COMPMID-806 Add NHWC data format support format for NEON pooling

Change-Id: I7ab174c72f3d56134fcec259a137739061fd12e9
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/123065
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>

9 files changed, 146 insertions, 95 deletions

diff --git a/tests/validation/CL/GlobalPoolingLayer.cpp b/tests/validation/CL/GlobalPoolingLayer.cpp
index 31e3fe0eb7..46752c4913 100644
--- a/tests/validation/CL/GlobalPoolingLayer.cpp
+++ b/tests/validation/CL/GlobalPoolingLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -59,7 +59,9 @@ using CLGlobalPoolingLayerFixture = GlobalPoolingLayerValidationFixture<CLTensor
 
 TEST_SUITE(Float)
 TEST_SUITE(FP32)
-FIXTURE_DATA_TEST_CASE(RunGlobalPooling, CLGlobalPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(GlobalPoolingLayerDataset, framework::dataset::make("DataType", DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunGlobalPooling, CLGlobalPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(combine(GlobalPoolingLayerDataset, framework::dataset::make("DataType",
+                                                                                                                  DataType::F32)),
+                                                                                                                  framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
@@ -67,8 +69,9 @@ FIXTURE_DATA_TEST_CASE(RunGlobalPooling, CLGlobalPoolingLayerFixture<float>, fra
 TEST_SUITE_END()
 
 TEST_SUITE(FP16)
-FIXTURE_DATA_TEST_CASE(RunGlobalPooling, CLGlobalPoolingLayerFixture<half>, framework::DatasetMode::ALL, combine(GlobalPoolingLayerDataset, framework::dataset::make("DataType",
-                                                                                                                 DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunGlobalPooling, CLGlobalPoolingLayerFixture<half>, framework::DatasetMode::ALL, combine(combine(GlobalPoolingLayerDataset, framework::dataset::make("DataType",
+                                                                                                                 DataType::F16)),
+                                                                                                                 framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f16);
diff --git a/tests/validation/CL/PoolingLayer.cpp b/tests/validation/CL/PoolingLayer.cpp
index 9da4c55c78..79e526b9ec 100644
--- a/tests/validation/CL/PoolingLayer.cpp
+++ b/tests/validation/CL/PoolingLayer.cpp
@@ -57,7 +57,7 @@ const auto PoolingLayerDatasetQS = combine(combine(combine(framework::dataset::m
 /** Input data set for asymmetric data type */
 const auto PoolingLayerDatasetQASYMM8 = combine(combine(combine(framework::dataset::make("PoolingType", { PoolingType::MAX, PoolingType::AVG }), framework::dataset::make("PoolingSize", { Size2D(2, 2), Size2D(3, 3), Size2D(5, 7), Size2D(8, 9) })),
                                                         framework::dataset::make("PadStride", { PadStrideInfo(1, 1, 0, 0), PadStrideInfo(2, 1, 0, 0), PadStrideInfo(1, 2, 1, 1), PadStrideInfo(2, 2, 1, 0) })),
-                                                framework::dataset::make("ExcludePadding", { true, false }));
+                                                framework::dataset::make("ExcludePadding", { true }));
 
 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(0.01f);  /**< Tolerance value for comparing reference's output against implementation's output for 16-bit floating-point type */
@@ -126,35 +126,40 @@ FIXTURE_DATA_TEST_CASE(RunSpecial, CLSpecialPoolingLayerFixture<float>, framewor
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunSmall, CLPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP, framework::dataset::make("DataType",
-                                                                                                    DataType::F32))))
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP, framework::dataset::make("DataType",
+                                                                                                            DataType::F32))),
+                                                                                                    framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, CLPoolingLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP, framework::dataset::make("DataType",
-                                                                                                        DataType::F32))))
+FIXTURE_DATA_TEST_CASE(RunLarge, CLPoolingLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP,
+                                                                                                                framework::dataset::make("DataType",
+                                                                                                                        DataType::F32))),
+                                                                                                        framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
 }
-TEST_SUITE_END()
+TEST_SUITE_END() // FP32
 
