COMPMID-1725: Implement Pack

Change-Id: I13f6e4c600f39355f69e015409bf30dafdc5e3aa
Reviewed-on: https://review.mlplatform.org/332
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>

14 files changed, 1256 insertions, 0 deletions

diff --git a/arm_compute/core/CL/CLKernels.h b/arm_compute/core/CL/CLKernels.h
index c707265c23..0fe1e04ccd 100644
--- a/arm_compute/core/CL/CLKernels.h
+++ b/arm_compute/core/CL/CLKernels.h
@@ -121,6 +121,7 @@
 #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/CLStackLayerKernel.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/CLStackLayerKernel.h b/arm_compute/core/CL/kernels/CLStackLayerKernel.h
new file mode 100644
index 0000000000..4d377daf8b
--- /dev/null
+++ b/arm_compute/core/CL/kernels/CLStackLayerKernel.h
@@ -0,0 +1,83 @@
+/*
+ * 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_CLSTACKLAYERKERNEL_H__
+#define __ARM_COMPUTE_CLSTACKLAYERKERNEL_H__
+
+#include "arm_compute/core/CL/ICLKernel.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ICLTensor;
+
+/** OpenCL kernel to stacks a rank-R tensor into one with rank-(R+1) along the axis dimension.*/
+class CLStackLayerKernel : public ICLKernel
+{
+public:
+    /** Default constructor */
+    CLStackLayerKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLStackLayerKernel(const CLStackLayerKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLStackLayerKernel &operator=(const CLStackLayerKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    CLStackLayerKernel(CLStackLayerKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    CLStackLayerKernel &operator=(CLStackLayerKernel &&) = default;
+    /** Default destructor */
+    ~CLStackLayerKernel() = default;
+    /** Initialise the kernel's inputs and output
+     *
+     * @param[in]  input       Input tensor. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32
+     * @param[in]  axis        The dimension to stack the tensors along. It must be smaller than the number of input dimensions.
+     * @param[in]  idx_input   Index of the input tensor in the list of tensors to stack.
+     *                         All tensors in the list must have the same shape
+     * @param[in]  num_tensors Number of tensors to stack
+     * @param[out] output      Output tensor. Data types supported: Same as @p input.
+     *
+     */
+    void configure(const ICLTensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ICLTensor *output);
+    /**  Static function to check if given info will lead to a valid configuration of @ref CLStackLayerKernel
+     *
+     * @param[in] input       Input tensor info. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32
+     * @param[in] axis        The dimension to stack the tensors along. It must be smaller than the number of input dimensions.
+     * @param[in] idx_input   Index of the input tensor in the list of tensors to stack
+     *                        All tensors in the list must have the same shape
+     * @param[in] num_tensors Number of tensors to stack
+     * @param[in] output      Output tensor info. Data types supported: Same as @p input.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, const ITensorInfo *output);
+
+    // Inherited methods overridden:
+    void run(const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    const ICLTensor *_input;
+    ICLTensor       *_output;
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_CLSTACKLAYERKERNEL_H__ */
diff --git a/arm_compute/core/Helpers.h b/arm_compute/core/Helpers.h
index ef59323073..8f4220fb80 100644
--- a/arm_compute/core/Helpers.h
+++ b/arm_compute/core/Helpers.h
@@ -710,6 +710,18 @@ inline Size2D compute_winograd_convolution_tiles(const Size2D &in_dims, const Si
     return Size2D(num_tiles_x, num_tiles_y);
 }
 
+/** Wrap-around a number within the range 0 <= x < m
+ *
+ * @param[in] x Input value
+ * @param[in] m Range
+ *
+ * @return the wrapped-around number
+ */
+template <typename T>
+inline T wrap_around(T x, T m)
+{
+    return x >= 0 ? x % m : (x % m + m) % m;
+}
 } // namespace arm_compute
 
 #include "arm_compute/core/Helpers.inl"
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index 38906dfc9b..c625a07a7f 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -754,6 +754,28 @@ inline TensorShape calculate_width_concatenate_shape(const std::vector<T *> &inp
 
     return out_shape;
 }
+
+inline TensorShape compute_stack_shape(const ITensorInfo &a, unsigned int axis, unsigned int num_tensors)
+{
+    ARM_COMPUTE_ERROR_ON(axis > a.num_dimensions());
+    ARM_COMPUTE_ERROR_ON(a.num_dimensions() > 4);
+
+    TensorShape shape_out{ a.tensor_shape() };
+    shape_out.set(axis, num_tensors);
+
+    unsigned int i_shift = 0;
+
+    for(unsigned int i = 0; i < a.num_dimensions(); ++i)
+    {
+        if(i == axis)
+        {
+            i_shift++;
+        }
+
