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

Change-Id: I036558d832c697da1fe9ea04ada0df38dc793914
Signed-off-by: giuros01 <giuseppe.rossini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/923
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

3 files changed, 114 insertions, 6 deletions

diff --git a/tests/datasets/ShapeDatasets.h b/tests/datasets/ShapeDatasets.h
index b6923c15c9..f3e3220fd6 100644
--- a/tests/datasets/ShapeDatasets.h
+++ b/tests/datasets/ShapeDatasets.h
@@ -580,6 +580,102 @@ public:
     }
 };
 
+/** Data set containing small 1x7 tensor shapes. */
+class Small1x7Shapes final : public ShapeDataset
+{
+public:
+    Small1x7Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 1U, 7U, 7U, 4U },
+                     TensorShape{ 1U, 7U, 4U, 13U },
+                     TensorShape{ 1U, 7U, 9U, 2U },
+                     TensorShape{ 1U, 7U, 3U, 5U },
+    })
+    {
+    }
+};
+
+/** Data set containing large 1x7 tensor shapes. */
+class Large1x7Shapes final : public ShapeDataset
+{
+public:
+    Large1x7Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 1U, 7U, 32U, 64U },
+                     TensorShape{ 1U, 7U, 51U, 13U },
+                     TensorShape{ 1U, 7U, 53U, 47U },
+                     TensorShape{ 1U, 7U, 128U, 384U },
+    })
+    {
+    }
+};
+
+/** Data set containing small 7x7 tensor shapes. */
+class Small7x7Shapes final : public ShapeDataset
+{
+public:
+    Small7x7Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 7U, 7U, 7U, 4U },
+                     TensorShape{ 7U, 7U, 4U, 13U },
+                     TensorShape{ 7U, 7U, 9U, 2U },
+                     TensorShape{ 7U, 7U, 3U, 5U },
+    })
+    {
+    }
+};
+
+/** Data set containing large 7x7 tensor shapes. */
+class Large7x7Shapes final : public ShapeDataset
+{
+public:
+    Large7x7Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 7U, 7U, 32U, 64U },
+                     TensorShape{ 7U, 7U, 51U, 13U },
+                     TensorShape{ 7U, 7U, 53U, 47U },
+                     TensorShape{ 7U, 7U, 128U, 384U },
+    })
+    {
+    }
+};
+
+/** Data set containing small 7x1 tensor shapes. */
+class Small7x1Shapes final : public ShapeDataset
+{
+public:
+    Small7x1Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 7U, 1U, 7U, 4U },
+                     TensorShape{ 7U, 1U, 4U, 13U },
+                     TensorShape{ 7U, 1U, 9U, 2U },
+                     TensorShape{ 7U, 1U, 3U, 5U },
+    })
+    {
+    }
+};
+
+/** Data set containing large 7x1 tensor shapes. */
+class Large7x1Shapes final : public ShapeDataset
+{
+public:
+    Large7x1Shapes()
+        : ShapeDataset("Shape",
+    {
+        TensorShape{ 7U, 1U, 32U, 64U },
+                     TensorShape{ 7U, 1U, 51U, 13U },
+                     TensorShape{ 7U, 1U, 53U, 47U },
+                     TensorShape{ 7U, 1U, 128U, 384U },
+    })
+    {
+    }
+};
+
 /** Data set containing small tensor shapes for deconvolution. */
 class SmallDeconvolutionShapes final : public ShapeDataset
 {
diff --git a/tests/validation/CL/Winograd.cpp b/tests/validation/CL/Winograd.cpp
index f933a287cc..1042dd7e08 100644
--- a/tests/validation/CL/Winograd.cpp
+++ b/tests/validation/CL/Winograd.cpp
@@ -118,7 +118,7 @@ const auto SmallWinogradFilterTransformDatasetNCHW =
            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 =
+const auto SmallWinogradFilterTransformDatasetNHWC_F16 =
            framework::dataset::concat(combine(datasets::Small3x3Shapes(), 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) })),
@@ -126,6 +126,11 @@ const auto SmallWinogradFilterTransformDatasetNHWC =
            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_F32 =
+           framework::dataset::concat(SmallWinogradFilterTransformDatasetNHWC_F16,
+           framework::dataset::concat(combine(datasets::Small7x7Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 2U) })),
+           framework::dataset::concat(combine(datasets::Small7x1Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 1U) })),
+                                      combine(datasets::Small1x7Shapes(), framework::dataset::make("OutputTile", { Size2D(1U, 2U) })))));
 
