COMPMID-1227 Implementing Space to Batch on OpenCL

Change-Id: I6fd83d6584c56a4fd2470948f1987e23237c16d3
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/145577
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: bsgcomp <bsgcomp@arm.com>

14 files changed, 1215 insertions, 0 deletions

diff --git a/arm_compute/core/CL/CLKernels.h b/arm_compute/core/CL/CLKernels.h
index 9586d8cb34..2fd6ab2e1c 100644
--- a/arm_compute/core/CL/CLKernels.h
+++ b/arm_compute/core/CL/CLKernels.h
@@ -112,6 +112,7 @@
 #include "arm_compute/core/CL/kernels/CLSobel5x5Kernel.h"
 #include "arm_compute/core/CL/kernels/CLSobel7x7Kernel.h"
 #include "arm_compute/core/CL/kernels/CLSoftmaxLayerKernel.h"
+#include "arm_compute/core/CL/kernels/CLSpaceToBatchLayerKernel.h"
 #include "arm_compute/core/CL/kernels/CLStridedSliceKernel.h"
 #include "arm_compute/core/CL/kernels/CLTableLookupKernel.h"
 #include "arm_compute/core/CL/kernels/CLThresholdKernel.h"
diff --git a/arm_compute/core/CL/kernels/CLSpaceToBatchLayerKernel.h b/arm_compute/core/CL/kernels/CLSpaceToBatchLayerKernel.h
new file mode 100644
index 0000000000..81095f9aef
--- /dev/null
+++ b/arm_compute/core/CL/kernels/CLSpaceToBatchLayerKernel.h
@@ -0,0 +1,101 @@
+/*
+ * 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.
+ */
+#ifndef __ARM_COMPUTE_CLSPACETOBATCHLAYERKERNEL_H__
+#define __ARM_COMPUTE_CLSPACETOBATCHLAYERKERNEL_H__
+
+#include "arm_compute/core/CL/ICLKernel.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ICLTensor;
+
+/** Interface for the space to batch kernel */
+class CLSpaceToBatchLayerKernel : public ICLKernel
+{
+public:
+    /** Default constructor */
+    CLSpaceToBatchLayerKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLSpaceToBatchLayerKernel(const CLSpaceToBatchLayerKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLSpaceToBatchLayerKernel &operator=(const CLSpaceToBatchLayerKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    CLSpaceToBatchLayerKernel(CLSpaceToBatchLayerKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    CLSpaceToBatchLayerKernel &operator=(CLSpaceToBatchLayerKernel &&) = default;
+    /** Default destructor */
+    ~CLSpaceToBatchLayerKernel() = 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[in]  block_shape 1-D tensor with shape [M]. Data types supported: S32
+     * @param[in]  paddings    2-D tensor with shape [2, M]. Data types supported: S32
+     * @param[out] output      Tensor output. Data types supported: same as @p input
+     */
+    void configure(const ICLTensor *input, const ICLTensor *block_shape, const ICLTensor *paddings, ICLTensor *output);
+    /** Initialise the kernel's input and output. (Static block shape and paddings)
+     *
+     * @param[in]  input         Tensor input. Supported tensor rank: 4. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[in]  block_shape_x Block shape x value.
+     * @param[in]  block_shape_y Block shape y value.
+     * @param[in]  padding_left  The left padding of the output tensor.
+     * @param[in]  padding_right The right padding of the output tensor.
+     * @param[out] output        Tensor output. Data types supported: same as @p input
+     */
+    void configure(const ICLTensor *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right, ICLTensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLSpaceToBatchLayerKernel
+     *
+     * @param[in] input       Tensor input. Supported tensor rank: 4. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[in] block_shape 1-D tensor with shape [M]. Data types supported: S32
+     * @param[in] paddings    2-D tensor with shape [2, M]. Data types supported: S32
+     * @param[in] output      Tensor output. Data types supported: same as @p input
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *block_shape, const ITensorInfo *paddings, const ITensorInfo *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLSpaceToBatchLayerKernel (Static block shape and paddings)
+     *
+     * @param[in] input         Tensor input. Supported tensor rank: 4. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[in] block_shape_x Block shape x value.
+     * @param[in] block_shape_y Block shape y value.
+     * @param[in] padding_left  The left padding of the output tensor.
+     * @param[in] padding_right The right padding of the output tensor.
+     * @param[in] output        Tensor output. Data types supported: same as @p input
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right, const ITensorInfo *output);
+
+    // Inherited methods overridden:
+    void run(const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    const ICLTensor *_input;       /**< Source tensor */
+    const ICLTensor *_block_shape; /**< Block shape tensor */
+    const ICLTensor *_paddings;    /**< Paddings tensor */
+    ICLTensor       *_output;      /**< Destination tensor */
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_CLSPACETOBATCHLAYERKERNEL_H__ */
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index cb04182c21..806149f83f 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -557,6 +557,16 @@ inline TensorShape compute_split_shape(const ITensorInfo *input, unsigned int ax
     return out_shape;
 }
 
+inline TensorShape compute_space_to_batch_shape(const ITensorInfo *input, const int block_x, const int block_y, const Size2D &padding_left, const Size2D &padding_right)
+{
+    TensorShape output_shape{ input->tensor_shape() };
+    output_shape.set(0, input->tensor_shape()[0] * block_x + padding_left.x() + padding_right.x());
+    output_shape.set(1, input->tensor_shape()[1] * block_y + padding_left.y() + padding_right.y());
+    output_shape.set(3, input->tensor_shape()[3] / (block_x * block_y));
+
+    return output_shape;
+}
+
 template <typename T>
 inline TensorShape extract_shape(T *data)
 {
diff --git a/arm_compute/runtime/CL/CLFunctions.h b/arm_compute/runtime/CL/CLFunctions.h
index 935a1ae5d0..cf8911b163 100644
--- a/arm_compute/runtime/CL/CLFunctions.h
+++ b/arm_compute/runtime/CL/CLFunctions.h
@@ -113,6 +113,7 @@
 #include "arm_compute/runtime/CL/functions/CLSobel5x5.h"
 #include "arm_compute/runtime/CL/functions/CLSobel7x7.h"
 #include "arm_compute/runtime/CL/functions/CLSoftmaxLayer.h"
+#include "arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h"
 #include "arm_compute/runtime/CL/functions/CLSplit.h"
 #include "arm_compute/runtime/CL/functions/CLStridedSlice.h"
 #include "arm_compute/runtime/CL/functions/CLTableLookup.h"
diff --git a/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h b/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h
new file mode 100644
index 0000000000..6478774701
--- /dev/null
+++ b/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h
@@ -0,0 +1,103 @@
+/*
+ * 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.
