Port MaxUnpoolingLayer kernel and add KernelSelect vaidation test

Resolves COMPMID-4958
Change-Id: Ibed5155f2e3ece46635f6ea9617bf11cefc402b1
Signed-off-by: Dana Zlotnik <dana.zlotnik@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7028
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

20 files changed, 431 insertions, 304 deletions

diff --git a/Android.bp b/Android.bp
index 6add3f3c2d..ebea73329e 100644
--- a/Android.bp
+++ b/Android.bp
@@ -278,7 +278,6 @@ cc_library_static {
         "src/core/NEON/kernels/NEInstanceNormalizationLayerKernel.cpp",
         "src/core/NEON/kernels/NEL2NormalizeLayerKernel.cpp",
         "src/core/NEON/kernels/NELogicalKernel.cpp",
-        "src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.cpp",
         "src/core/NEON/kernels/NEMeanStdDevNormalizationKernel.cpp",
         "src/core/NEON/kernels/NENormalizationLayerKernel.cpp",
         "src/core/NEON/kernels/NEPadLayerKernel.cpp",
@@ -414,6 +413,7 @@ cc_library_static {
         "src/cpu/kernels/CpuGemmMatrixMultiplyKernel.cpp",
         "src/cpu/kernels/CpuGemmTranspose1xWKernel.cpp",
         "src/cpu/kernels/CpuIm2ColKernel.cpp",
+        "src/cpu/kernels/CpuMaxUnpoolingLayerKernel.cpp",
         "src/cpu/kernels/CpuMulKernel.cpp",
         "src/cpu/kernels/CpuPermuteKernel.cpp",
         "src/cpu/kernels/CpuPool2dKernel.cpp",
@@ -558,6 +558,7 @@ cc_library_static {
         "src/cpu/operators/CpuGemmDirectConv2d.cpp",
         "src/cpu/operators/CpuGemmLowpMatrixMultiplyCore.cpp",
         "src/cpu/operators/CpuGemmLowpOutputStage.cpp",
+        "src/cpu/operators/CpuMaxUnpooling.cpp",
         "src/cpu/operators/CpuMul.cpp",
         "src/cpu/operators/CpuPermute.cpp",
         "src/cpu/operators/CpuPool2d.cpp",
diff --git a/arm_compute/runtime/NEON/functions/NEMaxUnpoolingLayer.h b/arm_compute/runtime/NEON/functions/NEMaxUnpoolingLayer.h
index 41ea040457..2f77540e1e 100644
--- a/arm_compute/runtime/NEON/functions/NEMaxUnpoolingLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEMaxUnpoolingLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020-2021 Arm Limited.
+ * Copyright (c) 2020-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -33,12 +33,10 @@ namespace arm_compute
 class ITensor;
 class ITensorInfo;
 class NEFill;
-class NEMaxUnpoolingLayerKernel;
 
 /** Function to perform MaxUnpooling. This function calls the following kernels:
  *
  * -# @ref NEFill
- * -# @ref NEMaxUnpoolingLayerKernel
  */
 class NEMaxUnpoolingLayer : public IFunction
 {
@@ -94,8 +92,9 @@ public:
     void run() override;
 
 private:
-    std::unique_ptr<NEFill>                    _fill_func;
-    std::unique_ptr<NEMaxUnpoolingLayerKernel> _unpooling_layer_kernel;
+    std::unique_ptr<NEFill> _fill_func;
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 }
 #endif /* ARM_COMPUTE_NEMAXUNPOOLINGLAYER_H */
diff --git a/docs/user_guide/release_version_and_change_log.dox b/docs/user_guide/release_version_and_change_log.dox
index 97a531c470..56fd277ac4 100644
--- a/docs/user_guide/release_version_and_change_log.dox
+++ b/docs/user_guide/release_version_and_change_log.dox
@@ -612,7 +612,7 @@ v20.08 Public major release
  - New OpenCL kernels / functions:
    - @ref CLMaxUnpoolingLayerKernel
  - New Arm® Neon™ kernels / functions:
-   - @ref NEMaxUnpoolingLayerKernel
+   - NEMaxUnpoolingLayerKernel
  - New graph example:
    - graph_yolov3_output_detector
  - GEMMTuner improvements:
diff --git a/filelist.json b/filelist.json
index bba3d568a6..b672a832b1 100644
--- a/filelist.json
+++ b/filelist.json
@@ -1612,15 +1612,16 @@
         "deps": [ "Fill" ],
         "files": {
           "common": [
-            "src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.cpp",
-            "src/runtime/NEON/functions/NEMaxUnpoolingLayer.cpp"
+            "src/cpu/kernels/CpuMaxUnpoolingLayerKernel.cpp",
+            "src/runtime/NEON/functions/NEMaxUnpoolingLayer.cpp", 
+            "src/cpu/operators/CpuMaxUnpooling.cpp"
           ],
           "neon":{
-            "common":["src/cpu/kernels/maxunpool/generic/neon/impl.cpp"],
             "fp32":["src/cpu/kernels/maxunpool/generic/neon/fp32.cpp"],
             "fp16":["src/cpu/kernels/maxunpool/generic/neon/fp16.cpp"],
             "qasymm8":["src/cpu/kernels/maxunpool/generic/neon/qasymm8.cpp"],
-            "qasymm8_signed":["src/cpu/kernels/maxunpool/generic/neon/qasymm8_signed.cpp"]
+            "qasymm8_signed":[ "src/cpu/kernels/maxunpool/generic/neon/qasymm8_signed.cpp"],
+            "common":["src/cpu/kernels/maxunpool/generic/neon/impl.cpp"]
           }
         }
       },
