COMPMID-2240 Implement DEPTH_TO_SPACE for NEON

Change-Id: I705aa0f804093c3628c691e46cca475f2819dc65
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1198
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Giuseppe Rossini <giuseppe.rossini@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

12 files changed, 851 insertions, 0 deletions

diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index b9716b1e43..d19b0c71ba 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -50,6 +50,7 @@
 #include "arm_compute/core/NEON/kernels/NECumulativeDistributionKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthConcatenateLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthConvertLayerKernel.h"
+#include "arm_compute/core/NEON/kernels/NEDepthToSpaceLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthwiseIm2ColKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthwiseVectorToTensorKernel.h"
diff --git a/arm_compute/core/NEON/kernels/NEDepthToSpaceLayerKernel.h b/arm_compute/core/NEON/kernels/NEDepthToSpaceLayerKernel.h
new file mode 100644
index 0000000000..a62da049a5
--- /dev/null
+++ b/arm_compute/core/NEON/kernels/NEDepthToSpaceLayerKernel.h
@@ -0,0 +1,79 @@
+/*
+ * 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_NEDEPTHTOSPACELAYERKERNEL_H__
+#define __ARM_COMPUTE_NEDEPTHTOSPACELAYERKERNEL_H__
+
+#include "arm_compute/core/NEON/INEKernel.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Interface for the depth to space kernel */
+class NEDepthToSpaceLayerKernel : public INEKernel
+{
+public:
+    const char *name() const override
+    {
+        return "NEDepthToSpaceLayerKernel";
+    }
+    /** Default constructor */
+    NEDepthToSpaceLayerKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEDepthToSpaceLayerKernel(const NEDepthToSpaceLayerKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEDepthToSpaceLayerKernel &operator=(const NEDepthToSpaceLayerKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    NEDepthToSpaceLayerKernel(NEDepthToSpaceLayerKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    NEDepthToSpaceLayerKernel &operator=(NEDepthToSpaceLayerKernel &&) = default;
+    /** Default destructor */
+    ~NEDepthToSpaceLayerKernel() = 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 x 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 NEDepthToSpaceLayerKernel.
+     *
+     * @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_NEDEPTHTOSPACELAYERKERNEL_H__ */
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index d66c87fb42..8394bbaed8 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -975,6 +975,30 @@ inline TensorShape compute_batch_to_space_shape(const ITensorInfo *input, const
     return output_shape;
 }
 
+/** Calculate the depth to space output shape of a tensor
+ *
+ * @param[in] input Input tensor info
+ * @param[in] block Block shape value
+ *
+ * @return the calculated shape
+ */
+inline TensorShape compute_depth_to_space_shape(const ITensorInfo *input, int block)
+{
+    ARM_COMPUTE_ERROR_ON(block < 2);
+
+    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);
+
+    TensorShape output_shape{ input->tensor_shape() };
+    output_shape.set(idx_width, input->dimension(idx_width) * block);
+    output_shape.set(idx_height, input->dimension(idx_height) * block);
+    output_shape.set(idx_channel, input->dimension(idx_channel) / (block * block));
+
+    return output_shape;
+}
+
 /** Calculate the split 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 9094ee4bdf..971ddfa703 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -53,6 +53,7 @@
 #include "arm_compute/runtime/NEON/functions/NECropResize.h"
 #include "arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDepthConvertLayer.h"
+#include "arm_compute/runtime/NEON/functions/NEDepthToSpaceLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDepthwiseConvolutionLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDepthwiseSeparableConvolutionLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDequantizationLayer.h"
diff --git a/arm_compute/runtime/NEON/functions/NEDepthToSpaceLayer.h b/arm_compute/runtime/NEON/functions/NEDepthToSpaceLayer.h
new file mode 100644
index 0000000000..54a6ddaefc
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NEDepthToSpaceLayer.h
@@ -0,0 +1,59 @@
+/*
+ * 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_NEDEPTHTOSPACELAYER_H__
+#define __ARM_COMPUTE_NEDEPTHTOSPACELAYER_H__
+
+#include "arm_compute/runtime/IFunction.h"
+
+#include "arm_compute/core/NEON/kernels/NEDepthToSpaceLayerKernel.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/INESimpleFunctionNoBorder.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Basic function to run @ref NEDepthToSpaceLayerKernel. */
+class NEDepthToSpaceLayer : public INESimpleFunctionNoBorder
+{
+public:
+    /** 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 NEDepthToSpaceLayer.
