COMPMID-1339 - Implementing Winograd Convolution Layer 1x5 and 5x1 kernels on OpenCL NCHW

Change-Id: Ia293cd89651146a0e27e5f7c74ca9c924807e83c
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/138707
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

7 files changed, 268 insertions, 4 deletions

diff --git a/tests/datasets/LargeConvolutionLayerDataset.h b/tests/datasets/LargeConvolutionLayerDataset.h
index ae25c8cd66..3eb98dbeea 100644
--- a/tests/datasets/LargeConvolutionLayerDataset.h
+++ b/tests/datasets/LargeConvolutionLayerDataset.h
@@ -122,6 +122,42 @@ public:
     }
 };
 
+class LargeWinogradConvolutionLayer5x1Dataset final : public ConvolutionLayerDataset
+{
+public:
+    LargeWinogradConvolutionLayer5x1Dataset()
+    {
+        // Batch size 1
+        add_config(TensorShape(224U, 224U, 3U), TensorShape(5U, 1U, 3U, 64U), TensorShape(64U), TensorShape(222U, 224U, 64U), PadStrideInfo(1, 1, 1, 0));
+        add_config(TensorShape(123U, 134U, 16U), TensorShape(5U, 1U, 16U, 7U), TensorShape(7U), TensorShape(121U, 134U, 7U), PadStrideInfo(1, 1, 1, 0));
+        add_config(TensorShape(181U, 152U, 42U), TensorShape(5U, 1U, 42U, 100U), TensorShape(100U), TensorShape(177U, 152U, 100U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(200U, 201U, 24U), TensorShape(5U, 1U, 24U, 61), TensorShape(61U), TensorShape(200U, 201U, 61), PadStrideInfo(1, 1, 2, 0));
+
+        // Batch size 2, 3 and 4
+        add_config(TensorShape(224U, 224U, 3U, 2U), TensorShape(5U, 1U, 3U, 64U), TensorShape(64U), TensorShape(222U, 224U, 64U, 2U), PadStrideInfo(1, 1, 1, 0));
+        add_config(TensorShape(123U, 134U, 16U, 3U), TensorShape(5U, 1U, 16U, 7U), TensorShape(7U), TensorShape(121U, 134U, 7U, 3U), PadStrideInfo(1, 1, 1, 0));
+        add_config(TensorShape(181U, 152U, 42U, 4U), TensorShape(5U, 1U, 42U, 100U), TensorShape(100U), TensorShape(177U, 152U, 100U, 4U), PadStrideInfo(1, 1, 0, 0));
+    }
+};
+
+class LargeWinogradConvolutionLayer1x5Dataset final : public ConvolutionLayerDataset
+{
+public:
+    LargeWinogradConvolutionLayer1x5Dataset()
+    {
+        // Batch size 1
+        add_config(TensorShape(224U, 224U, 3U), TensorShape(1U, 5U, 3U, 64U), TensorShape(64U), TensorShape(224U, 222U, 64U), PadStrideInfo(1, 1, 0, 1));
+        add_config(TensorShape(123U, 134U, 16U), TensorShape(1U, 5U, 16U, 7U), TensorShape(7U), TensorShape(123U, 130U, 7U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(181U, 152U, 42U), TensorShape(1U, 5U, 42U, 100U), TensorShape(100U), TensorShape(181U, 148U, 100U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(200U, 201U, 24U), TensorShape(1U, 5U, 24U, 61), TensorShape(61U), TensorShape(200U, 201U, 61), PadStrideInfo(1, 1, 0, 2));
+
+        // Batch size 2, 3 and 4
+        add_config(TensorShape(224U, 224U, 3U, 2U), TensorShape(1U, 5U, 3U, 64U), TensorShape(64U), TensorShape(224U, 222U, 64U, 2U), PadStrideInfo(1, 1, 0, 1));
+        add_config(TensorShape(123U, 134U, 16U, 3U), TensorShape(1U, 5U, 16U, 7U), TensorShape(7U), TensorShape(123U, 130U, 7U, 3U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(181U, 152U, 42U, 4U), TensorShape(1U, 5U, 42U, 100U), TensorShape(100U), TensorShape(181U, 148U, 100U, 4U), PadStrideInfo(1, 1, 0, 0));
+    }
+};
+
 class LargeConvolutionLayerDataset final : public ConvolutionLayerDataset
 {
 public:
diff --git a/tests/datasets/ShapeDatasets.h b/tests/datasets/ShapeDatasets.h
index 766530b0d7..bc98b1e471 100644
--- a/tests/datasets/ShapeDatasets.h
+++ b/tests/datasets/ShapeDatasets.h
@@ -500,6 +500,70 @@ public:
     }
 };
 
