COMPMID-1201 - Implementing Winograd Convolution Layer 1x3 and 3x1 kernels on OpenCL

Change-Id: I39667bab49daa4da009694163274a59fd3574c73
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/137595
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

5 files changed, 324 insertions, 7 deletions

diff --git a/tests/datasets/LargeConvolutionLayerDataset.h b/tests/datasets/LargeConvolutionLayerDataset.h
index 36b3d60d57..ae25c8cd66 100644
--- a/tests/datasets/LargeConvolutionLayerDataset.h
+++ b/tests/datasets/LargeConvolutionLayerDataset.h
@@ -59,6 +59,50 @@ public:
     }
 };
 
+class LargeWinogradConvolutionLayer3x1Dataset final : public ConvolutionLayerDataset
+{
+public:
+    LargeWinogradConvolutionLayer3x1Dataset()
+    {
+        // Kernel size 3
+        // Batch size 1
+        add_config(TensorShape(224U, 222U, 64U), TensorShape(3U, 1U, 64U, 64U), TensorShape(64U), TensorShape(224U, 222U, 64U), PadStrideInfo(1, 1, 1, 0));
+        add_config(TensorShape(112U, 113U, 64U), TensorShape(3U, 1U, 64U, 128U), TensorShape(128U), TensorShape(112U, 113U, 128U), PadStrideInfo(1, 1, 1, 0));
+        add_config(TensorShape(112U, 112U, 128U), TensorShape(3U, 1U, 128U, 129U), TensorShape(129U), TensorShape(112U, 112U, 129U), PadStrideInfo(1, 1, 1, 0));
+        add_config(TensorShape(53U, 56U, 125U), TensorShape(3U, 1U, 125U, 256U), TensorShape(256U), TensorShape(51U, 56U, 256U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(56U, 56U, 256U), TensorShape(3U, 1U, 256U, 256U), TensorShape(256U), TensorShape(56U, 56U, 256U), PadStrideInfo(1, 1, 1, 0));
+        add_config(TensorShape(28U, 28U, 257U), TensorShape(3U, 1U, 257U, 512U), TensorShape(512U), TensorShape(26U, 28U, 512U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(28U, 28U, 512U), TensorShape(3U, 1U, 512U, 512U), TensorShape(512U), TensorShape(28U, 28U, 512U), PadStrideInfo(1, 1, 1, 0));
+        add_config(TensorShape(14U, 14U, 512U), TensorShape(3U, 1U, 512U, 512U), TensorShape(512U), TensorShape(12U, 14U, 512U), PadStrideInfo(1, 1, 0, 0));
+        // Batch size 3, 2 and 4
+        add_config(TensorShape(224U, 222U, 64U, 3U), TensorShape(3U, 1U, 64U, 64U), TensorShape(64U), TensorShape(224U, 222U, 64U, 3U), PadStrideInfo(1, 1, 1, 0));
+        add_config(TensorShape(112U, 113U, 64U, 2U), TensorShape(3U, 1U, 64U, 128U), TensorShape(128U), TensorShape(110U, 113U, 128U, 2U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(111U, 112U, 127U, 4U), TensorShape(3U, 1U, 127U, 128U), TensorShape(128U), TensorShape(111U, 112U, 128U, 4U), PadStrideInfo(1, 1, 1, 0));
+    }
+};
+
+class LargeWinogradConvolutionLayer1x3Dataset final : public ConvolutionLayerDataset
+{
+public:
+    LargeWinogradConvolutionLayer1x3Dataset()
+    {
+        // Kernel size 3
+        // Batch size 1
+        add_config(TensorShape(224U, 222U, 64U), TensorShape(1U, 3U, 64U, 64U), TensorShape(64U), TensorShape(224U, 222U, 64U), PadStrideInfo(1, 1, 0, 1));
+        add_config(TensorShape(112U, 113U, 64U), TensorShape(1U, 3U, 64U, 128U), TensorShape(128U), TensorShape(112U, 113U, 128U), PadStrideInfo(1, 1, 0, 1));
+        add_config(TensorShape(112U, 112U, 128U), TensorShape(1U, 3U, 128U, 129U), TensorShape(129U), TensorShape(112U, 110U, 129U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(53U, 56U, 125U), TensorShape(1U, 3U, 125U, 256U), TensorShape(256U), TensorShape(53U, 56U, 256U), PadStrideInfo(1, 1, 0, 1));
+        add_config(TensorShape(56U, 56U, 256U), TensorShape(1U, 3U, 256U, 256U), TensorShape(256U), TensorShape(56U, 54U, 256U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(28U, 28U, 257U), TensorShape(1U, 3U, 257U, 512U), TensorShape(512U), TensorShape(28U, 28U, 512U), PadStrideInfo(1, 1, 0, 1));
+        add_config(TensorShape(28U, 28U, 512U), TensorShape(1U, 3U, 512U, 512U), TensorShape(512U), TensorShape(28U, 28U, 512U), PadStrideInfo(1, 1, 0, 1));
+        add_config(TensorShape(14U, 14U, 512U), TensorShape(1U, 3U, 512U, 512U), TensorShape(512U), TensorShape(14U, 12U, 512U), PadStrideInfo(1, 1, 0, 0));
+        // Batch size 3, 2 and 4
+        add_config(TensorShape(224U, 222U, 64U, 3U), TensorShape(1U, 3U, 64U, 64U), TensorShape(64U), TensorShape(224U, 222U, 64U, 3U), PadStrideInfo(1, 1, 0, 1));
+        add_config(TensorShape(112U, 113U, 64U, 2U), TensorShape(1U, 3U, 64U, 128U), TensorShape(128U), TensorShape(112U, 113U, 128U, 2U), PadStrideInfo(1, 1, 0, 1));
+        add_config(TensorShape(111U, 112U, 127U, 4U), TensorShape(1U, 3U, 127U, 128U), TensorShape(128U), TensorShape(111U, 112U, 128U, 4U), PadStrideInfo(1, 1, 0, 1));
+    }
+};
+
 class LargeWinogradConvolutionLayer5x5Dataset final : public ConvolutionLayerDataset
 {
 public:
diff --git a/tests/datasets/ShapeDatasets.h b/tests/datasets/ShapeDatasets.h
index a5620ff7cf..68263c7793 100644
--- a/tests/datasets/ShapeDatasets.h
+++ b/tests/datasets/ShapeDatasets.h
@@ -388,6 +388,38 @@ public:
     }
 };
 
