Implementation of ClPooling3d

 - For NDHWC layout
 - For F16 and F32 data types
 - Mixed Precision stil not supported

Resolves: COMPMID-4670
Signed-off-by: ramy.elgammal@arm.com
Change-Id: I0e14a13e4625569e8e5ee67e6033bd1efe0da469
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7262
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

6 files changed, 605 insertions, 17 deletions

diff --git a/tests/datasets/Pooling3dLayerDataset.h b/tests/datasets/Pooling3dLayerDataset.h
new file mode 100644
index 0000000000..cfe970e8be
--- /dev/null
+++ b/tests/datasets/Pooling3dLayerDataset.h
@@ -0,0 +1,121 @@
+/*
+ * Copyright (c) 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_TEST_POOLING_3D_LAYER_DATASET
+#define ARM_COMPUTE_TEST_POOLING_3D_LAYER_DATASET
+
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+#include "utils/TypePrinter.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace datasets
+{
+class Pooling3dLayerDataset
+{
+public:
+    using type = std::tuple<TensorShape, Pooling3dLayerInfo>;
+
+    struct iterator
+    {
+        iterator(std::vector<TensorShape>::const_iterator        src_it,
+                 std::vector<Pooling3dLayerInfo>::const_iterator infos_it)
+            : _src_it{ std::move(src_it) },
+              _infos_it{ std::move(infos_it) }
+        {
+        }
+
+        std::string description() const
+        {
+            std::stringstream description;
+            description << "In=" << *_src_it << ":";
+            description << "Info=" << *_infos_it << ":";
+            return description.str();
+        }
+
+        Pooling3dLayerDataset::type operator*() const
+        {
+            return std::make_tuple(*_src_it, *_infos_it);
+        }
+
+        iterator &operator++()
+        {
+            ++_src_it;
+            ++_infos_it;
+
+            return *this;
+        }
+
+    private:
+        std::vector<TensorShape>::const_iterator        _src_it;
+        std::vector<Pooling3dLayerInfo>::const_iterator _infos_it;
+    };
+
+    iterator begin() const
+    {
+        return iterator(_src_shapes.begin(), _infos.begin());
+    }
+
+    int size() const
+    {
+        return std::min(_src_shapes.size(), _infos.size());
+    }
+
+    void add_config(TensorShape src, Pooling3dLayerInfo info)
+    {
+        _src_shapes.emplace_back(std::move(src));
+        _infos.emplace_back(std::move(info));
+    }
+
+protected:
+    Pooling3dLayerDataset()                         = default;
+    Pooling3dLayerDataset(Pooling3dLayerDataset &&) = default;
+
+private:
+    std::vector<TensorShape>        _src_shapes{};
+    std::vector<Pooling3dLayerInfo> _infos{};
+};
+
+// Special pooling dataset
+class Pooling3dLayerDatasetSpecial final : public Pooling3dLayerDataset
+{
+public:
+    Pooling3dLayerDatasetSpecial()
+    {
+        // Special cases
+        add_config(TensorShape(2U, 3U, 4U, 2U, 4U), Pooling3dLayerInfo(PoolingType::AVG, /*pool size*/ Size3D(2, 2, 1), /*pool strides*/ Size3D(3, 3, 1), /*pool padding*/ Padding3D(0, 0, 0), true));
+        add_config(TensorShape(20U, 22U, 10U, 2U), Pooling3dLayerInfo(PoolingType::AVG, Size3D(100, 100, 100), Size3D(5, 5, 5), Padding3D(50, 50, 50), true));
+        add_config(TensorShape(10U, 20U, 32U, 3U, 2U), Pooling3dLayerInfo(PoolingType::MAX, /*pool size*/ 3, /*pool strides*/ Size3D(2, 2, 2), Padding3D(1, 1, 1, 1, 1, 1), false, false,
+                                                                    DimensionRoundingType::FLOOR));
+        add_config(TensorShape(14U, 10U, 10U, 3U, 5U), Pooling3dLayerInfo(PoolingType::AVG,  Size3D(3, 3, 3), /*pool strides*/ Size3D(3, 3, 3), Padding3D(2, 1, 2), true, false, DimensionRoundingType::CEIL));
+        add_config(TensorShape(14U, 10U, 10U, 2U, 4U), Pooling3dLayerInfo(PoolingType::AVG,  Size3D(3, 3, 3), /*pool strides*/ Size3D(3, 3, 3), Padding3D(2, 1, 2), false, false, DimensionRoundingType::CEIL));
+        add_config(TensorShape(15U, 13U, 13U, 3U, 5U), Pooling3dLayerInfo(PoolingType::AVG,  Size3D(4, 4, 4), /*pool strides*/ Size3D(2, 2, 2), Padding3D(2, 2, 2), true, false, DimensionRoundingType::CEIL));
+    }
+};
+} // namespace datasets
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_POOLING_3D_LAYER_DATASET */
diff --git a/tests/datasets/ShapeDatasets.h b/tests/datasets/ShapeDatasets.h
index e21589946b..31c9b9e051 100644
--- a/tests/datasets/ShapeDatasets.h
+++ b/tests/datasets/ShapeDatasets.h
@@ -540,6 +540,21 @@ public:
     }
 };
 