+/** Data set containing small 5x1 tensor shapes. */
+class Small5x1Shapes final : public ShapeDataset
+{
+public:
+    Small5x1Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 5U, 1U, 7U, 4U },
+                     TensorShape{ 5U, 1U, 4U, 13U },
+                     TensorShape{ 5U, 1U, 9U, 2U },
+                     TensorShape{ 5U, 1U, 3U, 5U },
+    })
+    {
+    }
+};
+
+/** Data set containing large 5x1 tensor shapes. */
+class Large5x1Shapes final : public ShapeDataset
+{
+public:
+    Large5x1Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 5U, 1U, 32U, 64U },
+                     TensorShape{ 5U, 1U, 51U, 13U },
+                     TensorShape{ 5U, 1U, 53U, 47U },
+                     TensorShape{ 5U, 1U, 128U, 384U },
+    })
+    {
+    }
+};
+
+/** Data set containing small 1x5 tensor shapes. */
+class Small1x5Shapes final : public ShapeDataset
+{
+public:
+    Small1x5Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 1U, 5U, 7U, 4U },
+                     TensorShape{ 1U, 5U, 4U, 13U },
+                     TensorShape{ 1U, 5U, 9U, 2U },
+                     TensorShape{ 1U, 5U, 3U, 5U },
+    })
+    {
+    }
+};
+
+/** Data set containing large 1x5 tensor shapes. */
+class Large1x5Shapes final : public ShapeDataset
+{
+public:
+    Large1x5Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 1U, 5U, 32U, 64U },
+                     TensorShape{ 1U, 5U, 51U, 13U },
+                     TensorShape{ 1U, 5U, 53U, 47U },
+                     TensorShape{ 1U, 5U, 128U, 384U },
+    })
+    {
+    }
+};
+
 /** Data set containing small tensor shapes for deconvolution. */
 class SmallDeconvolutionShapes final : public ShapeDataset
 {
diff --git a/tests/datasets/SmallConvolutionLayerDataset.h b/tests/datasets/SmallConvolutionLayerDataset.h
index f05cc15c06..ae12dd4b16 100644
--- a/tests/datasets/SmallConvolutionLayerDataset.h
+++ b/tests/datasets/SmallConvolutionLayerDataset.h
@@ -92,6 +92,26 @@ public:
     }
 };
 
+class SmallWinogradConvolutionLayer5x1Dataset final : public ConvolutionLayerDataset
+{
+public:
+    SmallWinogradConvolutionLayer5x1Dataset()
+    {
+        add_config(TensorShape(8U, 8U, 2U), TensorShape(5U, 1U, 2U, 1U), TensorShape(1U), TensorShape(4U, 8U, 1U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(8U, 8U, 2U), TensorShape(5U, 1U, 2U), TensorShape(1U), TensorShape(8U, 8U, 1U), PadStrideInfo(1, 1, 2, 0));
+    }
+};
+
+class SmallWinogradConvolutionLayer1x5Dataset final : public ConvolutionLayerDataset
+{
+public:
+    SmallWinogradConvolutionLayer1x5Dataset()
+    {
+        add_config(TensorShape(8U, 8U, 2U), TensorShape(1U, 5U, 2U, 1U), TensorShape(1U), TensorShape(8U, 4U, 1U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(8U, 8U, 2U), TensorShape(1U, 5U, 2U), TensorShape(1U), TensorShape(8U, 8U, 1U), PadStrideInfo(1, 1, 0, 2));
+    }
+};
+
 class SmallConvolutionLayerDataset final : public ConvolutionLayerDataset
 {
 public:
diff --git a/tests/datasets/WinogradInputTransformDataset.h b/tests/datasets/WinogradInputTransformDataset.h
index ca23984a1d..bd910fb055 100644
--- a/tests/datasets/WinogradInputTransformDataset.h
+++ b/tests/datasets/WinogradInputTransformDataset.h
@@ -202,6 +202,36 @@ public:
     }
 };
 
