COMPMID-1017: Implement dilated convolution in NEON, OpenCL, and GC

Change-Id: If4626ec9e215e14dffe22e80812da5bac84a52e2
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/125734
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>

7 files changed, 211 insertions, 9 deletions

diff --git a/tests/benchmark/CL/DirectConvolutionLayer.cpp b/tests/benchmark/CL/DirectConvolutionLayer.cpp
index 27994b4458..c7b07807e9 100644
--- a/tests/benchmark/CL/DirectConvolutionLayer.cpp
+++ b/tests/benchmark/CL/DirectConvolutionLayer.cpp
@@ -27,7 +27,7 @@
 #include "arm_compute/runtime/CL/CLTensorAllocator.h"
 #include "arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h"
 #include "tests/CL/CLAccessor.h"
-#include "tests/benchmark/fixtures/ConvolutionLayerFixture.h"
+#include "tests/benchmark/fixtures/DirectConvolutionLayerFixture.h"
 #include "tests/datasets/system_tests/alexnet/AlexNetConvolutionLayerDataset.h"
 #include "tests/datasets/system_tests/googlenet/inceptionv1/GoogLeNetInceptionV1ConvolutionLayerDataset.h"
 #include "tests/datasets/system_tests/googlenet/inceptionv4/GoogLeNetInceptionV4ConvolutionLayerDataset.h"
@@ -49,7 +49,7 @@ namespace
 const auto data_types = framework::dataset::make("DataType", { DataType::F16, DataType::F32 });
 } // namespace
 
-using CLConvolutionLayerFixture = ConvolutionLayerFixture<CLTensor, CLDirectConvolutionLayer, CLAccessor>;
+using CLConvolutionLayerFixture = DirectConvolutionLayerFixture<CLTensor, CLDirectConvolutionLayer, CLAccessor>;
 
 TEST_SUITE(CL)
 
diff --git a/tests/benchmark/GLES_COMPUTE/DirectConvolutionLayer.cpp b/tests/benchmark/GLES_COMPUTE/DirectConvolutionLayer.cpp
index 784f8e85ef..d319c41088 100644
--- a/tests/benchmark/GLES_COMPUTE/DirectConvolutionLayer.cpp
+++ b/tests/benchmark/GLES_COMPUTE/DirectConvolutionLayer.cpp
@@ -27,7 +27,7 @@
 #include "arm_compute/runtime/GLES_COMPUTE/GCTensorAllocator.h"
 #include "arm_compute/runtime/GLES_COMPUTE/functions/GCDirectConvolutionLayer.h"
 #include "tests/GLES_COMPUTE/GCAccessor.h"
-#include "tests/benchmark/fixtures/ConvolutionLayerFixture.h"
+#include "tests/benchmark/fixtures/DirectConvolutionLayerFixture.h"
 #include "tests/datasets/system_tests/alexnet/AlexNetConvolutionLayerDataset.h"
 #include "tests/datasets/system_tests/googlenet/inceptionv1/GoogLeNetInceptionV1ConvolutionLayerDataset.h"
 #include "tests/datasets/system_tests/googlenet/inceptionv4/GoogLeNetInceptionV4ConvolutionLayerDataset.h"
@@ -49,7 +49,7 @@ namespace
 const auto data_types = framework::dataset::make("DataType", { DataType::F32, DataType::F16 });
 } // namespace
 
-using GCConvolutionLayerFixture = ConvolutionLayerFixture<GCTensor, GCDirectConvolutionLayer, GCAccessor>;
+using GCConvolutionLayerFixture = DirectConvolutionLayerFixture<GCTensor, GCDirectConvolutionLayer, GCAccessor>;
 
 TEST_SUITE(GC)
 
diff --git a/tests/benchmark/NEON/ConvolutionLayer.cpp b/tests/benchmark/NEON/ConvolutionLayer.cpp
index 9914d08c72..a425d95a6e 100644
--- a/tests/benchmark/NEON/ConvolutionLayer.cpp
+++ b/tests/benchmark/NEON/ConvolutionLayer.cpp
@@ -29,6 +29,7 @@
 #include "arm_compute/runtime/TensorAllocator.h"
 #include "tests/NEON/Accessor.h"
 #include "tests/benchmark/fixtures/ConvolutionLayerFixture.h"
+#include "tests/benchmark/fixtures/WinogradLayerFixture.h"
 #include "tests/datasets/system_tests/alexnet/AlexNetConvolutionLayerDataset.h"
 #include "tests/datasets/system_tests/googlenet/inceptionv1/GoogLeNetInceptionV1ConvolutionLayerDataset.h"
 #include "tests/datasets/system_tests/googlenet/inceptionv4/GoogLeNetInceptionV4ConvolutionLayerDataset.h"
@@ -61,7 +62,7 @@ using NEGEMMConvolutionLayerFixture = ConvolutionLayerFixture<Tensor, NEGEMMConv
 