+/** Data set containing small 5D tensor shapes. */
+class Small5dShapes final : public ShapeDataset
+{
+public:
+    Small5dShapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 5U, 5U, 7U, 4U, 3U },
+                     TensorShape{ 5U, 5U, 4U, 13U, 2U },
+                     TensorShape{ 5U, 5U, 3U, 5U , 2U},
+    })
+    {
+    }
+};
+
 /** Data set containing large 5x5 tensor shapes. */
 class Large5x5Shapes final : public ShapeDataset
 {
@@ -553,6 +568,19 @@ public:
     }
 };
 
+/** Data set containing large 5D tensor shapes. */
+class Large5dShapes final : public ShapeDataset
+{
+public:
+    Large5dShapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 30U, 40U, 30U, 32U, 3U }
+    })
+    {
+    }
+};
+
 /** Data set containing small 5x1 tensor shapes. */
 class Small5x1Shapes final : public ShapeDataset
 {
diff --git a/tests/validation/CL/Pooling3dLayer.cpp b/tests/validation/CL/Pooling3dLayer.cpp
new file mode 100644
index 0000000000..5c80351da2
--- /dev/null
+++ b/tests/validation/CL/Pooling3dLayer.cpp
@@ -0,0 +1,283 @@
+/*
+ * Copyright (c) 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 "arm_compute/core/TensorShape.h"
+#include "tests/framework/datasets/Datasets.h"
+
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/CLTensorAllocator.h"
+#include "arm_compute/runtime/CL/functions/CLPooling3dLayer.h"
+#include "tests/CL/CLAccessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/Pooling3dLayerDataset.h"
+#include "tests/datasets/PoolingTypesDataset.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/Pooling3dLayerFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+/** Input data sets for floating-point data types */
+const auto Pooling3dLayerDatasetFP = combine(combine(combine(combine(datasets::PoolingTypes(), framework::dataset::make("PoolingSize", { Size3D(2, 3, 2) })),
+                                                             framework::dataset::make("Stride", { Size3D(1, 1, 1), Size3D(2, 1, 1), Size3D(1, 2, 1), Size3D(2, 2, 1) })),
+                                                     framework::dataset::make("Padding", { Padding3D(0, 1, 0), Padding3D(1, 1, 1) })),
+                                             framework::dataset::make("ExcludePadding", { true, false }));
+
+const auto Pooling3dLayerDatasetFPSmall = combine(combine(combine(combine(datasets::PoolingTypes(), framework::dataset::make("PoolingSize", { Size3D(2, 2, 2), Size3D(3, 3, 3) })),
+                                                                  framework::dataset::make("Stride", { Size3D(2, 2, 2), Size3D(2, 1, 1) })),
+                                                          framework::dataset::make("Padding", { Padding3D(0, 0, 0), Padding3D(1, 1, 1), Padding3D(1, 0, 0) })),
+                                                  framework::dataset::make("ExcludePadding", { true, false }));
+
+using ShapeDataset = framework::dataset::ContainerDataset<std::vector<TensorShape>>;
+
+constexpr AbsoluteTolerance<float> tolerance_f32(0.001f); /**< Tolerance value for comparing reference's output against implementation's output for 32-bit floating-point type */
+constexpr AbsoluteTolerance<float> tolerance_f16(0.1f);  /**< Tolerance value for comparing reference's output against implementation's output for 16-bit floating-point type */
+
+} // namespace
+
+TEST_SUITE(CL)
+TEST_SUITE(Pooling3dLayer)
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
+               framework::dataset::make("InputInfo", { TensorInfo(TensorShape(2U, 27U, 13U, 4U, 3U), 1, DataType::F32, DataLayout::NDHWC), // Mismatching data type
+                                                       TensorInfo(TensorShape(2U, 27U, 13U, 4U, 2U), 1, DataType::F32, DataLayout::NDHWC), // Invalid pad/size combination
+                                                       TensorInfo(TensorShape(2U, 27U, 13U, 4U, 2U), 1, DataType::F32, DataLayout::NDHWC), // Invalid pad/size combination
+                                                       TensorInfo(TensorShape(2U, 27U, 13U, 4U, 3U), 1, DataType::F32, DataLayout::NDHWC), // Invalid output shape
+                                                       TensorInfo(TensorShape(5U, 13U, 15U, 2U, 3U), 1, DataType::F32, DataLayout::NDHWC), // Global Pooling
+                                                       TensorInfo(TensorShape(13U,13U, 5U, 1U, 2U), 1, DataType::F32, DataLayout::NDHWC),  // Invalid output Global Pooling
+                                                       TensorInfo(TensorShape(5U, 13U, 13U, 4U, 4U), 1, DataType::F32, DataLayout::NDHWC), // Invalid data type
