COMPMID-2765 Add support for QASYMM8_SIGNED in NEDeconvolutionLayer

Signed-off-by: Luca Foschiani <luca.foschiani@arm.com>
Change-Id: I8295fadee15311a9ab846aa24c031b82c0b799eb
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2952
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sheri Zhang <sheri.zhang@arm.com>

10 files changed, 131 insertions, 328 deletions

diff --git a/Android.bp b/Android.bp
index 528467a44e..6e34cb5cca 100644
--- a/Android.bp
+++ b/Android.bp
@@ -212,7 +212,6 @@ cc_library_static {
         "src/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.cpp",
         "src/core/CPP/kernels/CPPCornerCandidatesKernel.cpp",
         "src/core/CPP/kernels/CPPDetectionWindowNonMaximaSuppressionKernel.cpp",
-        "src/core/CPP/kernels/CPPFlipWeightsKernel.cpp",
         "src/core/CPP/kernels/CPPNonMaximumSuppressionKernel.cpp",
         "src/core/CPP/kernels/CPPPermuteKernel.cpp",
         "src/core/CPP/kernels/CPPSortEuclideanDistanceKernel.cpp",
diff --git a/arm_compute/core/CPP/CPPKernels.h b/arm_compute/core/CPP/CPPKernels.h
index d12cb2857a..c7b40baf22 100644
--- a/arm_compute/core/CPP/CPPKernels.h
+++ b/arm_compute/core/CPP/CPPKernels.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -28,7 +28,6 @@
 #include "arm_compute/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.h"
 #include "arm_compute/core/CPP/kernels/CPPCornerCandidatesKernel.h"
 #include "arm_compute/core/CPP/kernels/CPPDetectionWindowNonMaximaSuppressionKernel.h"
-#include "arm_compute/core/CPP/kernels/CPPFlipWeightsKernel.h"
 #include "arm_compute/core/CPP/kernels/CPPNonMaximumSuppressionKernel.h"
 #include "arm_compute/core/CPP/kernels/CPPPermuteKernel.h"
 #include "arm_compute/core/CPP/kernels/CPPSortEuclideanDistanceKernel.h"
diff --git a/arm_compute/core/CPP/kernels/CPPFlipWeightsKernel.h b/arm_compute/core/CPP/kernels/CPPFlipWeightsKernel.h
deleted file mode 100644
index 285aa585be..0000000000
--- a/arm_compute/core/CPP/kernels/CPPFlipWeightsKernel.h
+++ /dev/null
@@ -1,83 +0,0 @@
-/*
- * Copyright (c) 2018-2019 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_CPP_FLIP_WEIGHTS_KERNEL_H
-#define ARM_COMPUTE_CPP_FLIP_WEIGHTS_KERNEL_H
-
-#include "arm_compute/core/CPP/ICPPKernel.h"
-
-namespace arm_compute
-{
-class ITensor;
-
-/** CPP kernel to perform 180 degrees flipping on deconvolution weights. */
-class CPPFlipWeightsKernel : public ICPPKernel
-{
-public:
-    const char *name() const override
-    {
-        return "CPPFlipWeightsKernel";
-    }
-    /** Default constructor */
-    CPPFlipWeightsKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CPPFlipWeightsKernel(const CPPFlipWeightsKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CPPFlipWeightsKernel &operator=(const CPPFlipWeightsKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    CPPFlipWeightsKernel(CPPFlipWeightsKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    CPPFlipWeightsKernel &operator=(CPPFlipWeightsKernel &&) = default;
-    /** Default destructor */
-    ~CPPFlipWeightsKernel() = default;
-
-    /** Set the input and output of the kernel.
-     *
-     * @param[in]  input  The input tensor to flip. Data types supported: QASYMM8/F16/F32. Data layouts supported: NCHW/NHWC.
-     * @param[out] output The output tensor. Data types supported: Same as @p input
-     */
-    void configure(const ITensor *input, ITensor *output);
-
-    // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
-
-    /** Function to perform flipping.
-     *
-     * @param[in] window_input Input region on which to execute the kernel.
-     */
-    template <typename T>
-    void flip_weights(const Window &window_input);
-
-    /** Common signature for all the specialised Flip functions
-     *
-     * @param[in] window_input Input region on which to execute the kernel.
-     */
-    using FlipWeightsFunction = void (CPPFlipWeightsKernel::*)(const Window &window_input);
-
-private:
-    const ITensor      *_input;
-    ITensor            *_output;
-    FlipWeightsFunction _func;
-};
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CPP_FLIP_WEIGHTS_KERNEL_H */
diff --git a/arm_compute/runtime/CL/functions/CLDirectDeconvolutionLayer.h b/arm_compute/runtime/CL/functions/CLDirectDeconvolutionLayer.h
index 76cd0fb0c2..6632bfce80 100644
--- a/arm_compute/runtime/CL/functions/CLDirectDeconvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLDirectDeconvolutionLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 ARM Limited.
+ * Copyright (c) 2019-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -61,7 +61,7 @@ class ICLTensor;
  *      stride_x and stride_y is the input stride of the first and second dimension.
  *
  * The weights used by Deconvolution are supposed to be the same as the ones used for Convolution. Therefore, it will be necessary to use the weights in the
- * reverse order to perform an actual convolution. This is achieved by using the @ref CPPFlipWeightsKernel.
+ * reverse order to perform an actual convolution. This is achieved by using @ref CLReverse.
  *
  * This function calls the following OpenCL kernels/functions:
  *
diff --git a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
index d031076ee7..c4c1664f20 100644
--- a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -27,8 +27,8 @@
 #include "arm_compute/runtime/CPP/functions/CPPUpsample.h"
 #include "arm_compute/runtime/NEON/functions/NEConvolutionLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h"
+#include "arm_compute/runtime/NEON/functions/NEReverse.h"
 