+ */
+#ifndef __ARM_COMPUTE_CLSPACETOBATCHLAYER_H__
+#define __ARM_COMPUTE_CLSPACETOBATCHLAYER_H__
+
+#include "arm_compute/runtime/IFunction.h"
+
+#include "arm_compute/core/CL/kernels/CLSpaceToBatchLayerKernel.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
+
+namespace arm_compute
+{
+class ICLTensor;
+
+/** Basic function to run @ref CLSpaceToBatchLayerKernel. */
+class CLSpaceToBatchLayer : public IFunction
+{
+public:
+    /** Default constructor */
+    CLSpaceToBatchLayer();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLSpaceToBatchLayer(const CLSpaceToBatchLayer &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLSpaceToBatchLayer &operator=(const CLSpaceToBatchLayer &) = delete;
+    /** Allow instances of this class to be moved */
+    CLSpaceToBatchLayer(CLSpaceToBatchLayer &&) = default;
+    /** Allow instances of this class to be moved */
+    CLSpaceToBatchLayer &operator=(CLSpaceToBatchLayer &&) = default;
+    /** Default destructor */
+    virtual ~CLSpaceToBatchLayer() = 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[in]  block_shape 1-D tensor with shape [M]. Data types supported: S32
+     * @param[in]  paddings    2-D tensor with shape [2, M]. Data types supported: S32
+     * @param[out] output      Tensor output. Data types supported: same as @p input
+     */
+    void configure(const ICLTensor *input, const ICLTensor *block_shape, const ICLTensor *paddings, ICLTensor *output);
+    /** Set the input and output tensors. (Static block shape and paddings)
+     *
+     * @param[in]  input         Tensor input. Supported tensor rank: 4. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[in]  block_shape_x Block shape x value.
+     * @param[in]  block_shape_y Block shape y value.
+     * @param[in]  padding_left  The left padding of the output tensor.
+     * @param[in]  padding_right The right padding of the output tensor.
+     * @param[out] output        Tensor output. Data types supported: same as @p input
+     */
+    void configure(const ICLTensor *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right, ICLTensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLSpaceToBatchLayer
+     *
+     * @param[in]  input       Tensor input info. Supported tensor rank: 4. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[in]  block_shape block shape tensor info with shape [M]. Data types supported: S32
+     * @param[in]  paddings    paddings tensor info with shape [2, M]. Data types supported: S32
+     * @param[out] output      Tensor output info. Data types supported: same as @p input
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *block_shape, const ITensorInfo *paddings, const ITensorInfo *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLSpaceToBatchLayer (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]  block_shape_x Block shape x value.
+     * @param[in]  block_shape_y Block shape y value.
+     * @param[in]  padding_left  The left padding of the output tensor.
+     * @param[in]  padding_right The right padding of the output tensor.
+     * @param[out] output        Tensor output info. Data types supported: same as @p input
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right, const ITensorInfo *output);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    CLSpaceToBatchLayerKernel _space_to_batch_kernel; /**< SpaceToBatch kernel to run */
+    ICLTensor                *_output;                /**< Output tensor */
+    bool                      _has_padding;           /**< Flag to check if the output has padding */
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_CLSPACETOBATCHLAYER_H__ */
diff --git a/src/core/CL/CLKernelLibrary.cpp b/src/core/CL/CLKernelLibrary.cpp
index 75ff2482c8..ef3a431f1a 100644
--- a/src/core/CL/CLKernelLibrary.cpp
+++ b/src/core/CL/CLKernelLibrary.cpp
@@ -364,6 +364,8 @@ const std::map<std::string, std::string> CLKernelLibrary::_kernel_program_map =
     { "softmax_layer_max_shift_exp_sum_quantized_serial", "softmax_layer_quantized.cl" },
     { "softmax_layer_max_shift_exp_sum_quantized_parallel", "softmax_layer_quantized.cl" },
     { "softmax_layer_max_shift_exp_sum_serial", "softmax_layer.cl" },
+    { "space_to_batch", "space_to_batch.cl" },
+    { "space_to_batch_static", "space_to_batch.cl" },
     { "softmax_layer_max_shift_exp_sum_parallel", "softmax_layer.cl" },
     { "strided_slice", "slice_ops.cl" },
     { "suppress_non_maximum", "canny.cl" },
@@ -760,6 +762,10 @@ const std::map<std::string, std::string> CLKernelLibrary::_program_source_map =
 #include "./cl_kernels/slice_ops.clembed"
     },
     {
+        "space_to_batch.cl",
+#include "./cl_kernels/space_to_batch.clembed"
+    },
+    {
         "tablelookup.cl",
 #include "./cl_kernels/tablelookup.clembed"
     },
diff --git a/src/core/CL/cl_kernels/space_to_batch.cl b/src/core/CL/cl_kernels/space_to_batch.cl
new file mode 100644
index 0000000000..1343695ed1
--- /dev/null
+++ b/src/core/CL/cl_kernels/space_to_batch.cl
@@ -0,0 +1,151 @@
+/*
+ * 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 withoutput restriction, including withoutput 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 KOUTD, EXPRESS OR
+ * IMPLIED, OUTCLUDOUTG BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONOUTFROUTGEMENT. OUT NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER OUT AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISOUTG FROM,
+ * OUT OF OR OUT CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALOUTGS OUT THE
+ * SOFTWARE.
