COMPMID-2237

Implement SPACE_TO_DEPTH for NEON

Change-Id: I9f427bceca6da52671e0096be08772612f4be152
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1227
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>

13 files changed, 880 insertions, 6 deletions

diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index d19b0c71ba..119f9b8702 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -129,6 +129,7 @@
 #include "arm_compute/core/NEON/kernels/NESobel7x7Kernel.h"
 #include "arm_compute/core/NEON/kernels/NESoftmaxLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NESpaceToBatchLayerKernel.h"
+#include "arm_compute/core/NEON/kernels/NESpaceToDepthLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEStackLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEStridedSliceKernel.h"
 #include "arm_compute/core/NEON/kernels/NETableLookupKernel.h"
diff --git a/arm_compute/core/NEON/kernels/NESpaceToDepthLayerKernel.h b/arm_compute/core/NEON/kernels/NESpaceToDepthLayerKernel.h
new file mode 100644
index 0000000000..c9ecdd26f8
--- /dev/null
+++ b/arm_compute/core/NEON/kernels/NESpaceToDepthLayerKernel.h
@@ -0,0 +1,80 @@
+/*
+ * Copyright (c) 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_NESPACETODEPTHLAYERKERNEL_H__
+#define __ARM_COMPUTE_NESPACETODEPTHLAYERKERNEL_H__
+
+#include "arm_compute/core/NEON/INEKernel.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Interface for the space to depth kernel */
+class NESpaceToDepthLayerKernel : public INEKernel
+{
+public:
+    const char *name() const override
+    {
+        return "NESpaceToDepthLayerKernel";
+    }
+    /** Default constructor */
+    NESpaceToDepthLayerKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NESpaceToDepthLayerKernel(const NESpaceToDepthLayerKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NESpaceToDepthLayerKernel &operator=(const NESpaceToDepthLayerKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    NESpaceToDepthLayerKernel(NESpaceToDepthLayerKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    NESpaceToDepthLayerKernel &operator=(NESpaceToDepthLayerKernel &&) = default;
+    /** Default destructor */
+    ~NESpaceToDepthLayerKernel() = default;
+    /** Initialise the kernel's inputs and output.
+     *
+     * @param[in]  input       Tensor input. Supported tensor rank: 4. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[out] output      Tensor output. Data types supported: same as @p input
+     * @param[in]  block_shape Block shape value
+     */
+    void configure(const ITensor *input, ITensor *output, int32_t block_shape);
+    /** Static function to check if given info will lead to a valid configuration of @ref NESpaceToDepthLayerKernel
+     *
+     * @param[in] input       Tensor input info. Supported tensor rank: 4. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[in] output      Tensor output info. Data types supported: same as @p input
+     * @param[in] block_shape Block shape value
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, int32_t block_shape);
+
+    // Inherited methods overridden:
+    void run(const Window &window, const ThreadInfo &info) override;
+
+private:
+    const ITensor *_input;       /**< Source tensor */
+    ITensor       *_output;      /**< Destination tensor */
+    int32_t        _block_shape; /**< Block shape */
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_NESPACETODEPTHLAYERKERNEL_H__ */
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index 8394bbaed8..7eab17ba11 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -1058,6 +1058,29 @@ inline TensorShape compute_space_to_batch_shape(const ITensorInfo *input, const
     return output_shape;
 }
 
+/** Calculate the space to batch output shape of a tensor
+ *
+ * @param[in] input       Input tensor info
+ * @param[in] block_shape Block shape value
+ *
+ * @return the calculated shape
+ */
+inline TensorShape compute_space_to_depth_shape(const ITensorInfo *input, int32_t block_shape)
+{
+    TensorShape output_shape{ input->tensor_shape() };
+
+    const DataLayout data_layout = input->data_layout();
+    const int        idx_width   = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
+    const int        idx_height  = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
+    const int        idx_depth   = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
+
+    output_shape.set(idx_width, input->tensor_shape()[idx_width] * block_shape);
+    output_shape.set(idx_height, input->tensor_shape()[idx_height] * block_shape);
