COMPMID-2421: Add 5x5 DepthwiseConv2d in the optimized path for NEON

Change-Id: Idecc8e7ce4404af83f0f99ed993be8f961f7b9af
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1527
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 148 insertions, 79 deletions

diff --git a/tests/datasets/DepthwiseConvolutionLayerDataset.h b/tests/datasets/DepthwiseConvolutionLayerDataset.h
index 440cb88ac2..d0617275c0 100644
--- a/tests/datasets/DepthwiseConvolutionLayerDataset.h
+++ b/tests/datasets/DepthwiseConvolutionLayerDataset.h
@@ -211,11 +211,15 @@ public:
     {
         // Stride 1
         add_config(TensorShape(7U, 7U, 16U), Size2D(3U, 3U), PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL));
+        add_config(TensorShape(7U, 7U, 16U), Size2D(3U, 3U), PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL), Size2D(2U, 2U));
         add_config(TensorShape(7U, 7U, 16U), Size2D(3U, 3U), PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL));
+        add_config(TensorShape(7U, 7U, 16U), Size2D(3U, 3U), PadStrideInfo(1, 1, 2, 2, DimensionRoundingType::CEIL), Size2D(2U, 2U));
         // Stride 2
-        add_config(TensorShape(7U, 7U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL));
-        add_config(TensorShape(7U, 7U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 1, 1, 1, 1, DimensionRoundingType::CEIL));
-        add_config(TensorShape(9U, 9U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 1, 1, 1, 1, DimensionRoundingType::CEIL), Size2D(2U, 2U));
+        add_config(TensorShape(9U, 9U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL));
+        add_config(TensorShape(9U, 9U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL), Size2D(2U, 2U));
+        add_config(TensorShape(9U, 9U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 1, 1, DimensionRoundingType::CEIL));
+        // TODO(COMPMID-2464): Enable once dilated conv with stride 2 is supported
+        // add_config(TensorShape(9U, 9U, 1U), Size2D(3U, 3U), PadStrideInfo(2, 2, 2, 2, DimensionRoundingType::CEIL), Size2D(2U, 2U));
     }
 };
 /** Dataset containing optimized, 3x3 depthwise convolution shapes. */
@@ -239,6 +243,27 @@ public:
         add_config(TensorShape(64U, 64U, 128U), Size2D(3U, 3U), PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::CEIL));
     }
 };
+
+/** Dataset containing optimized, 5x5 depthwise convolution shapes. */
+class SmallOptimizedDepthwiseConvolutionLayerDataset5x5 final : public DepthwiseConvolutionLayerDataset
+{
+public:
+    SmallOptimizedDepthwiseConvolutionLayerDataset5x5()
+    {
+        // Stride 1
+        add_config(TensorShape(7U, 7U, 16U), Size2D(5U, 5U), PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL));
+        add_config(TensorShape(11U, 11U, 16U), Size2D(5U, 5U), PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL), Size2D(2U, 2U));
+        add_config(TensorShape(7U, 7U, 16U), Size2D(5U, 5U), PadStrideInfo(1, 1, 2, 2, DimensionRoundingType::CEIL));
+        add_config(TensorShape(7U, 7U, 16U), Size2D(5U, 5U), PadStrideInfo(1, 1, 4, 4, DimensionRoundingType::CEIL), Size2D(2U, 2U));
+        // Stride 2
+        add_config(TensorShape(9U, 9U, 32U), Size2D(5U, 5U), PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL));
+        // TODO(COMPMID-2464): Enable once dilated conv with stride 2 is supported
+        // add_config(TensorShape(9U, 9U, 32U), Size2D(5U, 5U), PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL), Size2D(2U, 2U));
+        add_config(TensorShape(9U, 9U, 32U), Size2D(5U, 5U), PadStrideInfo(2, 2, 2, 2, 2, 2, DimensionRoundingType::CEIL));
+        // TODO(COMPMID-2464): Enable once dilated conv with stride 2 is supported
+        // add_config(TensorShape(9U, 9U, 32U), Size2D(5U, 5U), PadStrideInfo(2, 2, 4, 4, 4, 4, DimensionRoundingType::CEIL), Size2D(2U, 2U));
+    }
+};
 } // namespace datasets
 } // namespace test
 } // namespace arm_compute
diff --git a/tests/validation/NEON/DepthwiseConvolutionLayer.cpp b/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
index 2ffe540fbc..ec9eb107aa 100644
--- a/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
+++ b/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
@@ -294,38 +294,41 @@ FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixture<float>, fram
 TEST_SUITE_END() // Dilation
 TEST_SUITE_END() // Generic
 