+     *
+     * @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 x value.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, int32_t block_shape);
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_NEDEPTHTOSPACELAYER_H__ */
diff --git a/src/core/NEON/kernels/NEDepthToSpaceLayerKernel.cpp b/src/core/NEON/kernels/NEDepthToSpaceLayerKernel.cpp
new file mode 100644
index 0000000000..df631c3c03
--- /dev/null
+++ b/src/core/NEON/kernels/NEDepthToSpaceLayerKernel.cpp
@@ -0,0 +1,158 @@
+/*
+ * 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/NEDepthToSpaceLayerKernel.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 < 2);
+
+    const DataLayout data_layout = input->data_layout();
+    const int        idx_channel = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
+    ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[idx_channel] % (block_shape * block_shape) != 0);
+    // Validate output if initialized
+    if(output->total_size() != 0)
+    {
+        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);
+        ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape()[idx_width] != (block_shape * input->tensor_shape()[idx_width]));
+        ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape()[idx_height] != (block_shape * input->tensor_shape()[idx_height]));
+        ARM_COMPUTE_RETURN_ERROR_ON(output->num_dimensions() > 4);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    }
+
+    return Status{};
+}
+} // namespace
+
+NEDepthToSpaceLayerKernel::NEDepthToSpaceLayerKernel()
+    : _input(nullptr), _output(nullptr), _block_shape()
+{
+}
+
+void NEDepthToSpaceLayerKernel::configure(const ITensor *input, ITensor *output, int32_t block_shape)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+    TensorShape output_shape = compute_depth_to_space_shape(input->info(), block_shape);
+    // Output auto inizialitation if not yet initialized
+    auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
+
+    // Perform validation step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), block_shape));
+
+    _input       = input;
+    _output      = output;
+    _block_shape = block_shape;
+
+    // Configure kernel window
+    Window win = calculate_max_window(*input->info(), Steps());
+    ICPPKernel::configure(win);
+}
+
+Status NEDepthToSpaceLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, int32_t block_shape)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, block_shape));
+    return Status{};
+}
+
+void NEDepthToSpaceLayerKernel::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 int idx_channel  = get_data_layout_dimension_index(_input->info()->data_layout(), DataLayoutDimension::CHANNEL);
+    const int depth_size   = _input->info()->dimension(idx_channel);
+    const int r            = (depth_size / (_block_shape * _block_shape));
+    const int element_size = _input->info()->element_size();
+
+    Window slice_out = window.first_slice_window_3D();
+
+    // The slice_out slice does not move
+    slice_out.set(Window::DimX, Window::Dimension(0, 0, 0));
+    slice_out.set(Window::DimY, Window::Dimension(0, 0, 0));
+    slice_out.set(Window::DimZ, Window::Dimension(0, 0, 0));
+
+    // Main loop for NCHW and NHWC
+    if(_input->info()->data_layout() == DataLayout::NCHW)
+    {
+        Window slice_in = window.first_slice_window_2D();
+        do
+        {
+            Iterator in(_input, slice_in);
+            execute_window_loop(slice_in, [&](const Coordinates & id)
+            {
+                const int x = id.x();
+                const int y = id.y();
+
+                const int   z     = id.z() % r;
+                const int   out_x = x * _block_shape + (id.z() / r) % _block_shape;
+                const int   out_y = y * _block_shape + (id.z() / r) / _block_shape;
+                Coordinates output_coords{ out_x, out_y, z, id[3] };
+                memcpy(_output->ptr_to_element(output_coords), in.ptr(), element_size);
+            },
+            in);
+        }
+        while(window.slide_window_slice_2D(slice_in));
+    }
+    else
+    {
+        Window slice_in = window.first_slice_window_3D();
+        do
+        {
+            Iterator in(_input, slice_in);
+            execute_window_loop(slice_in, [&](const Coordinates & id)
+            {
+                const int x = id.y();
+                const int y = id.z();
+
+                const int   z     = id.x() % r;
+                const int   out_x = x * _block_shape + (id.x() / r) % _block_shape;
+                const int   out_y = y * _block_shape + (id.x() / r) / _block_shape;
