COMPMID-2322: PRELU support in CLActivationLayer

Change-Id: I3aa8d4964a1861c2b8acef59dc863df7c48f3316
Signed-off-by: giuros01 <giuseppe.rossini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1146
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

3 files changed, 321 insertions, 0 deletions

diff --git a/tests/validation/CL/PReluLayer.cpp b/tests/validation/CL/PReluLayer.cpp
new file mode 100644
index 0000000000..32fb2a113b
--- /dev/null
+++ b/tests/validation/CL/PReluLayer.cpp
@@ -0,0 +1,258 @@
+/*
+ * Copyright (c) 2019 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.
+ */
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/CLTensorAllocator.h"
+#include "arm_compute/runtime/CL/functions/CLPReluLayer.h"
+#include "tests/CL/CLAccessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/ConvertPolicyDataset.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/ElementwiseOperationsFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+RelativeTolerance<float> tolerance_fp32(0.000001f);
+RelativeTolerance<float> tolerance_fp16(0.001f);
+
+constexpr unsigned int num_elems_processed_per_iteration = 16;
+/** Input data sets **/
+const auto PReluLayerU8Dataset = combine(combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::U8)),
+                                         framework::dataset::make("DataType",
+                                                                  DataType::U8));
+const auto PReluLayerQASYMM8Dataset = combine(combine(framework::dataset::make("DataType", DataType::QASYMM8), framework::dataset::make("DataType", DataType::QASYMM8)),
+                                              framework::dataset::make("DataType",
+                                                                       DataType::QASYMM8));
+const auto PReluLayerS16Dataset = combine(combine(framework::dataset::make("DataType", { DataType::U8, DataType::S16 }), framework::dataset::make("DataType", DataType::S16)),
+                                          framework::dataset::make("DataType", DataType::S16));
+const auto PReluLayerFP16Dataset = combine(combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::F16)),
+                                           framework::dataset::make("DataType", DataType::F16));
+const auto PReluLayerFP32Dataset = combine(combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::F32)),
+                                           framework::dataset::make("DataType", DataType::F32));
+} // namespace
+
+TEST_SUITE(CL)
+TEST_SUITE(PReluLayer)
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
+               framework::dataset::make("Input1Info", { TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                                        TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::U8),      // Window shrink
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),      // Invalid data type combination
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),     // Mismatching shapes
+                                                      }),
+               framework::dataset::make("Input2Info",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                                       TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::U8),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S16),
+                                                       TensorInfo(TensorShape(48U, 11U, 2U), 1, DataType::F32),
+                                                     })),
+               framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S16),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                                       TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::U8),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                                       TensorInfo(TensorShape(48U, 11U, 2U), 1, DataType::F32),
+                                                     })),
+               framework::dataset::make("Expected", { true, true, false, false, false})),
+               input1_info, input2_info, output_info, expected)
+{
+    ARM_COMPUTE_EXPECT(bool(CLPReluLayer::validate(&input1_info.clone()->set_is_resizable(false), &input2_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false))) == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
+template <typename T>
+using CLPReluLayerFixture = PReluLayerValidationFixture<CLTensor, CLAccessor, CLPReluLayer, T>;
+
+TEST_SUITE(U8)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, datasets::SmallShapes(),
+               shape)
+{
+    // Create tensors
+    CLTensor ref_src1 = create_tensor<CLTensor>(shape, DataType::U8);
+    CLTensor ref_src2 = create_tensor<CLTensor>(shape, DataType::U8);
+    CLTensor dst      = create_tensor<CLTensor>(shape, DataType::U8);
+
+    // Create and Configure function
+    CLPReluLayer prelu;
+    prelu.configure(&ref_src1, &ref_src2, &dst);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(shape);
