COMPMID-1763 : NEON: Implement Gather

Change-Id: I9a3808315290bd395f5acce4530ab8daccddf8be
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/167195
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/520
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez <pablo.tello@arm.com>

7 files changed, 636 insertions, 0 deletions

diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index b7810cbd51..cc6af39a54 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -79,6 +79,7 @@
 #include "arm_compute/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h"
+#include "arm_compute/core/NEON/kernels/NEGatherKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGaussian3x3Kernel.h"
 #include "arm_compute/core/NEON/kernels/NEGaussian5x5Kernel.h"
 #include "arm_compute/core/NEON/kernels/NEGaussianPyramidKernel.h"
diff --git a/arm_compute/core/NEON/kernels/NEGatherKernel.h b/arm_compute/core/NEON/kernels/NEGatherKernel.h
new file mode 100644
index 0000000000..667e9812d8
--- /dev/null
+++ b/arm_compute/core/NEON/kernels/NEGatherKernel.h
@@ -0,0 +1,112 @@
+/*
+ * Copyright (c) 2019 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef __ARM_COMPUTE_NEGATHERKERNEL_H__
+#define __ARM_COMPUTE_NEGATHERKERNEL_H__
+
+#include "arm_compute/core/NEON/INEKernel.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Kernel to perform other operation on NEON */
+class NEGatherKernel : public INEKernel
+{
+public:
+    /** Default constructor. */
+    NEGatherKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers). */
+    NEGatherKernel(const NEGatherKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers). */
+    NEGatherKernel &operator=(const NEGatherKernel &) = delete;
+    /** Allow instances of this class to be moved. */
+    NEGatherKernel(NEGatherKernel &&) = default;
+    /** Allow instances of this class to be moved. */
+    NEGatherKernel &operator=(NEGatherKernel &&) = default;
+    /** Default detructor */
+    ~NEGatherKernel() = default;
+
+    /** Name of the kernel
+     *
+     * @return Kernel name
+     */
+    const char *name() const override
+    {
+        return "NEGatherKernel";
+    }
+    /** Initialise the kernel's inputs and outputs
+     *
+     * @param[in]  input   Source tensor. Supported tensor rank: up to 4. Data type supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+     * @param[in]  indices Indices tensor. Supported tensor rank: up to 1. Must be one of the following type: U32/S32. Each value Must be in range [0, input.shape[@p axis])
+     * @param[out] output  Destination tensor. Data type supported: Same as @p input
+     * @param[in]  axis    (Optional) The axis in @p input to gather @p indices from. Negative values wrap around. Defaults to 0
+     */
+    void configure(const ITensor *input, const ITensor *indices, ITensor *output, int axis = 0);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEGatherKernel
+     *
+     * @param[in] input   Source tensor info. Supported tensor rank: up to 4. Data type supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+     * @param[in] indices Indices tensor info. Supported tensor rank: up to 1. Must be one of the following type: U32/S32. Each value Must be in range [0, input.shape[@p axis])
+     * @param[in] output  Destination tensor info. Data type supported: Same as @p input
+     * @param[in] axis    (Optional) The axis in @p input to gather @p indices from. Negative values wrap around. Defaults to 0
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, int axis);
+
+    // Inherited methods overridden:
+    void run(const Window &window, const ThreadInfo &info) override;
+
+private:
+    /** Implementation of the gather operation for 0 axis.
+     *
+     * For gather on the 0 axis an element by element copy is performed.
+     *
+     * @param[in] window Region on which to execute the kernel. (Must be a region of the window returned by window())
+     * @param[in] info   Info about executing thread and CPU.
+     */
+    template <typename U>
+    void gather_0_axis(const Window &window, const ThreadInfo &info);
+
+    /** Implementation of the gather operation.
+     *
+     * For 1<=axis a row-wise copy is taking place.
+     *
+     * @param[in] window Region on which to execute the kernel. (Must be a region of the window returned by window())
+     * @param[in] info   Info about executing thread and CPU.
