COMPMID-873: Integrate RSH NEON Depthwise Convolution routine

Change-Id: Ida1e9a836bc518bfe5563e16bf7f92bde5fc13f7
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/118472
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>

3 files changed, 27 insertions, 4 deletions

diff --git a/tests/datasets/DepthwiseConvolutionLayerDataset.h b/tests/datasets/DepthwiseConvolutionLayerDataset.h
index d3eb2c5d9e..f1dfb981aa 100644
--- a/tests/datasets/DepthwiseConvolutionLayerDataset.h
+++ b/tests/datasets/DepthwiseConvolutionLayerDataset.h
@@ -174,6 +174,23 @@ public:
         add_config(TensorShape(177U, 311U, 22U), TensorShape(3U, 3U, 22U), TensorShape(89U, 311U, 22U), PadStrideInfo(2, 1, 1, 1));
     }
 };
+class OptimizedDepthwiseConvolutionLayerDataset3x3 final : public DepthwiseConvolutionLayerDataset
+{
+public:
+    OptimizedDepthwiseConvolutionLayerDataset3x3()
+    {
+        // Stride 1
+        add_config(TensorShape(7U, 7U, 16U), TensorShape(3U, 3U, 16U), TensorShape(5U, 5U, 16U), PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL));
+        add_config(TensorShape(7U, 7U, 16U), TensorShape(3U, 3U, 16U), TensorShape(7U, 7U, 16U), PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL));
+        add_config(TensorShape(28U, 28U, 16U), TensorShape(3U, 3U, 16U), TensorShape(26U, 26U, 16U), PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL));
+        add_config(TensorShape(28U, 28U, 16U), TensorShape(3U, 3U, 16U), TensorShape(28U, 28U, 16U), PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL));
+        // Stride 2
+        add_config(TensorShape(7U, 7U, 32U), TensorShape(3U, 3U, 32U), TensorShape(3U, 3U, 32U), PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL));
+        add_config(TensorShape(7U, 7U, 32U), TensorShape(3U, 3U, 32U), TensorShape(4U, 4U, 32U), PadStrideInfo(2, 2, 1, 1, 1, 1, DimensionRoundingType::CEIL));
+        add_config(TensorShape(8U, 8U, 32U), TensorShape(3U, 3U, 32U), TensorShape(3U, 3U, 32U), PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL));
+        add_config(TensorShape(8U, 8U, 32U), TensorShape(3U, 3U, 32U), TensorShape(4U, 4U, 32U), PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::CEIL));
+    }
+};
 } // namespace datasets
 } // namespace test
 } // namespace arm_compute
diff --git a/tests/validation/NEON/DepthwiseConvolutionLayer.cpp b/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
index f8c04dab3e..47e8896fd6 100644
--- a/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
+++ b/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
@@ -121,6 +121,12 @@ FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixture3x3<float>, f
 {
     validate(Accessor(_target), _reference, tolerance_f32);
 }
+FIXTURE_DATA_TEST_CASE(RunOptimized, NEDepthwiseConvolutionLayerFixture3x3<float>, framework::DatasetMode::ALL, combine(datasets::OptimizedDepthwiseConvolutionLayerDataset3x3(),
+                                                                                                                        framework::dataset::make("DataType",
+                                                                                                                                DataType::F32)))
+{
+    validate(Accessor(_target), _reference, tolerance_f32);
+}
 TEST_SUITE_END()
 TEST_SUITE_END()
 
diff --git a/tests/validation/reference/DepthwiseConvolutionLayer.cpp b/tests/validation/reference/DepthwiseConvolutionLayer.cpp
index 66e3a4b783..ffea1bcf89 100644
--- a/tests/validation/reference/DepthwiseConvolutionLayer.cpp
+++ b/tests/validation/reference/DepthwiseConvolutionLayer.cpp
@@ -67,10 +67,10 @@ SimpleTensor<T> depthwise_convolution(const SimpleTensor<T> &src, const SimpleTe
     const int filter_half_width  = filter_width / 2;
     const int filter_half_height = filter_height / 2;
 
-    const int pad_left   = std::min(static_cast<int>(conv_info.pad_left()), filter_half_width);
-    const int pad_top    = std::min(static_cast<int>(conv_info.pad_top()), filter_half_height);
-    const int pad_right  = std::min(static_cast<int>(conv_info.pad_right()), filter_half_width);
-    const int pad_bottom = std::min(static_cast<int>(conv_info.pad_bottom()), filter_half_height);
+    const int pad_left   = conv_info.pad_left();
+    const int pad_top    = conv_info.pad_top();
+    const int pad_right  = conv_info.pad_right();
+    const int pad_bottom = conv_info.pad_bottom();
 
     const int minimum_x = -pad_left + filter_half_width;
     const int minimum_y = -pad_top + filter_half_height;