+ */
+#include "helpers.h"
+
+#if defined(BATCH_SIZE) && defined(DATA_TYPE)
+/** Calculate the space to batch conversion.
+ *
+ * @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=float
+ * @note The block shape tensor rank must be passed at compile time using -DBLOCK_SHAPE_DIM. e.g. -DBLOCK_SHAPE_DIM=2
+ *
+ * @param[in]  input_ptr                                 Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32
+ * @param[in]  input_stride_x                            Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  input_step_x                              input_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  input_stride_y                            Stride of the source image in Y dimension (in bytes)
+ * @param[in]  input_step_y                              input_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  input_stride_z                            Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  input_step_z                              input_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  input_offset_first_element_in_bytes       The offset of the first element in the first source image
+ * @param[in]  paddings_ptr                              Pointer to the second source image. Supported data types: S32
+ * @param[in]  paddings_stride_x                         Stride of the paddinds tensor in X dimension (in bytes)
+ * @param[in]  paddings_step_x                           paddings_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  paddings_stride_y                         Stride of the paddinds tensor in Y dimension (in bytes)
+ * @param[in]  paddings_step_y                           paddings_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  paddingse_offset_first_element_in_bytes   The offset of the first element in the second source image
+ * @param[in]  block_shape_ptr                           Pointer to the block shape tensor. Supported data types: S32
+ * @param[in]  block_shape_stride_x                      Stride of the block shape tensor in X dimension (in bytes)
+ * @param[in]  block_shape_step_x                        block_shape_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  block_shape_stride_y                      Stride of the block shape tensor in Y dimension (in bytes)
+ * @param[in]  block_shape_step_y                        block_shape_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  block_shape_offset_first_element_in_bytes The offset of the first element in the block shapetensor
+ * @param[in]  batch_id                                  The output tensor batch id
+ * @param[out] output_ptr                                Pointer to the destination tensor. Supported data types: same as @p input_ptr
+ * @param[in]  output_stride_x                           Stride of the destination tensor in X dimension (in bytes)
+ * @param[in]  output_step_x                             output_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  output_stride_y                           Stride of the destination tensor in Y dimension (in bytes)
+ * @param[in]  output_step_y                             output_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  output_stride_z                           Stride of the destination tensor in Z dimension (in bytes)
+ * @param[in]  output_step_z                             output_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  output_offset_first_element_in_bytes      The offset of the first element in the destination image
+ */
+__kernel void space_to_batch(
+    TENSOR4D_DECLARATION(input),
+    IMAGE_DECLARATION(paddings),
+    VECTOR_DECLARATION(block_shape),
+    const int batch_id,
+    TENSOR3D_DECLARATION(output))
+{
+    Tensor4D in    = CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(input, 0);
+    Image    pad   = CONVERT_TO_IMAGE_STRUCT_NO_STEP(paddings);
+    Vector   block = CONVERT_TO_VECTOR_STRUCT_NO_STEP(block_shape);
+    Tensor3D out   = CONVERT_TO_TENSOR3D_STRUCT(output);
+
+    const int PAD_LEFT_X  = *((__global int *)offset(&pad, 0, 0));
+    const int PAD_RIGHT_X = *((__global int *)offset(&pad, 1, 0));
+    const int PAD_LEFT_Y  = *((__global int *)offset(&pad, 0, 1));
+    const int PAD_RIGHT_Y = *((__global int *)offset(&pad, 1, 1));
+
+    int block_x = *((__global int *)vector_offset(&block, 0));
+    int block_y = *((__global int *)vector_offset(&block, 1));
+
+    const int out_x = get_global_id(0);
+    const int out_y = get_global_id(1);
+    const int z     = get_global_id(2);
+
+    if((out_x >= PAD_LEFT_X && out_x < WIDTH_OUT - PAD_RIGHT_X) && (out_y >= PAD_LEFT_Y && out_y < HEIGHT_OUT - PAD_RIGHT_Y))
+    {
+        const int r                      = (BATCH_SIZE / (block_x * block_y));
+        const int w                      = batch_id % r;
+        const int in_x                   = (out_x - PAD_LEFT_X) * block_x + (batch_id / r) % block_x;
+        const int in_y                   = (out_y - PAD_LEFT_Y) * block_y + (batch_id / r) / block_x;
+        *((__global DATA_TYPE *)out.ptr) = *((__global DATA_TYPE *)tensor4D_offset(&in, in_x, in_y, z, w));
+    }
+}
+#endif // defined(BATCH_SIZE) && defined(DATA_TYPE)
+
+#if defined(BATCH_SIZE) && defined(DATA_TYPE) && defined(BLOCK_SHAPE_X) && defined(BLOCK_SHAPE_Y) && defined(PAD_LEFT_X) && defined(PAD_RIGHT_X) && defined(PAD_LEFT_Y) && defined(PAD_RIGHT_Y)
+/** Calculate the space to batch conversion.