+        shape_out.set(i + i_shift, a.tensor_shape()[i]);
+    }
+    return shape_out;
+}
 } // namespace shape_calculator
 } // namespace misc
 } // namespace arm_compute
diff --git a/arm_compute/runtime/CL/CLFunctions.h b/arm_compute/runtime/CL/CLFunctions.h
index e68e719a13..d340ce1e0c 100644
--- a/arm_compute/runtime/CL/CLFunctions.h
+++ b/arm_compute/runtime/CL/CLFunctions.h
@@ -125,6 +125,7 @@
 #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/CLStackLayer.h"
 #include "arm_compute/runtime/CL/functions/CLStridedSlice.h"
 #include "arm_compute/runtime/CL/functions/CLTableLookup.h"
 #include "arm_compute/runtime/CL/functions/CLThreshold.h"
diff --git a/arm_compute/runtime/CL/functions/CLStackLayer.h b/arm_compute/runtime/CL/functions/CLStackLayer.h
new file mode 100644
index 0000000000..9794014889
--- /dev/null
+++ b/arm_compute/runtime/CL/functions/CLStackLayer.h
@@ -0,0 +1,77 @@
+/*
+ * 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_CLSTACKLAYER_H__
+#define __ARM_COMPUTE_CLSTACKLAYER_H__
+
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/IFunction.h"
+
+#include "arm_compute/core/CL/kernels/CLStackLayerKernel.h"
+
+#include <memory>
+#include <vector>
+
+namespace arm_compute
+{
+class ICLTensor;
+
+/** Basic function to stack tensors along an axis. This function calls the following kernel:
+ *
+ * -# @ref CLStackLayerKernel
+ *
+ */
+class CLStackLayer : public IFunction
+{
+public:
+    /** Default constructor */
+    CLStackLayer();
+    /** Initialise the kernel's inputs vector and output.
+     *
+     * @param[in]  input  The vectors containing all the tensors with the same shape to stack. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32
+     * @param[in]  axis   The dimension to stack the tensors along. It must be smaller than the number of input dimensions.
+     *                    Negative values wrap around
+     * @param[out] output Output tensor. Data types supported: Same as @p input.
+     */
+    void configure(const std::vector<ICLTensor *> &input, int axis, ICLTensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLDepthConcatenateLayer
+     *
+     * @param[in] input  The vectors containing all the tensors info with the same shape to stack. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32
+     * @param[in] axis   The dimension to stack the tensors along. It must be smaller than the number of input dimensions.
+     *                   Negative values wrap around
+     * @param[in] output Output tensor info. Data types supported: Same as @p input.
+     *
+     * @return a status
+     */
+    static Status validate(const std::vector<ITensorInfo *> &input, int axis, const ITensorInfo *output);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    std::vector<ICLTensor *>              _input;
+    std::unique_ptr<CLStackLayerKernel[]> _stack_kernels;
+    unsigned int                          _num_inputs;
+};
+}
+#endif /* __ARM_COMPUTE_CLSTACKLAYER_H__ */
diff --git a/src/core/CL/CLKernelLibrary.cpp b/src/core/CL/CLKernelLibrary.cpp
index ac1d4b349e..e48ff03e05 100644
--- a/src/core/CL/CLKernelLibrary.cpp
+++ b/src/core/CL/CLKernelLibrary.cpp
@@ -423,6 +423,7 @@ const std::map<std::string, std::string> CLKernelLibrary::_kernel_program_map =
     { "space_to_batch_nhwc", "space_to_batch.cl" },
     { "space_to_batch_static_nhwc", "space_to_batch.cl" },
     { "softmax_layer_max_shift_exp_sum_parallel", "softmax_layer.cl" },
+    { "stack_layer", "stack_layer.cl" },
     { "strided_slice", "slice_ops.cl" },
     { "suppress_non_maximum", "canny.cl" },
     { "tablelookup_U8", "tablelookup.cl" },
@@ -865,6 +866,10 @@ const std::map<std::string, std::string> CLKernelLibrary::_program_source_map =
 #include "./cl_kernels/space_to_batch.clembed"
     },
     {
+        "stack_layer.cl",
+#include "./cl_kernels/stack_layer.clembed"
+    },
+    {
         "tablelookup.cl",
 #include "./cl_kernels/tablelookup.clembed"
     },
diff --git a/src/core/CL/cl_kernels/stack_layer.cl b/src/core/CL/cl_kernels/stack_layer.cl
new file mode 100644
index 0000000000..bed62662ad
--- /dev/null
+++ b/src/core/CL/cl_kernels/stack_layer.cl
@@ -0,0 +1,113 @@
+/*
+ * 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 "helpers.h"
+
+#if defined(DATA_TYPE) && defined(AXIS) && defined(SRC_DIM2) && defined(DST_DIM3)
+
+#if AXIS == 0
+#define X_DST (idx_input)
+#define Y_DST (x_src)
+#define Z_DST (y_src)
+#define W_DST (z_src)
+#define K_DST (w_src)
+#elif AXIS == 1 // AXIS == 1
+#define X_DST (x_src)
+#define Y_DST (idx_input)
+#define Z_DST (y_src)
+#define W_DST (z_src)
+#define K_DST (w_src)
+#elif AXIS == 2 // AXIS == 2
+#define X_DST (x_src)
+#define Y_DST (y_src)
+#define Z_DST (idx_input)
+#define W_DST (z_src)
+#define K_DST (w_src)
+#elif AXIS == 3 // AXIS == 3
+#define X_DST (x_src)
+#define Y_DST (y_src)