-TEST_SUITE(W3x3)
 template <typename T>
-using NEDepthwiseConvolutionLayerFixture3x3 = DepthwiseConvolutionLayerValidationFixture<Tensor, Accessor, NEDepthwiseConvolutionLayerOptimized, T>;
-FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerFixture3x3<float>, framework::DatasetMode::ALL, combine(combine(combine(combine(datasets::SmallDepthwiseConvolutionLayerDataset3x3(),
-                                                                                                                    depth_multipliers),
-                                                                                                                    framework::dataset::make("DataType",
-                                                                                                                            DataType::F32)),
-                                                                                                                    framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
-                                                                                                                    ActivationFunctionsDataset))
+using NEDepthwiseConvolutionLayerFixtureOptimized = DepthwiseConvolutionLayerValidationFixture<Tensor, Accessor, NEDepthwiseConvolutionLayerOptimized, T>;
+
+TEST_SUITE(W3x3)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerFixtureOptimized<float>, framework::DatasetMode::ALL, combine(combine(combine(combine(datasets::SmallDepthwiseConvolutionLayerDataset3x3(),
+                       depth_multipliers),
+                       framework::dataset::make("DataType",
+                                                DataType::F32)),
+                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
+                       ActivationFunctionsDataset))
 {
     validate(Accessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixture3x3<float>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(datasets::LargeDepthwiseConvolutionLayerDataset3x3(),
-                                                                                                                        depth_multipliers),
-                                                                                                                        framework::dataset::make("DataType",
-                                                                                                                                DataType::F32)),
-                                                                                                                        framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
-                                                                                                                        ActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixtureOptimized<float>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(datasets::LargeDepthwiseConvolutionLayerDataset3x3(),
+                                                       depth_multipliers),
+                                               framework::dataset::make("DataType",
+                                                                        DataType::F32)),
+                                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
+                               ActivationFunctionsDataset))
 {
     validate(Accessor(_target), _reference, tolerance_f32);
 }
 TEST_SUITE(Dilation)
-FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerFixture3x3<float>, framework::DatasetMode::ALL, combine(combine(combine(combine(datasets::SmallDepthwiseDilatedConvolutionLayerDataset3x3(),
-                                                                                                                    depth_multipliers),
-                                                                                                                    framework::dataset::make("DataType",
-                                                                                                                            DataType::F32)),
-                                                                                                                    framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
-                                                                                                                    ActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerFixtureOptimized<float>, framework::DatasetMode::ALL,
+                       combine(combine(combine(combine(datasets::SmallDepthwiseDilatedConvolutionLayerDataset3x3(),
+                                                       depth_multipliers),
+                                               framework::dataset::make("DataType",
+                                                                        DataType::F32)),
+                                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
+                               ActivationFunctionsDataset))
 {
     validate(Accessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixture3x3<float>, framework::DatasetMode::NIGHTLY,
+FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixtureOptimized<float>, framework::DatasetMode::NIGHTLY,
                        combine(combine(combine(combine(datasets::LargeDepthwiseDilatedConvolutionLayerDataset3x3(),
                                                        depth_multipliers),
                                                framework::dataset::make("DataType",
@@ -337,8 +340,10 @@ FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixture3x3<float>, f
 }
 