+     */
+    template <typename U>
+    void gather_n_axis(const Window &window, const ThreadInfo &info);
+
+    using kernel_ptr = void (NEGatherKernel::*)(const Window &window, const ThreadInfo &info);
+
+    const ITensor *_input;
+    const ITensor *_indices;
+    int            _axis;
+    ITensor       *_output;
+    kernel_ptr     _func;
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_NEGATHERKERNEL_H__ */
diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index 606f282e16..cc13ff5d5e 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -75,6 +75,7 @@
 #include "arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMMTranspose1xW.h"
+#include "arm_compute/runtime/NEON/functions/NEGather.h"
 #include "arm_compute/runtime/NEON/functions/NEGaussian3x3.h"
 #include "arm_compute/runtime/NEON/functions/NEGaussian5x5.h"
 #include "arm_compute/runtime/NEON/functions/NEGaussianPyramid.h"
diff --git a/arm_compute/runtime/NEON/functions/NEGather.h b/arm_compute/runtime/NEON/functions/NEGather.h
new file mode 100644
index 0000000000..f6de961d2a
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NEGather.h
@@ -0,0 +1,62 @@
+/*
+ * Copyright (c) 2019 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef __ARM_COMPUTE_NEGATHER_H__
+#define __ARM_COMPUTE_NEGATHER_H__
+
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/INESimpleFunctionNoBorder.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Basic function to run @ref NEGatherKernel */
+class NEGather : public INESimpleFunctionNoBorder
+{
+public:
+    /** Initialise the kernel's inputs and outputs
+     *
+     * @param[in]  input   Source tensor. Supported tensor rank: up to 4. Data type supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+     * @param[in]  indices Indices tensor. Supported tensor rank: up to 1. Must be one of the following type: U32/S32. Each value Must be in range [0, input.shape[@p axis])
+     * @param[out] output  Destination tensor. Data type supported: Same as @p input
+     * @param[in]  axis    (Optional) The axis in @p input to gather @p indices from. Defaults to 0
+     */
+    void configure(const ITensor *input, const ITensor *indices, ITensor *output, int axis = 0);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref NEGatherKernel
+     *
+     * @param[in] input   Source tensor info. Supported tensor rank: up to 4. Data type supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+     * @param[in] indices Indices tensor info. Supported tensor rank: up to 1. Must be one of the following types: U32/S32. Each value Must be in range [0, input.shape[@p axis])
+     * @param[in] output  Destination tensor info. Data type supported: Same as @p input
+     * @param[in] axis    (Optional) The axis in @p input to gather @p indices from. Defaults to 0
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, int axis);
+};
+
+} // namespace arm_compute
+
+#endif /* __ARM_COMPUTE_NEGATHER_H__ */
diff --git a/src/core/NEON/kernels/NEGatherKernel.cpp b/src/core/NEON/kernels/NEGatherKernel.cpp
new file mode 100644
index 0000000000..a05059d3b4
--- /dev/null
+++ b/src/core/NEON/kernels/NEGatherKernel.cpp
@@ -0,0 +1,203 @@
+/*
+ * Copyright (c) 2019 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/core/NEON/kernels/NEGatherKernel.h"
+
+#include "arm_compute/core/Coordinates.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/Validate.h"
+#include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
+namespace arm_compute
+{
+namespace
+{
+/** Validate the indices
+ *
+ * Validate that indices are not negative
+ *
+ * @param[in] indices Indices tensor info.