+#define Z_DST (z_src)
+#define W_DST (idx_input)
+#define K_DST (w_src)
+#elif AXIS == 4 // AXIS == 4
+#define X_DST (x_src)
+#define Y_DST (y_src)
+#define Z_DST (z_src)
+#define W_DST (w_src)
+#define K_DST (idx_input)
+#else // AXIS not supported
+#error "Not supported axis"
+#endif // AXIS == 0
+
+/** OpenCL kernel to stack a rank-R tensor into one with rank-(R+1) along the axis dimension
+ *
+ * @note The data type has to be passed at compile time using -DDATA_TYPE. i.e. -DDATA_TYPE=float
+ * @note The dimension to stack the tensors along has to be passed at compile time using -DAXIS. i.e. -DAXIS=1
+ * @note Dimension 2 of the input tensor must be passed at compile time using -DSRC_DIM2 (e.g. -DSRC_DIM2=112)
+ * @note Dimension 3 of the output tensor must be passed at compile time using -DDST_DIM3 (e.g. -DDST_DIM3=112)
+ *
+ * @param[in]  src_ptr                           Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32
+ * @param[in]  src_stride_x                      Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  src_step_x                        src_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  src_stride_y                      Stride of the source tensor in Y dimension (in bytes)
+ * @param[in]  src_step_y                        src_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  src_stride_z                      Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  src_step_z                        src_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  src_stride_w                      Stride of the source tensor in W dimension (in bytes)
+ * @param[in]  src_step_w                        src_stride_w * number of elements along W processed per workitem(in bytes)
+ * @param[in]  src_offset_first_element_in_bytes The offset of the first element in the source tensor
+ * @param[out] dst_ptr                           Pointer to the destination tensor. Supported data types: same as @p src_ptr
+ * @param[in]  dst_stride_x                      Stride of the destination tensor in X dimension (in bytes)
+ * @param[in]  dst_step_x                        dst_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  dst_stride_y                      Stride of the destination tensor in Y dimension (in bytes)
+ * @param[in]  dst_step_y                        dst_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  dst_stride_z                      Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  dst_step_z                        dst_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  dst_stride_w                      Stride of the source tensor in W dimension (in bytes)
+ * @param[in]  dst_step_w                        dst_stride_w * number of elements along W processed per workitem(in bytes)
+ * @param[in]  dst_offset_first_element_in_bytes The offset of the first element in the destination tensor
+ * @param[in]  idx_input                         Index of the input tensor in the list of tensors to stack
+ */
+__kernel void stack_layer(
+    TENSOR4D_DECLARATION(src),
+    TENSOR4D_DECLARATION(dst),
+    unsigned int idx_input)
+{
+    uint x_src = get_global_id(0);
+    uint y_src = get_global_id(1);
+    uint z_src = (get_global_id(2) % SRC_DIM2);
+    uint w_src = (get_global_id(2) / SRC_DIM2);
+
+    __global DATA_TYPE *src = (__global DATA_TYPE *)(src_ptr + src_offset_first_element_in_bytes + x_src * sizeof(DATA_TYPE) + y_src * src_stride_y + z_src * src_stride_z + w_src * src_stride_w);
+
+    __global DATA_TYPE *dst = (__global DATA_TYPE *)(dst_ptr + dst_offset_first_element_in_bytes + X_DST * sizeof(DATA_TYPE) + Y_DST * dst_stride_y + Z_DST * dst_stride_z + W_DST * dst_stride_w + K_DST *
+                                                     dst_stride_w * (uint)DST_DIM3);
+
+    *dst = *src;
+}
+
+#undef X_DST
+#undef Y_DST
+#undef Z_DST
+#undef W_DST
+#endif // defined(DATA_TYPE) && defined(AXIS) && defined(SRC_DIM2) && defined(DST_DIM3)
diff --git a/src/core/CL/kernels/CLStackLayerKernel.cpp b/src/core/CL/kernels/CLStackLayerKernel.cpp
new file mode 100644
index 0000000000..bac8992f7b
--- /dev/null
+++ b/src/core/CL/kernels/CLStackLayerKernel.cpp
@@ -0,0 +1,145 @@
+/*
+ * 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/CLStackLayerKernel.h"
+
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/CL/CLValidate.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/CL/OpenCL.h"
+#include "arm_compute/core/Error.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/IAccessWindow.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
+#include "support/ToolchainSupport.h"
+
+using namespace arm_compute;
+using namespace arm_compute::misc::shape_calculator;
+
+namespace
+{
+Status validate_arguments(const ITensorInfo *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::U8, DataType::S8,