diff --git a/src/core/NEON/NEKernels.h b/src/core/NEON/NEKernels.h
index cd01659c0d..12a1bb9a11 100644
--- a/src/core/NEON/NEKernels.h
+++ b/src/core/NEON/NEKernels.h
@@ -45,7 +45,6 @@
 #include "src/core/NEON/kernels/NEInstanceNormalizationLayerKernel.h"
 #include "src/core/NEON/kernels/NEL2NormalizeLayerKernel.h"
 #include "src/core/NEON/kernels/NELogicalKernel.h"
-#include "src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.h"
 #include "src/core/NEON/kernels/NEMeanStdDevNormalizationKernel.h"
 #include "src/core/NEON/kernels/NENormalizationLayerKernel.h"
 #include "src/core/NEON/kernels/NEPadLayerKernel.h"
diff --git a/src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.cpp b/src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.cpp
deleted file mode 100644
index 93da8a24c5..0000000000
--- a/src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.cpp
+++ /dev/null
@@ -1,171 +0,0 @@
-/*
- * Copyright (c) 2020-2022 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 "src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.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"
-#include "src/core/CPP/Validate.h"
-#include "src/core/common/Registrars.h"
-#include "src/core/helpers/AutoConfiguration.h"
-#include "src/core/helpers/WindowHelpers.h"
-#include "src/cpu/kernels/maxunpool/list.h"
-#include "support/ToolchainSupport.h"
-
-namespace arm_compute
-{
-using namespace misc::shape_calculator;
-
-namespace
-{
-struct MaxUnpoolingSelectorData
-{
-    DataType dt;
-};
-
-using MaxUnpoolingSelctorPtr = std::add_pointer<bool(const MaxUnpoolingSelectorData &data)>::type;
-using MaxUnpoolingUKernelPtr = std::add_pointer<void(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window)>::type;
-
-struct MaxUnpoolingKernel
-{
-    const char                  *name;
-    const MaxUnpoolingSelctorPtr is_selected;
-    MaxUnpoolingUKernelPtr       ukernel;
-};
-
-static const MaxUnpoolingKernel available_kernels[] =
-{
-    {
-        "fp32_neon_maxunpooling",
-        [](const MaxUnpoolingSelectorData & data) { return data.dt == DataType::F32; },
-        REGISTER_FP32_NEON(arm_compute::cpu::neon_fp32_maxunpooling)
-    },
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-    {
-        "fp16_neon_maxunpooling",
-        [](const MaxUnpoolingSelectorData & data) { return data.dt == DataType::F16; },
-        REGISTER_FP16_NEON(arm_compute::cpu::neon_fp16_maxunpooling)
-    },
-#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-#if defined(ARM_COMPUTE_ENABLE_NEON)
-    {
-        "qs8_neon_maxunpooling",
-        [](const MaxUnpoolingSelectorData & data) { return data.dt == DataType::QASYMM8; },
-        REGISTER_QASYMM8_NEON(arm_compute::cpu::neon_qs8_maxunpooling)
-    },
-    {
-        "qu8_neon_maxunpooling",
-        [](const MaxUnpoolingSelectorData & data) { return data.dt == DataType::QASYMM8_SIGNED; },
-        REGISTER_QASYMM8_SIGNED_NEON(arm_compute::cpu::neon_qu8_maxunpooling)
-    },
-#endif //defined(ARM_COMPUTE_ENABLE_NEON)
-};
-
-/** Micro-kernel selector
- *
- * @param[in] data Selection data passed to help pick the appropriate micro-kernel
- *
- * @return A matching micro-kernel else nullptr
- */
-const MaxUnpoolingKernel *get_implementation(const MaxUnpoolingSelectorData &data)
-{
-    for(const auto &uk : available_kernels)
-    {
-        if(uk.is_selected(data))
-        {
-            return &uk;
-        }
-    }
-    return nullptr;
-}
-
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const PoolingLayerInfo &pool_info, const ITensorInfo *indices)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output, indices);
-    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(indices, 1, DataType::U32);
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, indices);
-
-    int                 pool_stride_x   = 0;
-    int                 pool_stride_y   = 0;
-    PoolingType         pool_type       = pool_info.pool_type;
-    const PadStrideInfo pad_stride_info = pool_info.pad_stride_info;
-    std::tie(pool_stride_x, pool_stride_y) = pad_stride_info.stride();
-    const int    pool_size_x = pool_info.pool_size.width;
-    const int    pool_size_y = pool_info.pool_size.height;
-    const Size2D pool_size(pool_size_x, pool_size_y);
-
-    ARM_COMPUTE_RETURN_ERROR_ON_MSG(pool_type != PoolingType::MAX, "Pooling indices only supported for MAX pooling method");
-    ARM_COMPUTE_RETURN_ERROR_ON_MSG((pool_size != Size2D(2, 2)), "Pooling indices only supported for pool size 2x2");
-    if(output->total_size() != 0)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
-    }
-
-    return Status{};
