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>

12 files changed, 452 insertions, 2 deletions

diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index 65db06b878..5a2ac51308 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -574,6 +574,7 @@ enum class ArithmeticOperation
     MAX,          /**< Max(x, y) */
     SQUARED_DIFF, /**< (x - y)^2 */
     POWER,        /**< x ^ y */
+    PRELU,        /**< y*x if x < 0, x otherwise */
 };
 
 /** Available element wise unary operations */
diff --git a/arm_compute/runtime/CL/CLFunctions.h b/arm_compute/runtime/CL/CLFunctions.h
index fbaab35414..11ef390c10 100644
--- a/arm_compute/runtime/CL/CLFunctions.h
+++ b/arm_compute/runtime/CL/CLFunctions.h
@@ -105,6 +105,7 @@
 #include "arm_compute/runtime/CL/functions/CLNormalizationLayer.h"
 #include "arm_compute/runtime/CL/functions/CLNormalizePlanarYUVLayer.h"
 #include "arm_compute/runtime/CL/functions/CLOpticalFlow.h"
+#include "arm_compute/runtime/CL/functions/CLPReluLayer.h"
 #include "arm_compute/runtime/CL/functions/CLPadLayer.h"
 #include "arm_compute/runtime/CL/functions/CLPermute.h"
 #include "arm_compute/runtime/CL/functions/CLPhase.h"
diff --git a/arm_compute/runtime/CL/functions/CLPReluLayer.h b/arm_compute/runtime/CL/functions/CLPReluLayer.h
new file mode 100644
index 0000000000..871b48e23e
--- /dev/null
+++ b/arm_compute/runtime/CL/functions/CLPReluLayer.h
@@ -0,0 +1,61 @@
+/*
+ * 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.
+ */
+#ifndef __ARM_COMPUTE_CLPRELULAYER_H__
+#define __ARM_COMPUTE_CLPRELULAYER_H__
+
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
+
+namespace arm_compute
+{
+class ICLTensor;
+
+/** Basic function to run @ref CLArithmeticOperationKernel for PRELU
+ *
+ * @note The function implements an activation layer with the PRELU activation function.
+ */
+class CLPReluLayer : public ICLSimpleFunction
+{
+public:
+    /** Set the input and output tensor.
+     *
+     * @note If the output tensor is a nullptr or is equal to the input, the activation function will be performed in-place
+     *
+     * @param[in]  input  Source tensor. Data types supported: QASYMM8/F16/F32.
+     * @param[in]  alpha  PRelu layer parameters. Data types supported: same of @p input.
+     * @param[out] output Destination tensor. Data type supported: same as @p input
+     */
+    void configure(ICLTensor *input, ICLTensor *alpha, ICLTensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLPReluLayer
+     *
+     * @param[in] input  Source tensor info. Data types supported: QASYMM8/F16/F32.
+     * @param[in] alpha  PRelu layer parameters. Data types supported: same of @p input.
+     * @param[in] output Destination tensor info. Data type supported: same as @p input
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *alpha, const ITensorInfo *output);
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_CLPRELULAYER_H__ */
diff --git a/src/core/CL/CLKernelLibrary.cpp b/src/core/CL/CLKernelLibrary.cpp
index 28152168bc..23eac1664a 100644
--- a/src/core/CL/CLKernelLibrary.cpp
+++ b/src/core/CL/CLKernelLibrary.cpp
@@ -253,12 +253,14 @@ const std::map<std::string, std::string> CLKernelLibrary::_kernel_program_map =
     { "elementwise_operation_DIV", "elementwise_operation.cl" },
     { "elementwise_operation_SQUARED_DIFF", "elementwise_operation.cl" },
     { "elementwise_operation_POWER", "elementwise_operation.cl" },
+    { "elementwise_operation_PRELU", "elementwise_operation.cl" },
     { "elementwise_operation_ADD_quantized", "elementwise_operation_quantized.cl" },
     { "elementwise_operation_SUB_quantized", "elementwise_operation_quantized.cl" },
     { "elementwise_operation_MAX_quantized", "elementwise_operation_quantized.cl" },
     { "elementwise_operation_MIN_quantized", "elementwise_operation_quantized.cl" },
     { "elementwise_operation_DIV_quantized", "elementwise_operation_quantized.cl" },
     { "elementwise_operation_SQUARED_DIFF_quantized", "elementwise_operation_quantized.cl" },
+    { "elementwise_operation_PRELU_quantized", "elementwise_operation_quantized.cl" },
     { "elementwise_unary", "elementwise_unary.cl" },
     { "erode", "erode.cl" },
     { "fast_corners", "fast_corners.cl" },
diff --git a/src/core/CL/cl_kernels/elementwise_operation.cl b/src/core/CL/cl_kernels/elementwise_operation.cl
index 0b660e4012..42d6d33e03 100644
--- a/src/core/CL/cl_kernels/elementwise_operation.cl
+++ b/src/core/CL/cl_kernels/elementwise_operation.cl
@@ -38,6 +38,7 @@
 #define SQUARED_DIFF(x, y) (x - y) * (x - y)
 #define DIV(x, y) (x / y)
 #define POWER(x, y) pow(x, y)
+#define PRELU(x, y) (select(y * x, x, x > (DATA_TYPE_OUT)0))
 