 TEST_SUITE(FP16)
-FIXTURE_DATA_TEST_CASE(RunSmall, CLPoolingLayerFixture<half>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP,
-                                                                                                   framework::dataset::make("DataType", DataType::F16))))
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPoolingLayerFixture<half>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP,
+                                                                                                           framework::dataset::make("DataType", DataType::F16))),
+                                                                                                   framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f16);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, CLPoolingLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP,
-                                                                                                       framework::dataset::make("DataType", DataType::F16))))
+FIXTURE_DATA_TEST_CASE(RunLarge, CLPoolingLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP,
+                                                                                                               framework::dataset::make("DataType", DataType::F16))),
+                                                                                                       framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f16);
 }
-TEST_SUITE_END()
-TEST_SUITE_END()
+TEST_SUITE_END() // FP16
+TEST_SUITE_END() // Float
 
 template <typename T>
 using CLPoolingLayerFixedPointFixture = PoolingLayerValidationFixedPointFixture<CLTensor, CLAccessor, CLPoolingLayer, T>;
@@ -175,7 +180,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLPoolingLayerFixedPointFixture<int8_t>, framew
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_qs8);
 }
-TEST_SUITE_END()
+TEST_SUITE_END() // QS8
 
 TEST_SUITE(QS16)
 FIXTURE_DATA_TEST_CASE(RunTiny, CLPoolingLayerFixedPointFixture<int16_t>, framework::DatasetMode::ALL, combine(combine(datasets::TinyShapes(), combine(PoolingLayerDatasetQS,
@@ -192,8 +197,8 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLPoolingLayerFixedPointFixture<int16_t>, frame
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_qs16);
 }
-TEST_SUITE_END()
-TEST_SUITE_END()
+TEST_SUITE_END() // QS16
+TEST_SUITE_END() // fixedPoint
 
 TEST_SUITE(Quantized)
 
@@ -201,27 +206,29 @@ template <typename T>
 using CLPoolingLayerQuantizedFixture = PoolingLayerValidationQuantizedFixture<CLTensor, CLAccessor, CLPoolingLayer, T>;
 
 TEST_SUITE(QASYMM8)
-FIXTURE_DATA_TEST_CASE(RunSmall, CLPoolingLayerQuantizedFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), combine(PoolingLayerDatasetQASYMM8,
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPoolingLayerQuantizedFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(combine(datasets::SmallShapes(), combine(PoolingLayerDatasetQASYMM8,
                                                                                                                        framework::dataset::make("DataType", DataType::QASYMM8))),
-                                                                                                               framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255, 127),
-                                                                                                                       QuantizationInfo(7.f / 255, 123)
-                                                                                                                                                            })))
+                                                                                                                       framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255, 127),
+                                                                                                                               QuantizationInfo(7.f / 255, 123)
+                                                                                                                                                                    })),
+                                                                                                               framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_qasymm8);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, CLPoolingLayerQuantizedFixture<uint8_t>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), combine(PoolingLayerDatasetQASYMM8,
+FIXTURE_DATA_TEST_CASE(RunLarge, CLPoolingLayerQuantizedFixture<uint8_t>, framework::DatasetMode::NIGHTLY, combine(combine(combine(datasets::LargeShapes(), combine(PoolingLayerDatasetQASYMM8,
                                                                                                                    framework::dataset::make("DataType", DataType::QASYMM8))),
-                                                                                                                   framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f / 255, 0) })))
+                                                                                                                   framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f / 255, 0) })),
+                                                                                                                   framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_qasymm8);
 }
-TEST_SUITE_END()
-TEST_SUITE_END()
+TEST_SUITE_END() // QASYMM8
+TEST_SUITE_END() // Quantized
 