+                                                         DataType::U16, DataType::S16, DataType::U32, DataType::S32,
+                                                         DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON(idx_input >= num_tensors);
+    ARM_COMPUTE_RETURN_ERROR_ON(axis > input->num_dimensions());
+    ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
+
+    if(output->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), compute_stack_shape(*input, axis, num_tensors));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    }
+
+    return Status{};
+}
+
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, unsigned int axis, unsigned int num_tensors, ITensorInfo *output)
+{
+    // Output auto inizialitation if not yet initialized
+    auto_init_if_empty(*output, input->clone()->set_tensor_shape(compute_stack_shape(*input, axis, num_tensors)));
+
+    // Configure kernel window
+    constexpr unsigned int num_elems_processed_per_iteration = 1;
+
+    // The window needs to be based on input as we copy all the depths of input
+    Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration));
+
+    AccessWindowHorizontal input_access(input, 0, num_elems_processed_per_iteration);
+    AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration);
+
+    bool window_changed = update_window_and_padding(win, input_access, output_access);
+    output_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
+
+CLStackLayerKernel::CLStackLayerKernel()
+    : _input(nullptr), _output(nullptr)
+{
+}
+
+void CLStackLayerKernel::configure(const ICLTensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ICLTensor *output)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), axis, idx_input, num_tensors, output->info()));
+
+    _input  = input;
+    _output = output;
+
+    // Configure kernel window
+    auto win_config = validate_and_configure_window(input->info(), axis, num_tensors, output->info());
+
+    // Add build options
+    CLBuildOptions build_opts;
+    build_opts.add_option("-DDATA_TYPE=" + get_underlying_cl_type_from_data_type(input->info()->data_type()));
+    build_opts.add_option("-DAXIS=" + support::cpp11::to_string(axis));
+    build_opts.add_option("-DSRC_DIM2=" + support::cpp11::to_string(input->info()->dimension(2)));
+    build_opts.add_option("-DDST_DIM3=" + support::cpp11::to_string(output->info()->dimension(3)));
+
+    // Create kernel
+    _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("stack_layer", build_opts.options()));
+
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    ICLKernel::configure_internal(win_config.second);
+
+    const unsigned int idx = 2 * num_arguments_per_4D_tensor();
+    _kernel.setArg<cl_uint>(idx, idx_input);
+}
+
+Status CLStackLayerKernel::validate(const ITensorInfo *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, axis, idx_input, num_tensors, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), axis, num_tensors, output->clone().get()).first);
+    return Status{};
+}
+
+void CLStackLayerKernel::run(const Window &window, cl::CommandQueue &queue)
+{
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
+
+    Window window_out;
+    window_out.use_tensor_dimensions(_output->info()->tensor_shape());
+
+    Window collapsed = window.collapse(ICLKernel::window(), Window::DimZ);
+
+    Window slice_in  = collapsed.first_slice_window_4D();
+    Window slice_out = window_out.first_slice_window_4D();
+
+    unsigned int idx = 0;
+    add_4D_tensor_argument(idx, _input, slice_in);
+    add_4D_tensor_argument(idx, _output, slice_out);
+    enqueue(queue, *this, slice_in);
+}
diff --git a/src/runtime/CL/functions/CLStackLayer.cpp b/src/runtime/CL/functions/CLStackLayer.cpp
new file mode 100644
index 0000000000..85adcad90c
--- /dev/null
+++ b/src/runtime/CL/functions/CLStackLayer.cpp
@@ -0,0 +1,85 @@
+/*
+ * 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 <complex>
+
+#include "arm_compute/runtime/CL/functions/CLStackLayer.h"
+
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/Error.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "arm_compute/runtime/CL/CLScheduler.h"
+
+#include "support/ToolchainSupport.h"
+
+using namespace arm_compute;
+
+CLStackLayer::CLStackLayer() // NOLINT
+    : _input(),
+      _stack_kernels(),
+      _num_inputs(0)
+{
+}
+
+void CLStackLayer::configure(const std::vector<ICLTensor *> &input, int axis, ICLTensor *output)
+{
+    _num_inputs    = input.size();
+    _stack_kernels = arm_compute::support::cpp14::make_unique<CLStackLayerKernel[]>(_num_inputs);
+
+    // Wrap around negative values
+    const unsigned int axis_u = wrap_around(axis, static_cast<int>(input[0]->info()->num_dimensions() + 1));
+
+    for(unsigned int i = 0; i < _num_inputs; i++)
+    {
+        _stack_kernels[i].configure(input[i], axis_u, i, _num_inputs, output);