+    output_shape.set(idx_depth, input->tensor_shape()[idx_depth] / (block_shape * block_shape));
+
+    return output_shape;
+}
+
 /** Calculate the prior box output shape of a tensor
  *
  * @param[in] input Input tensor info
diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index 971ddfa703..4e0cdd7a0a 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -132,6 +132,7 @@
 #include "arm_compute/runtime/NEON/functions/NESobel7x7.h"
 #include "arm_compute/runtime/NEON/functions/NESoftmaxLayer.h"
 #include "arm_compute/runtime/NEON/functions/NESpaceToBatchLayer.h"
+#include "arm_compute/runtime/NEON/functions/NESpaceToDepthLayer.h"
 #include "arm_compute/runtime/NEON/functions/NESplit.h"
 #include "arm_compute/runtime/NEON/functions/NEStackLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEStridedSlice.h"
diff --git a/arm_compute/runtime/NEON/functions/NESpaceToDepthLayer.h b/arm_compute/runtime/NEON/functions/NESpaceToDepthLayer.h
new file mode 100644
index 0000000000..abcfed44b8
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NESpaceToDepthLayer.h
@@ -0,0 +1,80 @@
+/*
+ * Copyright (c) 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_NESPACETODEPTHLAYER_H__
+#define __ARM_COMPUTE_NESPACETODEPTHLAYER_H__
+
+#include "arm_compute/runtime/IFunction.h"
+
+#include "arm_compute/core/NEON/kernels/NEMemsetKernel.h"
+#include "arm_compute/core/NEON/kernels/NESpaceToDepthLayerKernel.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** This function calls the following NEON kernels/functions:
+ *
+ *  -# @ref NESpaceToDepthLayerKernel
+ */
+class NESpaceToDepthLayer : public IFunction
+{
+public:
+    /** Default constructor */
+    NESpaceToDepthLayer();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NESpaceToDepthLayer(const NESpaceToDepthLayer &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NESpaceToDepthLayer &operator=(const NESpaceToDepthLayer &) = delete;
+    /** Allow instances of this class to be moved */
+    NESpaceToDepthLayer(NESpaceToDepthLayer &&) = default;
+    /** Allow instances of this class to be moved */
+    NESpaceToDepthLayer &operator=(NESpaceToDepthLayer &&) = default;
+    /** Default destructor */
+    virtual ~NESpaceToDepthLayer() = default;
+    /** Set the input and output tensors.
+     *
+     * @param[in]  input       Tensor input. Supported tensor rank: 4. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[out] output      Tensor output. Data types supported: same as @p input
+     * @param[in]  block_shape Block shape value
+     */
+    void configure(const ITensor *input, ITensor *output, int32_t block_shape);
+    /** Static function to check if given info will lead to a valid configuration of @ref NESpaceToDepthLayer (Static block shape and paddings)
+     *
+     * @param[in] input       Tensor input info. Supported tensor rank: 4. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[in] output      Tensor output info. Data types supported: same as @p input
+     * @param[in] block_shape Block shape value
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, int32_t block_shape);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    NESpaceToDepthLayerKernel _space_to_depth_kernel; /**< SpaceToDepth kernel to run */
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_NESPACETODEPTHLAYER_H__ */
diff --git a/src/core/NEON/kernels/NESpaceToBatchLayerKernel.cpp b/src/core/NEON/kernels/NESpaceToBatchLayerKernel.cpp
index 2e46b149e3..511a109777 100644
--- a/src/core/NEON/kernels/NESpaceToBatchLayerKernel.cpp
+++ b/src/core/NEON/kernels/NESpaceToBatchLayerKernel.cpp
@@ -167,12 +167,6 @@ void NESpaceToBatchLayerKernel::run(const Window &window, const ThreadInfo &info
     const size_t batch_size = _input->info()->dimension(3);
 
     Window slice_out = window.first_slice_window_3D();
-    Window slice_in  = window.first_slice_window_4D();
-
-    slice_in.set(Window::DimX, Window::Dimension(0, 0, 0));
-    slice_in.set(Window::DimY, Window::Dimension(0, 0, 0));
-    slice_in.set(Window::DimZ, Window::Dimension(0, 0, 0));
-    slice_in.set(3, Window::Dimension(0, 0, 0));
 
     int batch_id = 0;
 
diff --git a/src/core/NEON/kernels/NESpaceToDepthLayerKernel.cpp b/src/core/NEON/kernels/NESpaceToDepthLayerKernel.cpp
new file mode 100644
index 0000000000..4803365013
--- /dev/null
+++ b/src/core/NEON/kernels/NESpaceToDepthLayerKernel.cpp
@@ -0,0 +1,153 @@
+/*
+ * Copyright (c) 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.