-#include "arm_compute/core/CPP/kernels/CPPFlipWeightsKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/IMemoryManager.h"
@@ -62,12 +62,14 @@ namespace arm_compute
  *      stride_x and stride_y is the input stride of the first and second dimension.
  *
  * The weights used by Deconvolution are supposed to be the same as the ones used for Convolution. Therefore, it will be necessary to use the weights in the
- * reverse order to perform an actual convolution. This is achieved by using the @ref CPPFlipWeightsKernel.
+ * reverse order to perform an actual convolution. This is achieved by using @ref NEReverse.
  *
  * This function calls the following NEON kernels/functions:
  *
  * -# @ref CPPUpsample
  * -# @ref NEConvolutionLayer
+ * -# @ref NEPermute
+ * -# @ref NEReverse
  *
  */
 class NEDeconvolutionLayer : public IFunction
@@ -89,9 +91,9 @@ public:
 
     /** Set the input, weights, biases and output tensors.
      *
-     * @param[in,out] input   Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32/F16/QASYMM8.
+     * @param[in,out] input   Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32/F16/QASYMM8/QASYMM8_SIGNED.
      * @param[in]     weights The 4d weights with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input.
-     * @param[in]     bias    Optional, ignored if NULL. The biases have one dimension. Data type supported: Data types supported: S32 for QASYMM8 input, F32 for F32 input, F16 for F16 input.
+     * @param[in]     bias    Optional, ignored if NULL. The biases have one dimension. Data type supported: Data types supported: S32 for QASYMM8 and QASYMM8_SIGNED input, F32 for F32 input, F16 for F16 input.
      * @param[out]    output  Output tensor. The output has the same number of dimensions as the @p input.
      * @param[in]     info    Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo.
      *
@@ -99,9 +101,9 @@ public:
     void configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &info);
     /** Static function to check if given info will lead to a valid configuration of @ref NEDeconvolutionLayer
      *
-     * @param[in] input   Input tensor info. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32/F16/QASYMM8.
+     * @param[in] input   Input tensor info. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32/F16/QASYMM8/QASYMM8_SIGNED.
      * @param[in] weights The 4d weights info with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input.
-     * @param[in] bias    (Optional) The biases have one dimension. Data type supported: Data types supported: S32 for QASYMM8 input, F32 for F32 input, F16 for F16 input.
+     * @param[in] bias    (Optional) The biases have one dimension. Data type supported: Data types supported: S32 for QASYMM8 and QASYMM8_SIGNED input, F32 for F32 input, F16 for F16 input.
      * @param[in] output  Output tensor info. The output has the same number of dimensions as the @p input.
      * @param[in] info    Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo.
      *
@@ -114,23 +116,24 @@ public:
     void prepare() override;
 