+    }
+}
+
+Status CLStackLayer::validate(const std::vector<ITensorInfo *> &input, int axis, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
+    ARM_COMPUTE_RETURN_ERROR_ON(input.size() < 2);
+
+    // Wrap around negative values
+    const unsigned int axis_u = wrap_around(axis, static_cast<int>(input[0]->num_dimensions() + 1));
+
+    const unsigned int num_inputs = input.size();
+
+    for(unsigned int i = 0; i < num_inputs; i++)
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(CLStackLayerKernel::validate(input[i], axis_u, i, num_inputs, output));
+    }
+
+    return Status{};
+}
+
+void CLStackLayer::run()
+{
+    for(unsigned i = 0; i < _num_inputs; i++)
+    {
+        CLScheduler::get().enqueue(_stack_kernels[i], false);
+    }
+}
diff --git a/tests/validation/CL/StackLayer.cpp b/tests/validation/CL/StackLayer.cpp
new file mode 100644
index 0000000000..089911272a
--- /dev/null
+++ b/tests/validation/CL/StackLayer.cpp
@@ -0,0 +1,405 @@
+/*
+ * 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/Helpers.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/CLTensorAllocator.h"
+#include "arm_compute/runtime/CL/functions/CLStackLayer.h"
+#include "tests/CL/CLAccessor.h"
+#include "tests/CL/Helper.h"
+#include "tests/PaddingCalculator.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/StackLayerFixture.h"
+
+#include <vector>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+// *INDENT-OFF*
+// clang-format off
+/** Data types */
+const auto data_types = framework::dataset::make("DataType", { DataType::QASYMM8, DataType::F16, DataType::F32 });
+
+/** Num tensors values to test */
+const auto n_values = framework::dataset::make("NumTensors", { 3, 4 });
+
+/** Shapes 1D to test */
+const auto shapes_1d_small = combine(datasets::Small1DShapes(), framework::dataset::make("Axis", -1, 2));
+
+/** Shapes 2D to test */
+const auto shapes_2d_small = combine(datasets::Small2DShapes(), framework::dataset::make("Axis", -2, 3));
+
+/** Shapes 3D to test */
+const auto shapes_3d_small = combine(datasets::Small3DShapes(), framework::dataset::make("Axis", -3, 4));
+
+/** Shapes 4D to test */
+const auto shapes_4d_small = combine(datasets::Small4DShapes(), framework::dataset::make("Axis", -4, 5));
+
+/** Shapes 1D to test */
+const auto shapes_1d_large = combine(datasets::Large1DShapes(), framework::dataset::make("Axis", -1, 2));
+
+/** Shapes 2D to test */
+const auto shapes_2d_large = combine(datasets::Large2DShapes(), framework::dataset::make("Axis", -2, 3));
+
+/** Shapes 3D to test */
+const auto shapes_3d_large = combine(datasets::Large3DShapes(), framework::dataset::make("Axis", -3, 4));
+
+/** Shapes 4D to test */
+const auto shapes_4d_large = combine(datasets::Large4DShapes(), framework::dataset::make("Axis", -4, 5));
+
+/** Configuration test */
+void validate_configuration(TensorShape shape_in, int axis, DataType data_type, int num_tensors)
+{
+    // Wrap around negative values
+    const unsigned int axis_u = wrap_around(axis, static_cast<int>(shape_in.num_dimensions() + 1));
+
+    const TensorShape shape_dst = compute_stack_shape(TensorInfo(shape_in, 1, data_type), axis_u, num_tensors);
+
+    std::vector<CLTensor>   tensors(num_tensors);
+    std::vector<ICLTensor*> src(num_tensors);
+
+    // Create vector of input tensors
+    for(int i = 0; i < num_tensors; ++i)
+    {
+        tensors[i] = create_tensor<CLTensor>(shape_in, data_type);
+        src[i]     = &(tensors[i]);
+        ARM_COMPUTE_EXPECT(src[i]->info()->is_resizable(), framework::LogLevel::ERRORS);
+    }
+
+    // Create tensors
+    CLTensor dst = create_tensor<CLTensor>(shape_dst, data_type);
+
+    ARM_COMPUTE_EXPECT(dst.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+    // Create and configure function
+    CLStackLayer stack;
+    stack.configure(src, axis, &dst);
+}
+} // namespace
+
+/** Fixture to use */
+template<typename T>
+using CLStackLayerFixture = StackLayerValidationFixture<CLTensor, ICLTensor, CLAccessor, CLStackLayer, T>;
+
+using namespace arm_compute::misc::shape_calculator;
+
+TEST_SUITE(CL)
+TEST_SUITE(StackLayer)
+TEST_SUITE(Shapes1D)
+
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(shapes_1d_small,
+                                                                           data_types),
+                                                                           n_values),
+shape_in, axis, data_type, num_tensors)
+{
+    validate_configuration(shape_in, axis, data_type, num_tensors);
+}
+
+TEST_SUITE(S32)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<int>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_1d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S32 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<int>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_1d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S32 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S32