 TEST_SUITE_END() // Dilation
+TEST_SUITE_END() // W3x3
 
-FIXTURE_DATA_TEST_CASE(RunOptimizedSmall, NEDepthwiseConvolutionLayerFixture3x3<float>, framework::DatasetMode::PRECOMMIT,
+TEST_SUITE(Optimized)
+FIXTURE_DATA_TEST_CASE(RunSmall3x3, NEDepthwiseConvolutionLayerFixtureOptimized<float>, framework::DatasetMode::PRECOMMIT,
                        combine(combine(combine(combine(datasets::SmallOptimizedDepthwiseConvolutionLayerDataset3x3(),
                                                        framework::dataset::make("DepthMultiplier", 1)),
                                                framework::dataset::make("DataType",
@@ -348,7 +353,17 @@ FIXTURE_DATA_TEST_CASE(RunOptimizedSmall, NEDepthwiseConvolutionLayerFixture3x3<
 {
     validate(Accessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunOptimizedLarge, NEDepthwiseConvolutionLayerFixture3x3<float>, framework::DatasetMode::NIGHTLY,
+FIXTURE_DATA_TEST_CASE(RunSmall5x5, NEDepthwiseConvolutionLayerFixtureOptimized<float>, framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(combine(datasets::SmallOptimizedDepthwiseConvolutionLayerDataset5x5(),
+                                                       framework::dataset::make("DepthMultiplier", 1)),
+                                               framework::dataset::make("DataType",
+                                                                        DataType::F32)),
+                                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
+                               ActivationFunctionsDataset))
+{
+    validate(Accessor(_target), _reference, tolerance_f32);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge3x3, NEDepthwiseConvolutionLayerFixtureOptimized<float>, framework::DatasetMode::NIGHTLY,
                        combine(combine(combine(combine(datasets::LargeOptimizedDepthwiseConvolutionLayerDataset3x3(),
                                                        framework::dataset::make("DepthMultiplier", 1)),
                                                framework::dataset::make("DataType",
@@ -358,7 +373,7 @@ FIXTURE_DATA_TEST_CASE(RunOptimizedLarge, NEDepthwiseConvolutionLayerFixture3x3<
 {
     validate(Accessor(_target), _reference, tolerance_f32);
 }
-TEST_SUITE_END() // W3x3
+TEST_SUITE_END() // Optimized
 TEST_SUITE_END() // F32
 
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
@@ -407,40 +422,42 @@ FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixture<half>, frame
 TEST_SUITE_END() // Dilation
 
 TEST_SUITE_END() // Generic
-TEST_SUITE(W3x3)
 template <typename T>
-using NEDepthwiseConvolutionLayerFixture3x3 = DepthwiseConvolutionLayerValidationFixture<Tensor, Accessor, NEDepthwiseConvolutionLayerOptimized, T>;
-FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerFixture3x3<half>, framework::DatasetMode::ALL, combine(combine(combine(combine(datasets::SmallDepthwiseConvolutionLayerDataset3x3(),
-                                                                                                                   depth_multipliers),
-                                                                                                                   framework::dataset::make("DataType",
-                                                                                                                           DataType::F16)),
-                                                                                                                   framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
-                                                                                                                   ActivationFunctionsDataset))
+using NEDepthwiseConvolutionLayerFixtureOptimized = DepthwiseConvolutionLayerValidationFixture<Tensor, Accessor, NEDepthwiseConvolutionLayerOptimized, T>;
+TEST_SUITE(W3x3)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerFixtureOptimized<half>, framework::DatasetMode::ALL, combine(combine(combine(combine(datasets::SmallDepthwiseConvolutionLayerDataset3x3(),
+                       depth_multipliers),
+                       framework::dataset::make("DataType",
+                                                DataType::F16)),
+                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
+                       ActivationFunctionsDataset))
 {
     validate(Accessor(_target), _reference, tolerance_f16);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixture3x3<half>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(datasets::LargeDepthwiseConvolutionLayerDataset3x3(),
-                                                                                                                       depth_multipliers),
-                                                                                                                       framework::dataset::make("DataType",
-                                                                                                                               DataType::F16)),
-                                                                                                                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
-                                                                                                                       ActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixtureOptimized<half>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(datasets::LargeDepthwiseConvolutionLayerDataset3x3(),
+                                                       depth_multipliers),
+                                               framework::dataset::make("DataType",
+                                                                        DataType::F16)),
+                                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
+                               ActivationFunctionsDataset))
 {
     validate(Accessor(_target), _reference, tolerance_f16);
 }
 
