COMPMID-1048 Add NHWC data format support to Winograd input transform 4x4_3x3

https://confluence.arm.com/display/MLENG/Winograd+Input+Transform%3A+NCHW+vs+NHWC+on+OpenCL

Change-Id: Iac35a54389266701b7d8f5434a7a37df85b7b187
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/133315
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>

3 files changed, 53 insertions, 18 deletions

diff --git a/tests/datasets/WinogradInputTransformDataset.h b/tests/datasets/WinogradInputTransformDataset.h
index 59fb2adbd6..e365f9657f 100644
--- a/tests/datasets/WinogradInputTransformDataset.h
+++ b/tests/datasets/WinogradInputTransformDataset.h
@@ -97,12 +97,11 @@ private:
     std::vector<WinogradInfo> _infos{};
 };
 
-class SmallWinogradInputTransformDataset final : public WinogradInputTransformDataset
+class SmallWinogradInputTransformDataset2x2_3x3 final : public WinogradInputTransformDataset
 {
 public:
-    SmallWinogradInputTransformDataset()
+    SmallWinogradInputTransformDataset2x2_3x3()
     {
-        // (2x2, 3x3)
         add_config(TensorShape(9U, 9U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(9U, 9U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(27U, 13U, 2U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(27U, 13U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(128U, 64U, 1U, 3U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(128U, 64U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
@@ -110,8 +109,14 @@ public:
         add_config(TensorShape(27U, 13U, 2U, 4U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(27U, 13U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(9U, 9U, 3U, 5U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(14U, 14U, 512U, 2U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+    }
+};
 
-        // (4x4, 3x3)
+class SmallWinogradInputTransformDataset4x4_3x3 final : public WinogradInputTransformDataset
+{
+public:
+    SmallWinogradInputTransformDataset4x4_3x3()
+    {
         add_config(TensorShape(9U, 9U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(9U, 9U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(27U, 13U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(27U, 13U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(128U, 64U, 1U, 3U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(128U, 64U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
@@ -119,8 +124,14 @@ public:
         add_config(TensorShape(27U, 13U, 2U, 4U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(27U, 13U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(9U, 9U, 3U, 5U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(14U, 14U, 512U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+    }
+};
 
-        // (4x4, 5x5)
+class SmallWinogradInputTransformDataset4x4_5x5 final : public WinogradInputTransformDataset
+{
+public:
+    SmallWinogradInputTransformDataset4x4_5x5()
+    {
         add_config(TensorShape(9U, 9U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(9U, 9U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(27U, 13U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(27U, 13U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(128U, 64U, 1U, 3U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(128U, 64U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
@@ -131,24 +142,35 @@ public:
     }
 };
 
-class LargeWinogradInputTransformDataset final : public WinogradInputTransformDataset
+class LargeWinogradInputTransformDataset2x2_3x3 final : public WinogradInputTransformDataset
 {
 public:
-    LargeWinogradInputTransformDataset()
+    LargeWinogradInputTransformDataset2x2_3x3()
     {
-        // (2x2, 3x3)
         add_config(TensorShape(42U, 37U, 8U, 15U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(42U, 37U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(57U, 60U, 13U, 8U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(57U, 60U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(128U, 64U, 21U, 13U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(128U, 64U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(83U, 72U, 14U, 5U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(83U, 72U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+    }
+};
 
-        // (4x4, 3x3)
+class LargeWinogradInputTransformDataset4x4_3x3 final : public WinogradInputTransformDataset
+{
+public:
+    LargeWinogradInputTransformDataset4x4_3x3()
+    {
         add_config(TensorShape(42U, 37U, 8U, 15U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(42U, 37U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(57U, 60U, 13U, 8U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(57U, 60U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(128U, 64U, 21U, 13U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(128U, 64U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(83U, 72U, 14U, 5U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(83U, 72U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+    }
+};
 
-        // (4x4, 5x5)
+class LargeWinogradInputTransformDataset4x4_5x5 final : public WinogradInputTransformDataset
+{
+public:
+    LargeWinogradInputTransformDataset4x4_5x5()
+    {
         add_config(TensorShape(42U, 37U, 8U, 15U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(42U, 37U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(57U, 60U, 13U, 8U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(57U, 60U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(128U, 64U, 21U, 13U), WinogradInfo(Size2D(4U, 4U), Size2D(5U, 5U), Size2D(128U, 64U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
diff --git a/tests/validation/CL/Winograd.cpp b/tests/validation/CL/Winograd.cpp
index 4565c2d59d..59fe25db73 100644
--- a/tests/validation/CL/Winograd.cpp
+++ b/tests/validation/CL/Winograd.cpp
@@ -53,6 +53,10 @@ namespace
 {
 constexpr AbsoluteTolerance<float> tolerance_f32(0.001f);
 constexpr AbsoluteTolerance<float> tolerance_convolution_layer_f32(0.1f);
+const auto                         SmallWinogradInputTransformDataset = framework::dataset::concat(datasets::SmallWinogradInputTransformDataset2x2_3x3(),
+                                                                                                   framework::dataset::concat(datasets::SmallWinogradInputTransformDataset4x4_3x3(), datasets::SmallWinogradInputTransformDataset4x4_5x5()));
+const auto LargeWinogradInputTransformDataset = framework::dataset::concat(datasets::LargeWinogradInputTransformDataset2x2_3x3(),
+                                                                           framework::dataset::concat(datasets::LargeWinogradInputTransformDataset4x4_3x3(), datasets::LargeWinogradInputTransformDataset4x4_5x5()));
 } // namespace
 