+
+TEST_SUITE(S16)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<short>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_1d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S16 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<short>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_1d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S16 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S16
+
+TEST_SUITE(S8)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<char>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_1d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S8 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<char>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_1d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S8 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S8
+TEST_SUITE_END() // Shapes1D
+
+TEST_SUITE(Shapes2D)
+
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(shapes_2d_small,
+                                                                           data_types),
+                                                                           n_values),
+shape_in, axis, data_type, num_tensors)
+{
+    validate_configuration(shape_in, axis, data_type, num_tensors);
+}
+
+TEST_SUITE(S32)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<int>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_2d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S32 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<int>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_2d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S32 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S32
+
+TEST_SUITE(S16)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<short>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_2d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S16 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<short>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_2d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S16 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S16
+
+TEST_SUITE(S8)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<char>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_2d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S8 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<char>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_2d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S8 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S8
+TEST_SUITE_END() // Shapes2D
+
+TEST_SUITE(Shapes3D)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(shapes_3d_small,
+                                                                           data_types),
+                                                                           n_values),
+shape_in, axis, data_type, num_tensors)
+{
+    validate_configuration(shape_in, axis, data_type, num_tensors);
+}
+
+TEST_SUITE(S32)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<int>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_3d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S32 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<int>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_3d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S32 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S32
+
+TEST_SUITE(S16)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<short>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_3d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S16 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<short>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_3d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S16 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S16
+
+TEST_SUITE(S8)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<char>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_3d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S8 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<char>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_3d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S8 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S8
+TEST_SUITE_END() // Shapes3D
+
+TEST_SUITE(Shapes4D)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(shapes_4d_small,
+                                                                           data_types),
+                                                                           n_values),
+shape_in, axis, data_type, num_tensors)
+{
+    validate_configuration(shape_in, axis, data_type, num_tensors);
+}
+