-TEST_SUITE_END()
-TEST_SUITE_END()
+TEST_SUITE_END() // PoolingLayer
+TEST_SUITE_END() // CL
 } // namespace validation
 } // namespace test
 } // namespace arm_compute
diff --git a/tests/validation/GLES_COMPUTE/GlobalPoolingLayer.cpp b/tests/validation/GLES_COMPUTE/GlobalPoolingLayer.cpp
index 88372ffe24..162f189af5 100644
--- a/tests/validation/GLES_COMPUTE/GlobalPoolingLayer.cpp
+++ b/tests/validation/GLES_COMPUTE/GlobalPoolingLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -59,7 +59,9 @@ using GCGlobalPoolingLayerFixture = GlobalPoolingLayerValidationFixture<GCTensor
 
 TEST_SUITE(Float)
 TEST_SUITE(FP32)
-FIXTURE_DATA_TEST_CASE(RunGlobalPooling, GCGlobalPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(GlobalPoolingLayerDataset, framework::dataset::make("DataType", DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunGlobalPooling, GCGlobalPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(combine(GlobalPoolingLayerDataset, framework::dataset::make("DataType",
+                                                                                                                  DataType::F32)),
+                                                                                                                  framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(GCAccessor(_target), _reference, tolerance_f32);
@@ -67,8 +69,9 @@ FIXTURE_DATA_TEST_CASE(RunGlobalPooling, GCGlobalPoolingLayerFixture<float>, fra
 TEST_SUITE_END()
 
 TEST_SUITE(FP16)
-FIXTURE_DATA_TEST_CASE(RunGlobalPooling, GCGlobalPoolingLayerFixture<half>, framework::DatasetMode::ALL, combine(GlobalPoolingLayerDataset, framework::dataset::make("DataType",
-                                                                                                                 DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunGlobalPooling, GCGlobalPoolingLayerFixture<half>, framework::DatasetMode::ALL, combine(combine(GlobalPoolingLayerDataset, framework::dataset::make("DataType",
+                                                                                                                 DataType::F16)),
+                                                                                                                 framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(GCAccessor(_target), _reference, tolerance_f16);
diff --git a/tests/validation/GLES_COMPUTE/PoolingLayer.cpp b/tests/validation/GLES_COMPUTE/PoolingLayer.cpp
index 1496ceec1c..ac1bd724ac 100644
--- a/tests/validation/GLES_COMPUTE/PoolingLayer.cpp
+++ b/tests/validation/GLES_COMPUTE/PoolingLayer.cpp
@@ -87,14 +87,17 @@ using GCPoolingLayerFixture = PoolingLayerValidationFixture<GCTensor, GCAccessor
 
 TEST_SUITE(Float)
 TEST_SUITE(FP32)
-FIXTURE_DATA_TEST_CASE(RunSmall, GCPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP, framework::dataset::make("DataType",
-                                                                                                    DataType::F32))))
+FIXTURE_DATA_TEST_CASE(RunSmall, GCPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP, framework::dataset::make("DataType",
+                                                                                                            DataType::F32))),
+                                                                                                    framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(GCAccessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, GCPoolingLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP, framework::dataset::make("DataType",
-                                                                                                        DataType::F32))))
+FIXTURE_DATA_TEST_CASE(RunLarge, GCPoolingLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP,
+                                                                                                                framework::dataset::make("DataType",
+                                                                                                                        DataType::F32))),
+                                                                                                        framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(GCAccessor(_target), _reference, tolerance_f32);
@@ -102,14 +105,16 @@ FIXTURE_DATA_TEST_CASE(RunLarge, GCPoolingLayerFixture<float>, framework::Datase
 TEST_SUITE_END()
 