+                                                       TensorInfo(TensorShape(5U, 13U, 13U, 4U, 4U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(5U, 13U, 13U, 5U, 4U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(1U, 16U,  1U, 3U, 4U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(5U, 13U, 13U, 4U, 3U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(5U, 13U, 13U, 4U, 2U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(5U, 13U, 13U, 4U, 3U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(5U, 13U, 13U, 4U, 3U), 1, DataType::F32, DataLayout::NDHWC),
+                                                     }),
+               framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(2U, 25U, 11U, 3U, 3U), 1, DataType::F16, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(2U, 30U, 11U, 3U, 2U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(2U, 25U, 16U, 3U, 2U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(2U, 27U, 13U, 3U, 3U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(5U,  1U,  1U, 1U, 3U), 1, DataType::F32, DataLayout::NDHWC), // Global pooling applied
+                                                       TensorInfo(TensorShape(5U,  2U,  2U, 2U, 2U), 1, DataType::F32, DataLayout::NDHWC), // Invalid output Global Pooling
+                                                       TensorInfo(TensorShape(5U, 12U, 12U, 3U, 4U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(5U, 12U, 12U, 3U, 4U), 1, DataType::QASYMM8, DataLayout::NDHWC), // Invalid data type
+                                                       TensorInfo(TensorShape(5U,  1U, 1U, 1U, 4U), 1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(1U, 15U, 1U, 2U, 4U), 1, DataType::F32, DataLayout::NDHWC), // Output width larger than input
+                                                       TensorInfo(TensorShape(5U, 6U, 6U, 2U, 3U),  1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(5U, 6U, 6U, 2U, 2U),  1, DataType::F32, DataLayout::NDHWC), 
+                                                       TensorInfo(TensorShape(5U, 6U, 6U, 2U, 3U),  1, DataType::F32, DataLayout::NDHWC),
+                                                       TensorInfo(TensorShape(5U, 6U, 6U, 2U, 3U),  1, DataType::F32, DataLayout::NDHWC),
+                                                     })),
+               framework::dataset::make("PoolInfo",  { Pooling3dLayerInfo(PoolingType::AVG, 3, Size3D(1, 1, 1), Padding3D(0, 0, 0)),
+                                                       Pooling3dLayerInfo(PoolingType::AVG, 2, Size3D(1, 1, 1), Padding3D(2, 0, 0)),
+                                                       Pooling3dLayerInfo(PoolingType::AVG, 2, Size3D(1, 1, 1), Padding3D(0, 0, 0)),
+                                                       Pooling3dLayerInfo(PoolingType::L2,  3, Size3D(1, 1, 1), Padding3D(0, 0, 0)),
+                                                       Pooling3dLayerInfo(PoolingType::AVG),
+                                                       Pooling3dLayerInfo(PoolingType::MAX),
+                                                       Pooling3dLayerInfo(PoolingType::AVG, 2, Size3D(), Padding3D(), false),
+                                                       Pooling3dLayerInfo(PoolingType::AVG, 2, Size3D(1U, 1U, 1U), Padding3D(), false),
+                                                       Pooling3dLayerInfo(PoolingType::AVG),
+                                                       Pooling3dLayerInfo(PoolingType::MAX, 2, Size3D(1, 1, 2), Padding3D(0, 0, 0), false),
+                                                       Pooling3dLayerInfo(PoolingType::AVG, 2, Size3D(2U, 2U, 2U), Padding3D(), false),
+                                                       Pooling3dLayerInfo(PoolingType::AVG, 1, Size3D(2U, 2U, 2U), Padding3D(2, 2, 2), true), // Pool size is smaller than the padding size with padding excluded
+                                                       Pooling3dLayerInfo(PoolingType::AVG, 1, Size3D(2U, 2U, 2U), Padding3D(2, 2, 2), false), // Pool size is smaller than the padding size with padding included
+                                                       Pooling3dLayerInfo(PoolingType::AVG, 3, Size3D(2U, 2U, 2U), Padding3D(2,1,2,2,1,2), false, false, DimensionRoundingType::CEIL), // CEIL with asymmetric Padding
+                                                      })),
+               framework::dataset::make("Expected", { false, false, false, false, true, false, false, false, true , false, true, false, false, false})),