+/** Data set containing small 3x1 tensor shapes. */
+class Small3x1Shapes final : public ShapeDataset
+{
+public:
+    Small3x1Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 3U, 1U, 7U, 4U },
+                     TensorShape{ 3U, 1U, 4U, 13U },
+                     TensorShape{ 3U, 1U, 9U, 2U },
+                     TensorShape{ 3U, 1U, 3U, 5U },
+    })
+    {
+    }
+};
+
+/** Data set containing small 1x3 tensor shapes. */
+class Small1x3Shapes final : public ShapeDataset
+{
+public:
+    Small1x3Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 1U, 3U, 7U, 4U },
+                     TensorShape{ 1U, 3U, 4U, 13U },
+                     TensorShape{ 1U, 3U, 9U, 2U },
+                     TensorShape{ 1U, 3U, 3U, 5U },
+    })
+    {
+    }
+};
+
 /** Data set containing large 3x3 tensor shapes. */
 class Large3x3Shapes final : public ShapeDataset
 {
@@ -404,6 +436,38 @@ public:
     }
 };
 
+/** Data set containing large 3x1 tensor shapes. */
+class Large3x1Shapes final : public ShapeDataset
+{
+public:
+    Large3x1Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 3U, 1U, 32U, 64U },
+                     TensorShape{ 3U, 1U, 51U, 13U },
+                     TensorShape{ 3U, 1U, 53U, 47U },
+                     TensorShape{ 3U, 1U, 128U, 384U },
+    })
+    {
+    }
+};
+
+/** Data set containing large 1x3 tensor shapes. */
+class Large1x3Shapes final : public ShapeDataset
+{
+public:
+    Large1x3Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 1U, 3U, 32U, 64U },
+                     TensorShape{ 1U, 3U, 51U, 13U },
+                     TensorShape{ 1U, 3U, 53U, 47U },
+                     TensorShape{ 1U, 3U, 128U, 384U },
+    })
+    {
+    }
+};
+
 /** Data set containing small 5x5 tensor shapes. */
 class Small5x5Shapes final : public ShapeDataset
 {
diff --git a/tests/datasets/SmallConvolutionLayerDataset.h b/tests/datasets/SmallConvolutionLayerDataset.h
index fed36de3dd..f05cc15c06 100644
--- a/tests/datasets/SmallConvolutionLayerDataset.h
+++ b/tests/datasets/SmallConvolutionLayerDataset.h
@@ -52,6 +52,36 @@ public:
     }
 };
 