 const auto LargeWinogradFilterTransformDatasetNCHW =
            framework::dataset::concat(combine(datasets::Large3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 2U), Size2D(4U, 4U) })),
@@ -135,7 +140,7 @@ const auto LargeWinogradFilterTransformDatasetNCHW =
            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 =
+const auto LargeWinogradFilterTransformDatasetNHWC_F16 =
            framework::dataset::concat(combine(datasets::Large3x3Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 4U) })),
            framework::dataset::concat(combine(datasets::Large3x1Shapes(), framework::dataset::make("OutputTile", { Size2D(4U, 1U) })),
            framework::dataset::concat(combine(datasets::Large1x3Shapes(), framework::dataset::make("OutputTile", { Size2D(1U, 4U) })),
@@ -143,6 +148,12 @@ const auto LargeWinogradFilterTransformDatasetNHWC =
            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_F32 =
+           framework::dataset::concat(LargeWinogradFilterTransformDatasetNHWC_F16,
+           framework::dataset::concat(combine(datasets::Large7x7Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 2U) })),
+           framework::dataset::concat(combine(datasets::Large7x1Shapes(), framework::dataset::make("OutputTile", { Size2D(2U, 1U) })),
+                                      combine(datasets::Large1x7Shapes(), framework::dataset::make("OutputTile", { Size2D(1U, 2U) })))));
+
 // Output transform
 const auto SmallWinogradOutputTransformDatasetNCHW = datasets::SmallWinogradOutputTransformDatasetNCHW();
 
@@ -364,7 +375,7 @@ TEST_SUITE_END() // NCHW
 TEST_SUITE(NHWC)
 TEST_SUITE(FP16)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradFilterTransformFixtureFP16, framework::DatasetMode::PRECOMMIT,
-                       combine(combine(SmallWinogradFilterTransformDatasetNHWC,
+                       combine(combine(SmallWinogradFilterTransformDatasetNHWC_F16,
                                        framework::dataset::make("DataLayout", { DataLayout::NHWC })),
                                        framework::dataset::make("DataType", { DataType::F16 })))
 {
@@ -373,7 +384,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradFilterTransformFixtureFP16, framework
 }
 
 FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradFilterTransformFixtureFP16, framework::DatasetMode::NIGHTLY,
-                       combine(combine(LargeWinogradFilterTransformDatasetNHWC,
+                       combine(combine(LargeWinogradFilterTransformDatasetNHWC_F16,
                                        framework::dataset::make("DataLayout", { DataLayout::NHWC })),
                                        framework::dataset::make("DataType", { DataType::F16 })))
 {
@@ -383,7 +394,7 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradFilterTransformFixtureFP16, framework
 TEST_SUITE_END() // FP16
 TEST_SUITE(FP32)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradFilterTransformFixtureFP32, framework::DatasetMode::PRECOMMIT,
-                       combine(combine(SmallWinogradFilterTransformDatasetNHWC,
+                       combine(combine(SmallWinogradFilterTransformDatasetNHWC_F32,
                                        framework::dataset::make("DataLayout", { DataLayout::NHWC })),
                                        framework::dataset::make("DataType", { DataType::F32 })))
 {
@@ -392,7 +403,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradFilterTransformFixtureFP32, framework
 }
 
 FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradFilterTransformFixtureFP32, framework::DatasetMode::NIGHTLY,
-                       combine(combine(LargeWinogradFilterTransformDatasetNHWC,
+                       combine(combine(LargeWinogradFilterTransformDatasetNHWC_F32,
                                        framework::dataset::make("DataLayout", { DataLayout::NHWC })),
                                        framework::dataset::make("DataType", { DataType::F32 })))
 {
diff --git a/tests/validation/reference/Winograd.cpp b/tests/validation/reference/Winograd.cpp
index f09b2205d9..5525bc4535 100644
--- a/tests/validation/reference/Winograd.cpp
+++ b/tests/validation/reference/Winograd.cpp
@@ -193,6 +193,7 @@ 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::FILTER), fmatrix4x4_5x5 },
         { WinogradKey(std::pair<int, int>(2, 1), std::pair<int, int>(7, 1), WinogradTransformType::FILTER), fmatrix2x1_7x7 },
         { WinogradKey(std::pair<int, int>(1, 2), std::pair<int, int>(1, 7), WinogradTransformType::FILTER), fmatrix2x1_7x7 },
+        { WinogradKey(std::pair<int, int>(2, 2), std::pair<int, int>(7, 7), WinogradTransformType::FILTER), fmatrix2x1_7x7 },
         { 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 },