+               input_info, output_info, pool_info, expected)
+{
+    ARM_COMPUTE_EXPECT(bool(CLPooling3dLayer::validate(&input_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false), pool_info)) == expected, framework::LogLevel::ERRORS);
+}
+
+
+template <typename T>
+using CLPooling3dLayerFixture = Pooling3dLayerValidationFixture<CLTensor, CLAccessor, CLPooling3dLayer, T>;
+
+template <typename T>
+using CLSpecialPooling3dLayerFixture = SpecialPooling3dLayerValidationFixture<CLTensor, CLAccessor, CLPooling3dLayer, T>;
+
+template <typename T>
+using CLPooling3dLayerGlobalFixture = Pooling3dLayerGlobalValidationFixture<CLTensor, CLAccessor, CLPooling3dLayer, T>;
+
+// clang-format on
+// *INDENT-ON*
+TEST_SUITE(Float)
+TEST_SUITE(FP32)
+
+FIXTURE_DATA_TEST_CASE(RunSpecial, CLSpecialPooling3dLayerFixture<float>, framework::DatasetMode::ALL, datasets::Pooling3dLayerDatasetSpecial() * framework::dataset::make("DataType", DataType::F32))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPooling3dLayerFixture<float>, framework::DatasetMode::PRECOMMIT, combine(datasets::Small5dShapes(), combine(Pooling3dLayerDatasetFPSmall,
+                                                                                                            framework::dataset::make("DataType", DataType::F32))))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLPooling3dLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(datasets::Large5dShapes(), combine(Pooling3dLayerDatasetFP,
+                                                                                                            framework::dataset::make("DataType",
+                                                                                                                    DataType::F32))))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+TEST_SUITE(GlobalPooling)
+// *INDENT-OFF*
+// clang-format off
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPooling3dLayerFixture<float>, framework::DatasetMode::ALL,
+                       combine(combine(combine(combine(combine(combine(
+                                    framework::dataset::make("InputShape", { TensorShape(3U, 27U, 13U, 4U),
+                                                                             TensorShape(4U, 27U, 13U, 4U, 2U)
+                                                                           }),
+                                    framework::dataset::make("PoolingType", { PoolingType::AVG, PoolingType::L2, PoolingType::MAX })),
+                                    framework::dataset::make("PoolingSize", { Size3D(27, 13, 4) })),
+                                    framework::dataset::make("Strides",  Size3D(1, 1, 1))),
+                                    framework::dataset::make("Paddings", Padding3D(0, 0, 0))),
+                                    framework::dataset::make("ExcludePadding", false)),
+                                    framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmallGlobal, CLPooling3dLayerGlobalFixture<float>, framework::DatasetMode::ALL,
+                       combine(combine(
+                                    framework::dataset::make("InputShape", { TensorShape(27U, 13U, 4U, 3U),
+                                                                             TensorShape(27U, 13U, 4U, 4U, 2U)
+                                                                           }),
+                                    framework::dataset::make("PoolingType", { PoolingType::AVG, PoolingType::L2, PoolingType::MAX })),
+                                    framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge, CLPooling3dLayerFixture<float>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(
+                                    framework::dataset::make("InputShape", { TensorShape(4U, 79U, 37U, 11U),
+                                                                             TensorShape(4U, 79U, 37U, 11U, 2U)
+                                                                           }),
+                                    framework::dataset::make("PoolingType", { PoolingType::AVG, PoolingType::L2, PoolingType::MAX })),
+                                    framework::dataset::make("PoolingSize", { Size3D(79, 37, 11) })),
+                                    framework::dataset::make("Strides",  Size3D(1, 1, 1))),
+                                    framework::dataset::make("Paddings", Padding3D(0, 0, 0))),