 TEST_SUITE(FP16)
-FIXTURE_DATA_TEST_CASE(RunSmall, GCPoolingLayerFixture<half>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP,
-                                                                                                   framework::dataset::make("DataType", DataType::F16))))
+FIXTURE_DATA_TEST_CASE(RunSmall, GCPoolingLayerFixture<half>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP,
+                                                                                                           framework::dataset::make("DataType", DataType::F16))),
+                                                                                                   framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(GCAccessor(_target), _reference, tolerance_f16);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, GCPoolingLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP,
-                                                                                                       framework::dataset::make("DataType", DataType::F16))))
+FIXTURE_DATA_TEST_CASE(RunLarge, GCPoolingLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP,
+                                                                                                               framework::dataset::make("DataType", DataType::F16))),
+                                                                                                       framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(GCAccessor(_target), _reference, tolerance_f16);
diff --git a/tests/validation/Helpers.cpp b/tests/validation/Helpers.cpp
index 3d554f0d25..6e7ec8ddc1 100644
--- a/tests/validation/Helpers.cpp
+++ b/tests/validation/Helpers.cpp
@@ -119,7 +119,8 @@ HarrisCornersParameters harris_corners_parameters()
 SimpleTensor<float> convert_from_asymmetric(const SimpleTensor<uint8_t> &src)
 {
     const QuantizationInfo &quantization_info = src.quantization_info();
-    SimpleTensor<float>     dst{ src.shape(), DataType::F32, 1, 0 };
+    SimpleTensor<float>     dst{ src.shape(), DataType::F32, 1, 0, QuantizationInfo(), src.data_layout() };
+
     for(int i = 0; i < src.num_elements(); ++i)
     {
         dst[i] = quantization_info.dequantize(src[i]);
diff --git a/tests/validation/NEON/GlobalPoolingLayer.cpp b/tests/validation/NEON/GlobalPoolingLayer.cpp
index 37950b059f..769780647b 100644
--- a/tests/validation/NEON/GlobalPoolingLayer.cpp
+++ b/tests/validation/NEON/GlobalPoolingLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -58,7 +58,9 @@ using NEGlobalPoolingLayerFixture = GlobalPoolingLayerValidationFixture<Tensor,
 
 TEST_SUITE(Float)
 TEST_SUITE(FP32)
-FIXTURE_DATA_TEST_CASE(RunGlobalPooling, NEGlobalPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(GlobalPoolingLayerDataset, framework::dataset::make("DataType", DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunGlobalPooling, NEGlobalPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(combine(GlobalPoolingLayerDataset, framework::dataset::make("DataType",
+                                                                                                                  DataType::F32)),
+                                                                                                                  framework::dataset::make("DataLayout", DataLayout::NCHW)))
 {
     // Validate output
     validate(Accessor(_target), _reference, tolerance_f32);
diff --git a/tests/validation/NEON/PoolingLayer.cpp b/tests/validation/NEON/PoolingLayer.cpp
index 350a7b883b..b44f945eb7 100644
--- a/tests/validation/NEON/PoolingLayer.cpp
+++ b/tests/validation/NEON/PoolingLayer.cpp
@@ -59,7 +59,7 @@ const auto PoolingLayerDatasetQS = combine(combine(combine(framework::dataset::m
 
 const auto PoolingLayerDatasetQASYMM8 = combine(combine(combine(framework::dataset::make("PoolingType", { PoolingType::MAX, PoolingType::AVG }), framework::dataset::make("PoolingSize", { Size2D(2, 2), Size2D(3, 3), Size2D(4, 4), Size2D(9, 9), Size2D(3, 7), Size2D(7, 8) })),
                                                         framework::dataset::make("PadStride", { PadStrideInfo(1, 1, 0, 0), PadStrideInfo(2, 1, 0, 0), PadStrideInfo(1, 2, 1, 1), PadStrideInfo(2, 2, 1, 0) })),
-                                                framework::dataset::make("ExcludePadding", { true, false }));
+                                                framework::dataset::make("ExcludePadding", { true }));
 
 constexpr AbsoluteTolerance<float> tolerance_f32(0.001f); /**< Tolerance value for comparing reference's output against implementation's output for float types */
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
@@ -130,37 +130,42 @@ FIXTURE_DATA_TEST_CASE(RunSpecial, NESpecialPoolingLayerFixture<float>, framewor
     // Validate output
     validate(Accessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunSmall, NEPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP, framework::dataset::make("DataType",
-                                                                                                    DataType::F32))))
+FIXTURE_DATA_TEST_CASE(RunSmall, NEPoolingLayerFixture<float>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP, framework::dataset::make("DataType",
+                                                                                                            DataType::F32))),
+                                                                                                    framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
 {
     // Validate output
     validate(Accessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, NEPoolingLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP, framework::dataset::make("DataType",
-                                                                                                        DataType::F32))))
+FIXTURE_DATA_TEST_CASE(RunLarge, NEPoolingLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP,
+                                                                                                                framework::dataset::make("DataType",
+                                                                                                                        DataType::F32))),
+                                                                                                        framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
 {
     // Validate output
     validate(Accessor(_target), _reference, tolerance_f32);
 }
-TEST_SUITE_END()
+TEST_SUITE_END() // FP32
 