 TEST_SUITE(Dilation)
 
-FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerFixture3x3<half>, framework::DatasetMode::ALL, combine(combine(combine(combine(datasets::SmallDepthwiseDilatedConvolutionLayerDataset3x3(),
-                                                                                                                   depth_multipliers),
-                                                                                                                   framework::dataset::make("DataType",
-                                                                                                                           DataType::F16)),
-                                                                                                                   framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
-                                                                                                                   ActivationFunctionsDataset))
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerFixtureOptimized<half>, framework::DatasetMode::ALL,
+                       combine(combine(combine(combine(datasets::SmallDepthwiseDilatedConvolutionLayerDataset3x3(),
+                                                       depth_multipliers),
+                                               framework::dataset::make("DataType",
+                                                                        DataType::F16)),
+                                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
+                               ActivationFunctionsDataset))
 {
     validate(Accessor(_target), _reference, tolerance_f16);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixture3x3<half>, framework::DatasetMode::NIGHTLY,
+FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixtureOptimized<half>, framework::DatasetMode::NIGHTLY,
                        combine(combine(combine(combine(datasets::LargeDepthwiseDilatedConvolutionLayerDataset3x3(),
                                                        depth_multipliers),
                                                framework::dataset::make("DataType",
@@ -452,8 +469,10 @@ FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixture3x3<half>, fr
 }
 
 TEST_SUITE_END() // Dilation
+TEST_SUITE_END() // W3x3
 
-FIXTURE_DATA_TEST_CASE(RunOptimizedSmall, NEDepthwiseConvolutionLayerFixture3x3<half>, framework::DatasetMode::PRECOMMIT,
+TEST_SUITE(Optimized)
+FIXTURE_DATA_TEST_CASE(RunSmallW3x3, NEDepthwiseConvolutionLayerFixtureOptimized<half>, framework::DatasetMode::PRECOMMIT,
                        combine(combine(combine(combine(datasets::SmallOptimizedDepthwiseConvolutionLayerDataset3x3(),
                                                        framework::dataset::make("DepthMultiplier", 1)),
                                                framework::dataset::make("DataType",
@@ -463,7 +482,17 @@ FIXTURE_DATA_TEST_CASE(RunOptimizedSmall, NEDepthwiseConvolutionLayerFixture3x3<
 {
     validate(Accessor(_target), _reference, tolerance_f16);
 }
-FIXTURE_DATA_TEST_CASE(RunOptimizedLarge, NEDepthwiseConvolutionLayerFixture3x3<half>, framework::DatasetMode::NIGHTLY,
+FIXTURE_DATA_TEST_CASE(RunSmallW5x5, NEDepthwiseConvolutionLayerFixtureOptimized<half>, framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(combine(datasets::SmallOptimizedDepthwiseConvolutionLayerDataset5x5(),
+                                                       framework::dataset::make("DepthMultiplier", 1)),
+                                               framework::dataset::make("DataType",
+                                                                        DataType::F16)),
+                                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
+                               ActivationFunctionsDataset))
+{
+    validate(Accessor(_target), _reference, tolerance_f16);
+}
+FIXTURE_DATA_TEST_CASE(RunLargeW3x3, NEDepthwiseConvolutionLayerFixtureOptimized<half>, framework::DatasetMode::NIGHTLY,
                        combine(combine(combine(combine(datasets::LargeOptimizedDepthwiseConvolutionLayerDataset3x3(),
                                                        framework::dataset::make("DepthMultiplier", 1)),
                                                framework::dataset::make("DataType",
@@ -473,14 +502,14 @@ FIXTURE_DATA_TEST_CASE(RunOptimizedLarge, NEDepthwiseConvolutionLayerFixture3x3<
 {
     validate(Accessor(_target), _reference, tolerance_f16);
 }
-TEST_SUITE_END() // W3x3
+TEST_SUITE_END() // Optimized
 TEST_SUITE_END() // FP16
 #endif           // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 
 TEST_SUITE_END() // Float
 