+                                    framework::dataset::make("ExcludePadding", false)),
+                                    framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+// clang-format on
+// *INDENT-ON*
+TEST_SUITE_END() // GlobalPooling
+TEST_SUITE_END() // FP32
+
+TEST_SUITE(FP16)
+
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPooling3dLayerFixture<half>, framework::DatasetMode::PRECOMMIT, combine(datasets::Small5x5Shapes(), combine(Pooling3dLayerDatasetFPSmall,
+                                                                                                           framework::dataset::make("DataType", DataType::F16))))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLPooling3dLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(datasets::Large5dShapes(), combine(Pooling3dLayerDatasetFP,
+                                                                                                           framework::dataset::make("DataType",
+                                                                                                                   DataType::F16))))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+TEST_SUITE(GlobalPooling)
+// *INDENT-OFF*
+// clang-format off
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPooling3dLayerFixture<half>, framework::DatasetMode::ALL,
+                       combine(combine(combine(combine(combine(combine(
+                                    framework::dataset::make("InputShape", { TensorShape(3U, 27U, 13U, 4U),
+                                                                             TensorShape(4U, 27U, 13U, 4U, 2U)
+                                                                           }),
+                                    framework::dataset::make("PoolingType", { PoolingType::AVG, PoolingType::L2, PoolingType::MAX })),
+                                    framework::dataset::make("PoolingSize", { Size3D(27, 13, 4) })),
+                                    framework::dataset::make("Strides",  Size3D(1, 1, 1))),
+                                    framework::dataset::make("Paddings", Padding3D(0, 0, 0))),
+                                    framework::dataset::make("ExcludePadding", false)),
+                                    framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmallGlobal, CLPooling3dLayerGlobalFixture<half>, framework::DatasetMode::ALL,
+                       combine(combine(
+                                    framework::dataset::make("InputShape", { TensorShape(27U, 13U, 4U, 3U),
+                                                                             TensorShape(27U, 13U, 4U, 4U, 2U)
+                                                                           }),
+                                    framework::dataset::make("PoolingType", { PoolingType::AVG, PoolingType::L2, PoolingType::MAX })),
+                                    framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge, CLPooling3dLayerFixture<half>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(
+                                    framework::dataset::make("InputShape", { TensorShape(4U, 79U, 37U, 11U),
+                                                                             TensorShape(4U, 79U, 37U, 11U, 2U)
+                                                                           }),
+                                    framework::dataset::make("PoolingType", { PoolingType::AVG, PoolingType::L2, PoolingType::MAX })),
+                                    framework::dataset::make("PoolingSize", { Size3D(79, 37, 11) })),
+                                    framework::dataset::make("Strides",  Size3D(1, 1, 1))),
+                                    framework::dataset::make("Paddings", Padding3D(0, 0, 0))),
+                                    framework::dataset::make("ExcludePadding", false)),
+                                    framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+// clang-format on
+// *INDENT-ON*
+TEST_SUITE_END() // GlobalPooling
+TEST_SUITE_END() // FP16
+TEST_SUITE_END() // Float
+TEST_SUITE_END() // Pooling3dLayer
+TEST_SUITE_END() // CL
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/Pooling3dLayerFixture.h b/tests/validation/fixtures/Pooling3dLayerFixture.h
new file mode 100644
index 0000000000..c1b3519e80
--- /dev/null
+++ b/tests/validation/fixtures/Pooling3dLayerFixture.h
@@ -0,0 +1,156 @@
+/*
+ * Copyright (c) 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_TEST_POOLING_3D_LAYER_FIXTURE