+                Coordinates output_coords{ z, out_x, out_y, id[3] };
+                memcpy(_output->ptr_to_element(output_coords), in.ptr(), element_size);
+            },
+            in);
+        }
+        while(window.slide_window_slice_3D(slice_in));
+    }
+}
+} // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NEDepthToSpaceLayer.cpp b/src/runtime/NEON/functions/NEDepthToSpaceLayer.cpp
new file mode 100644
index 0000000000..3569eecf0a
--- /dev/null
+++ b/src/runtime/NEON/functions/NEDepthToSpaceLayer.cpp
@@ -0,0 +1,45 @@
+/*
+ * 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/NEDepthToSpaceLayer.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"
+
+namespace arm_compute
+{
+void NEDepthToSpaceLayer::configure(const ITensor *input, ITensor *output, int32_t block_shape)
+{
+    auto k = arm_compute::support::cpp14::make_unique<NEDepthToSpaceLayerKernel>();
+    k->configure(input, output, block_shape);
+    _kernel = std::move(k);
+}
+
+Status NEDepthToSpaceLayer::validate(const ITensorInfo *input, const ITensorInfo *output, int32_t block_shape)
+{
+    return NEDepthToSpaceLayerKernel::validate(input, output, block_shape);
+}
+} // namespace arm_compute
diff --git a/tests/datasets/DepthToSpaceDataset.h b/tests/datasets/DepthToSpaceDataset.h
new file mode 100644
index 0000000000..06ee5ba432
--- /dev/null
+++ b/tests/datasets/DepthToSpaceDataset.h
@@ -0,0 +1,139 @@
+/*
+ * 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_DEPTH_TO_SPACE_LAYER_DATASET
+#define ARM_COMPUTE_TEST_DEPTH_TO_SPACE_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 DepthToSpaceLayerDataset
+{
+public:
+    using type = std::tuple<TensorShape, int32_t, TensorShape>;
+
+    struct iterator
+    {
+        iterator(std::vector<TensorShape>::const_iterator src_it,
+                 std::vector<int32_t>::const_iterator     block_shape_it,
+                 std::vector<TensorShape>::const_iterator dst_it)
+            : _src_it{ std::move(src_it) },
+              _block_shape_it{ std::move(block_shape_it) },
+              _dst_it{ std::move(dst_it) }
+        {
+        }
+
+        std::string description() const
+        {
+            std::stringstream description;
+            description << "In=" << *_src_it << ":";
+            description << "BlockShape=" << *_block_shape_it << ":";
+            description << "Out=" << *_dst_it;
+            return description.str();
+        }
+
+        DepthToSpaceLayerDataset::type operator*() const
+        {
+            return std::make_tuple(*_src_it, *_block_shape_it, *_dst_it);
+        }
+
+        iterator &operator++()
+        {
+            ++_src_it;
+            ++_block_shape_it;
+            ++_dst_it;
+
+            return *this;
+        }
+
+    private:
+        std::vector<TensorShape>::const_iterator _src_it;
+        std::vector<int32_t>::const_iterator     _block_shape_it;
+        std::vector<TensorShape>::const_iterator _dst_it;
+    };
+
+    iterator begin() const
+    {
+        return iterator(_src_shapes.begin(), _block_shape.begin(), _dst_shapes.begin());
+    }
+
+    int size() const
+    {
+        return std::min(_src_shapes.size(), std::min(_block_shape.size(), _dst_shapes.size()));
+    }
+
+    void add_config(TensorShape src, int32_t block_shape, TensorShape dst)
+    {
+        _src_shapes.emplace_back(std::move(src));
+        _block_shape.emplace_back(std::move(block_shape));
+        _dst_shapes.emplace_back(std::move(dst));
+    }
+
+protected:
+    DepthToSpaceLayerDataset()                            = default;
+    DepthToSpaceLayerDataset(DepthToSpaceLayerDataset &&) = default;
+
+private:
+    std::vector<TensorShape> _src_shapes{};
+    std::vector<int32_t>     _block_shape{};
+    std::vector<TensorShape> _dst_shapes{};
+};
+
+class SmallDepthToSpaceLayerDataset final : public DepthToSpaceLayerDataset
+{
+public:
+    SmallDepthToSpaceLayerDataset()
+    {
+        add_config(TensorShape(1U, 1U, 4U, 4U), 2U, TensorShape(2U, 2U, 1U, 4U));
+        add_config(TensorShape(3U, 1U, 4U, 4U), 2U, TensorShape(6U, 2U, 1U, 4U));
+        add_config(TensorShape(1U, 2U, 4U, 4U), 2U, TensorShape(2U, 4U, 1U, 4U));
+        add_config(TensorShape(1U, 3U, 4U, 8U), 2U, TensorShape(2U, 6U, 1U, 8U));
+        add_config(TensorShape(3U, 4U, 4U, 4U), 2U, TensorShape(6U, 8U, 1U, 4U));
+        add_config(TensorShape(1U, 1U, 16U, 8U), 4U, TensorShape(4U, 4U, 1U, 8U));
+        add_config(TensorShape(1U, 1U, 8U), 2U, TensorShape(2U, 2U, 2U));
+    }
+};
+