 TEST_SUITE(NEON)
 #if defined(__aarch64__)
-using NEWinogradLayerFixture = ConvolutionLayerFixture<Tensor, NEWinogradLayer, Accessor>;
+using NEWinogradLayerFixture = WinogradLayerFixture<Tensor, NEWinogradLayer, Accessor>;
 
 REGISTER_FIXTURE_DATA_TEST_CASE(AlexNetWinogradLayer, NEWinogradLayerFixture, framework::DatasetMode::ALL,
                                 framework::dataset::combine(framework::dataset::combine(datasets::AlexNetWinogradLayerDataset(), framework::dataset::make("DataType", DataType::F32)),
diff --git a/tests/benchmark/NEON/DirectConvolutionLayer.cpp b/tests/benchmark/NEON/DirectConvolutionLayer.cpp
index 67b948568f..8a17f3c03c 100644
--- a/tests/benchmark/NEON/DirectConvolutionLayer.cpp
+++ b/tests/benchmark/NEON/DirectConvolutionLayer.cpp
@@ -27,7 +27,7 @@
 #include "arm_compute/runtime/Tensor.h"
 #include "arm_compute/runtime/TensorAllocator.h"
 #include "tests/NEON/Accessor.h"
-#include "tests/benchmark/fixtures/ConvolutionLayerFixture.h"
+#include "tests/benchmark/fixtures/DirectConvolutionLayerFixture.h"
 #include "tests/datasets/DirectConvolutionLayerDataset.h"
 #include "tests/datasets/system_tests/alexnet/AlexNetConvolutionLayerDataset.h"
 #include "tests/datasets/system_tests/googlenet/inceptionv1/GoogLeNetInceptionV1ConvolutionLayerDataset.h"
@@ -55,7 +55,7 @@ const auto data_types = framework::dataset::make("DataType", { DataType::F32 });
 #endif /* ARM_COMPUTE_ENABLE_F16 */
 } // namespace
 
-using NEConvolutionLayerFixture = ConvolutionLayerFixture<Tensor, NEDirectConvolutionLayer, Accessor>;
+using NEConvolutionLayerFixture = DirectConvolutionLayerFixture<Tensor, NEDirectConvolutionLayer, Accessor>;
 
 TEST_SUITE(NEON)
 
diff --git a/tests/benchmark/fixtures/ConvolutionLayerFixture.h b/tests/benchmark/fixtures/ConvolutionLayerFixture.h
index 9815040d42..7558b4c9a8 100644
--- a/tests/benchmark/fixtures/ConvolutionLayerFixture.h
+++ b/tests/benchmark/fixtures/ConvolutionLayerFixture.h
@@ -42,7 +42,7 @@ class ConvolutionLayerFixture : public framework::Fixture
 {
 public:
     template <typename...>
-    void setup(TensorShape src_shape, TensorShape weights_shape, TensorShape biases_shape, TensorShape dst_shape, PadStrideInfo info, DataType data_type, int batches)
+    void setup(TensorShape src_shape, TensorShape weights_shape, TensorShape biases_shape, TensorShape dst_shape, PadStrideInfo info, Size2D dilation, DataType data_type, int batches)
     {
         // Set batched in source and destination shapes
         const unsigned int fixed_point_position = 4;
@@ -57,7 +57,7 @@ public:
         dst     = create_tensor<TensorType>(dst_shape, data_type, 1, fixed_point_position);
 
         // Create and configure function
-        conv_layer.configure(&src, &weights, &biases, &dst, info);
+        conv_layer.configure(&src, &weights, &biases, &dst, info, WeightsInfo(), dilation);
 