+#define ARM_COMPUTE_TEST_POOLING_3D_LAYER_FIXTURE
+
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "arm_compute/runtime/Tensor.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/reference/Pooling3dLayer.h"
+#include <random>
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class Pooling3dLayerValidationGenericFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape shape, Pooling3dLayerInfo pool_info, DataType data_type)
+    {
+        _target    = compute_target(shape, pool_info, data_type);
+        _reference = compute_reference(shape, pool_info, data_type);
+    }
+
+protected:
+    template <typename U>
+    void fill(U &&tensor)
+    {
+        if(tensor.data_type() == DataType::F32)
+        {
+            std::uniform_real_distribution<float> distribution(-1.0f, 1.0f);
+            library->fill(tensor, distribution, 0);
+        }
+        else if(tensor.data_type() == DataType::F16)
+        {
+            arm_compute::utils::uniform_real_distribution_16bit<half> distribution{ -1.0f, 1.0f };
+            library->fill(tensor, distribution, 0);
+        }
+        else // data type is quantized_asymmetric
+        {
+            ARM_COMPUTE_ERROR("Passed Type Not Supported");
+        }
+    }
+
+    TensorType compute_target(TensorShape shape, Pooling3dLayerInfo info,
+                              DataType data_type)
+    {
+        // Create tensors
+        TensorType        src       = create_tensor<TensorType>(shape, data_type, 1, QuantizationInfo(), DataLayout::NDHWC);
+        const TensorShape dst_shape = misc::shape_calculator::compute_pool3d_shape((src.info()->tensor_shape()), info);
+        TensorType        dst       = create_tensor<TensorType>(dst_shape, data_type, 1, QuantizationInfo(), DataLayout::NDHWC);
+
+        // Create and configure function
+        FunctionType pool_layer;
+        pool_layer.validate(src.info(), dst.info(), info);
+        pool_layer.configure(&src, &dst, info);
+
+        ARM_COMPUTE_ASSERT(src.info()->is_resizable());
+        ARM_COMPUTE_ASSERT(dst.info()->is_resizable());
+
+        // Allocate tensors
+        src.allocator()->allocate();
+        dst.allocator()->allocate();
+
+        ARM_COMPUTE_ASSERT(!src.info()->is_resizable());
+        ARM_COMPUTE_ASSERT(!dst.info()->is_resizable());
+
+        // Fill tensors
+        fill(AccessorType(src));
+
+        // Compute function
+        pool_layer.run();
+        return dst;
+    }
+
+    SimpleTensor<T> compute_reference(TensorShape shape, Pooling3dLayerInfo info, DataType data_type)
+    {
+        // Create reference
+        SimpleTensor<T> src(shape, data_type, 1, QuantizationInfo(), DataLayout::NDHWC);
+        // Fill reference
+        fill(src);
+        return reference::pooling_3d_layer<T>(src, info);
+    }
+
+    TensorType             _target{};
+    SimpleTensor<T>        _reference{};
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class Pooling3dLayerValidationFixture : public Pooling3dLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(TensorShape shape, PoolingType pool_type, Size3D pool_size, Size3D stride, Padding3D padding, bool exclude_padding, DataType data_type)
+    {
+        Pooling3dLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, Pooling3dLayerInfo(pool_type, pool_size, stride, padding, exclude_padding),
+                                                                                                 data_type);
+    }
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class Pooling3dLayerGlobalValidationFixture : public Pooling3dLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(TensorShape shape, PoolingType pool_type, DataType data_type)
+    {
+        Pooling3dLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, Pooling3dLayerInfo(pool_type), data_type);
+    }
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class SpecialPooling3dLayerValidationFixture : public Pooling3dLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(TensorShape src_shape, Pooling3dLayerInfo pool_info, DataType data_type)
+    {
+        Pooling3dLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(src_shape, pool_info, data_type);
+    }
+};
+
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_POOLING_3D_LAYER_FIXTURE */
diff --git a/tests/validation/reference/Pool3D.cpp b/tests/validation/reference/Pooling3dLayer.cpp
index 85a594e262..2e8f3a0b92 100644
--- a/tests/validation/reference/Pool3D.cpp
+++ b/tests/validation/reference/Pooling3dLayer.cpp
@@ -22,7 +22,7 @@
  * SOFTWARE.
  */
 