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 TEST_SUITE(FP16)
-FIXTURE_DATA_TEST_CASE(RunSmall, NEPoolingLayerFixture<half>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP,
-                                                                                                   framework::dataset::make("DataType", DataType::F16))))
+FIXTURE_DATA_TEST_CASE(RunSmall, NEPoolingLayerFixture<half>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), combine(PoolingLayerDatasetFP,
+                                                                                                           framework::dataset::make("DataType", DataType::F16))),
+                                                                                                   framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
 {
     // Validate output
     validate(Accessor(_target), _reference, tolerance_f16);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, NEPoolingLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP,
-                                                                                                       framework::dataset::make("DataType", DataType::F16))))
+FIXTURE_DATA_TEST_CASE(RunLarge, NEPoolingLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), combine(PoolingLayerDatasetFP,
+                                                                                                               framework::dataset::make("DataType", DataType::F16))),
+                                                                                                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
 {
     // Validate output
     validate(Accessor(_target), _reference, tolerance_f16);
 }
-TEST_SUITE_END()
-#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-TEST_SUITE_END()
+TEST_SUITE_END() // FP16
+#endif           /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+TEST_SUITE_END() // Float
 
 template <typename T>
 using NEPoolingLayerFixedPointFixture = PoolingLayerValidationFixedPointFixture<Tensor, Accessor, NEPoolingLayer, T>;
@@ -181,7 +186,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NEPoolingLayerFixedPointFixture<int8_t>, framew
     // Validate output
     validate(Accessor(_target), _reference, tolerance_qs8);
 }
-TEST_SUITE_END()
+TEST_SUITE_END() // QS8
 
 TEST_SUITE(QS16)
 FIXTURE_DATA_TEST_CASE(RunTiny, NEPoolingLayerFixedPointFixture<int16_t>, framework::DatasetMode::ALL, combine(combine(datasets::TinyShapes(), combine(PoolingLayerDatasetQS,
@@ -198,8 +203,8 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NEPoolingLayerFixedPointFixture<int16_t>, frame
     // Validate output
     validate(Accessor(_target), _reference, tolerance_qs16);
 }
-TEST_SUITE_END()
-TEST_SUITE_END()
+TEST_SUITE_END() // QS16
+TEST_SUITE_END() // FixedPoint
 
 TEST_SUITE(Quantized)
 
@@ -207,27 +212,28 @@ template <typename T>
 using NEPoolingLayerQuantizedFixture = PoolingLayerValidationQuantizedFixture<Tensor, Accessor, NEPoolingLayer, T>;
 