+ */
+#include "arm_compute/core/NEON/kernels/NESpaceToDepthLayerKernel.h"
+
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/ITensor.h"
+#include "arm_compute/core/NEON/wrapper/wrapper.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/Validate.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include <arm_neon.h>
+#include <cstdint>
+
+using namespace arm_compute::misc::shape_calculator;
+
+namespace arm_compute
+{
+namespace
+{
+Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, int32_t block_shape)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
+
+    ARM_COMPUTE_RETURN_ERROR_ON(block_shape < 1);
+
+    // Validate output if initialized
+    if(output->total_size() != 0)
+    {
+        const DataLayout data_layout = input->data_layout();
+        const int        idx_width   = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
+        const int        idx_height  = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
+        const int        idx_channel = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
+        const int        idx_batch   = get_data_layout_dimension_index(data_layout, DataLayoutDimension::BATCHES);
+        ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[idx_width] % block_shape != 0);
+        ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[idx_height] % block_shape != 0);
+        ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[idx_batch] != output->tensor_shape()[idx_batch]);
+        ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape()[idx_channel] % (block_shape * block_shape) != 0);
+        ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape().total_size() != output->tensor_shape().total_size());
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    }
+
+    return Status{};
+}
+} // namespace
+
+NESpaceToDepthLayerKernel::NESpaceToDepthLayerKernel()
+    : _input(nullptr), _output(nullptr), _block_shape()
+{
+}
+
+void NESpaceToDepthLayerKernel::configure(const ITensor *input, ITensor *output, int32_t block_shape)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+    TensorShape output_shape = misc::shape_calculator::compute_space_to_depth_shape(input->info(), block_shape);
+    auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type());
+
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), block_shape));
+
+    _input       = input;
+    _block_shape = block_shape;
+    _output      = output;
+
+    // Configure kernel window
+    Window win = calculate_max_window(*output->info(), Steps());
+    INEKernel::configure(win);
+}
+
+Status NESpaceToDepthLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, int32_t block_shape)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, block_shape));
+    return Status{};
+}
+
+void NESpaceToDepthLayerKernel::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);
+
+    const DataLayout data_layout  = _input->info()->data_layout();
+    const int        channel_idx  = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
+    const int        element_size = _input->info()->element_size();
+
+    const size_t channel_size = _input->info()->dimension(channel_idx);
+
+    Window slice_out = window.first_slice_window_3D();
+
+    int batch_id = 0;
+
+    // Main loop for NCHW and NHWC
+    if(_output->info()->data_layout() == DataLayout::NCHW)
+    {
+        do
+        {
+            Iterator out(_output, slice_out);