-#include "Pool3D.h"
+#include "Pooling3dLayer.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "tests/validation/Helpers.h"
 
@@ -37,7 +37,7 @@ namespace reference
 using namespace arm_compute::misc::shape_calculator;
 
 template <typename T>
-SimpleTensor<T> pool3d_internal(const SimpleTensor<T> &src, const Pool3DInfo &pool3d_info, SimpleTensor<uint32_t> *indices)
+SimpleTensor<T> pooling_3d_layer_internal(const SimpleTensor<T> &src, const Pooling3dLayerInfo &pool3d_info, SimpleTensor<uint32_t> *indices)
 {
     TensorShape     pooled_shape = compute_pool3d_shape(src.shape(), pool3d_info);
     SimpleTensor<T> dst{ pooled_shape, src.data_type(), 1 };
@@ -57,9 +57,9 @@ SimpleTensor<T> pool3d_internal(const SimpleTensor<T> &src, const Pool3DInfo &po
     const int pool_size_height = pool3d_info.is_global_pooling ? src.shape()[idx_height] : pool3d_info.pool_size.height;
     const int pool_size_depth  = pool3d_info.is_global_pooling ? src.shape()[idx_depth] : pool3d_info.pool_size.depth;
 
-    const int pool_stride_width  = static_cast<int>(pool3d_info.strides.width);
-    const int pool_stride_height = static_cast<int>(pool3d_info.strides.height);
-    const int pool_stride_depth  = static_cast<int>(pool3d_info.strides.depth);
+    const int pool_stride_width  = static_cast<int>(pool3d_info.stride.width);
+    const int pool_stride_height = static_cast<int>(pool3d_info.stride.height);
+    const int pool_stride_depth  = static_cast<int>(pool3d_info.stride.depth);
 
     const int pad_left  = static_cast<int>(pool3d_info.padding.left);
     const int pad_top   = static_cast<int>(pool3d_info.padding.top);
@@ -144,7 +144,6 @@ SimpleTensor<T> pool3d_internal(const SimpleTensor<T> &src, const Pool3DInfo &po
                                 {
                                     const auto val = static_cast<T>(
                                                          src[batch_offset_src + depth_offset_src + height_offset_src + x * num_channels + c]);
-
                                     if(val > max_val)
                                     {
                                         max_val   = val;
@@ -189,33 +188,33 @@ SimpleTensor<T> pool3d_internal(const SimpleTensor<T> &src, const Pool3DInfo &po
     return dst;
 }
 