-}
-} // namespace
-
-NEMaxUnpoolingLayerKernel::NEMaxUnpoolingLayerKernel()
-    : _input(nullptr), _output(nullptr), _indices(nullptr)
-{
-}
-
-void NEMaxUnpoolingLayerKernel::configure(const ITensor *input, const ITensor *indices, ITensor *output, const PoolingLayerInfo &pool_info)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), pool_info, indices->info()));
-
-    _input   = input;
-    _output  = output;
-    _indices = indices;
-
-    const TensorShape output_shape = compute_unpool_shape(*input->info(), pool_info);
-    auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
-
-    auto window = calculate_max_window(*input->info(), Steps());
-    INEKernel::configure(window);
-}
-
-Status NEMaxUnpoolingLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, const PoolingLayerInfo &pool_info)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, indices, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, pool_info, indices));
-    return Status{};
-}
-
-void NEMaxUnpoolingLayerKernel::run(const Window &window, const ThreadInfo &info)
-{
-    ARM_COMPUTE_UNUSED(info);
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
-    const auto *uk = get_implementation(MaxUnpoolingSelectorData{ _input->info()->data_type() });
-    ARM_COMPUTE_ERROR_ON(uk == nullptr || uk->ukernel == nullptr);
-
-    uk->ukernel(_input, _output, _indices, window);
-}
-} // namespace arm_compute
diff --git a/src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.h b/src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.h
deleted file mode 100644
index f7f9a31f6d..0000000000
--- a/src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.h
+++ /dev/null
@@ -1,96 +0,0 @@
-/*
- * Copyright (c) 2020-2022 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_NEMAXUNPOOLINGLAYERKERNEL_H
-#define ARM_COMPUTE_NEMAXUNPOOLINGLAYERKERNEL_H
-
-#include "src/core/NEON/INEKernel.h"
-
-namespace arm_compute
-{
-class ITensor;
-
-/** Interface for the pooling layer kernel */
-class NEMaxUnpoolingLayerKernel : public INEKernel
-{
-public:
-    const char *name() const override
-    {
-        return "NEMaxUnpoolingLayerKernel";
-    }
-    /** Default constructor */
-    NEMaxUnpoolingLayerKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    NEMaxUnpoolingLayerKernel(const NEMaxUnpoolingLayerKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    NEMaxUnpoolingLayerKernel &operator=(const NEMaxUnpoolingLayerKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    NEMaxUnpoolingLayerKernel(NEMaxUnpoolingLayerKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    NEMaxUnpoolingLayerKernel &operator=(NEMaxUnpoolingLayerKernel &&) = default;
-    /** Default destructor */
-    ~NEMaxUnpoolingLayerKernel() = default;
-    /** Set the input and output tensors.
-     *
-     * @note Output shape must be equal to the shape of the original input to pool.
-     *
-     * @param[in]  input     Source tensor. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
-     * @param[in]  indices   Tensor containing the offset to store the input elements in the output tensor.
-     *                       @ref NEPoolingLayer with indices should precede this function in order to
-     *                       properly reconstruct the output tensor.
-     *                       The tensor shape of this tensor has to be equal to the input tensor shape. Data type supported: U32.
-     * @param[out] output    Destination tensor. Data types supported: Same as @p input.
-     * @param[in]  pool_info Contains pooling operation information described in @ref PoolingLayerInfo.
-     */
-    void configure(const ITensor *input, const ITensor *indices, ITensor *output, const PoolingLayerInfo &pool_info);
-    /** Static function to check if given info will lead to a valid configuration of @ref NEMaxUnpoolingLayerKernel
-     *
-     * @param[in] input     Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
-     * @param[in] output    Destination tensor info. Data types supported: Same as @p input.
-     * @param[in] indices   Tensor info of the indices of the maximal values. Data type supported: U32.
-     * @param[in] pool_info Contains pooling operation information described in @ref PoolingLayerInfo.
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, const PoolingLayerInfo &pool_info);
-
-    // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
-
-private:
-    /** Function to perform 2x2 pooling and compute the pooling indices. The indices can be used for max unpool.
-     *
-     * @param[in] window_input Input region on which to execute the kernel.