+ */
+template <typename U>
+void validate_indices(const ITensor *indices)
+{
+    for(size_t i = 0; i < indices->info()->tensor_shape()[0]; ++i)
+    {
+        ARM_COMPUTE_ERROR_ON(*(reinterpret_cast<U *>(indices->ptr_to_element(Coordinates(i)))) < 0);
+    }
+}
+
+} // namespace
+
+NEGatherKernel::NEGatherKernel()
+    : _input{}, _indices{}, _axis{}, _output{}, _func{}
+{
+}
+
+template <typename U>
+inline void NEGatherKernel::gather_0_axis(const Window &window, const ThreadInfo &info)
+{
+    ARM_COMPUTE_UNUSED(info);
+
+    // Validate that the indices are not negative
+    validate_indices<U>(_indices);
+
+    Iterator output_it(_output, window);
+    execute_window_loop(window, [&](const Coordinates & id)
+    {
+        Coordinates gather_id(id);
+
+        auto new_index = *(reinterpret_cast<U *>(_indices->ptr_to_element(Coordinates(id[0]))));
+        gather_id.set(0, new_index);
+
+        std::copy_n(_input->ptr_to_element(gather_id), _output->info()->element_size(), output_it.ptr());
+    },
+    output_it);
+}
+
+template <typename U>
+void NEGatherKernel::gather_n_axis(const Window &window, const ThreadInfo &info)
+{
+    ARM_COMPUTE_UNUSED(info);
+
+    // Validate that the indices are not negative
+    validate_indices<U>(_indices);
+
+    Window output_window{ window };
+    output_window.set(Window::DimX, Window::Dimension(0, 1, 1));
+
+    Iterator output_it(_output, output_window);
+    execute_window_loop(output_window, [&](const Coordinates & id)
+    {
+        Coordinates gather_id(id);
+
+        auto new_index = *(reinterpret_cast<U *>(_indices->ptr_to_element(Coordinates(id[_axis]))));
+        gather_id.set(_axis, new_index);
+
+        std::copy_n(_input->ptr_to_element(gather_id), _input->info()->dimension(0) * _output->info()->element_size(), output_it.ptr());
+    },
+    output_it);
+}
+
+void NEGatherKernel::configure(const ITensor *input, const ITensor *indices, ITensor *output, int axis)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, indices);
+    ARM_COMPUTE_ERROR_ON(indices->info()->num_dimensions() != 1);
+    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(indices, 1, DataType::U32, DataType::S32);
+    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QASYMM8, DataType::U16, DataType::S16, DataType::U32, DataType::S32, DataType::F16, DataType::F32);
+
+    _input   = input;
+    _indices = indices;
+    _output  = output;
+    _axis    = axis;
+
+    if(_axis < 0)
+    {
+        _axis += input->info()->num_dimensions();
+    }
+    ARM_COMPUTE_ERROR_ON(0 > _axis || _axis >= static_cast<int32_t>(input->info()->num_dimensions()));
+
+    if(0 == _axis)
+    {
+        switch(_indices->info()->data_type())
+        {
+            case DataType::U32:
+                _func = &NEGatherKernel::gather_0_axis<uint32_t>;
+                break;
+            case DataType::S32:
+                _func = &NEGatherKernel::gather_0_axis<int32_t>;
+                break;
+            default:
+                ARM_COMPUTE_ERROR("Not supported");
+                break;
+        }
+    }
+    else
+    {
+        switch(_indices->info()->data_type())
+        {
+            case DataType::U32:
+                _func = &NEGatherKernel::gather_n_axis<uint32_t>;
+                break;
+            case DataType::S32:
+                _func = &NEGatherKernel::gather_n_axis<int32_t>;
+                break;
+            default:
+                ARM_COMPUTE_ERROR("Not supported");
+                break;
+        }
+    }
+    // Output auto initialization if not yet initialized
+    TensorShape output_shape = arm_compute::misc::shape_calculator::compute_gather_shape(input->info()->tensor_shape(), indices->info()->tensor_shape(), _axis);
+    auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type());
+
+    // Create window
+    Window win = calculate_max_window(*output->info(), Steps());
+    output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
+
+    INEKernel::configure(win);
+}
+
+Status NEGatherKernel::validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, int axis)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, indices, output);
+    ARM_COMPUTE_RETURN_ERROR_ON(indices->num_dimensions() > 1);
+    ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
+
+    if(axis < 0)
+    {
+        axis += input->num_dimensions();
+    }
+
+    ARM_COMPUTE_RETURN_ERROR_ON(0 > axis || axis >= static_cast<int32_t>(input->num_dimensions()));
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QASYMM8,
+                                                         DataType::U16, DataType::S16,