 TEST_SUITE(QASYMM8)
-FIXTURE_DATA_TEST_CASE(RunSmall, NEPoolingLayerQuantizedFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(datasets::SmallShapes(), combine(PoolingLayerDatasetQASYMM8,
+FIXTURE_DATA_TEST_CASE(RunSmall, NEPoolingLayerQuantizedFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(combine(datasets::SmallShapes(), combine(PoolingLayerDatasetQASYMM8,
                                                                                                                        framework::dataset::make("DataType", DataType::QASYMM8))),
-                                                                                                               framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255, 127),
-                                                                                                                       QuantizationInfo(7.f / 255, 123)
-                                                                                                                                                            })))
+                                                                                                                       framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255, 127),
+                                                                                                                               QuantizationInfo(7.f / 255, 123)
+                                                                                                                                                                    })),
+                                                                                                               framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
 {
     // Validate output
     validate(Accessor(_target), _reference, tolerance_qasymm8);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, NEPoolingLayerQuantizedFixture<uint8_t>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), combine(PoolingLayerDatasetQASYMM8,
+FIXTURE_DATA_TEST_CASE(RunLarge, NEPoolingLayerQuantizedFixture<uint8_t>, framework::DatasetMode::NIGHTLY, combine(combine(combine(datasets::LargeShapes(), combine(PoolingLayerDatasetQASYMM8,
                                                                                                                    framework::dataset::make("DataType", DataType::QASYMM8))),
-                                                                                                                   framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f / 255, 0) })))
+                                                                                                                   framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f / 255, 0) })),
+                                                                                                                   framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
 {
     // Validate output
     validate(Accessor(_target), _reference, tolerance_qasymm8);
 }
-TEST_SUITE_END()
-TEST_SUITE_END()
-
-TEST_SUITE_END()
-TEST_SUITE_END()
+TEST_SUITE_END() // QASYMM8
+TEST_SUITE_END() // Quantized
+TEST_SUITE_END() // PoolingLayer
+TEST_SUITE_END() // NEON
 } // namespace validation
 } // namespace test
 } // namespace arm_compute
diff --git a/tests/validation/fixtures/PoolingLayerFixture.h b/tests/validation/fixtures/PoolingLayerFixture.h
index 3bbb403ae7..3c28b3b64d 100644
--- a/tests/validation/fixtures/PoolingLayerFixture.h
+++ b/tests/validation/fixtures/PoolingLayerFixture.h
@@ -47,14 +47,20 @@ class PoolingLayerValidationGenericFixture : public framework::Fixture
 {
 public:
     template <typename...>
-    void setup(TensorShape shape, PoolingLayerInfo pool_info, DataType data_type, int fractional_bits, QuantizationInfo quantization_info)
+    void setup(TensorShape shape, PoolingLayerInfo pool_info, DataType data_type, DataLayout data_layout, int fractional_bits, QuantizationInfo quantization_info)
     {
         _fractional_bits   = fractional_bits;
         _quantization_info = quantization_info;
         _pool_info         = pool_info;
 
-        _target    = compute_target(shape, pool_info, data_type, fractional_bits, quantization_info);
-        _reference = compute_reference(shape, pool_info, data_type, fractional_bits, quantization_info);
+        // Change shape in case of NHWC.
+        if(data_layout == DataLayout::NHWC)
+        {
+            permute(shape, PermutationVector(2U, 0U, 1U));
+        }
+
+        _target    = compute_target(shape, pool_info, data_type, data_layout, fractional_bits, quantization_info);
+        _reference = compute_reference(shape, pool_info, data_type, data_layout, fractional_bits, quantization_info);
     }
 
 protected:
@@ -79,10 +85,10 @@ protected:
     }
 
     TensorType compute_target(const TensorShape &shape, PoolingLayerInfo info,
-                              DataType data_type, int fixed_point_position, QuantizationInfo quantization_info)
+                              DataType data_type, DataLayout data_layout, int fixed_point_position, QuantizationInfo quantization_info)
     {
         // Create tensors
-        TensorType src = create_tensor<TensorType>(shape, data_type, 1, fixed_point_position, quantization_info);
+        TensorType src = create_tensor<TensorType>(shape, data_type, 1, fixed_point_position, quantization_info, data_layout);
         TensorType dst;
 
         // Create and configure function
@@ -109,10 +115,10 @@ protected:
     }
 
     SimpleTensor<T> compute_reference(const TensorShape &shape, PoolingLayerInfo info,
-                                      DataType data_type, int fixed_point_position, QuantizationInfo quantization_info)
+                                      DataType data_type, DataLayout data_layout, int fixed_point_position, QuantizationInfo quantization_info)
     {
         // Create reference
-        SimpleTensor<T> src{ shape, data_type, 1, fixed_point_position, quantization_info };
+        SimpleTensor<T> src{ shape, data_type, 1, fixed_point_position, quantization_info, data_layout };
 