-     */
-    template <typename T>
-    void unpooling2(const Window &window_input);
-
-    using UnpoolingFunction = void (NEMaxUnpoolingLayerKernel::*)(const Window &window);
-
-private:
-    const ITensor *_input;
-    ITensor       *_output;
-    const ITensor *_indices;
-};
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_NEMAXUNPOOLINGLAYERKERNEL_H */
diff --git a/src/cpu/kernels/CpuMaxUnpoolingLayerKernel.cpp b/src/cpu/kernels/CpuMaxUnpoolingLayerKernel.cpp
new file mode 100644
index 0000000000..604f22f6cc
--- /dev/null
+++ b/src/cpu/kernels/CpuMaxUnpoolingLayerKernel.cpp
@@ -0,0 +1,148 @@
+/*
+ * Copyright (c) 2020-2022 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 "src/cpu/kernels/CpuMaxUnpoolingLayerKernel.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"
+#include "src/core/CPP/Validate.h"
+#include "src/core/common/Registrars.h"
+#include "src/core/helpers/AutoConfiguration.h"
+#include "src/core/helpers/WindowHelpers.h"
+#include "src/cpu/kernels/maxunpool/list.h"
+#include "support/ToolchainSupport.h"
+
+namespace arm_compute
+{
+namespace cpu
+{
+namespace kernels
+{
+using namespace misc::shape_calculator;
+
+namespace
+{
+static const std::vector<CpuMaxUnpoolingLayerKernel::MaxUnpoolingKernel> available_kernels =
+{
+    {
+        "neon_fp32_maxunpooling",
+        [](const DataTypeISASelectorData & data) { return data.dt == DataType::F32; },
+        REGISTER_FP32_NEON(neon_fp32_maxunpooling)
+    },
+    {
+        "neon_fp16_maxunpooling",
+        [](const DataTypeISASelectorData & data) { return data.dt == DataType::F16 && data.isa.fp16; },
+        REGISTER_FP16_NEON(neon_fp16_maxunpooling)
+    },
+    {
+        "neon_qu8_maxunpooling",
+        [](const DataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8; },
+        REGISTER_QASYMM8_NEON(neon_qs8_maxunpooling)
+    },
+    {
+        "neon_qs8_maxunpooling",
+        [](const DataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8_SIGNED; },
+        REGISTER_QASYMM8_SIGNED_NEON(neon_qu8_maxunpooling)
+    },
+};
+
+Status validate_arguments(const ITensorInfo *src, const ITensorInfo *indices, const ITensorInfo *dst, const PoolingLayerInfo &pool_info)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, indices, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(src);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(indices, 1, DataType::U32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(src, indices);
+
+    int                 pool_stride_x   = 0;
+    int                 pool_stride_y   = 0;
+    PoolingType         pool_type       = pool_info.pool_type;
+    const PadStrideInfo pad_stride_info = pool_info.pad_stride_info;
+    std::tie(pool_stride_x, pool_stride_y) = pad_stride_info.stride();
+    const int    pool_size_x = pool_info.pool_size.width;
+    const int    pool_size_y = pool_info.pool_size.height;
+    const Size2D pool_size(pool_size_x, pool_size_y);
+
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(pool_type != PoolingType::MAX, "Pooling indices only supported for MAX pooling method");
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG((pool_size != Size2D(2, 2)), "Pooling indices only supported for pool size 2x2");
+    if(dst->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, dst);
+    }
+
+    return Status{};
+}
+} // namespace
+
+void CpuMaxUnpoolingLayerKernel::configure(const ITensorInfo *src, const ITensorInfo *indices, ITensorInfo *dst, const PoolingLayerInfo &pool_info)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst, indices);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, indices, dst, pool_info));
+    ARM_COMPUTE_UNUSED(indices);
+
+    const auto uk = CpuMaxUnpoolingLayerKernel::get_implementation(DataTypeISASelectorData{ src->data_type(), CPUInfo::get().get_isa() });
+    ARM_COMPUTE_ERROR_ON_NULLPTR(uk);
+    _run_method = uk->ukernel;
+
+    const TensorShape output_shape = compute_unpool_shape(*src, pool_info);
+    auto_init_if_empty(*dst, src->clone()->set_tensor_shape(output_shape));
+
+    auto window = calculate_max_window(*src, Steps());
+    ICpuKernel::configure(window);
+}
+
+Status CpuMaxUnpoolingLayerKernel::validate(const ITensorInfo *src, const ITensorInfo *indices, const ITensorInfo *dst, const PoolingLayerInfo &pool_info)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, indices, dst);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, indices, dst, pool_info));