 private:
-    MemoryGroup          _memory_group;
-    NEConvolutionLayer   _conv_f;
-    CPPUpsample          _upsample_f;
-    CPPFlipWeightsKernel _flip_weights;
-    NEPermute            _permute_input;
-    NEPermute            _permute_weights;
-    NEPermute            _permute_output;
-    Tensor               _scaled_output;
-    Tensor               _weights_flipped;
-    Tensor               _permuted_input;
-    Tensor               _permuted_weights;
-    Tensor               _permuted_output;
-    bool                 _is_nchw;
-    const ITensor       *_original_weights;
-    ITensor             *_input;
-    PadStrideInfo        _info;
-    bool                 _is_prepared;
+    MemoryGroup        _memory_group;
+    NEConvolutionLayer _conv_f;
+    CPPUpsample        _upsample_f;
+    NEReverse          _flip_weights;
+    NEPermute          _permute_input;
+    NEPermute          _permute_weights;
+    NEPermute          _permute_output;
+    Tensor             _scaled_output;
+    Tensor             _weights_flipped;
+    Tensor             _permuted_input;
+    Tensor             _permuted_weights;
+    Tensor             _permuted_output;
+    Tensor             _flip_axis;
+    bool               _is_nchw;
+    const ITensor     *_original_weights;
+    ITensor           *_input;
+    PadStrideInfo      _info;
+    bool               _is_prepared;
 };
 } // arm_compute
 #endif /* ARM_COMPUTE_NEDECONVOLUTIONLAYER_H */
diff --git a/src/core/CPP/kernels/CPPFlipWeightsKernel.cpp b/src/core/CPP/kernels/CPPFlipWeightsKernel.cpp
deleted file mode 100644
index 2d4c0ce5c8..0000000000
--- a/src/core/CPP/kernels/CPPFlipWeightsKernel.cpp
+++ /dev/null
@@ -1,113 +0,0 @@
-/*
- * 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/CPP/kernels/CPPFlipWeightsKernel.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 <cstddef>
-#include <cstdint>
-
-using namespace arm_compute;
-
-CPPFlipWeightsKernel::CPPFlipWeightsKernel()
-    : _input(nullptr), _output(nullptr), _func(nullptr)
-{
-}
-
-template <typename T>
-void CPPFlipWeightsKernel::flip_weights(const Window &window_input)
-{
-    // Create iterators
-    Iterator in(_input, window_input);
-
-    const DataLayout data_layout = _input->info()->data_layout();
-    const size_t     idx_w       = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
-    const size_t     idx_h       = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
-
-    const int kernel_width  = _input->info()->dimension(idx_w);
-    const int kernel_height = _input->info()->dimension(idx_h);
-
-    execute_window_loop(window_input, [&](const Coordinates & id)
-    {
-        const unsigned int x = kernel_width - id[idx_w] - 1;
-        const unsigned int y = kernel_height - id[idx_h] - 1;
-        Coordinates        output_coord(id);
-        output_coord.set(idx_w, x);
-        output_coord.set(idx_h, y);
-        *(reinterpret_cast<T *>(_output->ptr_to_element(output_coord))) = *(reinterpret_cast<const T *>(in.ptr()));
-    },
-    in);
-}
-
-void CPPFlipWeightsKernel::configure(const ITensor *input, ITensor *output)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-
-    _input  = input;
-    _output = output;
-
-    // Configure kernel window
-    Window win = calculate_max_window(*input->info(), Steps());
-
-    // The CPPFlipWeightsKernel doesn't need padding so update_window_and_padding() can be skipped
-    Coordinates coord;
-    coord.set_num_dimensions(output->info()->num_dimensions());
-    output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
-
-    ICPPKernel::configure(win);
-
-    switch(input->info()->data_type())
-    {
-        case DataType::F32:
-            _func = &CPPFlipWeightsKernel::flip_weights<float>;
-            break;
-        case DataType::F16:
-            _func = &CPPFlipWeightsKernel::flip_weights<half>;
-            break;
-        case DataType::QASYMM8:
-            _func = &CPPFlipWeightsKernel::flip_weights<uint8_t>;
-            break;
-        default:
-            ARM_COMPUTE_ERROR("Not supported");
-    }
-}
-
-void CPPFlipWeightsKernel::run(const Window &window, const ThreadInfo &info)
-{
-    ARM_COMPUTE_UNUSED(info);
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICPPKernel::window(), window);
-    ARM_COMPUTE_ERROR_ON(_func == nullptr);
-
-    (this->*_func)(window);
-}
diff --git a/src/core/NEON/kernels/NEReverseKernel.cpp b/src/core/NEON/kernels/NEReverseKernel.cpp
index 2f584164dc..5a8c446ddd 100644
--- a/src/core/NEON/kernels/NEReverseKernel.cpp
+++ b/src/core/NEON/kernels/NEReverseKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -106,33 +106,20 @@ void run_reverse(const Window &window, const ITensor *input, const ITensor *axis
     }
 
