COMPMID-1340 - Implementing Winograd Convolution Layer 1x5/5x1 on OpenCL NHWC

Change-Id: Id5e0795238f77c049df9c109dafc5ef878c1897d
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/139234
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

2 files changed, 42 insertions, 4 deletions

diff --git a/tests/datasets/WinogradOutputTransformDataset.h b/tests/datasets/WinogradOutputTransformDataset.h
index 4085e91854..617a67de07 100644
--- a/tests/datasets/WinogradOutputTransformDataset.h
+++ b/tests/datasets/WinogradOutputTransformDataset.h
@@ -206,6 +206,22 @@ public:
         add_config(TensorShape(7U, 2U, 64U, 3U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
         add_config(TensorShape(24U, 9U, 64U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NHWC));
         add_config(TensorShape(7U, 2U, 64U, 5U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+
+        // (4x1, 5x1)
+        add_config(TensorShape(13U, 6U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 22U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(5U, 462U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(53U, 33U), PadStrideInfo(1, 1, 2, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 10U, 8U, 3U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(24U, 42U, 8U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 20U, 8U, 5U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(8U, 10U), PadStrideInfo(1, 1, 2, 0), DataLayout::NHWC));
+
+        // (1x4, 1x5)
+        add_config(TensorShape(13U, 7U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 20U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(5U, 477U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(53U, 33U), PadStrideInfo(1, 1, 0, 2), DataLayout::NHWC));
+        add_config(TensorShape(7U, 16U, 8U, 3U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(24U, 42U, 8U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(14U, 14U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+        add_config(TensorShape(7U, 24U, 8U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 2), DataLayout::NHWC));
     }
 };
 
@@ -316,6 +332,18 @@ public:
         add_config(TensorShape(13U, 182U, 64U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
         add_config(TensorShape(32U, 756U, 64U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
         add_config(TensorShape(13U, 182U, 64U, 5U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+
+        // (4x1, 5x1)
+        add_config(TensorShape(32U, 3136U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(112U, 112U), PadStrideInfo(1, 1, 2, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+        add_config(TensorShape(32U, 3024U, 8U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(112U, 112U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 8U, 5U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+
+        // (1x4, 1x5)
+        add_config(TensorShape(32U, 3136U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 0, 2), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+        add_config(TensorShape(32U, 3024U, 8U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 8U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
     }
 };
 } // namespace datasets
diff --git a/tests/validation/CL/Winograd.cpp b/tests/validation/CL/Winograd.cpp
index c39cb4e790..004d7c98a0 100644
--- a/tests/validation/CL/Winograd.cpp
+++ b/tests/validation/CL/Winograd.cpp
@@ -72,7 +72,9 @@ const auto SmallWinogradInputTransformDatasetNCHW =
 const auto SmallWinogradInputTransformDatasetNHWC = framework::dataset::concat(datasets::SmallWinogradInputTransformDataset4x4_3x3(),
                                                     framework::dataset::concat(datasets::SmallWinogradInputTransformDataset4x1_3x1(),
                                                     framework::dataset::concat(datasets::SmallWinogradInputTransformDataset1x4_1x3(),
-                                                                               datasets::SmallWinogradInputTransformDataset4x4_5x5())));
+                                                    framework::dataset::concat(datasets::SmallWinogradInputTransformDataset4x4_5x5(),
+                                                    framework::dataset::concat(datasets::SmallWinogradInputTransformDataset4x1_5x1(),
+                                                                               datasets::SmallWinogradInputTransformDataset1x4_1x5())))));
 
 const auto LargeWinogradInputTransformDatasetNCHW =
            framework::dataset::concat(datasets::LargeWinogradInputTransformDataset2x2_3x3(),
@@ -87,7 +89,9 @@ const auto LargeWinogradInputTransformDatasetNCHW =
 
 const auto LargeWinogradInputTransformDatasetNHWC =
            framework::dataset::concat(datasets::LargeWinogradInputTransformDataset4x4_3x3(),
-                                      datasets::LargeWinogradInputTransformDataset4x4_5x5());
+           framework::dataset::concat(datasets::LargeWinogradInputTransformDataset4x4_5x5(),
+           framework::dataset::concat(datasets::LargeWinogradInputTransformDataset4x1_5x1(),
+                                      datasets::LargeWinogradInputTransformDataset1x4_1x5())));
 
 // Filter transform
 const auto SmallWinogradFilterTransformDatasetNCHW =
@@ -102,7 +106,9 @@ const auto SmallWinogradFilterTransformDatasetNHWC =
            framework::dataset::concat(combine(datasets::Small3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 4U) })),
            framework::dataset::concat(combine(datasets::Small3x1Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 1U) })),
            framework::dataset::concat(combine(datasets::Small1x3Shapes(), framework::dataset::make("OutputTile", { Size2D(1U, 4U) })),
-                                      combine(datasets::Small5x5Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 4U) })))));
+           framework::dataset::concat(combine(datasets::Small5x5Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 4U) })),
+           framework::dataset::concat(combine(datasets::Small5x1Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 1U) })),
+                                      combine(datasets::Small1x5Shapes(), framework::dataset::make("OutputTile", { Size2D(1U, 4U) })))))));
 
 const auto LargeWinogradFilterTransformDatasetNCHW =
            framework::dataset::concat(combine(datasets::Large3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 2U), Size2D(4U, 4U) })),
@@ -114,7 +120,11 @@ const auto LargeWinogradFilterTransformDatasetNCHW =
 
 const auto LargeWinogradFilterTransformDatasetNHWC =
            framework::dataset::concat(combine(datasets::Large3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 4U) })),
-                                      combine(datasets::Large5x5Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 4U) })));
+           framework::dataset::concat(combine(datasets::Large3x1Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 1U) })),
+           framework::dataset::concat(combine(datasets::Large1x3Shapes(), framework::dataset::make("OutputTile", { Size2D(1U, 4U) })),
+           framework::dataset::concat(combine(datasets::Large5x5Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 4U) })),
+           framework::dataset::concat(combine(datasets::Large5x1Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 1U) })),
+                                      combine(datasets::Large1x5Shapes(), framework::dataset::make("OutputTile", { Size2D(1U, 4U) })))))));
 
 // Output transform
 const auto SmallWinogradOutputTransformDatasetNCHW = datasets::SmallWinogradOutputTransformDatasetNCHW();