+class SmallWinogradConvolutionLayer3x1Dataset final : public ConvolutionLayerDataset
+{
+public:
+    SmallWinogradConvolutionLayer3x1Dataset()
+    {
+        // Channel size big enough to force multithreaded execution of the input transform
+        add_config(TensorShape(8U, 8U, 32U), TensorShape(3U, 1U, 32U, 1U), TensorShape(1U), TensorShape(6U, 8U, 1U), PadStrideInfo(1, 1, 0, 0));
+        // Batch size 1
+        add_config(TensorShape(8U, 8U, 2U), TensorShape(3U, 1U, 2U, 1U), TensorShape(1U), TensorShape(6U, 8U, 1U), PadStrideInfo(1, 1, 0, 0));
+        // Batch size 4
+        add_config(TensorShape(23U, 27U, 5U, 4U), TensorShape(3U, 1U, 5U, 21U), TensorShape(21U), TensorShape(21U, 27U, 21U, 4U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(8U, 8U, 2U), TensorShape(3U, 1U, 2U, 1U), TensorShape(1U), TensorShape(8U, 8U, 1U), PadStrideInfo(1, 1, 1, 0));
+    }
+};
+
+class SmallWinogradConvolutionLayer1x3Dataset final : public ConvolutionLayerDataset
+{
+public:
+    SmallWinogradConvolutionLayer1x3Dataset()
+    {
+        // Channel size big enough to force multithreaded execution of the input transform
+        add_config(TensorShape(8U, 8U, 32U), TensorShape(1U, 3U, 32U, 1U), TensorShape(1U), TensorShape(8U, 6U, 1U), PadStrideInfo(1, 1, 0, 0));
+        // Batch size 1
+        add_config(TensorShape(8U, 8U, 2U), TensorShape(1U, 3U, 2U, 1U), TensorShape(1U), TensorShape(8U, 6U, 1U), PadStrideInfo(1, 1, 0, 0));
+        // Batch size 4
+        add_config(TensorShape(23U, 27U, 5U, 4U), TensorShape(1U, 3U, 5U, 21U), TensorShape(21U), TensorShape(23U, 25U, 21U, 4U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(8U, 8U, 2U), TensorShape(1U, 3U, 2U, 1U), TensorShape(1U), TensorShape(8U, 8U, 1U), PadStrideInfo(1, 1, 0, 1));
+    }
+};
+
 class SmallWinogradConvolutionLayer5x5Dataset final : public ConvolutionLayerDataset
 {
 public:
diff --git a/tests/datasets/WinogradInputTransformDataset.h b/tests/datasets/WinogradInputTransformDataset.h
index e365f9657f..ca23984a1d 100644
--- a/tests/datasets/WinogradInputTransformDataset.h
+++ b/tests/datasets/WinogradInputTransformDataset.h
@@ -112,6 +112,36 @@ public:
     }
 };
 
+class SmallWinogradInputTransformDataset2x1_3x1 final : public WinogradInputTransformDataset
+{
+public:
+    SmallWinogradInputTransformDataset2x1_3x1()
+    {
+        add_config(TensorShape(9U, 9U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(9U, 9U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(27U, 13U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 1U, 3U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(128U, 64U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 4U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U, 4U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(27U, 13U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 5U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(14U, 14U, 512U, 2U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+    }
+};
+
+class SmallWinogradInputTransformDataset1x2_1x3 final : public WinogradInputTransformDataset
+{
+public:
+    SmallWinogradInputTransformDataset1x2_1x3()
+    {
+        add_config(TensorShape(9U, 9U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(9U, 9U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(27U, 13U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 1U, 3U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(128U, 64U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 4U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U, 4U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(27U, 13U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 5U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(14U, 14U, 512U, 2U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+    }
+};
+
 class SmallWinogradInputTransformDataset4x4_3x3 final : public WinogradInputTransformDataset
 {
 public:
@@ -127,6 +157,36 @@ public:
     }
 };
 
