Optimize CL and Neon depthwise convolution tests

Resolves: COMPMID-6465

Change-Id: I5bbf4596dd5e34e806dc51de9be14df9b6fa320a
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10452
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 58 insertions, 22 deletions

diff --git a/tests/datasets/DepthwiseConvolutionLayerDataset.h b/tests/datasets/DepthwiseConvolutionLayerDataset.h
index 0d58e95911..17e03368ac 100644
--- a/tests/datasets/DepthwiseConvolutionLayerDataset.h
+++ b/tests/datasets/DepthwiseConvolutionLayerDataset.h
@@ -141,12 +141,12 @@ public:
         add_config(TensorShape(17U, 31U, 2U), Size2D(13U, 9U), PadStrideInfo(1, 2, 1, 1));
         add_config(TensorShape(23U, 27U, 5U), Size2D(11U, 3U), PadStrideInfo(1, 2, 0, 0));
         add_config(TensorShape(17U, 31U, 2U, 3U), Size2D(5U, 9U), PadStrideInfo(1, 2, 1, 1));
-        add_config(TensorShape(233U, 277U, 55U), Size2D(1U, 1U), PadStrideInfo(2, 1, 0, 0));
-        add_config(TensorShape(333U, 277U, 77U), Size2D(1U, 1U), PadStrideInfo(3, 2, 1, 0));
-        add_config(TensorShape(177U, 311U, 22U), Size2D(3U, 4U), PadStrideInfo(1, 2, 1, 1));
-        add_config(TensorShape(233U, 277U, 55U), Size2D(3U, 4U), PadStrideInfo(1, 2, 0, 0));
-        add_config(TensorShape(333U, 277U, 77U), Size2D(3U, 4U), PadStrideInfo(2, 3, 0, 1));
-        add_config(TensorShape(177U, 311U, 22U), Size2D(3U, 4U), PadStrideInfo(2, 1, 1, 1));
+        add_config(TensorShape(133U, 127U, 55U), Size2D(1U, 1U), PadStrideInfo(2, 1, 0, 0));
+        add_config(TensorShape(233U, 109U, 77U), Size2D(1U, 1U), PadStrideInfo(3, 2, 1, 0));
+        add_config(TensorShape(177U, 111U, 22U), Size2D(3U, 4U), PadStrideInfo(1, 2, 1, 1));
+        add_config(TensorShape(233U, 87U, 55U), Size2D(3U, 4U), PadStrideInfo(1, 2, 0, 0));
+        add_config(TensorShape(333U, 79U, 77U), Size2D(3U, 4U), PadStrideInfo(2, 3, 0, 1));
+        add_config(TensorShape(67U, 211U, 22U), Size2D(3U, 4U), PadStrideInfo(2, 1, 1, 1));
         // Asymmetric padding
         add_config(TensorShape(33U, 27U, 7U), Size2D(5U, 7U), PadStrideInfo(3, 2, 2, 1, 2, 0, DimensionRoundingType::FLOOR));
         add_config(TensorShape(33U, 27U, 7U), Size2D(5U, 7U), PadStrideInfo(3, 2, 1, 3, 0, 2, DimensionRoundingType::FLOOR));
@@ -157,6 +157,24 @@ public:
     }
 };
 