+    validate(dst.info()->valid_region(), valid_region);
+
+    // Validate padding
+    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
+    validate(ref_src1.info()->padding(), padding);
+    validate(ref_src2.info()->padding(), padding);
+    validate(dst.info()->padding(), padding);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPReluLayerFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallShapes(), PReluLayerU8Dataset))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END()
+
+template <typename T>
+using CLPReluLayerQuantizedFixture = PReluLayerValidationQuantizedFixture<CLTensor, CLAccessor, CLPReluLayer, T>;
+
+TEST_SUITE(Quantized)
+TEST_SUITE(QASYMM8)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, datasets::SmallShapes(),
+               shape)
+{
+    // Create tensors
+    CLTensor ref_src1 = create_tensor<CLTensor>(shape, DataType::QASYMM8);
+    CLTensor ref_src2 = create_tensor<CLTensor>(shape, DataType::QASYMM8);
+    CLTensor dst      = create_tensor<CLTensor>(shape, DataType::QASYMM8);
+
+    // Create and Configure function
+    CLPReluLayer prelu;
+    prelu.configure(&ref_src1, &ref_src2, &dst);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(shape);
+    validate(dst.info()->valid_region(), valid_region);
+
+    // Validate padding
+    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
+    validate(ref_src1.info()->padding(), padding);
+    validate(ref_src2.info()->padding(), padding);
+    validate(dst.info()->padding(), padding);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPReluLayerQuantizedFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(datasets::SmallShapes(),
+                                                                                                                   PReluLayerQASYMM8Dataset),
+                                                                                                                   framework::dataset::make("QuantizationInfo", { QuantizationInfo(5.f / 255.f, 20) })),
+                                                                                                                   framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255.f, 10) })),
+                                                                                                                   framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f / 255.f, 5) }))
+
+                      )
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_fp32, 0.01);
+}
+TEST_SUITE_END()
+TEST_SUITE_END()
+
+TEST_SUITE(S16)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), framework::dataset::make("DataType", { DataType::U8, DataType::S16 })),
+               shape, data_type)
+{
+    // Create tensors
+    CLTensor ref_src1 = create_tensor<CLTensor>(shape, data_type);
+    CLTensor ref_src2 = create_tensor<CLTensor>(shape, DataType::S16);
+    CLTensor dst      = create_tensor<CLTensor>(shape, DataType::S16);
+
+    // Create and Configure function
+    CLPReluLayer prelu;
+    prelu.configure(&ref_src1, &ref_src2, &dst);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(shape);
+    validate(dst.info()->valid_region(), valid_region);
+
+    // Validate padding
+    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
+    validate(ref_src1.info()->padding(), padding);
+    validate(ref_src2.info()->padding(), padding);
+    validate(dst.info()->padding(), padding);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPReluLayerFixture<int16_t>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), PReluLayerS16Dataset))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END()
+
+TEST_SUITE(Float)
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPReluLayerFixture<half>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), PReluLayerFP16Dataset))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_fp16, 0.01);
+}
+TEST_SUITE_END()
+
+TEST_SUITE(FP32)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, datasets::SmallShapes(),
+               shape)
+{
+    // Create tensors
+    CLTensor ref_src1 = create_tensor<CLTensor>(shape, DataType::F32);
+    CLTensor ref_src2 = create_tensor<CLTensor>(shape, DataType::F32);
+    CLTensor dst      = create_tensor<CLTensor>(shape, DataType::F32);
+
+    // Create and Configure function
+    CLPReluLayer prelu;
+    prelu.configure(&ref_src1, &ref_src2, &dst);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(shape);
+    validate(dst.info()->valid_region(), valid_region);
+
+    // Validate padding
+    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
+    validate(ref_src1.info()->padding(), padding);
+    validate(ref_src2.info()->padding(), padding);
+    validate(dst.info()->padding(), padding);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall, CLPReluLayerFixture<float>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), PReluLayerFP32Dataset))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_fp32);
+}
+template <typename T>