+TEST_SUITE(S32)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<int>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_4d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S32 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<int>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_4d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S32 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S32
+
+TEST_SUITE(S16)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<short>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_4d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S16 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<short>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_4d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S16 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S16
+
+TEST_SUITE(S8)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLStackLayerFixture<char>, framework::DatasetMode::ALL,
+                                                           combine(combine(shapes_4d_small,
+                                                                           framework::dataset::make("DataType", { DataType::S8 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLStackLayerFixture<char>, framework::DatasetMode::NIGHTLY,
+                                                           combine(combine(shapes_4d_large,
+                                                                           framework::dataset::make("DataType", { DataType::S8 })),
+                                                                           n_values))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // S8
+TEST_SUITE_END() // Shapes4D
+TEST_SUITE_END() // StackLayer
+TEST_SUITE_END() // CL
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/StackLayerFixture.h b/tests/validation/fixtures/StackLayerFixture.h
new file mode 100644
index 0000000000..cab4350787
--- /dev/null
+++ b/tests/validation/fixtures/StackLayerFixture.h
@@ -0,0 +1,138 @@
+/*
+ * 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_STACK_LAYER_FIXTURE
+#define ARM_COMPUTE_TEST_STACK_LAYER_FIXTURE
+
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "tests/AssetsLibrary.h"
+#include "tests/Globals.h"
+#include "tests/IAccessor.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Fixture.h"
+#include "tests/validation/Helpers.h"
+#include "tests/validation/reference/StackLayer.h"
+#include "tests/validation/reference/Utils.h"
+
+#include <random>
+#include <vector>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+using namespace arm_compute::misc::shape_calculator;
+
+template <typename TensorType, typename AbstractTensorType, typename AccessorType, typename FunctionType, typename T>
+class StackLayerValidationFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape shape_src, int axis, DataType data_type, int num_tensors)
+    {
+        _target    = compute_target(shape_src, axis, data_type, num_tensors);
+        _reference = compute_reference(shape_src, axis, data_type, num_tensors);
+    }
+
+protected:
+    template <typename U>
+    void fill(U &&tensor, unsigned int i)
+    {
+        library->fill_tensor_uniform(tensor, i);
+    }
+
+    TensorType compute_target(TensorShape shape_src, int axis, DataType data_type, int num_tensors)
+    {
+        std::vector<TensorType>           tensors(num_tensors);
+        std::vector<AbstractTensorType *> src(num_tensors);
+
+        // Create vector of input tensors
+        for(int i = 0; i < num_tensors; ++i)
+        {
+            tensors[i] = create_tensor<TensorType>(shape_src, data_type);
+            src[i]     = &(tensors[i]);
+            ARM_COMPUTE_EXPECT(tensors[i].info()->is_resizable(), framework::LogLevel::ERRORS);
+        }
+
+        // Create tensors
+        CLTensor dst;
+
+        // The output tensor will be auto-initialized within the function
+
+        // Create and configure function
+        FunctionType stack;
+        stack.configure(src, axis, &dst);
+
+        // Allocate and fill the input tensors
+        for(int i = 0; i < num_tensors; ++i)
+        {
+            ARM_COMPUTE_EXPECT(tensors[i].info()->is_resizable(), framework::LogLevel::ERRORS);
+            tensors[i].allocator()->allocate();
+            ARM_COMPUTE_EXPECT(!tensors[i].info()->is_resizable(), framework::LogLevel::ERRORS);
+
+            // Fill input tensor
+            fill(AccessorType(tensors[i]), i);
+        }
+
+        // Allocate output tensor
+        dst.allocator()->allocate();
+
+        ARM_COMPUTE_EXPECT(!dst.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Compute stack function
+        stack.run();
+
+        return dst;
+    }
+
+    SimpleTensor<T> compute_reference(const TensorShape &shape_src, int axis, DataType data_type, int num_tensors)
+    {
+        std::vector<SimpleTensor<T>> src;
+
+        for(int i = 0; i < num_tensors; ++i)
+        {