+                                                         DataType::U32, DataType::S32, DataType::F16, DataType::F32);
+
+    if(output->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
+        TensorShape output_shape = arm_compute::misc::shape_calculator::compute_gather_shape(input->tensor_shape(), indices->tensor_shape(), axis);
+        ARM_COMPUTE_RETURN_ERROR_ON(output_shape.total_size() != output->tensor_shape().total_size());
+    }
+
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(indices, 1, DataType::U32, DataType::S32);
+
+    return Status{};
+}
+
+void NEGatherKernel::run(const Window &window, const ThreadInfo &info)
+{
+    ARM_COMPUTE_UNUSED(info);
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON(_func == nullptr);
+
+    (this->*_func)(window, info);
+}
+
+} // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NEGather.cpp b/src/runtime/NEON/functions/NEGather.cpp
new file mode 100644
index 0000000000..078bd5ab26
--- /dev/null
+++ b/src/runtime/NEON/functions/NEGather.cpp
@@ -0,0 +1,45 @@
+/*
+ * Copyright (c) 2019 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/runtime/NEON/functions/NEGather.h"
+
+#include "arm_compute/core/NEON/kernels/NEGatherKernel.h"
+#include "support/ToolchainSupport.h"
+
+#include <utility>
+
+namespace arm_compute
+{
+void NEGather::configure(const ITensor *input, const ITensor *indices, ITensor *output, int axis)
+{
+    auto k = arm_compute::support::cpp14::make_unique<NEGatherKernel>();
+    k->configure(input, indices, output, axis);
+    _kernel = std::move(k);
+}
+
+Status NEGather::validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, int axis)
+{
+    return NEGatherKernel::validate(input, indices, output, axis);
+}
+
+} // namespace arm_compute
diff --git a/tests/validation/NEON/Gather.cpp b/tests/validation/NEON/Gather.cpp
new file mode 100644
index 0000000000..2e6a3d4ed2
--- /dev/null
+++ b/tests/validation/NEON/Gather.cpp
@@ -0,0 +1,212 @@
+/*
+ * Copyright (c) 2019 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/functions/NEGather.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "arm_compute/runtime/TensorAllocator.h"
+
+#include "tests/NEON/Accessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/GatherDataset.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/GatherFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+TEST_SUITE(NEON)
+TEST_SUITE(Gather)
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(
+        framework::dataset::make("InputInfo", { TensorInfo(TensorShape(27U, 27U), 1, DataType::F16),
+                                                TensorInfo(TensorShape(27U, 27U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(27U, 27U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(27U, 27U), 1, DataType::F32),     // Invalid Indices data type
+                                                TensorInfo(TensorShape(27U, 27U), 1, DataType::F32),     // Invalid Indices dimensionality
+                                                TensorInfo(TensorShape(5U, 5U, 5U, 5U, 5U), 1, DataType::F32),    // Invalid Input dimensionality
+                                                TensorInfo(TensorShape(27U, 27U), 1, DataType::F16),     // Mismatching data type input/output
+                                                TensorInfo(TensorShape(27U, 27U), 1, DataType::F32),     // Invalid positive axis value
+                                                TensorInfo(TensorShape(27U, 27U), 1, DataType::F16),     // Invalid negative axis value
+        }),
+        framework::dataset::make("IndicesInfo", {
+                                                TensorInfo(TensorShape(10U), 1, DataType::U32),
+                                                TensorInfo(TensorShape(10U), 1, DataType::U32),
+                                                TensorInfo(TensorShape(10U), 1, DataType::U32),
+                                                TensorInfo(TensorShape(10U), 1, DataType::U8),
+                                                TensorInfo(TensorShape(10U, 10U), 1, DataType::U32),
+                                                TensorInfo(TensorShape(10U), 1, DataType::U32),
+                                                TensorInfo(TensorShape(10U), 1, DataType::U32),
+                                                TensorInfo(TensorShape(10U), 1, DataType::U32),
+                                                TensorInfo(TensorShape(10U), 1, DataType::U32),
+        })),
+        framework::dataset::make("OutputInfo", {
+                                                TensorInfo(TensorShape(10U, 27U), 1, DataType::F16),