-template SimpleTensor<float> pool3d(const SimpleTensor<float> &src, const Pool3DInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices);
-template SimpleTensor<half> pool3d(const SimpleTensor<half> &src, const Pool3DInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices);
+template SimpleTensor<float> pooling_3d_layer(const SimpleTensor<float> &src, const Pooling3dLayerInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices);
+template SimpleTensor<half> pooling_3d_layer(const SimpleTensor<half> &src, const Pooling3dLayerInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices);
 
 template <typename T>
-SimpleTensor<T> pool3d(const SimpleTensor<T> &src, const Pool3DInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices)
+SimpleTensor<T> pooling_3d_layer(const SimpleTensor<T> &src, const Pooling3dLayerInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices)
 {
     ARM_COMPUTE_UNUSED(output_qinfo);
-    return pool3d_internal<T>(src, pool3d_info, indices);
+    return pooling_3d_layer_internal<T>(src, pool3d_info, indices);
 }
 
 template <>
-SimpleTensor<int8_t> pool3d<int8_t>(const SimpleTensor<int8_t> &src, const Pool3DInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices)
+SimpleTensor<int8_t> pooling_3d_layer<int8_t>(const SimpleTensor<int8_t> &src, const Pooling3dLayerInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices)
 {
     SimpleTensor<float> src_tmp = convert_from_asymmetric(src);
-    SimpleTensor<float> dst_tmp = pool3d_internal<float>(src_tmp, pool3d_info, indices);
+    SimpleTensor<float> dst_tmp = pooling_3d_layer_internal<float>(src_tmp, pool3d_info, indices);
     return convert_to_asymmetric<int8_t>(dst_tmp, output_qinfo);
 }
 
 template <>
-SimpleTensor<uint8_t> pool3d<uint8_t>(const SimpleTensor<uint8_t> &src, const Pool3DInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices)
+SimpleTensor<uint8_t> pooling_3d_layer<uint8_t>(const SimpleTensor<uint8_t> &src, const Pooling3dLayerInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices)
 {
     SimpleTensor<float> src_tmp = convert_from_asymmetric(src);
-    SimpleTensor<float> dst_tmp = pool3d_internal<float>(src_tmp, pool3d_info, indices);
+    SimpleTensor<float> dst_tmp = pooling_3d_layer_internal<float>(src_tmp, pool3d_info, indices);
     return convert_to_asymmetric<uint8_t>(dst_tmp, output_qinfo);
 }
 
 } // namespace reference
 } // namespace validation
 } // namespace test
-} // namespace arm_compute
\ No newline at end of file
+} // namespace arm_compute
diff --git a/tests/validation/reference/Pool3D.h b/tests/validation/reference/Pooling3dLayer.h
index bdb5744ecc..481a0d3024 100644
--- a/tests/validation/reference/Pool3D.h
+++ b/tests/validation/reference/Pooling3dLayer.h
@@ -38,10 +38,11 @@ namespace validation
 namespace reference
 {
 template <typename T>
-SimpleTensor<T> pool3d_internal(const SimpleTensor<T> &src, const Pool3DInfo &pool3d_info, SimpleTensor<uint32_t> *indices);
+SimpleTensor<T> pooling_3d_layer_internal(const SimpleTensor<T> &src, const Pooling3dLayerInfo &pool3d_info, SimpleTensor<uint32_t> *indices = nullptr);
 
 template <typename T>
-SimpleTensor<T> pool3d(const SimpleTensor<T> &src, const Pool3DInfo &pool3d_info, const QuantizationInfo &output_qinfo, SimpleTensor<uint32_t> *indices);
+SimpleTensor<T> pooling_3d_layer(const SimpleTensor<T> &src, const Pooling3dLayerInfo &pool3d_info, const QuantizationInfo &output_qinfo = QuantizationInfo(),
+                                 SimpleTensor<uint32_t> *indices = nullptr);
 } // namespace reference
 } // namespace validation
 } // namespace test