+ *
+ * @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=float
+ * @note The input tensor batch size must be passed at compile time using -DBATCH_SIZE. e.g. -DBATCH_SIZE=2
+ * @note The block shape x must be passed at compile time using -DBLOCK_SHAPE_X. e.g. -DBLOCK_SHAPE_X=2
+ * @note The block shape y must be passed at compile time using -DBLOCK_SHAPE_Y. e.g. -DBLOCK_SHAPE_Y=2
+ * @note The starting pad value of x must be passed at compile time using -DPAD_LEFT_X. e.g. -DPAD_LEFT_X=2
+ * @note The ending pad value of x must be passed at compile time using -DPAD_RIGHT_X. e.g. -DPAD_RIGHT_X=2
+ * @note The starting pad value of y must be passed at compile time using -DPAD_LEFT_Y. e.g. -DPAD_LEFT_Y=2
+ * @note The ending pad value of  y must be passed at compile time using -DPAD_RIGHT_Y. e.g. -DPAD_RIGHT_X=2
+ *
+ * @param[in]  input_ptr                            Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32
+ * @param[in]  input_stride_x                       Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  input_step_x                         input_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  input_stride_y                       Stride of the source image in Y dimension (in bytes)
+ * @param[in]  input_step_y                         input_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  input_stride_z                       Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  input_step_z                         input_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  input_offset_first_element_in_bytes  The offset of the first element in the first source image
+ * @param[in]  batch_id                             The output tensor batch id
+ * @param[out] output_ptr                           Pointer to the destination tensor. Supported data types: same as @p input_ptr
+ * @param[in]  output_stride_x                      Stride of the destination tensor in X dimension (in bytes)
+ * @param[in]  output_step_x                        output_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  output_stride_y                      Stride of the destination tensor in Y dimension (in bytes)
+ * @param[in]  output_step_y                        output_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  output_stride_z                      Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  output_step_z                        output_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  output_offset_first_element_in_bytes The offset of the first element in the destination image
+ */
+__kernel void space_to_batch_static(
+    TENSOR4D_DECLARATION(input),
+    const int batch_id,
+    TENSOR3D_DECLARATION(output))
+{
+    Tensor4D in  = CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(input, 0);
+    Tensor3D out = CONVERT_TO_TENSOR3D_STRUCT(output);
+
+    int block_x = BLOCK_SHAPE_X;
+    int block_y = *((__global int *)vector_offset(&block, 1));
+
+    const int out_x = get_global_id(0);
+    const int out_y = get_global_id(1);
+    const int z     = get_global_id(2);
+
+    if((out_x >= PAD_LEFT_X && out_x < WIDTH_OUT - PAD_RIGHT_X) && (out_y >= PAD_LEFT_Y && out_y < HEIGHT_OUT - PAD_RIGHT_Y))
+    {
+        const int r                      = (BATCH_SIZE / (block_x * block_y));
+        const int w                      = batch_id % r;
+        const int in_x                   = (out_x - PAD_LEFT_X) * block_x + (batch_id / r) % block_x;
+        const int in_y                   = (out_y - PAD_LEFT_Y) * block_y + (batch_id / r) / block_x;
+        *((__global DATA_TYPE *)out.ptr) = *((__global DATA_TYPE *)tensor4D_offset(&in, in_x, in_y, z, w));
+    }
+}
+#endif // defined(BATCH_SIZE) && defined(DATA_TYPE) && defined(BLOCK_SHAPE_X) && defined(BLOCK_SHAPE_Y) && defined(PAD_LEFT_X) && defined(PAD_RIGHT_X) && defined(PAD_LEFT_Y) && defined(PAD_RIGHT_Y)
diff --git a/src/core/CL/kernels/CLSpaceToBatchLayerKernel.cpp b/src/core/CL/kernels/CLSpaceToBatchLayerKernel.cpp
new file mode 100644
index 0000000000..cda6e96806
--- /dev/null
+++ b/src/core/CL/kernels/CLSpaceToBatchLayerKernel.cpp
@@ -0,0 +1,182 @@
+/*
+ * 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/CL/kernels/CLSpaceToBatchLayerKernel.h"
+
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/CLValidate.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
+using namespace arm_compute::misc::shape_calculator;
+
+namespace arm_compute
+{
+namespace
+{
+Status validate_arguments(const ITensorInfo *input, const ITensorInfo *block_info, const ITensorInfo *padddings, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, block_info, padddings, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(block_info, 1, DataType::S32);
+    ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
+
+    // Validate output if initialized
+    if(output->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    }
+
+    return Status{};
+}
+Status validate_arguments_static(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
+                                 const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_RETURN_ERROR_ON(block_shape_x < 1 || block_shape_y < 1);
+    ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
+
+    // Validate output if initialized
+    if(output->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape()[0] < padding_left.x() + padding_right.y());
+        ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[0] / block_shape_x != (output->tensor_shape()[0] - padding_left.x() - padding_right.y()));
+        ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[1] / block_shape_y != (output->tensor_shape()[1] - padding_left.x() - padding_right.y()));
+        ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[2] != output->tensor_shape()[2]);
+        ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape()[3] % (block_shape_x * block_shape_y) != 0);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    }
+
+    return Status{};
+}
+} // namespace
+
+CLSpaceToBatchLayerKernel::CLSpaceToBatchLayerKernel()
+    : _input(nullptr), _block_shape(nullptr), _paddings(nullptr), _output(nullptr)
+{
+}
+
+void CLSpaceToBatchLayerKernel::configure(const ICLTensor *input, const ICLTensor *block_shape, const ICLTensor *paddings, ICLTensor *output)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), block_shape->info(), paddings->info(), output->info()));
+
+    _input       = input;
+    _block_shape = block_shape;
+    _paddings    = paddings;
+    _output      = output;
+
+    // Create kernel
+    CLBuildOptions build_opts;
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
+    build_opts.add_option("-DWIDTH_OUT=" + support::cpp11::to_string(output->info()->dimension(0)));
+    build_opts.add_option("-DHEIGHT_OUT=" + support::cpp11::to_string(output->info()->dimension(1)));
+    build_opts.add_option("-DBATCH_SIZE=" + support::cpp11::to_string(output->info()->dimension(3)));