+    return Status{};
+}
+
+void CpuMaxUnpoolingLayerKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
+{
+    ARM_COMPUTE_UNUSED(info);
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window);
+
+    const auto src     = tensors.get_const_tensor(TensorType::ACL_SRC_0);
+    const auto indices = tensors.get_const_tensor(TensorType::ACL_SRC_1);
+    const auto dst     = tensors.get_tensor(TensorType::ACL_DST);
+
+    _run_method(src, indices, dst, window);
+}
+
+const char *CpuMaxUnpoolingLayerKernel::name() const
+{
+    return "CpuMaxUnpoolingLayerKernel";
+}
+
+const std::vector<CpuMaxUnpoolingLayerKernel::MaxUnpoolingKernel> &CpuMaxUnpoolingLayerKernel::get_available_kernels()
+{
+    return available_kernels;
+}
+} // namespace kernels
+} // namespace cpu
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/cpu/kernels/CpuMaxUnpoolingLayerKernel.h b/src/cpu/kernels/CpuMaxUnpoolingLayerKernel.h
new file mode 100644
index 0000000000..d0c13471c8
--- /dev/null
+++ b/src/cpu/kernels/CpuMaxUnpoolingLayerKernel.h
@@ -0,0 +1,92 @@
+/*
+ * Copyright (c) 2020-2022 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_CPUMAXUNPOOLINGLAYERKERNEL_H
+#define ARM_COMPUTE_CPUMAXUNPOOLINGLAYERKERNEL_H
+
+#include "src/core/common/Macros.h"
+#include "src/cpu/ICpuKernel.h"
+
+namespace arm_compute
+{
+namespace cpu
+{
+namespace kernels
+{
+/** Interface for the pooling layer kernel */
+class CpuMaxUnpoolingLayerKernel : public ICpuKernel<CpuMaxUnpoolingLayerKernel>
+{
+private:
+    using MaxUnpoolingUKernelPtr = std::add_pointer<void(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window)>::type;
+
+public:
+    /** Default constructor */
+    CpuMaxUnpoolingLayerKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(CpuMaxUnpoolingLayerKernel);
+
+    /** Configure kernel for a given list of arguments
+     *
+     * @note Dst shape must be equal to the shape of the original src to pool.
+     *
+     * @param[in]  src       Source tensor to permute. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
+     * @param[in]  indices   Tensor containing the offset to store the src elements in the dst tensor.
+     *                       @ref CpuMaxUnpooling with indices should precede this function in order to
+     *                       properly reconstruct the output tensor.
+     *                       The tensor shape of this tensor has to be equal to the src tensor shape. Data type supported: U32.
+     * @param[out] dst       Destination tensor. Data types supported: Same as @p src
+     * @param[in]  pool_info Contains pooling operation information described in @ref PoolingLayerInfo.
+     */
+    void configure(const ITensorInfo *src, const ITensorInfo *indices, ITensorInfo *dst, const PoolingLayerInfo &pool_info);
+    /** Static function to check if given info will lead to a valid configuration of @ref CpuMaxUnpoolingLayerKernel
+     *
+     * @param[in]  src       Source tensor to permute. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
+     * @param[in]  indices   Tensor info of the indices of the maximal values. Data type supported: U32.
+     * @param[out] dst       Destination tensor. Data types supported: Same as @p src
+     * @param[in]  pool_info Contains pooling operation information described in @ref PoolingLayerInfo.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *indices, const ITensorInfo *dst, const PoolingLayerInfo &pool_info);
+
+    // Inherited methods overridden:
+    void run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info) override;
+
+    struct MaxUnpoolingKernel
+    {
+        const char                  *name;
+        const DataTypeISASelectorPtr is_selected;
+        MaxUnpoolingUKernelPtr       ukernel;
+    };
+
+    static const std::vector<MaxUnpoolingKernel> &get_available_kernels();
+
+    const char *name() const override;
+
+private:
+    MaxUnpoolingUKernelPtr _run_method{ nullptr };
+};
+
+} // namespace kernels
+} // namespace cpu
+} // namespace arm_compute
+#endif /*ARM_COMPUTE_CPUMAXUNPOOLINGLAYERKERNEL_H */
diff --git a/src/cpu/kernels/maxunpool/generic/neon/fp16.cpp b/src/cpu/kernels/maxunpool/generic/neon/fp16.cpp
index d43503aa2f..e81ff92311 100644
--- a/src/cpu/kernels/maxunpool/generic/neon/fp16.cpp
+++ b/src/cpu/kernels/maxunpool/generic/neon/fp16.cpp