+            execute_window_loop(slice_out, [&](const Coordinates & id)
+            {
+                const size_t channel_id = id.z();
+                const size_t in_x       = id.x() * _block_shape + (channel_id / channel_size) % _block_shape;
+                const size_t in_y       = id.y() * _block_shape + (channel_id / channel_size) / _block_shape;
+                const int    z          = channel_id % channel_size;
+                Coordinates  input_coords{ in_x, in_y, z, batch_id };
+                memcpy(out.ptr(), _input->ptr_to_element(input_coords), element_size);
+            },
+            out);
+            ++batch_id;
+        }
+        while(window.slide_window_slice_3D(slice_out));
+    }
+    else
+    {
+        do
+        {
+            Iterator out(_output, slice_out);
+            execute_window_loop(slice_out, [&](const Coordinates & id)
+            {
+                const size_t channel_id = id.x();
+                const size_t in_x       = id.y() * _block_shape + (channel_id / channel_size) % _block_shape;
+                const size_t in_y       = id.z() * _block_shape + (channel_id / channel_size) / _block_shape;
+                const int    z          = channel_id % channel_size;
+                Coordinates  input_coords{ z, in_x, in_y, batch_id };
+                memcpy(out.ptr(), _input->ptr_to_element(input_coords), element_size);
+            },
+            out);
+            ++batch_id;
+        }
+        while(window.slide_window_slice_3D(slice_out));
+    }
+}
+} // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NESpaceToDepthLayer.cpp b/src/runtime/NEON/functions/NESpaceToDepthLayer.cpp
new file mode 100644
index 0000000000..18d82918c7
--- /dev/null
+++ b/src/runtime/NEON/functions/NESpaceToDepthLayer.cpp
@@ -0,0 +1,56 @@
+/*
+ * Copyright (c) 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.
+ */
+
+#include "arm_compute/runtime/NEON/functions/NESpaceToDepthLayer.h"
+
+#include "arm_compute/core/Error.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/Validate.h"
+#include "arm_compute/runtime/NEON/NEScheduler.h"
+
+namespace arm_compute
+{
+NESpaceToDepthLayer::NESpaceToDepthLayer()
+    : _space_to_depth_kernel()
+{
+}
+
+void NESpaceToDepthLayer::configure(const ITensor *input, ITensor *output, int32_t block_shape)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+    _space_to_depth_kernel.configure(input, output, block_shape);
+}
+
+Status NESpaceToDepthLayer::validate(const ITensorInfo *input, const ITensorInfo *output, int32_t block_shape)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(NESpaceToDepthLayerKernel::validate(input, output, block_shape));
+    return Status{};
+}
+
+void NESpaceToDepthLayer::run()
+{
+    NEScheduler::get().schedule(&_space_to_depth_kernel, Window::DimY);
+}
+} // namespace arm_compute
diff --git a/tests/datasets/SpaceToDepthDataset.h b/tests/datasets/SpaceToDepthDataset.h
new file mode 100644
index 0000000000..563a475d79
--- /dev/null
+++ b/tests/datasets/SpaceToDepthDataset.h
@@ -0,0 +1,135 @@
+/*
+ * Copyright (c) 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_TEST_SPACE_TO_DEPTH_LAYER_DATASET
+#define ARM_COMPUTE_TEST_SPACE_TO_DEPTH_LAYER_DATASET
+
+#include "utils/TypePrinter.h"
+
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace datasets
+{
+class SpaceToDepthLayerDataset
+{
+public:
+    using type = std::tuple<TensorShape, TensorShape, int32_t>;
+
+    struct iterator
+    {