+    _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("space_to_batch", build_opts.options()));
+
+    // Configure kernel window
+    Window win = calculate_max_window(*output->info(), Steps());
+    ICLKernel::configure_internal(win);
+}
+
+void CLSpaceToBatchLayerKernel::configure(const ICLTensor *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
+                                          ICLTensor *output)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+    TensorShape output_shape = misc::shape_calculator::compute_space_to_batch_shape(input->info(), block_shape_x, block_shape_y, padding_left, padding_right);
+    auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type());
+
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_static(input->info(), block_shape_x, block_shape_y, padding_left, padding_right, output->info()));
+
+    _input  = input;
+    _output = output;
+
+    // Create kernel
+    CLBuildOptions build_opts;
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
+    build_opts.add_option("-DWIDTH_OUT=" + support::cpp11::to_string(output->info()->dimension(0)));
+    build_opts.add_option("-DHEIGHT_OUT=" + support::cpp11::to_string(output->info()->dimension(1)));
+    build_opts.add_option("-DBATCH_SIZE=" + support::cpp11::to_string(output->info()->dimension(3)));
+    build_opts.add_option("-DBLOCK_SHAPE_X=" + support::cpp11::to_string(block_shape_x));
+    build_opts.add_option("-DBLOCK_SHAPE_Y=" + support::cpp11::to_string(block_shape_y));
+    build_opts.add_option("-DPAD_START_X=" + support::cpp11::to_string(padding_left.x()));
+    build_opts.add_option("-DPAD_END_X=" + support::cpp11::to_string(padding_right.x()));
+    build_opts.add_option("-DPAD_START_Y=" + support::cpp11::to_string(padding_left.y()));
+    build_opts.add_option("-DPAD_END_Y=" + support::cpp11::to_string(padding_right.y()));
+    _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("space_to_batch_static", build_opts.options()));
+
+    // Configure kernel window
+    Window win = calculate_max_window(*output->info(), Steps());
+    ICLKernel::configure_internal(win);
+}
+
+Status CLSpaceToBatchLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *block_shape, const ITensorInfo *paddings, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, block_shape, paddings, output));
+    return Status{};
+}
+Status CLSpaceToBatchLayerKernel::validate(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
+                                           const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_static(input, block_shape_x, block_shape_y, padding_left, padding_right, output));
+    return Status{};
+}
+
+void CLSpaceToBatchLayerKernel::run(const Window &window, cl::CommandQueue &queue)
+{
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
+
+    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));
+
+    Window vector_slice = window.first_slice_window_1D();
+    vector_slice.set(Window::DimX, Window::Dimension(0, 0, 0));
+
+    Window padding_slice = window.first_slice_window_2D();
+    padding_slice.set(Window::DimX, Window::Dimension(0, 0, 0));
+    padding_slice.set(Window::DimY, Window::Dimension(0, 0, 0));
+
+    int batch_id = 0;
+    do
+    {
+        unsigned int idx = 0;
+        add_4D_tensor_argument(idx, _input, slice_in);
+        if(_paddings != nullptr && _block_shape != nullptr)
+        {
+            add_2D_tensor_argument(idx, _paddings, padding_slice);
+            add_1D_tensor_argument(idx, _block_shape, vector_slice);
+        }
+        add_argument(idx, batch_id);
+        add_3D_tensor_argument(idx, _output, slice_out);
+        enqueue(queue, *this, slice_out);
+        ++batch_id;
+    }
+    while(window.slide_window_slice_3D(slice_out));
+}
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/runtime/CL/functions/CLSpaceToBatchLayer.cpp b/src/runtime/CL/functions/CLSpaceToBatchLayer.cpp
new file mode 100644
index 0000000000..76c1e188e6
--- /dev/null
+++ b/src/runtime/CL/functions/CLSpaceToBatchLayer.cpp
@@ -0,0 +1,98 @@
+/*
+ * 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/runtime/CL/functions/CLSpaceToBatchLayer.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/CL/CLScheduler.h"
+
+namespace arm_compute
+{
+CLSpaceToBatchLayer::CLSpaceToBatchLayer()
+    : _space_to_batch_kernel(), _output(nullptr), _has_padding(false)
+{
+}
+
+void CLSpaceToBatchLayer::configure(const ICLTensor *input, const ICLTensor *block_shape, const ICLTensor *paddings, ICLTensor *output)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+    if(input->info()->tensor_shape().total_size() != output->info()->tensor_shape().total_size())
+    {
+        _has_padding = true;
+    }
+
+    _output = output;
+    _space_to_batch_kernel.configure(input, block_shape, paddings, output);
+}
+
+void CLSpaceToBatchLayer::configure(const ICLTensor *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right, ICLTensor *output)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+    if(input->info()->tensor_shape().total_size() != output->info()->tensor_shape().total_size())
+    {
+        _has_padding = true;
+    }
+
+    _output = output;
+    _space_to_batch_kernel.configure(input, block_shape_x, block_shape_y, padding_left, padding_right, output);
+}
+
+Status CLSpaceToBatchLayer::validate(const ITensorInfo *input, const ITensorInfo *block_shape, const ITensorInfo *paddings, const ITensorInfo *output)
+{
+    return CLSpaceToBatchLayerKernel::validate(input, block_shape, paddings, output);
+}
+
+Status CLSpaceToBatchLayer::validate(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
+                                     const ITensorInfo *output)
+{
+    return CLSpaceToBatchLayerKernel::validate(input, block_shape_x, block_shape_y, padding_left, padding_right, output);
+}
+
+void CLSpaceToBatchLayer::run()
+{
+    // Zero out output only if we have paddings
+    // TODO(micspy01): replace with memset once ready
+    if(_has_padding)
+    {
+        _output->map(CLScheduler::get().queue(), true);
+        if(is_data_type_quantized_asymmetric(_output->info()->data_type()))
+        {
+            const uint8_t quantized_zero = _output->info()->quantization_info().offset;
+            std::fill_n(_output->buffer(), _output->info()->total_size(), quantized_zero);
+        }
+        else
+        {
+            memset(_output->buffer(), 0, _output->info()->total_size());
+        }
+        _output->unmap(CLScheduler::get().queue());
+    }
+
+    CLScheduler::get().enqueue(_space_to_batch_kernel, true);
+}
+} // namespace arm_compute
diff --git a/tests/datasets/SpaceToBatchDataset.h b/tests/datasets/SpaceToBatchDataset.h
new file mode 100644
index 0000000000..0ab06f9327
--- /dev/null
+++ b/tests/datasets/SpaceToBatchDataset.h
@@ -0,0 +1,142 @@
+/*
+ * 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.