     // Check if we need a left-over loop for the y dimension
-    const int window_step_x            = 16 / input->info()->element_size();
-    const int window_start_x           = window.x().start();
-    const int window_end_x             = std::min(window.x().end(), static_cast<int>(input->info()->dimension(0)));
-    const int window_end_x_multiple_of = ((window_end_x - window_start_x) / window_step_x) * window_step_x;
-    bool      left_over_loop_x         = (((window_end_x - window_start_x) % window_step_x) != 0);
+    const int window_step_x  = 16 / input->info()->element_size();
+    const int window_start_x = window.x().start();
+    const int window_end_x   = window.x().end();
 
-    Window slice = window.first_slice_window_4D();
+    Window win(window);
+    win.set(Window::DimX, Window::Dimension(0, 1, 1));
 
-    if(left_over_loop_x)
+    Iterator input_it(input, win);
+    execute_window_loop(win, [&](const Coordinates & id)
     {
-        // Check if window_end_y_multiple_of is greater than window_start_y
-        if(window_end_x_multiple_of > window_start_x)
+        int x = window_start_x;
+        for(; x <= (window_end_x - window_step_x); x += window_step_x)
         {
-            slice.set(Window::DimX, Window::Dimension(window_start_x, window_end_x_multiple_of, window_step_x));
-        }
-        else
-        {
-            slice.set(Window::DimX, Window::Dimension(0, 0, 1));
-        }
-    }
-
-    do
-    {
-        Iterator input_it(input, slice);
-        execute_window_loop(slice, [&](const Coordinates & id)
-        {
-            auto in = wrapper::vloadq(reinterpret_cast<T *>(input_it.ptr()));
+            auto in = wrapper::vloadq(reinterpret_cast<T *>(input_it.ptr()) + x);
 
             // Reverse 0 axis
             if(axis_bit & 0x1)
@@ -141,39 +128,29 @@ void run_reverse(const Window &window, const ITensor *input, const ITensor *axis
                 in = wrapper::vcombine(wrapper::vgethigh(in), wrapper::vgetlow(in));
             }
 
-            const int offset_x = (axis_bit & 0x1) ? output->info()->dimension(0) - id.x() - window_step_x : id.x();
+            const int offset_x = (axis_bit & 0x1) ? output->info()->dimension(0) - x - window_step_x : x;
             const int offset_y = (axis_bit & 0x2) ? output->info()->dimension(1) - id.y() - 1 : id.y();
             const int offset_z = (axis_bit & 0x4) ? output->info()->dimension(2) - id.z() - 1 : id.z();
             const int offset_w = (axis_bit & 0x8) ? output->info()->dimension(3) - id[3] - 1 : id[3];
 
             auto out_ptr = reinterpret_cast<T *>(output->ptr_to_element(Coordinates(offset_x, offset_y, offset_z, offset_w)));
             wrapper::vstore(out_ptr, in);
-        },
-        input_it);
+        }
 