+class LargeDepthToSpaceLayerDataset final : public DepthToSpaceLayerDataset
+{
+public:
+    LargeDepthToSpaceLayerDataset()
+    {
+        add_config(TensorShape(64U, 32U, 4U), 2U, TensorShape(128U, 64U, 1U));
+        add_config(TensorShape(128U, 16U, 16U), 4U, TensorShape(512U, 64U, 1U));
+        add_config(TensorShape(16U, 8U, 8U), 2U, TensorShape(32U, 16U, 2U));
+        add_config(TensorShape(8U, 16U, 16U), 2U, TensorShape(16U, 32U, 4U));
+    }
+};
+} // namespace datasets
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_DEPTH_TO_SPACE_LAYER_DATASET */
diff --git a/tests/validation/NEON/DepthToSpaceLayer.cpp b/tests/validation/NEON/DepthToSpaceLayer.cpp
new file mode 100644
index 0000000000..abc8c4f266
--- /dev/null
+++ b/tests/validation/NEON/DepthToSpaceLayer.cpp
@@ -0,0 +1,115 @@
+/*
+ * 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/NEDepthToSpaceLayer.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/DepthToSpaceDataset.h"
+#include "tests/datasets/ShapeDatasets.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/DepthToSpaceLayerFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+TEST_SUITE(NEON)
+TEST_SUITE(DepthToSpaceLayer)
+
+template <typename T>
+using NEDepthToSpaceLayerFixture = DepthToSpaceLayerValidationFixture<Tensor, Accessor, NEDepthToSpaceLayer, T>;
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
+               framework::dataset::make("InputInfo", { TensorInfo(TensorShape(16U, 8U, 4U, 4U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(16U, 8U, 4U, 4U), 1, DataType::F32),   // block < 2
+                                                       TensorInfo(TensorShape(16U, 8U, 2U, 4U), 1, DataType::F32),    // Mismatching data types
+                                                       TensorInfo(TensorShape(16U, 8U, 2U, 4U), 1, DataType::F32),    // Negative block shape
+                                                       TensorInfo(TensorShape(32U, 16U, 2U, 4U, 4U), 1, DataType::F32), // Wrong tensor shape
+                                                     }),
+               framework::dataset::make("BlockShape", { 2, 1, 2, 2, 2 })),
+               framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(32U, 16U, 1U, 4U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(64U, 16U, 1U, 4U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 16U, 2U, 1U), 1, DataType::F16),
+                                                       TensorInfo(TensorShape(32U, 16U, 2U, 1U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 8U, 2U, 1U), 1, DataType::F32),
+                                                     })),
+               framework::dataset::make("Expected", { true, false, false, false, false})),
+               input_info, block_shape, output_info, expected)
+{
+    bool has_error = bool(NEDepthToSpaceLayer::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(RunSmall, NEDepthToSpaceLayerFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallDepthToSpaceLayerDataset(), framework::dataset::make("DataType",
+                                                                                                                       DataType::F32)),
+                                                                                                               framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthToSpaceLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeDepthToSpaceLayerDataset(), framework::dataset::make("DataType",
+                                                                                                                     DataType::F32)),
+                                                                                                             framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END()
+
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthToSpaceLayerFixture<half>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallDepthToSpaceLayerDataset(), framework::dataset::make("DataType",
+                                                                                                                      DataType::F16)),
+                                                                                                              framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthToSpaceLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeDepthToSpaceLayerDataset(), framework::dataset::make("DataType",
+                                                                                                                    DataType::F16)),