+ */
+#ifndef ARM_COMPUTE_TEST_SPACE_TO_BATCH_LAYER_DATASET
+#define ARM_COMPUTE_TEST_SPACE_TO_BATCH_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 SpaceToBatchLayerDataset
+{
+public:
+    using type = std::tuple<TensorShape, TensorShape, TensorShape, TensorShape>;
+
+    struct iterator
+    {
+        iterator(std::vector<TensorShape>::const_iterator src_it,
+                 std::vector<TensorShape>::const_iterator block_shape_it,
+                 std::vector<TensorShape>::const_iterator paddings_shape_it,
+                 std::vector<TensorShape>::const_iterator dst_it)
+            : _src_it{ std::move(src_it) },
+              _block_shape_it{ std::move(block_shape_it) },
+              _paddings_shape_it{ std::move(paddings_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 << "PaddingsShape=" << *_paddings_shape_it << ":";
+            description << "Out=" << *_dst_it;
+            return description.str();
+        }
+
+        SpaceToBatchLayerDataset::type operator*() const
+        {
+            return std::make_tuple(*_src_it, *_block_shape_it, *_paddings_shape_it, *_dst_it);
+        }
+
+        iterator &operator++()
+        {
+            ++_src_it;
+            ++_block_shape_it;
+            ++_paddings_shape_it;
+            ++_dst_it;
+
+            return *this;
+        }
+
+    private:
+        std::vector<TensorShape>::const_iterator _src_it;
+        std::vector<TensorShape>::const_iterator _block_shape_it;
+        std::vector<TensorShape>::const_iterator _paddings_shape_it;
+        std::vector<TensorShape>::const_iterator _dst_it;
+    };
+
+    iterator begin() const
+    {
+        return iterator(_src_shapes.begin(), _block_shape_shapes.begin(), _padding_shapes.begin(), _dst_shapes.begin());
+    }
+
+    int size() const
+    {
+        return std::min(_src_shapes.size(), std::min(_block_shape_shapes.size(), std::min(_padding_shapes.size(), _dst_shapes.size())));
+    }
+
+    void add_config(TensorShape src, TensorShape block_shape, TensorShape padding_shapes, TensorShape dst)
+    {
+        _src_shapes.emplace_back(std::move(src));
+        _block_shape_shapes.emplace_back(std::move(block_shape));
+        _padding_shapes.emplace_back(std::move(padding_shapes));
+        _dst_shapes.emplace_back(std::move(dst));
+    }
+
+protected:
+    SpaceToBatchLayerDataset()                            = default;
+    SpaceToBatchLayerDataset(SpaceToBatchLayerDataset &&) = default;
+
+private:
+    std::vector<TensorShape> _src_shapes{};
+    std::vector<TensorShape> _block_shape_shapes{};
+    std::vector<TensorShape> _padding_shapes{};
+    std::vector<TensorShape> _dst_shapes{};
+};
+
+class SmallSpaceToBatchLayerDataset final : public SpaceToBatchLayerDataset
+{
+public:
+    SmallSpaceToBatchLayerDataset()
+    {
+        add_config(TensorShape(2U, 2U, 1U, 1U), TensorShape(2U), TensorShape(2U, 4U), TensorShape(1U, 1U, 1U, 4U));
+        add_config(TensorShape(6U, 2U, 1U, 1U), TensorShape(2U), TensorShape(2U, 4U), TensorShape(3U, 1U, 1U, 4U));
+        add_config(TensorShape(2U, 4U, 2U, 1U), TensorShape(2U), TensorShape(2U, 4U), TensorShape(1U, 2U, 2U, 4U));
+        add_config(TensorShape(2U, 6U, 1U, 2U), TensorShape(2U), TensorShape(2U, 4U), TensorShape(1U, 3U, 1U, 8U));
+        add_config(TensorShape(6U, 8U, 1U, 1U), TensorShape(2U), TensorShape(2U, 4U), TensorShape(3U, 4U, 1U, 4U));
+        add_config(TensorShape(6U, 8U, 15U, 5U), TensorShape(2U), TensorShape(2U, 4U), TensorShape(3U, 4U, 15U, 20U));
+    }
+};
+class LargeSpaceToBatchLayerDataset final : public SpaceToBatchLayerDataset
+{
+public:
+    LargeSpaceToBatchLayerDataset()
+    {
+        add_config(TensorShape(128U, 64U, 2U, 1U), TensorShape(2U), TensorShape(2U, 4U), TensorShape(64U, 32U, 2U, 4U));
+        add_config(TensorShape(512U, 64U, 2U, 1U), TensorShape(2U), TensorShape(2U, 4U), TensorShape(128U, 16U, 2U, 16U));
+    }
+};
+} // namespace datasets
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_SPACE_TO_BATCH_LAYER_DATASET */
diff --git a/tests/validation/CL/SpaceToBatchLayer.cpp b/tests/validation/CL/SpaceToBatchLayer.cpp
new file mode 100644
index 0000000000..63d4022d08
--- /dev/null
+++ b/tests/validation/CL/SpaceToBatchLayer.cpp
@@ -0,0 +1,161 @@
+/*
+ * 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/Types.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/CLTensorAllocator.h"
+#include "arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h"
+#include "tests/CL/CLAccessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/datasets/SpaceToBatchDataset.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Helpers.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/SpaceToBatchFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+TEST_SUITE(CL)
+TEST_SUITE(SpaceToBatchLayer)
+
+template <typename T>
+using CLSpaceToBatchLayerFixture = SpaceToBatchLayerValidationFixture<CLTensor, CLAccessor, CLSpaceToBatchLayer, T>;
+
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(datasets::SmallSpaceToBatchLayerDataset(),
+                                                                   framework::dataset::make("DataType", { DataType::F16, DataType::F32 })),
+               input_shape, block_shape_shape, paddings_shape, output_shape, dt)
+{
+    // Create tensors
+    CLTensor src         = create_tensor<CLTensor>(input_shape, dt, 1);
+    CLTensor dst         = create_tensor<CLTensor>(output_shape, dt, 1);
+    CLTensor block_shape = create_tensor<CLTensor>(block_shape_shape, DataType::S32, 1);
+    CLTensor paddings    = create_tensor<CLTensor>(paddings_shape, DataType::S32, 1);
+
+    // Create and Configure function
+    CLSpaceToBatchLayer batch_to_space;
+    batch_to_space.configure(&src, &block_shape, &paddings, &dst);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(output_shape);
+    validate(dst.info()->valid_region(), valid_region);