 template <typename T>
-using NEDepthwiseConvolutionLayerQuantizedFixture3x3 = DepthwiseConvolutionLayerValidationQuantizedFixture<Tensor, Accessor, NEDepthwiseConvolutionLayerOptimized, T>;
+using NEDepthwiseConvolutionLayerQuantizedFixtureOptimized = DepthwiseConvolutionLayerValidationQuantizedFixture<Tensor, Accessor, NEDepthwiseConvolutionLayerOptimized, T>;
 template <typename T>
 using NEDepthwiseConvolutionLayerQuantizedFixture = DepthwiseConvolutionLayerValidationQuantizedFixture<Tensor, Accessor, NEDepthwiseConvolutionLayer, T>;
 
@@ -522,10 +551,10 @@ FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerQuantizedFixture<uin
 {
     validate(Accessor(_target), _reference, tolerance_qasymm8);
 }
-TEST_SUITE_END() //Dilation
+TEST_SUITE_END() // Dilation
 TEST_SUITE_END() // Generic
 TEST_SUITE(W3x3)
-FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerQuantizedFixture3x3<uint8_t>, framework::DatasetMode::PRECOMMIT,
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerQuantizedFixtureOptimized<uint8_t>, framework::DatasetMode::PRECOMMIT,
                        combine(combine(combine(combine(combine(combine(datasets::SmallDepthwiseConvolutionLayerDataset3x3(), depth_multipliers),
                                                                framework::dataset::make("DataType", DataType::QASYMM8)),
                                                        framework::dataset::make("SrcQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
@@ -535,11 +564,10 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerQuantizedFixture3x3<
 {
     validate(Accessor(_target), _reference, tolerance_qasymm8);
 }
-FIXTURE_DATA_TEST_CASE(RunOptimizedSmall, NEDepthwiseConvolutionLayerQuantizedFixture3x3<uint8_t>, framework::DatasetMode::PRECOMMIT,
-                       combine(combine(combine(combine(combine(combine(datasets::SmallOptimizedDepthwiseConvolutionLayerDataset3x3(),
-                                                                       framework::dataset::make("DepthMultiplier", 1)),
-                                                               framework::dataset::make("DataType",
-                                                                                        DataType::QASYMM8)),
+FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerQuantizedFixtureOptimized<uint8_t>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(datasets::LargeDepthwiseConvolutionLayerDataset3x3(),
+                                                                       depth_multipliers),
+                                                               framework::dataset::make("DataType", DataType::QASYMM8)),
                                                        framework::dataset::make("SrcQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
                                                framework::dataset::make("DstQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
                                        framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
@@ -547,20 +575,21 @@ FIXTURE_DATA_TEST_CASE(RunOptimizedSmall, NEDepthwiseConvolutionLayerQuantizedFi
 {
     validate(Accessor(_target), _reference, tolerance_qasymm8);
 }
-FIXTURE_DATA_TEST_CASE(RunOptimizedLarge, NEDepthwiseConvolutionLayerQuantizedFixture3x3<uint8_t>, framework::DatasetMode::NIGHTLY,
-                       combine(combine(combine(combine(combine(combine(datasets::LargeOptimizedDepthwiseConvolutionLayerDataset3x3(),
-                                                                       framework::dataset::make("DepthMultiplier", 1)),
-                                                               framework::dataset::make("DataType",
-                                                                                        DataType::QASYMM8)),
+
+TEST_SUITE(Dilation)
+
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerQuantizedFixtureOptimized<uint8_t>, framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(combine(combine(combine(datasets::SmallDepthwiseDilatedConvolutionLayerDataset3x3(), depth_multipliers),
+                                                               framework::dataset::make("DataType", DataType::QASYMM8)),
                                                        framework::dataset::make("SrcQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
-                                               framework::dataset::make("DstQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
+                                               framework::dataset::make("DstQuantizationInfo", { QuantizationInfo(0.7f, 10) })),
                                        framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
                                ActivationFunctionsDataset))
 {
     validate(Accessor(_target), _reference, tolerance_qasymm8);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerQuantizedFixture3x3<uint8_t>, framework::DatasetMode::NIGHTLY,
-                       combine(combine(combine(combine(combine(combine(datasets::LargeDepthwiseConvolutionLayerDataset3x3(),
+FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerQuantizedFixtureOptimized<uint8_t>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(datasets::LargeDepthwiseDilatedConvolutionLayerDataset3x3(),
                                                                        depth_multipliers),
                                                                framework::dataset::make("DataType", DataType::QASYMM8)),
                                                        framework::dataset::make("SrcQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
@@ -570,23 +599,27 @@ FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerQuantizedFixture3x3<
 {
     validate(Accessor(_target), _reference, tolerance_qasymm8);
 }
+TEST_SUITE_END() // Dilation
+TEST_SUITE_END() // W3x3
 