+        iterator(std::vector<TensorShape>::const_iterator src_it,
+                 std::vector<TensorShape>::const_iterator dst_it,
+                 std::vector<int>::const_iterator         block_shape_it)
+            : _src_it{ std::move(src_it) },
+              _dst_it{ std::move(dst_it) },
+              _block_shape_it{ std::move(block_shape_it) }
+        {
+        }
+
+        std::string description() const
+        {
+            std::stringstream description;
+            description << "In=" << *_src_it << ":";
+            description << "Out=" << *_dst_it;
+            description << "BlockShape=" << *_block_shape_it << ":";
+            return description.str();
+        }
+
+        SpaceToDepthLayerDataset::type operator*() const
+        {
+            return std::make_tuple(*_src_it, *_dst_it, *_block_shape_it);
+        }
+
+        iterator &operator++()
+        {
+            ++_src_it;
+            ++_dst_it;
+            ++_block_shape_it;
+
+            return *this;
+        }
+
+    private:
+        std::vector<TensorShape>::const_iterator _src_it;
+        std::vector<TensorShape>::const_iterator _dst_it;
+        std::vector<int32_t>::const_iterator     _block_shape_it;
+    };
+
+    iterator begin() const
+    {
+        return iterator(_src_shapes.begin(), _dst_shapes.begin(), _block_shape.begin());
+    }
+
+    int size() const
+    {
+        return std::min(_src_shapes.size(), std::min(_dst_shapes.size(), _block_shape.size()));
+    }
+
+    void add_config(TensorShape src, TensorShape dst, int32_t block_shape)
+    {
+        _src_shapes.emplace_back(std::move(src));
+        _dst_shapes.emplace_back(std::move(dst));
+        _block_shape.emplace_back(std::move(block_shape));
+    }
+
+protected:
+    SpaceToDepthLayerDataset()                            = default;
+    SpaceToDepthLayerDataset(SpaceToDepthLayerDataset &&) = default;
+
+private:
+    std::vector<TensorShape> _src_shapes{};
+    std::vector<TensorShape> _dst_shapes{};
+    std::vector<int32_t>     _block_shape{};
+};
+
+class SmallSpaceToDepthLayerDataset final : public SpaceToDepthLayerDataset
+{
+public:
+    SmallSpaceToDepthLayerDataset()
+    {
+        add_config(TensorShape(2U, 2U, 1U, 1U), TensorShape(1U, 1U, 4U, 1U), 2);
+        add_config(TensorShape(6U, 2U, 1U, 1U), TensorShape(3U, 1U, 4U, 1U), 2);
+        add_config(TensorShape(2U, 4U, 2U, 1U), TensorShape(1U, 2U, 8U, 1U), 2);
+        add_config(TensorShape(2U, 6U, 1U, 2U), TensorShape(1U, 3U, 4U, 2U), 2);
+        add_config(TensorShape(6U, 8U, 1U, 1U), TensorShape(3U, 4U, 4U, 1U), 2);
+        add_config(TensorShape(6U, 8U, 15U, 5U), TensorShape(3U, 4U, 60U, 5U), 2);
+    }
+};
+class LargeSpaceToDepthLayerDataset final : public SpaceToDepthLayerDataset
+{
+public:
+    LargeSpaceToDepthLayerDataset()
+    {
+        add_config(TensorShape(128U, 64U, 2U, 1U), TensorShape(64U, 32U, 8U, 1U), 2);
+        add_config(TensorShape(512U, 64U, 2U, 1U), TensorShape(128U, 16U, 8U, 4U), 2);
+    }
+};
+} // namespace datasets
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_SPACE_TO_DEPTH_LAYER_DATASET */
diff --git a/tests/validation/NEON/SpaceToDepthLayer.cpp b/tests/validation/NEON/SpaceToDepthLayer.cpp
new file mode 100644
index 0000000000..8d2726168c
--- /dev/null
+++ b/tests/validation/NEON/SpaceToDepthLayer.cpp
@@ -0,0 +1,113 @@
+/*
+ * Copyright (c) 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.