@@ -27,9 +27,9 @@ namespace arm_compute
 {
 namespace cpu
 {
-void neon_fp16_maxunpooling(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window)
+void neon_fp16_maxunpooling(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window)
 {
-    return max_unpooling<float16_t>(input, output, indices, window);
+    return max_unpooling<float16_t>(input, indices, output, window);
 }
 } // namespace cpu
 } // namespace arm_compute
diff --git a/src/cpu/kernels/maxunpool/generic/neon/fp32.cpp b/src/cpu/kernels/maxunpool/generic/neon/fp32.cpp
index 2f96e86695..ba0d7851a9 100644
--- a/src/cpu/kernels/maxunpool/generic/neon/fp32.cpp
+++ b/src/cpu/kernels/maxunpool/generic/neon/fp32.cpp
@@ -26,9 +26,9 @@ namespace arm_compute
 {
 namespace cpu
 {
-void neon_fp32_maxunpooling(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window)
+void neon_fp32_maxunpooling(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window)
 {
-    return max_unpooling<float>(input, output, indices, window);
+    return max_unpooling<float>(input, indices, output, window);
 }
 } // namespace cpu
 } // namespace arm_compute
diff --git a/src/cpu/kernels/maxunpool/generic/neon/impl.cpp b/src/cpu/kernels/maxunpool/generic/neon/impl.cpp
index 8bbc8d128f..77e3b8594a 100644
--- a/src/cpu/kernels/maxunpool/generic/neon/impl.cpp
+++ b/src/cpu/kernels/maxunpool/generic/neon/impl.cpp
@@ -29,7 +29,7 @@ class Window;
 namespace cpu
 {
 template <typename T>
-void max_unpooling(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window)
+void max_unpooling(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window)
 {
     Iterator  input_itr(input, window);
     Iterator  indices_itr(indices, window);
@@ -43,12 +43,12 @@ void max_unpooling(const ITensor *input, ITensor *output, const ITensor *indices
     },
     input_itr, indices_itr);
 }
-template void max_unpooling<float>(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window);
-template void max_unpooling<int8_t>(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window);
-template void max_unpooling<uint8_t>(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window);
+template void max_unpooling<float>(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window);
+template void max_unpooling<int8_t>(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window);
+template void max_unpooling<uint8_t>(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window);
 
 #if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS)
-template void max_unpooling<float16_t>(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window);
+template void max_unpooling<float16_t>(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window);
 #endif //defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS)
 } // namespace cpu
 } // namespace arm_compute
diff --git a/src/cpu/kernels/maxunpool/generic/neon/impl.h b/src/cpu/kernels/maxunpool/generic/neon/impl.h
index 6a14c66b33..3fea9cfcf3 100644
--- a/src/cpu/kernels/maxunpool/generic/neon/impl.h
+++ b/src/cpu/kernels/maxunpool/generic/neon/impl.h
@@ -33,7 +33,7 @@ class Window;
 namespace cpu
 {
 template <typename T>
-void max_unpooling(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window);
+void max_unpooling(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window);
 } // namespace cpu
 } // namespace arm_compute
 #endif //define SRC_CORE_SVE_KERNELS_MAXUNPOOLING_IMPL_H
diff --git a/src/cpu/kernels/maxunpool/generic/neon/qasymm8.cpp b/src/cpu/kernels/maxunpool/generic/neon/qasymm8.cpp
index b6d0f48fda..53e601bba6 100644
--- a/src/cpu/kernels/maxunpool/generic/neon/qasymm8.cpp
+++ b/src/cpu/kernels/maxunpool/generic/neon/qasymm8.cpp
@@ -26,9 +26,9 @@ namespace arm_compute
 {
 namespace cpu
 {
-void neon_qs8_maxunpooling(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window)
+void neon_qs8_maxunpooling(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window)
 {
-    return max_unpooling<int8_t>(input, output, indices, window);
+    return max_unpooling<int8_t>(input, indices, output, window);
 }
 } // namespace cpu
 } // namespace arm_compute
diff --git a/src/cpu/kernels/maxunpool/generic/neon/qasymm8_signed.cpp b/src/cpu/kernels/maxunpool/generic/neon/qasymm8_signed.cpp
index 79f3013805..a3c346fba7 100644
--- a/src/cpu/kernels/maxunpool/generic/neon/qasymm8_signed.cpp
+++ b/src/cpu/kernels/maxunpool/generic/neon/qasymm8_signed.cpp
