COMPMID-1337 Implementing Winograd Convolution Layer 1x3 and 3x1 kernels on OpenCL NHWC

Change-Id: Ia07e0dfcbcd07366c4bcb956e298369fb12a0369
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/138759
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>

3 files changed, 36 insertions, 17 deletions

diff --git a/tests/datasets/WinogradOutputTransformDataset.h b/tests/datasets/WinogradOutputTransformDataset.h
index fc23e65258..4085e91854 100644
--- a/tests/datasets/WinogradOutputTransformDataset.h
+++ b/tests/datasets/WinogradOutputTransformDataset.h
@@ -178,6 +178,20 @@ class SmallWinogradOutputTransformDatasetNHWC final : public WinogradOutputTrans
 public:
     SmallWinogradOutputTransformDatasetNHWC()
     {
+        // (4x1, 3x1)
+        add_config(TensorShape(13U, 12U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 22U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(1U, 462U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(53U, 33U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 20U, 6U, 3U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(24U, 56U, 6U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+
+        // (1x4, 1x3)
+        add_config(TensorShape(13U, 7U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(7U, 6U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 30U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(1U, 477U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(53U, 33U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+        add_config(TensorShape(7U, 16U, 6U, 3U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(24U, 56U, 6U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(14U, 14U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+
         // (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));
@@ -200,7 +214,6 @@ class LargeWinogradOutputTransformDatasetNCHW final : public WinogradOutputTrans
 public:
     LargeWinogradOutputTransformDatasetNCHW()
     {
-        // NCHW
         // (2x2, 3x3)
         add_config(TensorShape(64U, 12544U, 16U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(224U, 224U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(32U, 3080U, 16U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
@@ -274,6 +287,22 @@ class LargeWinogradOutputTransformDatasetNHWC final : public WinogradOutputTrans
 public:
     LargeWinogradOutputTransformDatasetNHWC()
     {
+        // (4x1, 3x1)
+        add_config(TensorShape(64U, 12488U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(224U, 223U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+        add_config(TensorShape(32U, 3080U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(112U, 110U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 6U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+        add_config(TensorShape(64U, 12488U, 6U, 3U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(224U, 223U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(32U, 3080U, 6U, 2U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(112U, 110U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 6U, 5U), WinogradInfo(Size2D(4U, 1U), Size2D(3U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+
+        // (1x4, 1x3)
+        add_config(TensorShape(64U, 12544U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(224U, 223U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+        add_config(TensorShape(32U, 3136U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(112U, 110U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 6U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+        add_config(TensorShape(64U, 12544U, 6U, 3U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(224U, 223U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(32U, 3024U, 6U, 2U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(112U, 110U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 6U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+
         // (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));
diff --git a/tests/validation/CL/Winograd.cpp b/tests/validation/CL/Winograd.cpp
index 849d0c13bc..c39cb4e790 100644
--- a/tests/validation/CL/Winograd.cpp
+++ b/tests/validation/CL/Winograd.cpp
@@ -70,7 +70,9 @@ const auto SmallWinogradInputTransformDatasetNCHW =
                                       datasets::SmallWinogradInputTransformDataset1x4_1x5()))))))));
 
 const auto SmallWinogradInputTransformDatasetNHWC = framework::dataset::concat(datasets::SmallWinogradInputTransformDataset4x4_3x3(),
-                                                                               datasets::SmallWinogradInputTransformDataset4x4_5x5());
+                                                    framework::dataset::concat(datasets::SmallWinogradInputTransformDataset4x1_3x1(),
+                                                    framework::dataset::concat(datasets::SmallWinogradInputTransformDataset1x4_1x3(),
+                                                                               datasets::SmallWinogradInputTransformDataset4x4_5x5())));
 
 const auto LargeWinogradInputTransformDatasetNCHW =
            framework::dataset::concat(datasets::LargeWinogradInputTransformDataset2x2_3x3(),
@@ -98,7 +100,9 @@ const auto SmallWinogradFilterTransformDatasetNCHW =
 
 const auto SmallWinogradFilterTransformDatasetNHWC =
            framework::dataset::concat(combine(datasets::Small3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 4U) })),
-                                      combine(datasets::Small5x5Shapes(), 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) })))));
 
 const auto LargeWinogradFilterTransformDatasetNCHW =
            framework::dataset::concat(combine(datasets::Large3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 2U), Size2D(4U, 4U) })),
diff --git a/tests/validation/reference/Winograd.cpp b/tests/validation/reference/Winograd.cpp
index 026b30031c..132d252383 100644
--- a/tests/validation/reference/Winograd.cpp
+++ b/tests/validation/reference/Winograd.cpp
@@ -194,20 +194,6 @@ void initialize_matrix_transform(SimpleTensor<T> &src, const Size2D &output_tile
 } // namespace
 
 template <typename T>
-void print_tile(SimpleTensor<T> &in)
-{
-    for(int y = 0; y < in.shape()[1]; y++)
-    {
-        for(int x = 0; x < in.shape()[0]; x++)
-        {
-            std::cout << in[x + y * in.shape()[0]] << " ";
-        }
-
-        std::cout << std::endl;
-    }
-}
-
-template <typename T>
 SimpleTensor<T> winograd_input_transform(const SimpleTensor<T> &in, const TensorShape &output_shape, const WinogradInfo &winograd_info)
 {
     ARM_COMPUTE_ERROR_ON(in.data_layout() != DataLayout::NCHW);