+                                                TensorInfo(TensorShape(27U, 10U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(10U, 27U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(10U, 27U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(27U, 10U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(10U, 5U, 5U, 5U, 5U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(27U, 10U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(27U, 27U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(27U, 27U), 1, DataType::F16),
+        })),
+        framework::dataset::make("Axis", {
+                                            0,
+                                            1,
+                                            -2,
+                                            0,
+                                            1,
+                                            0,
+                                            1,
+                                            2,
+                                            -3,
+        })),
+        framework::dataset::make("Expected", { true, true, true, false, false, false, false, false, false })),
+        input_info, indices_info, output_info, axis, expected)
+{
+    const Status status = NEGather::validate(&input_info.clone()->set_is_resizable(true), &indices_info.clone()->set_is_resizable(true), &output_info.clone()->set_is_resizable(true), axis);
+    ARM_COMPUTE_EXPECT(bool(status) == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
+DATA_TEST_CASE(Configuration,
+               framework::DatasetMode::ALL,
+               combine(arm_compute::test::datasets::SmallGatherDataset(), framework::dataset::make("DataType", { DataType::F16, DataType::F32 })),
+               input_shape, indices_shape, axis, data_type)
+{
+    const uint32_t actual_axis = wrap_around(axis, static_cast<int>(input_shape.num_dimensions()));
+    Tensor         src         = create_tensor<Tensor>(input_shape, data_type);
+    Tensor         indices     = create_tensor<Tensor>(indices_shape, DataType::U32);
+    TensorShape    dst_shape   = arm_compute::misc::shape_calculator::compute_gather_shape(input_shape, indices_shape, actual_axis);
+    Tensor         dst         = create_tensor<Tensor>(dst_shape, data_type);
+
+    // Create and Configure function
+    NEGather gather;
+    gather.configure(&src, &indices, &dst, axis);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(dst.info()->tensor_shape());
+    validate(dst.info()->valid_region(), valid_region);
+}
+
+template <typename T>
+using NEGatherFixture = GatherFixture<Tensor, Accessor, NEGather, T>;
+
+TEST_SUITE(Float)
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       NEGatherFixture<half>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(datasets::SmallGatherDataset(), framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       NEGatherFixture<half>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(datasets::LargeGatherDataset(), framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // FP16
+
+TEST_SUITE(FP32)
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       NEGatherFixture<float>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(datasets::SmallGatherDataset(), framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       NEGatherFixture<float>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(datasets::LargeGatherDataset(), framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // FP32
+TEST_SUITE_END() // Float
+
+TEST_SUITE(U8)
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       NEGatherFixture<uint8_t>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(datasets::SmallGatherDataset(), framework::dataset::make("DataType", DataType::U8)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       NEGatherFixture<uint8_t>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(datasets::LargeGatherDataset(), framework::dataset::make("DataType", DataType::U8)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // U8
+
+TEST_SUITE(U16)
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       NEGatherFixture<uint16_t>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(datasets::SmallGatherDataset(), framework::dataset::make("DataType", DataType::U16)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       NEGatherFixture<uint16_t>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(datasets::LargeGatherDataset(), framework::dataset::make("DataType", DataType::U16)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // U16
+
+TEST_SUITE_END() // Gather
+TEST_SUITE_END() // NEON
+} // namespace validation
+} // namespace test
+} // namespace arm_compute