+}
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(
+               framework::dataset::make("InputInfo", { TensorInfo(TensorShape(32U, 13U, 2U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U, 2U), 1, DataType::F32),    // Mismatching data types
+                                                       TensorInfo(TensorShape(32U, 13U, 2U, 2U), 1, DataType::F32),    // Wrong data type block shape
+                                                       TensorInfo(TensorShape(32U, 13U, 2U, 2U, 4U), 1, DataType::F32),    // Wrong tensor shape
+                                                     }),
+               framework::dataset::make("BlockShapeInfo",{ TensorInfo(TensorShape(2U, 2U), 1, DataType::S32),
+                                                       TensorInfo(TensorShape(2U, 2U), 1, DataType::S32),
+                                                       TensorInfo(TensorShape(2U, 2U), 1, DataType::F16),
+                                                       TensorInfo(TensorShape(2U, 2U), 1, DataType::S32),
+                                                     })),
+               framework::dataset::make("PaddingsShapeInfo",{ TensorInfo(TensorShape(2U, 2U), 1, DataType::S32),
+                                                       TensorInfo(TensorShape(2U, 2U), 1, DataType::S32),
+                                                       TensorInfo(TensorShape(2U, 2U), 1, DataType::F16),
+                                                       TensorInfo(TensorShape(2U, 2U), 1, DataType::S32),
+                                                     })),
+               framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(32U, 13U, 2U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U, 2U), 1, DataType::F16),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U, 2U), 1, DataType::F32),
+                                                     })),
+               framework::dataset::make("Expected", { true, false, false, false})),
+               input_info, block_shape_info, paddings_info, output_info, expected)
+{
+    bool has_error = bool(CLSpaceToBatchLayer::validate(&input_info.clone()->set_is_resizable(false), &block_shape_info.clone()->set_is_resizable(false), &paddings_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false)));
+    ARM_COMPUTE_EXPECT(has_error == expected, framework::LogLevel::ERRORS);
+}
+DATA_TEST_CASE(ValidateStatic, framework::DatasetMode::ALL, zip(zip(zip(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
+                                                       TensorInfo(TensorShape(32U, 16U, 2U, 1U, 4U), 1, DataType::F32), // Wrong paddings
+                                                     }),
+               framework::dataset::make("BlockShapeX", { 2, 2, 2, 2, 2 })),
+               framework::dataset::make("BlockShapeY", { 2, 2, -2, 2, 2 })),
+               framework::dataset::make("PadLeft", { Size2D(0, 0), Size2D(0, 0), Size2D(0, 0), Size2D(0, 0), Size2D(3, 11) })),
+               framework::dataset::make("PadRight", { Size2D(0, 0), Size2D(0, 0), Size2D(0, 0), Size2D(0, 0), Size2D(3, 11) })),
+               framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(16U, 8U, 2U, 4U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 8U, 2U, 4U), 1, DataType::F16),
+                                                       TensorInfo(TensorShape(32U, 8U, 2U, 4U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 8U, 2U, 4U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 8U, 2U, 4U), 1, DataType::F32),
+                                                     })),
+               framework::dataset::make("Expected", { true, false, false, false, false})),
+               input_info, block_shape_x, block_shape_y, padding_left, padding_right, output_info, expected)
+{
+    bool has_error = bool(CLSpaceToBatchLayer::validate(&input_info.clone()->set_is_resizable(false), block_shape_x, block_shape_y, padding_left, padding_right, &output_info.clone()->set_is_resizable(false)));
+    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, CLSpaceToBatchLayerFixture<float>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallSpaceToBatchLayerDataset(), framework::dataset::make("DataType",
+                                                                                                            DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+FIXTURE_DATA_TEST_CASE(Large, CLSpaceToBatchLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeSpaceToBatchLayerDataset(), framework::dataset::make("DataType",
+                                                                                                          DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // FP32
+
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(Small, CLSpaceToBatchLayerFixture<half>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallSpaceToBatchLayerDataset(), framework::dataset::make("DataType",
+                                                                                                           DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+FIXTURE_DATA_TEST_CASE(Large, CLSpaceToBatchLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeSpaceToBatchLayerDataset(), framework::dataset::make("DataType",
+                                                                                                         DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // FP16
+TEST_SUITE_END() // Float
+
+TEST_SUITE_END() // SpaceToBatch
+TEST_SUITE_END() // CL
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/SpaceToBatchFixture.h b/tests/validation/fixtures/SpaceToBatchFixture.h
new file mode 100644
index 0000000000..26c24c1331
--- /dev/null
+++ b/tests/validation/fixtures/SpaceToBatchFixture.h
@@ -0,0 +1,133 @@
+/*
+ * 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.