-TEST_SUITE(Dilation)
-
-FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerQuantizedFixture3x3<uint8_t>, framework::DatasetMode::PRECOMMIT,
-                       combine(combine(combine(combine(combine(combine(datasets::SmallDepthwiseDilatedConvolutionLayerDataset3x3(), depth_multipliers),
-                                                               framework::dataset::make("DataType", DataType::QASYMM8)),
+TEST_SUITE(Optimized)
+FIXTURE_DATA_TEST_CASE(RunSmall3x3, NEDepthwiseConvolutionLayerQuantizedFixtureOptimized<uint8_t>, framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(combine(combine(combine(datasets::SmallOptimizedDepthwiseConvolutionLayerDataset3x3(),
+                                                                       framework::dataset::make("DepthMultiplier", 1)),
+                                                               framework::dataset::make("DataType",
+                                                                                        DataType::QASYMM8)),
                                                        framework::dataset::make("SrcQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
-                                               framework::dataset::make("DstQuantizationInfo", { QuantizationInfo(0.7f, 10) })),
+                                               framework::dataset::make("DstQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
                                        framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
                                ActivationFunctionsDataset))
 {
     validate(Accessor(_target), _reference, tolerance_qasymm8);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerQuantizedFixture3x3<uint8_t>, framework::DatasetMode::NIGHTLY,
-                       combine(combine(combine(combine(combine(combine(datasets::LargeDepthwiseDilatedConvolutionLayerDataset3x3(),
-                                                                       depth_multipliers),
-                                                               framework::dataset::make("DataType", DataType::QASYMM8)),
+FIXTURE_DATA_TEST_CASE(RunSmall5x5, NEDepthwiseConvolutionLayerQuantizedFixtureOptimized<uint8_t>, framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(combine(combine(combine(datasets::SmallOptimizedDepthwiseConvolutionLayerDataset5x5(),
+                                                                       framework::dataset::make("DepthMultiplier", 1)),
+                                                               framework::dataset::make("DataType",
+                                                                                        DataType::QASYMM8)),
                                                        framework::dataset::make("SrcQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
                                                framework::dataset::make("DstQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
                                        framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
@@ -594,8 +627,19 @@ FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerQuantizedFixture3x3<
 {
     validate(Accessor(_target), _reference, tolerance_qasymm8);
 }
-TEST_SUITE_END() // Dilation
-TEST_SUITE_END() // W3x3
+FIXTURE_DATA_TEST_CASE(RunLarge3x3, NEDepthwiseConvolutionLayerQuantizedFixtureOptimized<uint8_t>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(datasets::LargeOptimizedDepthwiseConvolutionLayerDataset3x3(),
+                                                                       framework::dataset::make("DepthMultiplier", 1)),
+                                                               framework::dataset::make("DataType",
+                                                                                        DataType::QASYMM8)),
+                                                       framework::dataset::make("SrcQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
+                                               framework::dataset::make("DstQuantizationInfo", { QuantizationInfo(0.5f, 10) })),
+                                       framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
+                               ActivationFunctionsDataset))
+{
+    validate(Accessor(_target), _reference, tolerance_qasymm8);
+}
+TEST_SUITE_END() // Optimized
 TEST_SUITE_END() // QASYMM8
 TEST_SUITE_END() // Quantized