         // Allocate tensors
         src.allocator()->allocate();
diff --git a/tests/benchmark/fixtures/DirectConvolutionLayerFixture.h b/tests/benchmark/fixtures/DirectConvolutionLayerFixture.h
new file mode 100644
index 0000000000..e3289b7fbc
--- /dev/null
+++ b/tests/benchmark/fixtures/DirectConvolutionLayerFixture.h
@@ -0,0 +1,101 @@
+/*
+ * Copyright (c) 2017-2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_TEST_DIRECTCONVOLUTIONLAYERFIXTURE
+#define ARM_COMPUTE_TEST_DIRECTCONVOLUTIONLAYERFIXTURE
+
+#include "arm_compute/core/Error.h"
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+#include "tests/Globals.h"
+#include "tests/Utils.h"
+#include "tests/framework/Fixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace benchmark
+{
+/** Fixture that can be used for NEON and CL */
+template <typename TensorType, typename Function, typename Accessor>
+class DirectConvolutionLayerFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape src_shape, TensorShape weights_shape, TensorShape biases_shape, TensorShape dst_shape, PadStrideInfo info, Size2D dilation, DataType data_type, int batches)
+    {
+        ARM_COMPUTE_UNUSED(dilation);
+
+        // Set batched in source and destination shapes
+        const unsigned int fixed_point_position = 4;
+        src_shape.set(3 /* batch */, batches);
+        dst_shape.set(3 /* batch */, batches);
+        DataType bias_data_type = is_data_type_quantized_asymmetric(data_type) ? DataType::S32 : data_type;
+
+        // Create tensors
+        src     = create_tensor<TensorType>(src_shape, data_type, 1, fixed_point_position);
+        weights = create_tensor<TensorType>(weights_shape, data_type, 1, fixed_point_position);
+        biases  = create_tensor<TensorType>(biases_shape, bias_data_type, 1, fixed_point_position);
+        dst     = create_tensor<TensorType>(dst_shape, data_type, 1, fixed_point_position);
+
+        // Create and configure function
+        conv_layer.configure(&src, &weights, &biases, &dst, info);
+
+        // Allocate tensors
+        src.allocator()->allocate();
+        weights.allocator()->allocate();
+        biases.allocator()->allocate();
+        dst.allocator()->allocate();
+    }
+
+    void run()
+    {
+        conv_layer.run();
+    }
+
+    void sync()
+    {
+        sync_if_necessary<TensorType>();
+        sync_tensor_if_necessary<TensorType>(dst);
+    }
+
+    void teardown()
+    {
+        src.allocator()->free();
+        weights.allocator()->free();
+        biases.allocator()->free();
+        dst.allocator()->free();
+    }
+
+private:
+    TensorType src{};
+    TensorType weights{};
+    TensorType biases{};
+    TensorType dst{};
+    Function   conv_layer{};
+};
+} // namespace benchmark
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_DIRECTCONVOLUTIONLAYERFIXTURE */
diff --git a/tests/benchmark/fixtures/WinogradLayerFixture.h b/tests/benchmark/fixtures/WinogradLayerFixture.h
new file mode 100644
index 0000000000..31a1eb8e1c
--- /dev/null
+++ b/tests/benchmark/fixtures/WinogradLayerFixture.h
@@ -0,0 +1,100 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_TEST_WINOGRADLAYERFIXTURE
+#define ARM_COMPUTE_TEST_WINOGRADLAYERFIXTURE
+
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+#include "tests/Globals.h"
+#include "tests/Utils.h"
+#include "tests/framework/Fixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace benchmark
+{
+/** Fixture that can be used for NEON and CL */
+template <typename TensorType, typename Function, typename Accessor>
+class WinogradLayerFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape src_shape, TensorShape weights_shape, TensorShape biases_shape, TensorShape dst_shape, PadStrideInfo info, Size2D dilation, DataType data_type, int batches)
+    {
+        ARM_COMPUTE_UNUSED(dilation);
+
+        // Set batched in source and destination shapes
+        const unsigned int fixed_point_position = 4;
+        src_shape.set(3 /* batch */, batches);
+        dst_shape.set(3 /* batch */, batches);
+        DataType bias_data_type = is_data_type_quantized_asymmetric(data_type) ? DataType::S32 : data_type;
+
+        // Create tensors
+        src     = create_tensor<TensorType>(src_shape, data_type, 1, fixed_point_position);
+        weights = create_tensor<TensorType>(weights_shape, data_type, 1, fixed_point_position);
+        biases  = create_tensor<TensorType>(biases_shape, bias_data_type, 1, fixed_point_position);
+        dst     = create_tensor<TensorType>(dst_shape, data_type, 1, fixed_point_position);
+
+        // Create and configure function
+        conv_layer.configure(&src, &weights, &biases, &dst, info);
+
+        // Allocate tensors
+        src.allocator()->allocate();
+        weights.allocator()->allocate();
+        biases.allocator()->allocate();
+        dst.allocator()->allocate();
+    }
+
+    void run()
+    {
+        conv_layer.run();
+    }
+
+    void sync()
+    {
+        sync_if_necessary<TensorType>();
+        sync_tensor_if_necessary<TensorType>(dst);
+    }
+
+    void teardown()
+    {
+        src.allocator()->free();
+        weights.allocator()->free();
+        biases.allocator()->free();
+        dst.allocator()->free();
+    }
+
+private:
+    TensorType src{};
+    TensorType weights{};
+    TensorType biases{};
+    TensorType dst{};
+    Function   conv_layer{};
+};
+} // namespace benchmark
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_WINOGRADLAYERFIXTURE */