-        if(left_over_loop_x)
+        // Compute left-over elements
+        for(; x < window_end_x; ++x)
         {
-            slice.set(Window::DimX, Window::Dimension(window_end_x_multiple_of, window_end_x, 1));
+            const auto in = *(reinterpret_cast<T *>(input_it.ptr()) + x);
 
-            Iterator input_it(input, slice);
-
-            // Compute left-over elements along the y dimension (1x1)
-            execute_window_loop(slice, [&](const Coordinates & id)
-            {
-                const auto in = *reinterpret_cast<T *>(input_it.ptr());
-
-                const int offset_x = (axis_bit & 0x1) ? output->info()->dimension(0) - id.x() - 1 : id.x();
-                const int offset_y = (axis_bit & 0x2) ? output->info()->dimension(1) - id.y() - 1 : id.y();
-                const int offset_z = (axis_bit & 0x4) ? output->info()->dimension(2) - id.z() - 1 : id.z();
-                const int offset_w = (axis_bit & 0x8) ? output->info()->dimension(3) - id[3] - 1 : id[3];
+            const int offset_x = (axis_bit & 0x1) ? output->info()->dimension(0) - x - 1 : x;
+            const int offset_y = (axis_bit & 0x2) ? output->info()->dimension(1) - id.y() - 1 : id.y();
+            const int offset_z = (axis_bit & 0x4) ? output->info()->dimension(2) - id.z() - 1 : id.z();
+            const int offset_w = (axis_bit & 0x8) ? output->info()->dimension(3) - id[3] - 1 : id[3];
 
-                *reinterpret_cast<T *>(output->ptr_to_element(Coordinates(offset_x, offset_y, offset_z, offset_w))) = in;
-            },
-            input_it);
+            *reinterpret_cast<T *>(output->ptr_to_element(Coordinates(offset_x, offset_y, offset_z, offset_w))) = in;
         }
-
-    }
-    while(window.slide_window_slice_4D(slice));
+    },
+    input_it);
 }
 
 void NEReverseKernel::run(const Window &window, const ThreadInfo &info)
diff --git a/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp b/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp
index 0411b41220..06885d59e5 100644
--- a/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -46,6 +46,7 @@ NEDeconvolutionLayer::NEDeconvolutionLayer(std::shared_ptr<IMemoryManager> memor
       _permuted_input(),
       _permuted_weights(),
       _permuted_output(),
+      _flip_axis(),
       _is_nchw(false),
       _original_weights(nullptr),
       _input(nullptr),
@@ -57,7 +58,7 @@ NEDeconvolutionLayer::NEDeconvolutionLayer(std::shared_ptr<IMemoryManager> memor
 Status NEDeconvolutionLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, const ITensorInfo *output, const PadStrideInfo &info)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32, DataType::F16, DataType::QASYMM8);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32, DataType::F16, DataType::QASYMM8, DataType::QASYMM8_SIGNED);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(weights, input);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(weights, input);
     const unsigned int width_idx  = get_data_layout_dimension_index(weights->data_layout(), DataLayoutDimension::WIDTH);
@@ -122,6 +123,7 @@ void NEDeconvolutionLayer::configure(ITensor *input, const ITensor *weights, con
     _info             = info;
     _is_prepared      = false;
     _is_nchw          = data_layout == DataLayout::NCHW;
+    _flip_axis.allocator()->init(TensorInfo(TensorShape(2U), 1, DataType::U32));
 
     const unsigned int pad_left   = info.pad_left();
     const unsigned int pad_right  = info.pad_right();
@@ -139,6 +141,7 @@ void NEDeconvolutionLayer::configure(ITensor *input, const ITensor *weights, con
     // Output auto initialization if not yet initialized
     auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type(), input->info()->quantization_info());
 
