COMPMID-1318: Implementing Winograd 7x7 NHWC on OpenCL - Part III

Change-Id: I7ebf09cc12fb117834faf88cdd556d2a66eacf07
Signed-off-by: giuros01 <giuseppe.rossini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/926
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>

3 files changed, 75 insertions, 12 deletions

diff --git a/tests/datasets/WinogradOutputTransformDataset.h b/tests/datasets/WinogradOutputTransformDataset.h
index 617a67de07..ea38e73adc 100644
--- a/tests/datasets/WinogradOutputTransformDataset.h
+++ b/tests/datasets/WinogradOutputTransformDataset.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -173,10 +173,10 @@ public:
     }
 };
 
-class SmallWinogradOutputTransformDatasetNHWC final : public WinogradOutputTransformDataset
+class SmallWinogradOutputTransformDatasetNHWC_F16 : public WinogradOutputTransformDataset
 {
 public:
-    SmallWinogradOutputTransformDatasetNHWC()
+    SmallWinogradOutputTransformDatasetNHWC_F16()
     {
         // (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));
@@ -225,7 +225,39 @@ public:
     }
 };
 
-class LargeWinogradOutputTransformDatasetNCHW final : public WinogradOutputTransformDataset
+class SmallWinogradOutputTransformDatasetNHWC_F32 : public SmallWinogradOutputTransformDatasetNHWC_F16
+{
+public:
+    SmallWinogradOutputTransformDatasetNHWC_F32()
+        : SmallWinogradOutputTransformDatasetNHWC_F16()
+    {
+        // (2x2, 7x7)
+        add_config(TensorShape(13U, 4U, 64U), WinogradInfo(Size2D(2U, 2U), Size2D(7U, 7U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 6U, 64U), WinogradInfo(Size2D(2U, 2U), Size2D(7U, 7U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(5U, 360U, 64U), WinogradInfo(Size2D(2U, 2U), Size2D(7U, 7U), Size2D(53U, 33U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+        add_config(TensorShape(7U, 2U, 64U, 3U), WinogradInfo(Size2D(2U, 2U), Size2D(7U, 7U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(24U, 25U, 64U, 2U), WinogradInfo(Size2D(2U, 2U), Size2D(7U, 7U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NHWC));
+        add_config(TensorShape(7U, 2U, 64U, 5U), WinogradInfo(Size2D(2U, 2U), Size2D(7U, 7U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+
+        // (2x1, 7x1)
+        add_config(TensorShape(13U, 18U, 8U), WinogradInfo(Size2D(2U, 1U), Size2D(7U, 1U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 22U, 8U), WinogradInfo(Size2D(2U, 1U), Size2D(7U, 1U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(5U, 858U, 8U), WinogradInfo(Size2D(2U, 1U), Size2D(7U, 1U), Size2D(53U, 33U), PadStrideInfo(1, 1, 2, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 10U, 8U, 3U), WinogradInfo(Size2D(2U, 1U), Size2D(7U, 1U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(24U, 70U, 8U, 2U), WinogradInfo(Size2D(2U, 1U), Size2D(7U, 1U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 30U, 8U, 5U), WinogradInfo(Size2D(2U, 1U), Size2D(7U, 1U), Size2D(8U, 10U), PadStrideInfo(1, 1, 2, 0), DataLayout::NHWC));
+
+        // (1x2, 1x7)
+        add_config(TensorShape(13U, 18U, 8U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 7U), Size2D(9U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(7U, 30U, 8U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 7U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(5U, 848U, 8U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 7U), Size2D(53U, 33U), PadStrideInfo(1, 1, 0, 2), DataLayout::NHWC));
+        add_config(TensorShape(7U, 16U, 8U, 3U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 7U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(24U, 70U, 8U, 2U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 7U), Size2D(14U, 14U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+        add_config(TensorShape(7U, 32U, 8U, 5U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 7U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 2), DataLayout::NHWC));
+    }
+};
+
+class LargeWinogradOutputTransformDatasetNCHW : public WinogradOutputTransformDataset
 {
 public:
     LargeWinogradOutputTransformDatasetNCHW()
@@ -298,10 +330,10 @@ public:
     }
 };
 