@@ -26,9 +26,9 @@ namespace arm_compute
 {
 namespace cpu
 {
-void neon_qu8_maxunpooling(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window)
+void neon_qu8_maxunpooling(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window)
 {
-    return max_unpooling<uint8_t>(input, output, indices, window);
+    return max_unpooling<uint8_t>(input, indices, output, window);
 }
 } // namespace cpu
 } // namespace arm_compute
diff --git a/src/cpu/kernels/maxunpool/list.h b/src/cpu/kernels/maxunpool/list.h
index 0f9bb499d3..2c4fe940d9 100644
--- a/src/cpu/kernels/maxunpool/list.h
+++ b/src/cpu/kernels/maxunpool/list.h
@@ -28,7 +28,7 @@ namespace arm_compute
 namespace cpu
 {
 #define DECLARE_MAXUNPOOL_KERNEL(func_name) \
-    void func_name(const ITensor *input, ITensor *output, const ITensor *indices, const Window &window)
+    void func_name(const ITensor *input, const ITensor *indices, ITensor *output, const Window &window)
 DECLARE_MAXUNPOOL_KERNEL(neon_fp32_maxunpooling);
 DECLARE_MAXUNPOOL_KERNEL(neon_fp16_maxunpooling);
 DECLARE_MAXUNPOOL_KERNEL(neon_qs8_maxunpooling);
diff --git a/src/cpu/operators/CpuMaxUnpooling.cpp b/src/cpu/operators/CpuMaxUnpooling.cpp
new file mode 100644
index 0000000000..24e9fd6d46
--- /dev/null
+++ b/src/cpu/operators/CpuMaxUnpooling.cpp
@@ -0,0 +1,45 @@
+/*
+ * Copyright (c) 2018-2022 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 "src/cpu/operators/CpuMaxUnpooling.h"
+#include "src/common/utils/Log.h"
+#include "src/cpu/kernels/CpuMaxUnpoolingLayerKernel.h"
+
+namespace arm_compute
+{
+namespace cpu
+{
+void CpuMaxUnpooling::configure(const ITensorInfo *src, const ITensorInfo *indices, ITensorInfo *dst, const PoolingLayerInfo &pool_info)
+{
+    ARM_COMPUTE_LOG_PARAMS(src, indices, dst, pool_info);
+    auto k = std::make_unique<kernels::CpuMaxUnpoolingLayerKernel>();
+    k->configure(src, indices, dst, pool_info);
+    _kernel = std::move(k);
+}
+
+Status CpuMaxUnpooling::validate(const ITensorInfo *src, const ITensorInfo *indices, const ITensorInfo *dst, const PoolingLayerInfo &pool_info)
+{
+    return kernels::CpuMaxUnpoolingLayerKernel::validate(src, indices, dst, pool_info);
+}
+} // namesapce cpu
+} // namespace arm_compute
diff --git a/src/cpu/operators/CpuMaxUnpooling.h b/src/cpu/operators/CpuMaxUnpooling.h
new file mode 100644
index 0000000000..aa1f1072a5
--- /dev/null
+++ b/src/cpu/operators/CpuMaxUnpooling.h
@@ -0,0 +1,58 @@
+/*
+ * Copyright (c) 2021-2022 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_CPU_MAXUNPOOLING_H
+#define ARM_COMPUTE_CPU_MAXUNPOOLING_H
+
+#include "src/cpu/ICpuOperator.h"
+
+namespace arm_compute
+{
+namespace cpu
+{
+/** Basic function to run @ref kernels::CpuMaxUnpoolingLayerKernel */
+class CpuMaxUnpooling : public ICpuOperator
+{
+public:
+    /** Configure operator for a given list of arguments
+     *
+     * @param[in]  src       Source tensor to permute. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
+     * @param[in]  indices   Tensor containing the offset to store the src elements in the dst tensor.
+     *                       @ref CpuMaxUnpooling with indices should precede this function in order to
+     *                       properly reconstruct the output tensor.
+     *                       The tensor shape of this tensor has to be equal to the src tensor shape. Data type supported: U32.
+     * @param[out] dst       Destination tensor. Data types supported: Same as @p src
+     * @param[in]  pool_info Contains pooling operation information described in @ref PoolingLayerInfo.
+     */
+    void configure(const ITensorInfo *src, const ITensorInfo *indices, ITensorInfo *dst, const PoolingLayerInfo &pool_info);
+    /** Static function to check if given info will lead to a valid configuration
+     *
+     * Similar to @ref CpuMaxUnpooling::configure()
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *indices, const ITensorInfo *dst, const PoolingLayerInfo &pool_info);
+};
+} // namespace cpu
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CPU_MAXUNPOOLING_H  */
diff --git a/src/runtime/NEON/functions/NEMaxUnpoolingLayer.cpp b/src/runtime/NEON/functions/NEMaxUnpoolingLayer.cpp
index 39f717545b..97ddaea41d 100644
--- a/src/runtime/NEON/functions/NEMaxUnpoolingLayer.cpp
+++ b/src/runtime/NEON/functions/NEMaxUnpoolingLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020-2021 Arm Limited.