+class SmallWinogradInputTransformDataset4x1_5x1 final : public WinogradInputTransformDataset
+{
+public:
+    SmallWinogradInputTransformDataset4x1_5x1()
+    {
+        add_config(TensorShape(9U, 9U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(9U, 9U), PadStrideInfo(1, 1, 2, 0), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(27U, 13U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 1U, 3U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(128U, 64U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 4U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(9U, 9U), PadStrideInfo(1, 1, 2, 0), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U, 4U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(27U, 13U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 5U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(14U, 14U, 512U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(14U, 14U), PadStrideInfo(1, 1, 2, 0), DataLayout::NCHW));
+    }
+};
+
+class SmallWinogradInputTransformDataset1x4_1x5 final : public WinogradInputTransformDataset
+{
+public:
+    SmallWinogradInputTransformDataset1x4_1x5()
+    {
+        add_config(TensorShape(9U, 9U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 2), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(27U, 13U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 1U, 3U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(128U, 64U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 4U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 2), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U, 4U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(27U, 13U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(14U, 14U, 512U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(14U, 14U), PadStrideInfo(1, 1, 0, 2), DataLayout::NCHW));
+    }
+};
+
 class LargeWinogradInputTransformDataset2x2_3x3 final : public WinogradInputTransformDataset
 {
 public:
@@ -285,6 +315,30 @@ public:
         add_config(TensorShape(83U, 72U, 14U, 5U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(83U, 72U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
     }
 };
+
+class LargeWinogradInputTransformDataset4x1_5x1 final : public WinogradInputTransformDataset
+{
+public:
+    LargeWinogradInputTransformDataset4x1_5x1()
+    {
+        add_config(TensorShape(42U, 37U, 8U, 15U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(42U, 37U), PadStrideInfo(1, 1, 2, 0), DataLayout::NCHW));
+        add_config(TensorShape(57U, 60U, 13U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(57U, 60U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 21U, 13U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(128U, 64U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(83U, 72U, 14U, 5U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(83U, 72U), PadStrideInfo(1, 1, 2, 0), DataLayout::NCHW));
+    }
+};
+
+class LargeWinogradInputTransformDataset1x4_1x5 final : public WinogradInputTransformDataset
+{
+public:
+    LargeWinogradInputTransformDataset1x4_1x5()
+    {
+        add_config(TensorShape(42U, 37U, 8U, 15U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(42U, 37U), PadStrideInfo(1, 1, 0, 2), DataLayout::NCHW));
+        add_config(TensorShape(57U, 60U, 13U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(57U, 60U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 21U, 13U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(128U, 64U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(83U, 72U, 14U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(83U, 72U), PadStrideInfo(1, 1, 0, 2), DataLayout::NCHW));
+    }
+};
 } // namespace datasets
 } // namespace test
 } // namespace arm_compute
diff --git a/tests/datasets/WinogradOutputTransformDataset.h b/tests/datasets/WinogradOutputTransformDataset.h
index a4689c6ef1..fc23e65258 100644
--- a/tests/datasets/WinogradOutputTransformDataset.h
+++ b/tests/datasets/WinogradOutputTransformDataset.h
@@ -154,6 +154,22 @@ public:
         add_config(TensorShape(7U, 2U, 64U, 3U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(24U, 9U, 64U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(7U, 2U, 64U, 5U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+
+        // (4x1, 5x1)
+        add_config(TensorShape(13U, 6U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 22U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(5U, 462U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(53U, 33U), PadStrideInfo(1, 1, 2, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 10U, 8U, 3U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(24U, 42U, 8U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 20U, 8U, 5U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(8U, 10U), PadStrideInfo(1, 1, 2, 0), DataLayout::NCHW));
+
+        // (1x4, 1x5)
+        add_config(TensorShape(13U, 7U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 20U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(5U, 477U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(53U, 33U), PadStrideInfo(1, 1, 0, 2), DataLayout::NCHW));
+        add_config(TensorShape(7U, 16U, 8U, 3U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(24U, 42U, 8U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(14U, 14U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(7U, 24U, 8U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 2), DataLayout::NCHW));
     }
 };
 
@@ -238,6 +254,18 @@ public:
         add_config(TensorShape(13U, 182U, 64U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
         add_config(TensorShape(32U, 756U, 64U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
         add_config(TensorShape(13U, 182U, 64U, 5U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+
+        // (4x1, 5x1)
+        add_config(TensorShape(32U, 3136U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(112U, 112U), PadStrideInfo(1, 1, 2, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 784U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(32U, 3024U, 8U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(112U, 112U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 784U, 8U, 5U), WinogradInfo(Size2D(4U, 1U), Size2D(5U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+
+        // (1x4, 1x5)
+        add_config(TensorShape(32U, 3136U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 0, 2), DataLayout::NCHW));
+        add_config(TensorShape(13U, 784U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(32U, 3024U, 8U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 784U, 8U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
     }
 };
 