+ */
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/functions/NESpaceToDepthLayer.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "arm_compute/runtime/TensorAllocator.h"
+#include "tests/NEON/Accessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/datasets/SpaceToDepthDataset.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/SpaceToDepthFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+TEST_SUITE(NEON)
+TEST_SUITE(SpaceToDepthLayer)
+
+template <typename T>
+using NESpaceToDepthLayerFixture = SpaceToDepthLayerValidationFixture<Tensor, Accessor, NESpaceToDepthLayer, T>;
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(ValidateStatic, framework::DatasetMode::ALL, zip(zip(zip(
+               framework::dataset::make("InputInfo", { TensorInfo(TensorShape(32U, 16U, 2U, 1U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 16U, 2U, 1U), 1, DataType::F32),    // Mismatching data types
+                                                       TensorInfo(TensorShape(32U, 16U, 2U, 1U), 1, DataType::F32),    // Negative block shapes
+                                                       TensorInfo(TensorShape(32U, 16U, 2U, 1U, 4U), 1, DataType::F32), // Wrong tensor shape
+                                                     }),
+               framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(16U, 8U, 8U, 1U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 8U, 8U, 1U), 1, DataType::F16),
+                                                       TensorInfo(TensorShape(32U, 8U, 8U, 1U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 8U, 8U, 1U), 1, DataType::F32),
+                                                     })),
+               framework::dataset::make("BlockShape", { 2, 2, -2, 2 })),
+               framework::dataset::make("Expected", { true, false, false, false})),
+               input_info, output_info, block_shape, expected)
+{
+    bool has_error = bool(NESpaceToDepthLayer::validate(&input_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false), block_shape));
+    ARM_COMPUTE_EXPECT(has_error == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
+TEST_SUITE(Float)
+TEST_SUITE(FP32)
+FIXTURE_DATA_TEST_CASE(Small, NESpaceToDepthLayerFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallSpaceToDepthLayerDataset(), framework::dataset::make("DataType",
+                                                                                                                    DataType::F32)),
+                                                                                                            framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+FIXTURE_DATA_TEST_CASE(Large, NESpaceToDepthLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeSpaceToDepthLayerDataset(), framework::dataset::make("DataType",
+                                                                                                                  DataType::F32)),
+                                                                                                          framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // FP32
+
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(Small, NESpaceToDepthLayerFixture<half>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallSpaceToDepthLayerDataset(), framework::dataset::make("DataType",
+                                                                                                                   DataType::F16)),
+                                                                                                           framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+FIXTURE_DATA_TEST_CASE(Large, NESpaceToDepthLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeSpaceToDepthLayerDataset(), framework::dataset::make("DataType",
+                                                                                                                 DataType::F16)),
+                                                                                                         framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // FP16
+TEST_SUITE_END() // Float
+
+TEST_SUITE_END() // SpaceToDepthLayer
+TEST_SUITE_END() // NEON
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/SpaceToDepthFixture.h b/tests/validation/fixtures/SpaceToDepthFixture.h
new file mode 100644
index 0000000000..170fdfa397
--- /dev/null
+++ b/tests/validation/fixtures/SpaceToDepthFixture.h
@@ -0,0 +1,110 @@
+/*
+ * Copyright (c) 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_TEST_SPACE_TO_DEPTH_LAYER_FIXTURE
+#define ARM_COMPUTE_TEST_SPACE_TO_DEPTH_LAYER_FIXTURE
+
+#include "tests/Globals.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Fixture.h"
+#include "tests/validation/reference/SpaceToDepth.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class SpaceToDepthLayerValidationFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape input_shape, TensorShape output_shape, const int block_shape, DataType data_type, DataLayout data_layout)
+    {
+        _target    = compute_target(input_shape, output_shape, block_shape, data_type, data_layout);