         // Fill reference
         fill(src);
@@ -132,10 +138,10 @@ class PoolingLayerValidationFixture : public PoolingLayerValidationGenericFixtur
 {
 public:
     template <typename...>
-    void setup(TensorShape shape, PoolingType pool_type, Size2D pool_size, PadStrideInfo pad_stride_info, bool exclude_padding, DataType data_type)
+    void setup(TensorShape shape, PoolingType pool_type, Size2D pool_size, PadStrideInfo pad_stride_info, bool exclude_padding, DataType data_type, DataLayout data_layout)
     {
         PoolingLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, PoolingLayerInfo(pool_type, pool_size, pad_stride_info, exclude_padding),
-                                                                                               data_type, 0, QuantizationInfo());
+                                                                                               data_type, data_layout, 0, QuantizationInfo());
     }
 };
 
@@ -147,7 +153,7 @@ public:
     void setup(TensorShape shape, PoolingType pool_type, Size2D pool_size, PadStrideInfo pad_stride_info, bool exclude_padding, DataType data_type, int fractional_bits)
     {
         PoolingLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, PoolingLayerInfo(pool_type, pool_size, pad_stride_info, exclude_padding),
-                                                                                               data_type, fractional_bits, QuantizationInfo());
+                                                                                               data_type, DataLayout::NCHW, fractional_bits, QuantizationInfo());
     }
 };
 
@@ -156,10 +162,11 @@ class PoolingLayerValidationQuantizedFixture : public PoolingLayerValidationGene
 {
 public:
     template <typename...>
-    void setup(TensorShape shape, PoolingType pool_type, Size2D pool_size, PadStrideInfo pad_stride_info, bool exclude_padding, DataType data_type, QuantizationInfo quantization_info)
+    void setup(TensorShape shape, PoolingType pool_type, Size2D pool_size, PadStrideInfo pad_stride_info, bool exclude_padding, DataType data_type,
+               QuantizationInfo quantization_info, DataLayout data_layout = DataLayout::NCHW)
     {
         PoolingLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, PoolingLayerInfo(pool_type, pool_size, pad_stride_info, exclude_padding),
-                                                                                               data_type, 0, quantization_info);
+                                                                                               data_type, data_layout, 0, quantization_info);
     }
 };
 
@@ -171,7 +178,7 @@ public:
     void setup(TensorShape src_shape, TensorShape dst_shape, PoolingLayerInfo pool_info, DataType data_type)
     {
         ARM_COMPUTE_UNUSED(dst_shape);
-        PoolingLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(src_shape, pool_info, data_type, 0, QuantizationInfo());
+        PoolingLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(src_shape, pool_info, data_type, DataLayout::NCHW, 0, QuantizationInfo());
     }
 };
 
@@ -180,9 +187,9 @@ class GlobalPoolingLayerValidationFixture : public PoolingLayerValidationGeneric
 {
 public:
     template <typename...>
-    void setup(TensorShape shape, PoolingType pool_type, DataType data_type)
+    void setup(TensorShape shape, PoolingType pool_type, DataType data_type, DataLayout data_layout = DataLayout::NCHW)
     {
-        PoolingLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, PoolingLayerInfo(pool_type), data_type, 0, QuantizationInfo());
+        PoolingLayerValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, PoolingLayerInfo(pool_type), data_type, DataLayout::NCHW, 0, QuantizationInfo());
     }
 };
 
diff --git a/tests/validation/reference/PoolingLayer.cpp b/tests/validation/reference/PoolingLayer.cpp
index c14ab98c28..071c20ed56 100644
--- a/tests/validation/reference/PoolingLayer.cpp
+++ b/tests/validation/reference/PoolingLayer.cpp
@@ -23,6 +23,7 @@
  */
 #include "PoolingLayer.h"
 
+#include "Permute.h"
 #include "arm_compute/core/Types.h"
 #include "tests/validation/FixedPoint.h"
 #include "tests/validation/Helpers.h"
@@ -54,8 +55,8 @@ TensorShape calculate_output_shape(TensorShape shape, const PoolingLayerInfo &in
 }
 } // namespace
 