+ * Copyright (c) 2020-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -28,14 +28,23 @@
 #include "arm_compute/runtime/NEON/NEScheduler.h"
 #include "arm_compute/runtime/NEON/functions/NEFill.h"
 #include "src/common/utils/Log.h"
-#include "src/core/NEON/kernels/NEMaxUnpoolingLayerKernel.h"
+#include "src/cpu/kernels/CpuMaxUnpoolingLayerKernel.h"
+#include "src/cpu/operators/CpuMaxUnpooling.h"
 
 namespace arm_compute
 {
+struct NEMaxUnpoolingLayer::Impl
+{
+    const ITensor                        *src{ nullptr };
+    const ITensor                        *indices{ nullptr };
+    ITensor                              *dst{ nullptr };
+    std::unique_ptr<cpu::CpuMaxUnpooling> op{ nullptr };
+};
+
 NEMaxUnpoolingLayer::~NEMaxUnpoolingLayer() = default;
 
 NEMaxUnpoolingLayer::NEMaxUnpoolingLayer()
-    : _fill_func(), _unpooling_layer_kernel()
+    : _fill_func(), _impl()
 {
 }
 
@@ -44,20 +53,32 @@ void NEMaxUnpoolingLayer::configure(ITensor *input, ITensor *indices, ITensor *o
     ARM_COMPUTE_LOG_PARAMS(input, indices, output, pool_info);
 
     const PixelValue zero_value(0.f);
-    _fill_func              = std::make_unique<NEFill>();
-    _unpooling_layer_kernel = std::make_unique<NEMaxUnpoolingLayerKernel>();
+    _fill_func     = std::make_unique<NEFill>();
+    _impl          = std::make_unique<Impl>();
+    _impl->src     = input;
+    _impl->indices = indices;
+    _impl->dst     = output;
+
+    _impl->op = std::make_unique<cpu::CpuMaxUnpooling>();
     _fill_func->configure(output, zero_value);
-    _unpooling_layer_kernel->configure(input, indices, output, pool_info);
+    _impl->op->configure(input->info(), indices->info(), output->info(), pool_info);
 }
 
 Status NEMaxUnpoolingLayer::validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, const PoolingLayerInfo &pool_info)
 {
-    return NEMaxUnpoolingLayerKernel::validate(input, indices, output, pool_info);
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output, indices);
+    ARM_COMPUTE_RETURN_ON_ERROR(cpu::CpuMaxUnpooling::validate(input, indices, output, pool_info));
+    return Status{};
 }
 
 void NEMaxUnpoolingLayer::run()
 {
+    ITensorPack pack;
+    pack.add_tensor(TensorType::ACL_SRC_0, _impl->src);
+    pack.add_tensor(TensorType::ACL_SRC_1, _impl->indices);
+    pack.add_tensor(TensorType::ACL_DST, _impl->dst);
+
     _fill_func->run();
-    NEScheduler::get().schedule(_unpooling_layer_kernel.get(), Window::DimY);
+    _impl->op->run(pack);
 }
 } /* namespace arm_compute */
diff --git a/tests/validation/NEON/MaxUnpoolingLayer.cpp b/tests/validation/NEON/MaxUnpoolingLayer.cpp
index 65838f5b30..8904993afe 100644
--- a/tests/validation/NEON/MaxUnpoolingLayer.cpp
+++ b/tests/validation/NEON/MaxUnpoolingLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020-2021 Arm Limited.
+ * Copyright (c) 2020-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,6 +26,7 @@
 #include "arm_compute/runtime/NEON/functions/NEPoolingLayer.h"
 #include "arm_compute/runtime/Tensor.h"
 #include "arm_compute/runtime/TensorAllocator.h"
+#include "src/cpu/kernels/CpuMaxUnpoolingLayerKernel.h"
 #include "tests/NEON/Accessor.h"
 #include "tests/datasets/ShapeDatasets.h"
 #include "tests/framework/Asserts.h"
@@ -33,7 +34,6 @@
 #include "tests/framework/datasets/Datasets.h"
 #include "tests/validation/Validation.h"
 #include "tests/validation/fixtures/MaxUnpoolingLayerFixture.h"
-
 namespace arm_compute
 {
 namespace test
@@ -74,7 +74,37 @@ FIXTURE_DATA_TEST_CASE(MaxUnpooling, NEMaxUnpoolingLayerFixture<half>, framework
 }
 TEST_SUITE_END() // FP16
 #endif           /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+
 TEST_SUITE_END() // Float
+
+TEST_SUITE(KernelSelection)
+
+DATA_TEST_CASE(KernelSelection, framework::DatasetMode::ALL,
+               combine(framework::dataset::make("CpuExt", std::string("NEON")),
+                       framework::dataset::make("DataType", { DataType::F32,
+                                                              DataType::F16,
+                                                              DataType::QASYMM8,
+                                                              DataType::QASYMM8_SIGNED
+                                                            })),
+               cpu_ext, data_type)
+{
+    using namespace cpu::kernels;
+
+    cpuinfo::CpuIsaInfo cpu_isa{};
+    cpu_isa.neon = (cpu_ext == "NEON");
+    cpu_isa.sve  = (cpu_ext == "SVE");
+    cpu_isa.fp16 = (data_type == DataType::F16);
+
+    const auto *selected_impl = CpuMaxUnpoolingLayerKernel::get_implementation(DataTypeISASelectorData{ data_type, cpu_isa }, cpu::KernelSelectionType::Preferred);
+
+    ARM_COMPUTE_ERROR_ON_NULLPTR(selected_impl);
+
+    std::string expected = lower_string(cpu_ext) + "_" + cpu_impl_dt(data_type) + "_maxunpooling";
+    std::string actual   = selected_impl->name;
+
+    ARM_COMPUTE_EXPECT_EQUAL(expected, actual, framework::LogLevel::ERRORS);
+}
+TEST_SUITE_END() // KernelSelection
 TEST_SUITE_END() // PoolingLayer
 TEST_SUITE_END() // Neon
 } // namespace validation