diff --git a/tests/validation/CL/Winograd.cpp b/tests/validation/CL/Winograd.cpp
index 501afaccf9..849d0c13bc 100644
--- a/tests/validation/CL/Winograd.cpp
+++ b/tests/validation/CL/Winograd.cpp
@@ -65,7 +65,9 @@ const auto SmallWinogradInputTransformDatasetNCHW =
            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 SmallWinogradInputTransformDatasetNHWC = framework::dataset::concat(datasets::SmallWinogradInputTransformDataset4x4_3x3(),
                                                                                datasets::SmallWinogradInputTransformDataset4x4_5x5());
@@ -77,7 +79,9 @@ const auto LargeWinogradInputTransformDatasetNCHW =
            framework::dataset::concat(datasets::LargeWinogradInputTransformDataset4x4_3x3(),
            framework::dataset::concat(datasets::LargeWinogradInputTransformDataset4x1_3x1(),
            framework::dataset::concat(datasets::LargeWinogradInputTransformDataset1x4_1x3(),
-                                      datasets::LargeWinogradInputTransformDataset4x4_5x5()))))));
+           framework::dataset::concat(datasets::LargeWinogradInputTransformDataset4x4_5x5(),
+           framework::dataset::concat(datasets::LargeWinogradInputTransformDataset4x1_5x1(),
+                                      datasets::LargeWinogradInputTransformDataset1x4_1x5()))))))));
 
 const auto LargeWinogradInputTransformDatasetNHWC =
            framework::dataset::concat(datasets::LargeWinogradInputTransformDataset4x4_3x3(),
@@ -88,7 +92,9 @@ const auto SmallWinogradFilterTransformDatasetNCHW =
            framework::dataset::concat(combine(datasets::Small3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 2U), Size2D(4U, 4U) })),
            framework::dataset::concat(combine(datasets::Small3x1Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 1U), Size2D(4U, 1U) })),
            framework::dataset::concat(combine(datasets::Small1x3Shapes(), framework::dataset::make("OutputTile", { Size2D(1U, 2U), 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 SmallWinogradFilterTransformDatasetNHWC =
            framework::dataset::concat(combine(datasets::Small3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 4U) })),
@@ -98,7 +104,9 @@ const auto LargeWinogradFilterTransformDatasetNCHW =
            framework::dataset::concat(combine(datasets::Large3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 2U), Size2D(4U, 4U) })),
            framework::dataset::concat(combine(datasets::Large3x1Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 1U), Size2D(4U, 1U) })),
            framework::dataset::concat(combine(datasets::Large1x3Shapes(), framework::dataset::make("OutputTile", { Size2D(1U, 2U), Size2D(1U, 4U) })),
-                                      combine(datasets::Large5x5Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 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) })))))));
 
 const auto LargeWinogradFilterTransformDatasetNHWC =
            framework::dataset::concat(combine(datasets::Large3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 4U) })),
@@ -643,6 +651,54 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradConvolutionLayerFastMathFixture, fram
 }
 TEST_SUITE_END() // Conv5x5
 