+    _flip_axis.allocator()->init(TensorInfo(TensorShape(2U), 1, DataType::U32));
     _memory_group.manage(&_scaled_output);
 
     if(!_is_nchw)
@@ -185,7 +188,7 @@ void NEDeconvolutionLayer::configure(ITensor *input, const ITensor *weights, con
 
         _weights_flipped.allocator()->init(*_permuted_weights.info()->clone());
         _weights_flipped.info()->set_quantization_info(weights->info()->quantization_info());
-        _flip_weights.configure(&_permuted_weights, &_weights_flipped);
+        _flip_weights.configure(&_permuted_weights, &_weights_flipped, &_flip_axis);
 
         // setup the function to convolve the upscaled output
         const PadStrideInfo conv_info(1, 1, 0, 0, 0, 0, DimensionRoundingType::CEIL);
@@ -230,13 +233,19 @@ void NEDeconvolutionLayer::configure(ITensor *input, const ITensor *weights, con
         _upsample_f.configure(input, &_scaled_output, upsample_info);
 
         _weights_flipped.allocator()->init(weights->info()->clone()->set_data_layout(data_layout));
-        _flip_weights.configure(weights, &_weights_flipped);
+        _flip_weights.configure(weights, &_weights_flipped, &_flip_axis);
 
         // setup the function to convolve the upscaled output
         const PadStrideInfo conv_info(1, 1, 0, 0, 0, 0, DimensionRoundingType::CEIL);
         _conv_f.configure(&_scaled_output, &_weights_flipped, bias, output, conv_info);
     }
     _scaled_output.allocator()->allocate();
+
+    // Setup flip axis data
+    _flip_axis.allocator()->allocate();
+    auto axis_data = reinterpret_cast<uint32_t *>(_flip_axis.buffer());
+    axis_data[0]   = 0;
+    axis_data[1]   = 1;
 }
 
 void NEDeconvolutionLayer::run()
@@ -276,16 +285,13 @@ void NEDeconvolutionLayer::prepare()
 
         // Run weights flipping and mark original weights tensor as unused
         _weights_flipped.allocator()->allocate();
-        NEScheduler::get().schedule(&_flip_weights, Window::DimZ);
+        _flip_weights.run();
         _original_weights->mark_as_unused();
 
         // Prepare convolution
         _conv_f.prepare();
 
-        if(!_weights_flipped.is_used())
-        {
-            _weights_flipped.allocator()->free();
-        }
+        // Unused weights are already released in _conv_f
 
         if(!_is_nchw)
         {
diff --git a/tests/validation/NEON/DeconvolutionLayer.cpp b/tests/validation/NEON/DeconvolutionLayer.cpp
index 89f9d98ed5..38256eb2ad 100644
--- a/tests/validation/NEON/DeconvolutionLayer.cpp
+++ b/tests/validation/NEON/DeconvolutionLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -43,8 +43,8 @@ namespace validation
 {
 namespace
 {
-constexpr AbsoluteTolerance<float> tolerance_fp32(0.001f);  /**< Tolerance for floating point tests */
-constexpr AbsoluteTolerance<float> tolerance_qasymm8(1.0f); /**< Tolerance value for comparing reference's output against implementation's output for quantized data types */
+constexpr AbsoluteTolerance<float> tolerance_fp32(0.001f);    /**< Tolerance for floating point tests */
+constexpr AbsoluteTolerance<float> tolerance_quantized(1.0f); /**< Tolerance value for comparing reference's output against implementation's output for quantized data types */
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 const RelativeTolerance<half_float::half> tolerance_fp16(half_float::half(0.2f)); /**< Relative tolerance value for comparing reference's output against implementation's output for DataType::F16 */
 #endif                                                                            /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC*/
@@ -85,46 +85,6 @@ const auto output_qinfo_dataset = framework::dataset::make("OutputQInfo",
 TEST_SUITE(NEON)
 TEST_SUITE(DeconvolutionLayer)
 
-DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, (combine(datasets::SmallDeconvolutionShapes(), framework::dataset::make("DataType", DataType::F32))),
-               input_shape, data_type)
-{
-    // Create shapes
-    const unsigned int  kernel_size_x = 3;
-    const unsigned int  kernel_size_y = 3;
-    const unsigned int  num_kernels   = 1;
-    const TensorShape   weights_shape(kernel_size_x, kernel_size_y, input_shape.z(), num_kernels);
-    const TensorShape   bias_shape(num_kernels);
-    const PadStrideInfo info(1, 1, 1, 1);
-    auto                out_dim      = deconvolution_output_dimensions(input_shape.x(), input_shape.y(), kernel_size_x, kernel_size_y, info);
-    TensorShape         output_shape = compute_deconvolution_output_shape(out_dim, TensorInfo(input_shape, 1, data_type), TensorInfo(weights_shape, 1, data_type));
-
-    // Create tensors
-    Tensor src     = create_tensor<Tensor>(input_shape, data_type, 1);
-    Tensor weights = create_tensor<Tensor>(weights_shape, data_type, 1);
-    Tensor bias    = create_tensor<Tensor>(bias_shape, data_type, 1);
-    Tensor dst     = create_tensor<Tensor>(output_shape, data_type, 1);
-
-    ARM_COMPUTE_EXPECT(src.info()->is_resizable(), framework::LogLevel::ERRORS);
-    ARM_COMPUTE_EXPECT(weights.info()->is_resizable(), framework::LogLevel::ERRORS);
-    ARM_COMPUTE_EXPECT(bias.info()->is_resizable(), framework::LogLevel::ERRORS);
-    ARM_COMPUTE_EXPECT(dst.info()->is_resizable(), framework::LogLevel::ERRORS);
-
-    // Create and configure function
-    NEDeconvolutionLayer deconv;
-    deconv.configure(&src, &weights, &bias, &dst, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL));
-
-    // Validate valid region
-    const ValidRegion src_valid_region     = shape_to_valid_region(input_shape);
-    const ValidRegion weights_valid_region = shape_to_valid_region(weights_shape);
-    const ValidRegion bias_valid_region    = shape_to_valid_region(bias_shape);
-    const ValidRegion dst_valid_region     = shape_to_valid_region(output_shape);
-
-    validate(src.info()->valid_region(), src_valid_region);
-    validate(weights.info()->valid_region(), weights_valid_region);
-    validate(bias.info()->valid_region(), bias_valid_region);
-    validate(dst.info()->valid_region(), dst_valid_region);
-}
-
 // *INDENT-OFF*
 // clang-format off
 DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(
@@ -294,7 +254,7 @@ FIXTURE_DATA_TEST_CASE(Run, NEDeconvolutionLayerQuantizedFixture4x4<uint8_t>, fr
                                                                                                                        add_bias_dataset))
 {
     // Validate output
-    validate(Accessor(_target), _reference, tolerance_qasymm8, tolerance_num);
+    validate(Accessor(_target), _reference, tolerance_quantized, tolerance_num);
 }
 TEST_SUITE_END() // W4x4
 
@@ -308,7 +268,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NEDeconvolutionLayerQuantizedFixture3x3<uint8_t
                        add_bias_dataset))
 {
     // Validate output
-    validate(Accessor(_target), _reference, tolerance_qasymm8, tolerance_num);
+    validate(Accessor(_target), _reference, tolerance_quantized, tolerance_num);
 }
 FIXTURE_DATA_TEST_CASE(RunLarge, NEDeconvolutionLayerQuantizedFixture3x3<uint8_t>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(data3x3,
                        framework::dataset::make("DataType",
@@ -319,7 +279,7 @@ FIXTURE_DATA_TEST_CASE(RunLarge, NEDeconvolutionLayerQuantizedFixture3x3<uint8_t
                        add_bias_dataset))
 {
     // Validate output
-    validate(Accessor(_target), _reference, tolerance_qasymm8, tolerance_num);
+    validate(Accessor(_target), _reference, tolerance_quantized, tolerance_num);
 }
 TEST_SUITE_END() // W3x3
 