 #define OP_FUN_NAME_STR(op) elementwise_operation_##op
 #define OP_FUN_NAME(op) OP_FUN_NAME_STR(op)
diff --git a/src/core/CL/cl_kernels/elementwise_operation_quantized.cl b/src/core/CL/cl_kernels/elementwise_operation_quantized.cl
index 1f0533be13..1b45da164f 100644
--- a/src/core/CL/cl_kernels/elementwise_operation_quantized.cl
+++ b/src/core/CL/cl_kernels/elementwise_operation_quantized.cl
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -28,6 +28,7 @@
 #define MAX(x, y) max((x), (y))
 #define MIN(x, y) min((x), (y))
 #define SQUARED_DIFF(x, y) (x - y) * (x - y)
+#define PRELU(x, y) (select(y * x, x, x > (DATA_TYPE_OUT)0))
 #define DIV(x, y) (x / y)
 
 #define CONVERT_RTE(x, type) (convert_##type##_rte((x)))
diff --git a/src/core/CL/kernels/CLElementwiseOperationKernel.cpp b/src/core/CL/kernels/CLElementwiseOperationKernel.cpp
index ce0c51dac5..414b040f4c 100644
--- a/src/core/CL/kernels/CLElementwiseOperationKernel.cpp
+++ b/src/core/CL/kernels/CLElementwiseOperationKernel.cpp
@@ -43,6 +43,7 @@ std::map<ArithmeticOperation, std::string> supported_arithmetic_ops =
     { ArithmeticOperation::MIN, "MIN" },
     { ArithmeticOperation::MAX, "MAX" },
     { ArithmeticOperation::POWER, "POWER" },
+    { ArithmeticOperation::PRELU, "PRELU" },
 };
 
 std::map<ArithmeticOperation, std::string> supported_sat_arithmetic_ops =
diff --git a/src/runtime/CL/functions/CLElementwiseOperations.cpp b/src/runtime/CL/functions/CLElementwiseOperations.cpp
index 15de56d24a..8317e0d938 100644
--- a/src/runtime/CL/functions/CLElementwiseOperations.cpp
+++ b/src/runtime/CL/functions/CLElementwiseOperations.cpp
@@ -21,10 +21,11 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#include "arm_compute/runtime/CL/functions/CLElementwiseOperations.h"
+
 #include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/core/CL/kernels/CLElementwiseOperationKernel.h"
 #include "support/ToolchainSupport.h"
-#include <arm_compute/runtime/CL/functions/CLElementwiseOperations.h>
 
 #include <utility>
 
diff --git a/src/runtime/CL/functions/CLPReluLayer.cpp b/src/runtime/CL/functions/CLPReluLayer.cpp
new file mode 100644
index 0000000000..d463ef9c50
--- /dev/null
+++ b/src/runtime/CL/functions/CLPReluLayer.cpp
@@ -0,0 +1,60 @@
+/*
+ * 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/CL/kernels/CLElementwiseOperationKernel.h"
+
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/runtime/CL/functions/CLPReluLayer.h"
+#include "support/ToolchainSupport.h"
+
+namespace arm_compute
+{
+namespace
+{
+void configure_border_handler(CLFillBorderKernel &border_handler, BorderSize border_size, ICLTensor *input1, ICLTensor *input2, const ICLTensor *output)
+{
+    if(output->info()->dimension(0) > 1)
+    {
+        ICLTensor *broadcasted_info = (input1->info()->dimension(0) == 1) ? input1 : input2;
+
+        if(broadcasted_info->info()->dimension(0) == 1)
+        {
+            border_handler.configure(broadcasted_info, border_size, BorderMode::REPLICATE);
+        }
+    }
+}
+} // namespace
+
+void CLPReluLayer::configure(ICLTensor *input, ICLTensor *alpha, ICLTensor *output)
+{
+    auto k = arm_compute::support::cpp14::make_unique<CLArithmeticOperationKernel>();
+    k->configure(ArithmeticOperation::PRELU, input, alpha, output);
+    _kernel = std::move(k);
+    configure_border_handler(_border_handler, _kernel->border_size(), input, alpha, output);
+}
+
+Status CLPReluLayer::validate(const ITensorInfo *input, const ITensorInfo *alpha, const ITensorInfo *output)
+{
+    return CLArithmeticOperationKernel::validate(ArithmeticOperation::PRELU, input, alpha, output);
+}
+} // namespace arm_compute
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");