-class LargeWinogradOutputTransformDatasetNHWC final : public WinogradOutputTransformDataset
+class LargeWinogradOutputTransformDatasetNHWC_F16 : public WinogradOutputTransformDataset
 {
 public:
-    LargeWinogradOutputTransformDatasetNHWC()
+    LargeWinogradOutputTransformDatasetNHWC_F16()
     {
         // (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));
@@ -346,6 +378,32 @@ public:
         add_config(TensorShape(13U, 784U, 8U, 5U), WinogradInfo(Size2D(1U, 4U), Size2D(1U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
     }
 };
+
+class LargeWinogradOutputTransformDatasetNHWC_F32 : public LargeWinogradOutputTransformDatasetNHWC_F16
+{
+public:
+    LargeWinogradOutputTransformDatasetNHWC_F32()
+    {
+        // (2x2, 7x7)
+        add_config(TensorShape(32U, 756U, 64U), WinogradInfo(Size2D(2U, 2U), Size2D(5U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 182U, 64U), WinogradInfo(Size2D(2U, 2U), Size2D(5U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+        add_config(TensorShape(32U, 756U, 64U, 2U), WinogradInfo(Size2D(2U, 2U), Size2D(5U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 182U, 64U, 5U), WinogradInfo(Size2D(2U, 2U), Size2D(5U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+
+        // (2x1, 7x1)
+        add_config(TensorShape(32U, 3136U, 8U), WinogradInfo(Size2D(2U, 1U), Size2D(5U, 1U), Size2D(112U, 112U), PadStrideInfo(1, 1, 2, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 8U), WinogradInfo(Size2D(2U, 1U), Size2D(5U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+        add_config(TensorShape(32U, 3024U, 8U, 2U), WinogradInfo(Size2D(2U, 1U), Size2D(5U, 1U), Size2D(112U, 112U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 8U, 5U), WinogradInfo(Size2D(2U, 1U), Size2D(5U, 1U), Size2D(56U, 56U), PadStrideInfo(1, 1, 1, 0), DataLayout::NHWC));
+
+        // (1x2, 1x7)
+        add_config(TensorShape(32U, 3136U, 8U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 0, 2), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 8U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+        add_config(TensorShape(32U, 3024U, 8U, 2U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 5U), Size2D(112U, 112U), PadStrideInfo(1, 1, 0, 0), DataLayout::NHWC));
+        add_config(TensorShape(13U, 784U, 8U, 5U), WinogradInfo(Size2D(1U, 2U), Size2D(1U, 5U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NHWC));
+    }
+};
+
 } // namespace datasets
 } // namespace test
 } // namespace arm_compute
diff --git a/tests/validation/CL/Winograd.cpp b/tests/validation/CL/Winograd.cpp
index 1042dd7e08..62f0335253 100644
--- a/tests/validation/CL/Winograd.cpp
+++ b/tests/validation/CL/Winograd.cpp
@@ -157,11 +157,15 @@ const auto LargeWinogradFilterTransformDatasetNHWC_F32 =
 // Output transform
 const auto SmallWinogradOutputTransformDatasetNCHW = datasets::SmallWinogradOutputTransformDatasetNCHW();
 
-const auto SmallWinogradOutputTransformDatasetNHWC = datasets::SmallWinogradOutputTransformDatasetNHWC();
+const auto SmallWinogradOutputTransformDatasetNHWC_F16 = datasets::SmallWinogradOutputTransformDatasetNHWC_F16();
+
+const auto SmallWinogradOutputTransformDatasetNHWC_F32 = datasets::SmallWinogradOutputTransformDatasetNHWC_F32();
 
 const auto LargeWinogradOutputTransformDatasetNCHW = datasets::LargeWinogradOutputTransformDatasetNCHW();
 