@@ -332,11 +292,66 @@ FIXTURE_DATA_TEST_CASE(Run, NEDeconvolutionLayerQuantizedFixture1x1<uint8_t>, fr
                                                                                                                        add_bias_dataset))
 {
     // Validate output
-    validate(Accessor(_target), _reference, tolerance_qasymm8, tolerance_num);
+    validate(Accessor(_target), _reference, tolerance_quantized, tolerance_num);
 }
 TEST_SUITE_END() // W1x1
 
 TEST_SUITE_END() // QASYMM8
+
+TEST_SUITE(QASYMM8_SIGNED)
+
+TEST_SUITE(W4x4)
+FIXTURE_DATA_TEST_CASE(Run, NEDeconvolutionLayerQuantizedFixture4x4<int8_t>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(data4x4, framework::dataset::make("DataType",
+                                                                                                                      DataType::QASYMM8_SIGNED)),
+                                                                                                                      data_layouts_dataset),
+                                                                                                                      input_qinfo_dataset),
+                                                                                                                      output_qinfo_dataset),
+                                                                                                                      add_bias_dataset))
+{
+    // Validate output
+    validate(Accessor(_target), _reference, tolerance_quantized, tolerance_num);
+}
+TEST_SUITE_END() // W4x4
+
+TEST_SUITE(W3x3)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDeconvolutionLayerQuantizedFixture3x3<int8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(combine(data3x3_precommit,
+                       framework::dataset::make("DataType",
+                                                DataType::QASYMM8_SIGNED)),
+                       data_layouts_dataset),
+                       input_qinfo_dataset),
+                       output_qinfo_dataset),
+                       add_bias_dataset))
+{
+    // Validate output
+    validate(Accessor(_target), _reference, tolerance_quantized, tolerance_num);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge, NEDeconvolutionLayerQuantizedFixture3x3<int8_t>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(data3x3,
+                       framework::dataset::make("DataType",
+                                                DataType::QASYMM8_SIGNED)),
+                       data_layouts_dataset),
+                       input_qinfo_dataset),
+                       output_qinfo_dataset),
+                       add_bias_dataset))
+{
+    // Validate output
+    validate(Accessor(_target), _reference, tolerance_quantized, tolerance_num);
+}
+TEST_SUITE_END() // W3x3
+
+TEST_SUITE(W1x1)
+FIXTURE_DATA_TEST_CASE(Run, NEDeconvolutionLayerQuantizedFixture1x1<int8_t>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(data1x1, framework::dataset::make("DataType",
+                                                                                                                      DataType::QASYMM8_SIGNED)),
+                                                                                                                      data_layouts_dataset),
+                                                                                                                      input_qinfo_dataset),
+                                                                                                                      output_qinfo_dataset),
+                                                                                                                      add_bias_dataset))
+{
+    // Validate output
+    validate(Accessor(_target), _reference, tolerance_quantized, tolerance_num);
+}
+TEST_SUITE_END() // W1x1
+
+TEST_SUITE_END() // QASYMM8_SIGNED
 TEST_SUITE_END() // Quantized
 
 TEST_SUITE_END() // DeconvolutionLayer
diff --git a/tests/validation/fixtures/DeconvolutionLayerFixture.h b/tests/validation/fixtures/DeconvolutionLayerFixture.h
index b9a478b1a9..57951c0f36 100644
--- a/tests/validation/fixtures/DeconvolutionLayerFixture.h
+++ b/tests/validation/fixtures/DeconvolutionLayerFixture.h
@@ -46,7 +46,7 @@ template <typename TensorType, typename AccessorType, typename FunctionType, typ
 class DeconvolutionLayerFixtureBase : public framework::Fixture
 {
 public:
-    using TBias = typename std::conditional < std::is_same<T, uint8_t>::value || std::is_same<T, int8_t>::value, int32_t, T >::type;
+    using TBias = typename std::conditional < std::is_same<typename std::decay<T>::type, uint8_t>::value || std::is_same<typename std::decay<T>::type, int8_t>::value, int32_t, T >::type;
 
 public:
     template <typename...>