+class SmallWinogradInputTransformDataset4x1_3x1 final : public WinogradInputTransformDataset
+{
+public:
+    SmallWinogradInputTransformDataset4x1_3x1()
+    {
+        add_config(TensorShape(9U, 9U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(9U, 9U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(27U, 13U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 1U, 3U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(128U, 64U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 4U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U, 4U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(27U, 13U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 5U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(14U, 14U, 512U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+    }
+};
+
+class SmallWinogradInputTransformDataset1x4_1x3 final : public WinogradInputTransformDataset
+{
+public:
+    SmallWinogradInputTransformDataset1x4_1x3()
+    {
+        add_config(TensorShape(9U, 9U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(9U, 9U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(27U, 13U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 1U, 3U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(128U, 64U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 4U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(27U, 13U, 2U, 4U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(27U, 13U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(9U, 9U, 3U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(14U, 14U, 512U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+    }
+};
+
 class SmallWinogradInputTransformDataset4x4_5x5 final : public WinogradInputTransformDataset
 {
 public:
@@ -154,6 +214,30 @@ public:
     }
 };
 
+class LargeWinogradInputTransformDataset2x1_3x1 final : public WinogradInputTransformDataset
+{
+public:
+    LargeWinogradInputTransformDataset2x1_3x1()
+    {
+        add_config(TensorShape(42U, 37U, 8U, 15U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(42U, 37U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(57U, 60U, 13U, 8U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(57U, 60U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 21U, 13U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(128U, 64U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(83U, 72U, 14U, 5U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(83U, 72U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+    }
+};
+
+class LargeWinogradInputTransformDataset1x2_1x3 final : public WinogradInputTransformDataset
+{
+public:
+    LargeWinogradInputTransformDataset1x2_1x3()
+    {
+        add_config(TensorShape(42U, 37U, 8U, 15U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(42U, 37U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(57U, 60U, 13U, 8U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(57U, 60U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 21U, 13U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(128U, 64U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(83U, 72U, 14U, 5U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(83U, 72U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+    }
+};
+
 class LargeWinogradInputTransformDataset4x4_3x3 final : public WinogradInputTransformDataset
 {
 public:
@@ -166,6 +250,30 @@ public:
     }
 };
 
+class LargeWinogradInputTransformDataset4x1_3x1 final : public WinogradInputTransformDataset
+{
+public:
+    LargeWinogradInputTransformDataset4x1_3x1()
+    {
+        add_config(TensorShape(42U, 37U, 8U, 15U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(42U, 37U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(57U, 60U, 13U, 8U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(57U, 60U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 21U, 13U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(128U, 64U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(83U, 72U, 14U, 5U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(83U, 72U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+    }
+};
+
+class LargeWinogradInputTransformDataset1x4_1x3 final : public WinogradInputTransformDataset
+{
+public:
+    LargeWinogradInputTransformDataset1x4_1x3()
+    {
+        add_config(TensorShape(42U, 37U, 8U, 15U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(42U, 37U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(57U, 60U, 13U, 8U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(57U, 60U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(128U, 64U, 21U, 13U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(128U, 64U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(83U, 72U, 14U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(83U, 72U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+    }
+};
+
 class LargeWinogradInputTransformDataset4x4_5x5 final : public WinogradInputTransformDataset
 {
 public:
diff --git a/tests/datasets/WinogradOutputTransformDataset.h b/tests/datasets/WinogradOutputTransformDataset.h
index c7ba3b2b7d..a4689c6ef1 100644
--- a/tests/datasets/WinogradOutputTransformDataset.h
+++ b/tests/datasets/WinogradOutputTransformDataset.h
@@ -99,12 +99,11 @@ private:
     std::vector<WinogradInfo> _info{};
 };
 
-class SmallWinogradOutputTransformDataset final : public WinogradOutputTransformDataset
+class SmallWinogradOutputTransformDatasetNCHW final : public WinogradOutputTransformDataset
 {
 public:
-    SmallWinogradOutputTransformDataset()
+    SmallWinogradOutputTransformDatasetNCHW()
     {
-        // NCHW
         // (2x2, 3x3)
         add_config(TensorShape(13U, 6U, 16U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(7U, 20U, 16U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
@@ -120,6 +119,34 @@ public:
         add_config(TensorShape(24U, 16U, 36U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(7U, 12U, 16U, 5U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
 
