groups)
 {
     _y_dim = get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::HEIGHT);
 
-    _kernel.configure(input, output, kernel_dims, conv_info, has_bias, dilation, num_groups, is_fully_connected, is_flatten);
+    _kernel.configure(input, output, kernel_dims, conv_info, has_bias, dilation, num_groups);
 }
 
 Status NEIm2Col::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, const Size2D &dilation,
-                          unsigned int num_groups, bool is_fully_connected, bool is_flatten)
+                          unsigned int num_groups)
 {
-    return NEIm2ColKernel::validate(input, output, kernel_dims, conv_info, has_bias, dilation, num_groups, is_fully_connected, is_flatten);
+    return NEIm2ColKernel::validate(input, output, kernel_dims, conv_info, has_bias, dilation, num_groups);
 }
 
 void NEIm2Col::run()