+ */
+#ifndef ARM_COMPUTE_TEST_SPACE_TO_BATCH_LAYER_FIXTURE
+#define ARM_COMPUTE_TEST_SPACE_TO_BATCH_LAYER_FIXTURE
+
+#include "tests/Globals.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Fixture.h"
+#include "tests/validation/reference/SpaceToBatch.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class SpaceToBatchLayerValidationFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape input_shape, TensorShape block_shape_shape, TensorShape paddings_shape, TensorShape output_shape, DataType data_type)
+    {
+        _target    = compute_target(input_shape, block_shape_shape, paddings_shape, output_shape, data_type);
+        _reference = compute_reference(input_shape, block_shape_shape, paddings_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);
+    }
+    template <typename U>
+    void fill_pad(U &&tensor, int i)
+    {
+        std::uniform_int_distribution<> distribution(0, 0);
+        library->fill(tensor, distribution, i);
+    }
+    TensorType compute_target(const TensorShape &input_shape, const TensorShape &block_shape_shape, const TensorShape &paddings_shape, const TensorShape &output_shape,
+                              DataType data_type)
+    {
+        // Create tensors
+        TensorType input       = create_tensor<TensorType>(input_shape, data_type);
+        TensorType block_shape = create_tensor<TensorType>(block_shape_shape, DataType::S32);
+        TensorType paddings    = create_tensor<TensorType>(paddings_shape, DataType::S32);
+        TensorType output      = create_tensor<TensorType>(output_shape, data_type);
+
+        // Create and configure function
+        FunctionType space_to_batch;
+        space_to_batch.configure(&input, &block_shape, &paddings, &output);
+
+        ARM_COMPUTE_EXPECT(input.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(block_shape.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(paddings.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(output.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Allocate tensors
+        input.allocator()->allocate();
+        block_shape.allocator()->allocate();
+        paddings.allocator()->allocate();
+        output.allocator()->allocate();
+
+        ARM_COMPUTE_EXPECT(!input.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(!block_shape.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(!paddings.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(!output.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Fill tensors
+        fill(AccessorType(input), 0);
+        fill_pad(AccessorType(paddings), 0);
+        {
+            auto block_shape_data = AccessorType(block_shape);
+            for(unsigned int i = 0; i < block_shape_shape.x(); ++i)
+            {
+                static_cast<int32_t *>(block_shape_data.data())[i] = input_shape[i] / output_shape[i];
+            }
+        }
+        // Compute function
+        space_to_batch.run();
+
+        return output;
+    }
+
+    SimpleTensor<T> compute_reference(const TensorShape &input_shape, const TensorShape &block_shape_shape, const TensorShape &paddings_shape,
+                                      const TensorShape &output_shape, DataType data_type)
+    {
+        // Create reference
+        SimpleTensor<T>       input{ input_shape, data_type };
+        SimpleTensor<int32_t> block_shape{ block_shape_shape, DataType::S32 };
+        SimpleTensor<int32_t> paddings{ paddings_shape, DataType::S32 };
+
+        // Fill reference
+        fill(input, 0);
+        fill_pad(paddings, 0);
+        for(unsigned int i = 0; i < block_shape_shape.x(); ++i)
+        {
+            block_shape[i] = input_shape[i] / output_shape[i];
+        }
+
+        // Compute reference
+        return reference::space_to_batch(input, block_shape, paddings, output_shape);
+    }
+
+    TensorType      _target{};
+    SimpleTensor<T> _reference{};
+};
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_SPACE_TO_BATCH_LAYER_FIXTURE */
diff --git a/tests/validation/reference/SpaceToBatch.cpp b/tests/validation/reference/SpaceToBatch.cpp
new file mode 100644
index 0000000000..979ab94b33
--- /dev/null
+++ b/tests/validation/reference/SpaceToBatch.cpp
@@ -0,0 +1,82 @@
+/*
+ * 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 "SpaceToBatch.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_batch(const SimpleTensor<T> &src, const SimpleTensor<int32_t> &block_shape, const SimpleTensor<int32_t> &paddings, const TensorShape &dst_shape)
+{
+    SimpleTensor<T> result(dst_shape, src.data_type());
+
+    auto width_out  = static_cast<int>(dst_shape[0]);
+    auto height_out = static_cast<int>(dst_shape[1]);
+    auto z_out      = static_cast<int>(dst_shape[2]);
+
+    int out_pos = 0;
+    for(int batch = 0; batch < static_cast<int>(dst_shape[3]); ++batch)
+    {
+        for(int z = 0; z < z_out; ++z)
+        {
+            for(int y = 0; y < height_out; ++y)
+            {
+                for(int x = 0; x < width_out; ++x)
+                {
+                    if(x < paddings[0] || x > width_out - paddings[1] - 1
+                       || y < paddings[2] || y > height_out - paddings[3] - 1)
+                    {
+                        result[out_pos] = 0;
+                    }
+                    else
+                    {
+                        const int r      = dst_shape[3] / (block_shape[0] * block_shape[1]);
+                        const int in_x   = (block_shape[0] * (x - paddings[0]) + (batch / r) % block_shape[0]);
+                        const int in_y   = (block_shape[1] * (y - paddings[2]) + (batch / r) / block_shape[0]);
+                        int       in_pos = in_x + src.shape()[0] * in_y + z * src.shape()[0] * src.shape()[1] + (batch % r) * src.shape()[0] * src.shape()[1] * src.shape()[2];
+                        result[out_pos]  = src[in_pos];
+                    }
+                    ++out_pos;
+                }
+            }
+        }
+    }
+
+    return result;
+}
+
+template SimpleTensor<float> space_to_batch(const SimpleTensor<float> &src, const SimpleTensor<int32_t> &block_shape, const SimpleTensor<int32_t> &paddings, const TensorShape &dst_shape);
+template SimpleTensor<half> space_to_batch(const SimpleTensor<half> &src, const SimpleTensor<int32_t> &block_shape, const SimpleTensor<int32_t> &paddings, const TensorShape &dst_shape);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/reference/SpaceToBatch.h b/tests/validation/reference/SpaceToBatch.h
new file mode 100644
index 0000000000..ba35211314
--- /dev/null
+++ b/tests/validation/reference/SpaceToBatch.h
@@ -0,0 +1,44 @@
+/*
+ * 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.
+ */
+#ifndef __ARM_COMPUTE_TEST_SPACE_TO_BATCH_LAYER_H__
+#define __ARM_COMPUTE_TEST_SPACE_TO_BATCH_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_batch(const SimpleTensor<T> &src, const SimpleTensor<int32_t> &block_shape, const SimpleTensor<int32_t> &paddings, const TensorShape &dst_shape);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_SPACE_TO_BATCH_LAYER_H__ */