 using namespace arm_compute::misc::shape_calculator;
@@ -102,7 +106,7 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
 
 using CLWinogradInputTransformFixture = WinogradInputTransformValidationFixture<CLTensor, CLAccessor, CLWinogradInputTransform, float>;
 
-DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(framework::dataset::concat(datasets::SmallWinogradInputTransformDataset(), datasets::LargeWinogradInputTransformDataset()),
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(framework::dataset::concat(SmallWinogradInputTransformDataset, LargeWinogradInputTransformDataset),
                                                                            framework::dataset::make("DataLayout", { DataLayout::NCHW })),
                                                                    framework::dataset::make("DataType", { DataType::F32 })),
                shape_in, winograd_info, data_layout, data_type)
@@ -123,15 +127,19 @@ DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(frame
     winograd_input_transform.configure(&in, &out, winograd_info);
 }
 
-FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradInputTransformFixture, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallWinogradInputTransformDataset(),
-                                                                                                                     framework::dataset::make("DataLayout", { DataLayout::NCHW })),
+FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradInputTransformFixture, framework::DatasetMode::PRECOMMIT, combine(framework::dataset::concat(combine(SmallWinogradInputTransformDataset,
+                                                                                                             framework::dataset::make("DataLayout", { DataLayout::NCHW })),
+                                                                                                             combine(datasets::SmallWinogradInputTransformDataset4x4_3x3(),
+                                                                                                                     framework::dataset::make("DataLayout", { DataLayout::NHWC }))),
                                                                                                              framework::dataset::make("DataType", { DataType::F32 })))
 {
     validate(CLAccessor(_target), _reference, tolerance_f32);
 }
 
-FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradInputTransformFixture, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeWinogradInputTransformDataset(),
-                                                                                                                   framework::dataset::make("DataLayout", { DataLayout::NCHW })),
+FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradInputTransformFixture, framework::DatasetMode::NIGHTLY, combine(framework::dataset::concat(combine(LargeWinogradInputTransformDataset,
+                                                                                                           framework::dataset::make("DataLayout", { DataLayout::NCHW })),
+                                                                                                           combine(datasets::LargeWinogradInputTransformDataset4x4_3x3(),
+                                                                                                                   framework::dataset::make("DataLayout", { DataLayout::NHWC }))),
                                                                                                            framework::dataset::make("DataType", { DataType::F32 })))
 {
     validate(CLAccessor(_target), _reference, tolerance_f32);
diff --git a/tests/validation/fixtures/WinogradConvolutionLayerFixture.h b/tests/validation/fixtures/WinogradConvolutionLayerFixture.h
index f40f3d2e43..07795c2361 100644
--- a/tests/validation/fixtures/WinogradConvolutionLayerFixture.h
+++ b/tests/validation/fixtures/WinogradConvolutionLayerFixture.h
@@ -314,10 +314,15 @@ protected:
         }
     }
 
-    TensorType compute_target(const TensorShape &input_shape, const TensorShape &output_shape, const WinogradInfo &winograd_info, DataLayout data_layout, DataType data_type)
+    TensorType compute_target(TensorShape input_shape, const TensorShape &output_shape, const WinogradInfo &winograd_info, DataLayout data_layout, DataType data_type)
     {
+        if(data_layout == DataLayout::NHWC)
+        {
+            permute(input_shape, PermutationVector(2U, 0U, 1U));
+        }
+
         TensorType src = create_tensor<TensorType>(input_shape, data_type, 1, 0, QuantizationInfo(), data_layout);
-        TensorType dst = create_tensor<TensorType>(output_shape, data_type, 1, 0, QuantizationInfo(), data_layout);
+        TensorType dst = create_tensor<TensorType>(output_shape, data_type, 1, 0, QuantizationInfo());
 
         // Create and configure function
         FunctionType transf;
@@ -345,7 +350,7 @@ protected:
     SimpleTensor<T> compute_reference(const TensorShape &input_shape, const TensorShape &output_shape, const WinogradInfo &winograd_info, DataLayout data_layout, DataType data_type)
     {
         // Create reference
-        SimpleTensor<T> src{ input_shape, data_type, 1, 0, QuantizationInfo(), data_layout };
+        SimpleTensor<T> src{ input_shape, data_type, 1, 0, QuantizationInfo() };
 
         // Fill reference
         fill(src, 0, -1.f, 1.f);