+TEST_SUITE(Conv5x1)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradConvolutionLayerFastMathFixture, framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(datasets::SmallWinogradConvolutionLayer5x1Dataset(),
+                                               framework::dataset::make("DataType", { DataType::F32 })),
+                                               framework::dataset::make("ActivationLayerInfo", { ActivationLayerInfo() })),
+                                               framework::dataset::make("DataLayout", { DataLayout::NCHW })))
+
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_convolution_layer_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradConvolutionLayerFastMathFixture, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(datasets::LargeWinogradConvolutionLayer5x1Dataset(),
+                                               framework::dataset::make("DataType", { DataType::F32 })),
+                                               framework::dataset::make("ActivationLayerInfo", { ActivationLayerInfo() })),
+                                               framework::dataset::make("DataLayout", { DataLayout::NCHW })))
+
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_convolution_layer_f32);
+}
+TEST_SUITE_END() // Conv5x1
+
+TEST_SUITE(Conv1x5)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradConvolutionLayerFastMathFixture, framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(datasets::SmallWinogradConvolutionLayer1x5Dataset(),
+                                               framework::dataset::make("DataType", { DataType::F32 })),
+                                               framework::dataset::make("ActivationLayerInfo", { ActivationLayerInfo() })),
+                                               framework::dataset::make("DataLayout", { DataLayout::NCHW })))
+
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_convolution_layer_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradConvolutionLayerFastMathFixture, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(datasets::LargeWinogradConvolutionLayer1x5Dataset(),
+                                               framework::dataset::make("DataType", { DataType::F32 })),
+                                               framework::dataset::make("ActivationLayerInfo", { ActivationLayerInfo() })),
+                                               framework::dataset::make("DataLayout", { DataLayout::NCHW })))
+
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_convolution_layer_f32);
+}
+TEST_SUITE_END() // Conv1x5
+
 TEST_SUITE_END() // ConvolutionLayer
 TEST_SUITE_END() // Winograd
 TEST_SUITE_END() // CL
diff --git a/tests/validation/reference/Winograd.cpp b/tests/validation/reference/Winograd.cpp
index 5be4fe274b..026b30031c 100644
--- a/tests/validation/reference/Winograd.cpp
+++ b/tests/validation/reference/Winograd.cpp
@@ -148,6 +148,8 @@ void initialize_matrix_transform(SimpleTensor<T> &src, const Size2D &output_tile
         { WinogradKey(std::pair<int, int>(1, 2), std::pair<int, int>(1, 3), WinogradTransformType::INPUT), imatrix2x2_3x3 },
         { WinogradKey(std::pair<int, int>(1, 4), std::pair<int, int>(1, 3), WinogradTransformType::INPUT), imatrix4x4_3x3 },
         { WinogradKey(std::pair<int, int>(4, 4), std::pair<int, int>(5, 5), WinogradTransformType::INPUT), imatrix4x4_5x5 },
+        { WinogradKey(std::pair<int, int>(4, 1), std::pair<int, int>(5, 1), WinogradTransformType::INPUT), imatrix4x4_5x5 },
+        { WinogradKey(std::pair<int, int>(1, 4), std::pair<int, int>(1, 5), WinogradTransformType::INPUT), imatrix4x4_5x5 },
         { WinogradKey(std::pair<int, int>(2, 2), std::pair<int, int>(3, 3), WinogradTransformType::FILTER), fmatrix2x2_3x3 },
         { WinogradKey(std::pair<int, int>(4, 4), std::pair<int, int>(3, 3), WinogradTransformType::FILTER), fmatrix4x4_3x3 },
         { WinogradKey(std::pair<int, int>(2, 1), std::pair<int, int>(3, 1), WinogradTransformType::FILTER), fmatrix2x2_3x3 },
@@ -155,6 +157,8 @@ void initialize_matrix_transform(SimpleTensor<T> &src, const Size2D &output_tile
         { WinogradKey(std::pair<int, int>(1, 2), std::pair<int, int>(1, 3), WinogradTransformType::FILTER), fmatrix2x2_3x3 },
         { WinogradKey(std::pair<int, int>(1, 4), std::pair<int, int>(1, 3), WinogradTransformType::FILTER), fmatrix4x4_3x3 },
         { WinogradKey(std::pair<int, int>(4, 4), std::pair<int, int>(5, 5), WinogradTransformType::FILTER), fmatrix4x4_5x5 },
+        { WinogradKey(std::pair<int, int>(4, 1), std::pair<int, int>(5, 1), WinogradTransformType::FILTER), fmatrix4x4_5x5 },
+        { WinogradKey(std::pair<int, int>(1, 4), std::pair<int, int>(1, 5), WinogradTransformType::FILTER), fmatrix4x4_5x5 },
         { WinogradKey(std::pair<int, int>(2, 2), std::pair<int, int>(3, 3), WinogradTransformType::OUTPUT), omatrix2x2_3x3 },
         { WinogradKey(std::pair<int, int>(4, 4), std::pair<int, int>(3, 3), WinogradTransformType::OUTPUT), omatrix4x4_3x3 },
         { WinogradKey(std::pair<int, int>(2, 1), std::pair<int, int>(3, 1), WinogradTransformType::OUTPUT), omatrix2x2_3x3 },
@@ -162,6 +166,8 @@ void initialize_matrix_transform(SimpleTensor<T> &src, const Size2D &output_tile
         { WinogradKey(std::pair<int, int>(1, 2), std::pair<int, int>(1, 3), WinogradTransformType::OUTPUT), omatrix2x2_3x3 },
         { WinogradKey(std::pair<int, int>(1, 4), std::pair<int, int>(1, 3), WinogradTransformType::OUTPUT), omatrix4x4_3x3 },
         { WinogradKey(std::pair<int, int>(4, 4), std::pair<int, int>(5, 5), WinogradTransformType::OUTPUT), omatrix4x4_5x5 },
+        { WinogradKey(std::pair<int, int>(4, 1), std::pair<int, int>(5, 1), WinogradTransformType::OUTPUT), omatrix4x4_5x5 },
+        { WinogradKey(std::pair<int, int>(1, 4), std::pair<int, int>(1, 5), WinogradTransformType::OUTPUT), omatrix4x4_5x5 },
     };
 
     // Find transformation matrix