+            src.emplace_back(std::move(SimpleTensor<T>(shape_src, data_type, 1)));
+
+            fill(src[i], i);
+        }
+
+        // Wrap around negative values
+        const unsigned int axis_u = wrap_around(axis, static_cast<int>(shape_src.num_dimensions() + 1));
+
+        const TensorShape shape_dst = compute_stack_shape(TensorInfo(shape_src, 1, data_type), axis_u, num_tensors);
+
+        return reference::stack_layer<T>(src, shape_dst, data_type, axis_u);
+    }
+
+    TensorType      _target{};
+    SimpleTensor<T> _reference{};
+};
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_STACK_LAYER_FIXTURE */
diff --git a/tests/validation/reference/StackLayer.cpp b/tests/validation/reference/StackLayer.cpp
new file mode 100644
index 0000000000..50e440c914
--- /dev/null
+++ b/tests/validation/reference/StackLayer.cpp
@@ -0,0 +1,125 @@
+/*
+ * 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 "StackLayer.h"
+
+#include "arm_compute/core/Types.h"
+
+#include "tests/validation/Helpers.h"
+
+#include <vector>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> stack_layer(const std::vector<SimpleTensor<T>> &in, const TensorShape &output_shape, DataType data_type, unsigned int axis)
+{
+    ARM_COMPUTE_ERROR_ON(output_shape.num_dimensions() > 5);
+    ARM_COMPUTE_ERROR_ON(in.size() < 2);
+    ARM_COMPUTE_ERROR_ON(axis > in[0].shape().num_dimensions());
+
+    SimpleTensor<T> out{ output_shape, data_type };
+
+    const int width       = in[0].shape()[0];
+    const int height      = in[0].shape()[1];
+    const int depth       = in[0].shape()[2];
+    const int batch_size  = in[0].shape()[3];
+    const int num_tensors = in.size();
+
+    // Array to store the input coordinates
+    // i_coordinates[0] = xi, i_coordinates[1] = yi, i_coordinates[2] = zi
+    // i_coordinates[3] = bi, i_coordinates[4] = i, i_coordinates[5] = 0
+    // i_coordinates[5] will be always zero and used for not incrementing the output when the input has less than 4 dimensions
+    int i_coordinates[6] = { 0 };
+
+    // Array of pointers used to map the output coordinates to the input ones accordingly with the axis
+    // This array is initialized with &i_coordinates[5] since this will be always zero
+    int *o_coordinates[5] = { &i_coordinates[5], &i_coordinates[5], &i_coordinates[5], &i_coordinates[5], &i_coordinates[5] };
+
+    // Set the axis coordinate
+    o_coordinates[axis] = &i_coordinates[4];
+
+    unsigned int k_shift = 0;
+
+    // Map the output coordinates
+    for(unsigned int k = 0; k < in[0].shape().num_dimensions(); ++k)
+    {
+        if(k == axis)
+        {
+            k_shift++;
+        }
+
+        o_coordinates[k + k_shift] = &i_coordinates[k];
+    }
+
+    // Use alias for the input coordinates
+    int &xi = i_coordinates[0];
+    int &yi = i_coordinates[1];
+    int &zi = i_coordinates[2];
+    int &bi = i_coordinates[3];
+    int &i  = i_coordinates[4];
+
+    // Use alias for the output coordinates
+    int &xo = *(o_coordinates[0]);
+    int &yo = *(o_coordinates[1]);
+    int &zo = *(o_coordinates[2]);
+    int &bo = *(o_coordinates[3]);
+    int &wo = *(o_coordinates[4]);
+
+    // Stack tensors
+    for(; i < num_tensors; ++(i))
+    {
+        bi = 0;
+        for(; bi < batch_size; ++(bi))
+        {
+            zi = 0;
+            for(; zi < depth; ++(zi))
+            {
+                yi = 0;
+                for(; yi < height; ++(yi))
+                {
+                    xi = 0;
+                    for(; xi < width; ++(xi))
+                    {
+                        *(reinterpret_cast<T *>(out(Coordinates(xo, yo, zo, bo, wo)))) = *(reinterpret_cast<const T *>(in[i](Coordinates(xi, yi, zi, bi))));
+                    }
+                }
+            }
+        }
+    }
+
+    return out;
+}
+template SimpleTensor<int> stack_layer(const std::vector<SimpleTensor<int>> &in, const TensorShape &output_shape, DataType data_type, unsigned int axis);
+template SimpleTensor<short> stack_layer(const std::vector<SimpleTensor<short>> &in, const TensorShape &output_shape, DataType data_type, unsigned int axis);
+template SimpleTensor<char> stack_layer(const std::vector<SimpleTensor<char>> &in, const TensorShape &output_shape, DataType data_type, unsigned int axis);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/reference/StackLayer.h b/tests/validation/reference/StackLayer.h
new file mode 100644
index 0000000000..453f176a9d
--- /dev/null
+++ b/tests/validation/reference/StackLayer.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_STACK_LAYER_H__
+#define __ARM_COMPUTE_TEST_STACK_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> stack_layer(const std::vector<SimpleTensor<T>> &in, const TensorShape &output_shape, DataType data_type, unsigned int axis);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_STACK_LAYER_H__ */