+                                                                                                            framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END()
+TEST_SUITE_END()
+
+TEST_SUITE_END() // DepthToSpace
+TEST_SUITE_END() // NEON
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/DepthToSpaceLayerFixture.h b/tests/validation/fixtures/DepthToSpaceLayerFixture.h
new file mode 100644
index 0000000000..b42bed52a7
--- /dev/null
+++ b/tests/validation/fixtures/DepthToSpaceLayerFixture.h
@@ -0,0 +1,111 @@
+/*
+ * 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_DEPTH_TO_SPACE_LAYER_FIXTURE
+#define ARM_COMPUTE_TEST_DEPTH_TO_SPACE_LAYER_FIXTURE
+
+#include "tests/Globals.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Fixture.h"
+#include "tests/validation/reference/DepthToSpaceLayer.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class DepthToSpaceLayerValidationFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape input_shape, int32_t block_shape, TensorShape output_shape, DataType data_type, DataLayout data_layout)
+    {
+        _target    = compute_target(input_shape, block_shape, output_shape, data_type, data_layout);
+        _reference = compute_reference(input_shape, block_shape, output_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, int32_t block_shape, TensorShape output_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 depth_to_space;
+        depth_to_space.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
+        depth_to_space.run();
+
+        return output;
+    }
+
+    SimpleTensor<T> compute_reference(const TensorShape &input_shape, int32_t block_shape,
+                                      const TensorShape &output_shape, DataType data_type)
+    {
+        // Create reference
+        SimpleTensor<T> input{ input_shape, data_type };
+
+        // Fill reference
+        fill(input, 0);
+
+        // Compute reference
+        return reference::depth_to_space(input, output_shape, block_shape);
+    }
+
+    TensorType      _target{};
+    SimpleTensor<T> _reference{};
+};
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_DEPTH_TO_SPACE_LAYER_FIXTURE */
diff --git a/tests/validation/reference/DepthToSpaceLayer.cpp b/tests/validation/reference/DepthToSpaceLayer.cpp
new file mode 100644
index 0000000000..4135ce5471
--- /dev/null
+++ b/tests/validation/reference/DepthToSpaceLayer.cpp
@@ -0,0 +1,75 @@
+/*
+ * 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 "DepthToSpaceLayer.h"
+
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+// Batch to Space
+template <typename T>
+SimpleTensor<T> depth_to_space(const SimpleTensor<T> &src, const TensorShape &dst_shape, int32_t block_shape)
+{
+    ARM_COMPUTE_ERROR_ON(block_shape <= 0);
+    SimpleTensor<T> result(dst_shape, src.data_type());
+
+    int        in_pos     = 0;
+    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_in   = static_cast<int>(src.shape()[3]);
+    const int  r          = channel_in / (block_shape * block_shape);
+
+    for(int b = 0; b < batch_in; ++b)
+    {
+        for(int z = 0; z < channel_in; ++z)
+        {
+            for(int y = 0; y < height_in; ++y)
+            {
+                for(int x = 0; x < width_in; ++x)
+                {
+                    const int out_x   = (block_shape * x + (z / r) % block_shape);
+                    const int out_y   = (block_shape * y + (z / r) / block_shape);
+                    const int out_pos = out_x + dst_shape[0] * out_y + (z % r) * dst_shape[0] * dst_shape[1] + b * dst_shape[0] * dst_shape[1] * dst_shape[2];
+                    result[out_pos]   = src[in_pos];
+                    ++in_pos;
+                }
+            }
+        }
+    }
+
+    return result;
+}
+template SimpleTensor<float> depth_to_space(const SimpleTensor<float> &src, const TensorShape &dst_shape, int32_t block_shape);
+template SimpleTensor<half> depth_to_space(const SimpleTensor<half> &src, const TensorShape &dst_shape, int32_t block_shape);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/reference/DepthToSpaceLayer.h b/tests/validation/reference/DepthToSpaceLayer.h
new file mode 100644
index 0000000000..3989401c08
--- /dev/null
+++ b/tests/validation/reference/DepthToSpaceLayer.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_DEPTH_TO_SPACE_LAYER_H__
+#define __ARM_COMPUTE_TEST_DEPTH_TO_SPACE_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> depth_to_space(const SimpleTensor<T> &src, const TensorShape &dst_shape, int32_t block_shape);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_DEPTH_TO_SPACE_LAYER_H__ */