+using CLPReluLayerBroadcastFixture = PReluLayerBroadcastValidationFixture<CLTensor, CLAccessor, CLPReluLayer, T>;
+
+FIXTURE_DATA_TEST_CASE(RunSmallBroadcast, CLPReluLayerBroadcastFixture<float>, framework::DatasetMode::ALL, combine(datasets::SmallShapesBroadcast(),
+                                                                                                                    PReluLayerFP32Dataset))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_fp32);
+}
+TEST_SUITE_END()
+TEST_SUITE_END()
+
+TEST_SUITE_END()
+TEST_SUITE_END()
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/ElementwiseOperationsFixture.h b/tests/validation/fixtures/ElementwiseOperationsFixture.h
index e86e7a0f20..de61c487e6 100644
--- a/tests/validation/fixtures/ElementwiseOperationsFixture.h
+++ b/tests/validation/fixtures/ElementwiseOperationsFixture.h
@@ -337,6 +337,62 @@ public:
 };
 
 template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class PReluLayerBroadcastValidationFixture : public ArithmeticOperationsGenericFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(const TensorShape &shape0, const TensorShape &shape1, DataType data_type0, DataType data_type1, DataType output_data_type)
+    {
+        ArithmeticOperationsGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(ArithmeticOperation::PRELU, shape0, shape1,
+                                                                                             data_type0, data_type1, output_data_type,
+                                                                                             QuantizationInfo(), QuantizationInfo(), QuantizationInfo());
+    }
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class PReluLayerValidationFixture : public ArithmeticOperationsGenericFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(const TensorShape &shape, DataType data_type0, DataType data_type1, DataType output_data_type)
+    {
+        ArithmeticOperationsGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(ArithmeticOperation::PRELU, shape, shape,
+                                                                                             data_type0, data_type1, output_data_type,
+                                                                                             QuantizationInfo(), QuantizationInfo(), QuantizationInfo());
+    }
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class PReluLayerValidationQuantizedFixture : public ArithmeticOperationsGenericFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(const TensorShape &shape, DataType data_type0, DataType data_type1, DataType output_data_type,
+               QuantizationInfo qinfo0, QuantizationInfo qinfo1, QuantizationInfo qinfo_out)
+
+    {
+        ArithmeticOperationsGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(ArithmeticOperation::PRELU, shape, shape,
+                                                                                             data_type0, data_type1, output_data_type,
+                                                                                             qinfo0, qinfo1, qinfo_out);
+    }
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class PReluLayerQuantizedBroadcastValidationFixture : public ArithmeticOperationsGenericFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(const TensorShape &shape0, const TensorShape &shape1, DataType data_type0, DataType data_type1, DataType output_data_type,
+               QuantizationInfo qinfo0, QuantizationInfo qinfo1, QuantizationInfo qinfo_out)
+
+    {
+        ArithmeticOperationsGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(ArithmeticOperation::PRELU, shape0, shape1,
+                                                                                             data_type0, data_type1, output_data_type,
+                                                                                             qinfo0, qinfo1, qinfo_out);
+    }
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
 class ElementwiseDivisionBroadcastValidationFixture : public ArithmeticOperationsGenericFixture<TensorType, AccessorType, FunctionType, T>
 {
 public:
diff --git a/tests/validation/reference/ElementwiseOperations.cpp b/tests/validation/reference/ElementwiseOperations.cpp
index 82f42a0c21..44eb417969 100644
--- a/tests/validation/reference/ElementwiseOperations.cpp
+++ b/tests/validation/reference/ElementwiseOperations.cpp
@@ -81,6 +81,13 @@ T arithm_op(ArithmeticOperation op, T src1, T src2, ConvertPolicy convert_policy
             val = std::pow(static_cast<intermediate_type>(src1), static_cast<intermediate_type>(src2));
             break;
         }
+        case ArithmeticOperation::PRELU:
+        {
+            const T x     = static_cast<intermediate_type>(src1);
+            const T alpha = static_cast<intermediate_type>(src2);
+            val           = (x > 0 ? x : alpha * x);
+            break;
+        }
         default:
         {
             ARM_COMPUTE_ERROR("Not handled");