+        _reference = compute_reference(input_shape, output_shape, block_shape, data_type);
+    }
+
+protected:
+    template <typename U>
+    void fill(U &&tensor, int i)
+    {
+        std::uniform_real_distribution<> distribution(-1.0f, 1.0f);
+        library->fill(tensor, distribution, i);
+    }
+    TensorType compute_target(TensorShape input_shape, TensorShape output_shape, const int block_shape,
+                              DataType data_type, DataLayout data_layout)
+    {
+        if(data_layout == DataLayout::NHWC)
+        {
+            permute(input_shape, PermutationVector(2U, 0U, 1U));
+            permute(output_shape, PermutationVector(2U, 0U, 1U));
+        }
+
+        // Create tensors
+        TensorType input  = create_tensor<TensorType>(input_shape, data_type, 1, QuantizationInfo(), data_layout);
+        TensorType output = create_tensor<TensorType>(output_shape, data_type, 1, QuantizationInfo(), data_layout);
+
+        // Create and configure function
+        FunctionType space_to_depth;
+        space_to_depth.configure(&input, &output, block_shape);
+
+        ARM_COMPUTE_EXPECT(input.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(output.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Allocate tensors
+        input.allocator()->allocate();
+        output.allocator()->allocate();
+
+        ARM_COMPUTE_EXPECT(!input.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(!output.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Fill tensors
+        fill(AccessorType(input), 0);
+        // Compute function
+        space_to_depth.run();
+
+        return output;
+    }
+
+    SimpleTensor<T> compute_reference(const TensorShape &input_shape, const TensorShape &output_shape,
+                                      const int block_shape, DataType data_type)
+    {
+        // Create reference
+        SimpleTensor<T> input{ input_shape, data_type };
+
+        // Fill reference
+        fill(input, 0);
+
+        // Compute reference
+        return reference::space_to_depth(input, output_shape, block_shape);
+    }
+
+    TensorType      _target{};
+    SimpleTensor<T> _reference{};
+};
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_SPACE_TO_DEPTH_LAYER_FIXTURE */
diff --git a/tests/validation/reference/SpaceToDepth.cpp b/tests/validation/reference/SpaceToDepth.cpp
new file mode 100644
index 0000000000..bd8e37a8bf
--- /dev/null
+++ b/tests/validation/reference/SpaceToDepth.cpp
@@ -0,0 +1,84 @@
+/*
+ * Copyright (c) 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.
+ */
+#include "SpaceToDepth.h"
+
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+// Space to Batch
+template <typename T>
+SimpleTensor<T> space_to_depth(const SimpleTensor<T> &src, const TensorShape &dst_shape, const int block_shape)
+{
+    SimpleTensor<T> result(dst_shape, src.data_type());
+
+    const auto width_out   = static_cast<int>(dst_shape[0]);
+    const auto height_out  = static_cast<int>(dst_shape[1]);
+    const auto channel_out = static_cast<int>(dst_shape[2]);
+
+    const auto width_in   = static_cast<int>(src.shape()[0]);
+    const auto height_in  = static_cast<int>(src.shape()[1]);
+    const auto channel_in = static_cast<int>(src.shape()[2]);
+
+    const auto batch = static_cast<int>(src.shape()[3]);
+
+    const auto block_width  = block_shape;
+    const auto block_height = block_shape;
+
+    int out_pos = 0;
+    for(int ba = 0; ba < batch; ++ba)
+    {
+        for(int outC = 0; outC < channel_out; ++outC)
+        {
+            unsigned int inC = outC % channel_in;
+
+            int shift_w = (outC / channel_in) % block_width;
+            int shift_h = (outC / channel_in) / block_width;
+
+            for(int outH = 0; outH < height_out; ++outH)
+            {
+                for(int outW = 0; outW < width_out; ++outW)
+                {
+                    const auto in_pos = ((ba * channel_in + inC) * height_in + ((outH * block_height + shift_h))) * width_in + (outW * block_width + shift_w);
+                    result[out_pos]   = src[in_pos];
+                    ++out_pos;
+                }
+            }
+        }
+    }
+    return result;
+}
+
+template SimpleTensor<float> space_to_depth(const SimpleTensor<float> &src, const TensorShape &dst_shape, const int block_shape);
+template SimpleTensor<half> space_to_depth(const SimpleTensor<half> &src, const TensorShape &dst_shape, const int block_shape);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/reference/SpaceToDepth.h b/tests/validation/reference/SpaceToDepth.h
new file mode 100644
index 0000000000..885c6153eb
--- /dev/null
+++ b/tests/validation/reference/SpaceToDepth.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 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_TEST_SPACE_TO_DEPTH_LAYER_H__
+#define __ARM_COMPUTE_TEST_SPACE_TO_DEPTH_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> space_to_depth(const SimpleTensor<T> &src, const TensorShape &dst_shape, const int block_shape);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_SPACE_TO_DEPTH_LAYER_H__ */