-const auto LargeWinogradOutputTransformDatasetNHWC = datasets::LargeWinogradOutputTransformDatasetNHWC();
+const auto LargeWinogradOutputTransformDatasetNHWC_F16 = datasets::LargeWinogradOutputTransformDatasetNHWC_F16();
+
+const auto LargeWinogradOutputTransformDatasetNHWC_F32 = datasets::LargeWinogradOutputTransformDatasetNHWC_F32();
 
 //Activation Functions
 const auto ActivationFunctionsDataset = framework::dataset::make("ActivationInfo",
@@ -514,7 +518,7 @@ TEST_SUITE_END() // NCHW
 TEST_SUITE(NHWC)
 TEST_SUITE(FP16)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradOutputTransformFixtureFP16, framework::DatasetMode::ALL,
-                       combine(combine(SmallWinogradOutputTransformDatasetNHWC,
+                       combine(combine(SmallWinogradOutputTransformDatasetNHWC_F16,
                                framework::dataset::make("DataType", { DataType::F16 })),
                                framework::dataset::make("ActivationInfo",{ ActivationLayerInfo() }) ))
 {
@@ -523,7 +527,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradOutputTransformFixtureFP16, framework
 }
 
 FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradOutputTransformFixtureFP16, framework::DatasetMode::NIGHTLY,
-                       combine(combine(LargeWinogradOutputTransformDatasetNHWC,
+                       combine(combine(LargeWinogradOutputTransformDatasetNHWC_F16,
                                framework::dataset::make("DataType", { DataType::F16 })),
                                framework::dataset::make("ActivationInfo",{ ActivationLayerInfo() }) ))
 {
@@ -533,7 +537,7 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradOutputTransformFixtureFP16, framework
 TEST_SUITE_END() // FP16
 TEST_SUITE(FP32)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradOutputTransformFixtureFP32, framework::DatasetMode::ALL,
-                       combine(combine(SmallWinogradOutputTransformDatasetNHWC,
+                       combine(combine(SmallWinogradOutputTransformDatasetNHWC_F32,
                                framework::dataset::make("DataType", { DataType::F32 })),
                                framework::dataset::make("ActivationInfo",{ ActivationLayerInfo() }) ))
 {
@@ -542,7 +546,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradOutputTransformFixtureFP32, framework
 }
 
 FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradOutputTransformFixtureFP32, framework::DatasetMode::NIGHTLY,
-                       combine(combine(LargeWinogradOutputTransformDatasetNHWC,
+                       combine(combine(LargeWinogradOutputTransformDatasetNHWC_F32,
                                framework::dataset::make("DataType", { DataType::F32 })),
                                framework::dataset::make("ActivationInfo",{ ActivationLayerInfo() }) ))
 {
diff --git a/tests/validation/reference/Winograd.cpp b/tests/validation/reference/Winograd.cpp
index 5525bc4535..06463d577f 100644
--- a/tests/validation/reference/Winograd.cpp
+++ b/tests/validation/reference/Winograd.cpp
@@ -205,6 +205,7 @@ void initialize_matrix_transform(SimpleTensor<T> &src, const Size2D &output_tile
         { WinogradKey(std::pair<int, int>(4, 1), std::pair<int, int>(5, 1), WinogradTransformType::OUTPUT), omatrix4x4_5x5 },
         { WinogradKey(std::pair<int, int>(2, 1), std::pair<int, int>(7, 1), WinogradTransformType::OUTPUT), omatrix2x1_7x7 },
         { WinogradKey(std::pair<int, int>(1, 2), std::pair<int, int>(1, 7), WinogradTransformType::OUTPUT), omatrix2x1_7x7 },
+        { WinogradKey(std::pair<int, int>(2, 2), std::pair<int, int>(7, 7), WinogradTransformType::OUTPUT), omatrix2x1_7x7 },
         { WinogradKey(std::pair<int, int>(1, 4), std::pair<int, int>(1, 5), WinogradTransformType::OUTPUT), omatrix4x4_5x5 },
     };