+class LargeDepthwiseConvolutionLayerDatasetFp16Subset final : public DepthwiseConvolutionLayerDataset
+{
+public:
+    LargeDepthwiseConvolutionLayerDatasetFp16Subset()
+    {
+        add_config(TensorShape(33U, 27U, 11U), Size2D(3U, 4U), PadStrideInfo(1, 2, 0, 1));
+        add_config(TensorShape(17U, 31U, 2U, 3U), Size2D(5U, 9U), PadStrideInfo(1, 2, 1, 1));
+        add_config(TensorShape(233U, 109U, 77U), Size2D(1U, 1U), PadStrideInfo(3, 2, 1, 0));
+        add_config(TensorShape(177U, 111U, 22U), Size2D(3U, 4U), PadStrideInfo(1, 2, 1, 1));
+        add_config(TensorShape(67U, 211U, 22U), Size2D(3U, 4U), PadStrideInfo(2, 1, 1, 1));
+        // Asymmetric padding
+        add_config(TensorShape(33U, 27U, 7U), Size2D(5U, 7U), PadStrideInfo(3, 2, 1, 3, 0, 2, DimensionRoundingType::FLOOR));
+        add_config(TensorShape(33U, 27U, 7U), Size2D(5U, 7U), PadStrideInfo(3, 2, 1, 0, 1, 0, DimensionRoundingType::FLOOR));
+        // Padding greater than kernel size.
+        add_config(TensorShape(128, 56, 56), Size2D(4, 4), PadStrideInfo(2, 2, 0, 10, 0, 10, DimensionRoundingType::FLOOR));
+    }
+};
+
 /** Dataset containing large kernel size for generic depthwise convolution. */
 class LargeKernelSizeDepthwiseConvolutionLayerNHWCDataset final : public DepthwiseConvolutionLayerDataset
 {
@@ -198,21 +216,39 @@ class LargeDepthwiseConvolutionLayerDataset3x3 final : public DepthwiseConvoluti
 public:
     LargeDepthwiseConvolutionLayerDataset3x3()
     {
-        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(1, 1, 1, 1));
         add_config(TensorShape(21U, 31U, 9U, 4U), Size2D(3U, 3U), PadStrideInfo(1, 2, 1, 0));
-        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(1, 2, 0, 1));
-        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(1, 2, 1, 1));
         add_config(TensorShape(21U, 31U, 9U, 4U), Size2D(3U, 3U), PadStrideInfo(2, 1, 1, 0));
+        add_config(TensorShape(21U, 31U, 9U, 4U), Size2D(3U, 3U), PadStrideInfo(2, 2, 1, 2));
+
+        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(1, 1, 1, 1));
+        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(1, 2, 1, 2));
+        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(2, 1, 0, 1));
+        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(2, 2, 2, 1));
+
+        add_config(TensorShape(77U, 209U, 22U), Size2D(3U, 3U), PadStrideInfo(1, 2, 1, 1));
+        add_config(TensorShape(123U, 76U, 55U), Size2D(3U, 3U), PadStrideInfo(1, 2, 0, 0));
+        add_config(TensorShape(133U, 277U, 77U), Size2D(3U, 3U), PadStrideInfo(2, 3, 0, 0));
+        add_config(TensorShape(77U, 95U, 22U), Size2D(3U, 3U), PadStrideInfo(2, 1, 1, 1));
+
+        // Width and height are a multiple of the processing tile size
+        add_config(TensorShape(32U, 21U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(1, 1, 0, 1));
+    }
+};
+
+class LargeDepthwiseConvolutionLayerDataset3x3Fp16Subset final : public DepthwiseConvolutionLayerDataset
+{
+public:
+    LargeDepthwiseConvolutionLayerDataset3x3Fp16Subset()
+    {
+        add_config(TensorShape(21U, 31U, 9U, 4U), Size2D(3U, 3U), PadStrideInfo(2, 2, 1, 2));
+
+        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(1, 2, 1, 2));
         add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(2, 1, 0, 1));
-        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(2, 1, 1, 1));
-        add_config(TensorShape(21U, 31U, 9U, 4U), Size2D(3U, 3U), PadStrideInfo(2, 2, 1, 0));
-        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(2, 2, 0, 1));
-        add_config(TensorShape(33U, 27U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(2, 2, 1, 1));
-        add_config(TensorShape(177U, 311U, 22U), Size2D(3U, 3U), PadStrideInfo(1, 2, 1, 1));
-        add_config(TensorShape(233U, 277U, 55U), Size2D(3U, 3U), PadStrideInfo(1, 2, 0, 0));
-        add_config(TensorShape(333U, 277U, 77U), Size2D(3U, 3U), PadStrideInfo(2, 3, 0, 0));
-        add_config(TensorShape(177U, 311U, 22U), Size2D(3U, 3U), PadStrideInfo(2, 1, 1, 1));
-        // Width and height are a multipile of the processing tile size
+
+        add_config(TensorShape(123U, 76U, 55U), Size2D(3U, 3U), PadStrideInfo(1, 2, 0, 0));
+        add_config(TensorShape(77U, 95U, 22U), Size2D(3U, 3U), PadStrideInfo(2, 1, 1, 1));
+
+        // Width and height are a multiple of the processing tile size
         add_config(TensorShape(32U, 21U, 11U, 3U), Size2D(3U, 3U), PadStrideInfo(1, 1, 0, 1));
     }
 };
@@ -241,14 +277,14 @@ public:
     LargeOptimizedDepthwiseConvolutionLayerDataset3x3()
     {
         // Stride 1
-        add_config(TensorShape(233U, 277U, 16U), Size2D(3U, 3U), PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL));
-        add_config(TensorShape(233U, 7U, 16U), Size2D(3U, 3U), PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL));
+        add_config(TensorShape(233U, 173U, 16U), Size2D(3U, 3U), PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL));
+        add_config(TensorShape(133U, 7U, 16U), Size2D(3U, 3U), PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL));
         add_config(TensorShape(7U, 7U, 21U), Size2D(3U, 3U), PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL));
         add_config(TensorShape(28U, 28U, 16U), Size2D(3U, 3U), PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL));
         add_config(TensorShape(28U, 28U, 16U), Size2D(3U, 3U), PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL));
         // Stride 2
-        add_config(TensorShape(233U, 277U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL));
-        add_config(TensorShape(233U, 277U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 1, 1, 1, 1, DimensionRoundingType::CEIL));
+        add_config(TensorShape(133U, 97U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL));
+        add_config(TensorShape(153U, 77U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 1, 1, 1, 1, DimensionRoundingType::CEIL));
         add_config(TensorShape(8U, 8U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::FLOOR));
         add_config(TensorShape(8U, 8U, 32U), Size2D(3U, 3U), PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::CEIL));
         add_config(TensorShape(8U, 8U, 33U), Size2D(3U, 3U), PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::CEIL));