-template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type>
-SimpleTensor<T> pooling_layer(const SimpleTensor<T> &src, const PoolingLayerInfo &info)
+template <typename T>
+SimpleTensor<T> pooling_layer_nchw(const SimpleTensor<T> &src, SimpleTensor<T> &dst, const PoolingLayerInfo &info)
 {
     ARM_COMPUTE_ERROR_ON(info.is_global_pooling() && (src.shape().x() != src.shape().y()));
 
@@ -74,9 +75,6 @@ SimpleTensor<T> pooling_layer(const SimpleTensor<T> &src, const PoolingLayerInfo
     const auto h_src      = static_cast<int>(src.shape()[1]);
     const int  upper_dims = src.shape().total_size() / (w_src * h_src);
 
-    // Create reference
-    SimpleTensor<T> dst{ calculate_output_shape(src.shape(), info), src.data_type(), 1, src.fixed_point_position() };
-
     const auto w_dst = static_cast<int>(dst.shape()[0]);
     const auto h_dst = static_cast<int>(dst.shape()[1]);
 
@@ -173,6 +171,10 @@ SimpleTensor<T> pooling_layer(const SimpleTensor<T> &src, const PoolingLayerInfo
 {
     ARM_COMPUTE_ERROR_ON(info.is_global_pooling() && (src.shape().x() != src.shape().y()));
 
+    const auto w_src      = static_cast<int>(src.shape()[0]);
+    const auto h_src      = static_cast<int>(src.shape()[1]);
+    const int  upper_dims = src.shape().total_size() / (w_src * h_src);
+
     const int   pool_size_x     = info.is_global_pooling() ? src.shape().x() : info.pool_size().width;
     const int   pool_size_y     = info.is_global_pooling() ? src.shape().y() : info.pool_size().height;
     PoolingType type            = info.pool_type();
@@ -184,10 +186,6 @@ SimpleTensor<T> pooling_layer(const SimpleTensor<T> &src, const PoolingLayerInfo
     int         pad_bottom      = info.pad_stride_info().pad_bottom();
     bool        exclude_padding = info.exclude_padding();
 
-    const auto w_src      = static_cast<int>(src.shape()[0]);
-    const auto h_src      = static_cast<int>(src.shape()[1]);
-    const int  upper_dims = src.shape().total_size() / (w_src * h_src);
-
     // Create reference
     SimpleTensor<T> dst{ calculate_output_shape(src.shape(), info), src.data_type(), 1, src.fixed_point_position() };
 
@@ -299,6 +297,25 @@ SimpleTensor<uint8_t> pooling_layer<uint8_t>(const SimpleTensor<uint8_t> &src, c
     return dst;
 }
 
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type>
+SimpleTensor<T> pooling_layer(const SimpleTensor<T> &src, const PoolingLayerInfo &info)
+{
+    if(src.data_layout() == DataLayout::NHWC)
+    {
+        SimpleTensor<T> src_nchw = reference::permute<T>(src, PermutationVector(1U, 2U, 0U));
+        SimpleTensor<T> dst{ calculate_output_shape(src_nchw.shape(), info), src_nchw.data_type(), 1, src_nchw.fixed_point_position() };
+
+        pooling_layer_nchw<T>(src_nchw, dst, info);
+
+        return reference::permute<T>(dst, PermutationVector(2U, 0U, 1U));
+    }
+    else
+    {
+        SimpleTensor<T> dst{ calculate_output_shape(src.shape(), info), src.data_type(), 1, src.fixed_point_position() };
+        return pooling_layer_nchw<T>(src, dst, info);
+    }
+}
+
 template SimpleTensor<float> pooling_layer(const SimpleTensor<float> &src, const PoolingLayerInfo &info);
 template SimpleTensor<half> pooling_layer(const SimpleTensor<half> &src, const PoolingLayerInfo &info);
 template SimpleTensor<qint8_t> pooling_layer(const SimpleTensor<qint8_t> &src, const PoolingLayerInfo &info);