+        // (2x1, 3x1)
+        add_config(TensorShape(13U, 18U, 4U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 44U, 4U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(1U, 891U, 4U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(53U, 33U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 30U, 4U, 3U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(24U, 98U, 4U, 2U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+
+        // (1x2, 1x3)
+        add_config(TensorShape(13U, 14U, 4U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 50U, 4U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(1U, 901U, 4U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(53U, 33U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(7U, 32U, 4U, 3U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(24U, 98U, 4U, 2U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(14U, 14U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+
+        // (4x1, 3x1)
+        add_config(TensorShape(13U, 12U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 22U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(1U, 462U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(53U, 33U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 20U, 6U, 3U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(24U, 56U, 6U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+
+        // (1x4, 1x3)
+        add_config(TensorShape(13U, 7U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 30U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(1U, 477U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(53U, 33U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(7U, 16U, 6U, 3U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(24U, 56U, 6U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(14U, 14U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+
         // (4x4, 5x5)
         add_config(TensorShape(13U, 1U, 64U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(7U, 4U, 64U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
@@ -127,8 +154,14 @@ 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));
+    }
+};
 
-        // NHWC
+class SmallWinogradOutputTransformDatasetNHWC final : public WinogradOutputTransformDataset
+{
+public:
+    SmallWinogradOutputTransformDatasetNHWC()
+    {
         // (4x4, 3x3)
         add_config(TensorShape(13U, 4U, 36U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(10U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
         add_config(TensorShape(13U, 6U, 36U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
@@ -146,10 +179,10 @@ public:
     }
 };
 
-class LargeWinogradOutputTransformDataset final : public WinogradOutputTransformDataset
+class LargeWinogradOutputTransformDatasetNCHW final : public WinogradOutputTransformDataset
 {
 public:
-    LargeWinogradOutputTransformDataset()
+    LargeWinogradOutputTransformDatasetNCHW()
     {
         // NCHW
         // (2x2, 3x3)
@@ -168,13 +201,51 @@ public:
         add_config(TensorShape(32U, 784U, 36U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
         add_config(TensorShape(13U, 196U, 36U, 5U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
 
+        // (2x1, 3x1)
+        add_config(TensorShape(64U, 24976U, 4U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(224U, 223U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(32U, 6160U, 4U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(112U, 110U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 1568U, 4U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(64U, 24753U, 4U, 3U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(224U, 223U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(32U, 6050U, 4U, 2U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(112U, 110U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 1512U, 4U, 5U), WinogradInfo(Size2D(2U, 1U), Size2D(3U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+
+        // (1x2, 1x3)
+        add_config(TensorShape(64U, 25088U, 4U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(224U, 223U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(32U, 6160U, 4U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(112U, 110U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(13U, 1568U, 4U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(64U, 24864U, 4U, 3U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(224U, 223U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(32U, 6048U, 4U, 2U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(112U, 110U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 1512U, 4U, 5U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+
+        // (4x1, 3x1)
+        add_config(TensorShape(64U, 12488U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(224U, 223U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(32U, 3080U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(112U, 110U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 784U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(64U, 12488U, 6U, 3U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(224U, 223U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(32U, 3080U, 6U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(112U, 110U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 784U, 6U, 5U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+
+        // (1x4, 1x3)
+        add_config(TensorShape(64U, 12544U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(224U, 223U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(32U, 3136U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(112U, 110U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(13U, 784U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(64U, 12544U, 6U, 3U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(224U, 223U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(32U, 3024U, 6U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(112U, 110U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 784U, 6U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+
         // (4x4, 5x5)
         add_config(TensorShape(32U, 756U, 64U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
         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));
+    }
+};
 
-        // NHWC
+class LargeWinogradOutputTransformDatasetNHWC final : public WinogradOutputTransformDataset
+{
+public:
+    LargeWinogradOutputTransformDatasetNHWC()
+    {
         // (4x4, 3x3)
         add_config(TensorShape(64U, 3136U, 36U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(224U, 224U), PadStrideInfo(1, 1, 1, 1), DataLayout::NHWC));
         add_config(TensorShape(32U, 784U, 36U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));