Add sigmoid and tanh for dynamic fusion

* Add sigmoid and tanh activation functions for dynamic fusion.
* Add corresponding tests, but both activation functions share
  the same fixture implementation.

Resolves: COMPMID-5939
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Change-Id: I0aae0eaa18b746ce89680d2773c66e09b0f854ce
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9257
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

3 files changed, 520 insertions, 0 deletions

diff --git a/tests/validation/dynamic_fusion/gpu/cl/Sigmoid.cpp b/tests/validation/dynamic_fusion/gpu/cl/Sigmoid.cpp
new file mode 100644
index 0000000000..aace23eff4
--- /dev/null
+++ b/tests/validation/dynamic_fusion/gpu/cl/Sigmoid.cpp
@@ -0,0 +1,158 @@
+/*
+ * Copyright (c) 2023 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/dynamic_fusion/sketch/gpu/operators/GpuSigmoid.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
+
+#include "tests/CL/CLAccessor.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/dynamic_fusion/operators/ActivationFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+constexpr AbsoluteTolerance<float> tolerance_f32(1e-6f);
+constexpr AbsoluteTolerance<float> tolerance_f16(0.001f);
+} // namespace
+
+TEST_SUITE(CL)
+TEST_SUITE(DYNAMIC_FUSION)
+TEST_SUITE(SIGMOID)
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(
+                framework::dataset::make("InputInfo", { TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F16),
+                                                        TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::QASYMM8), // Unsupported data type
+                                                    }),
+                framework::dataset::make("Expected", { true, true, false })),
+                input_info, expected)
+{
+    // Create a new workload sketch
+    CLCompileContext cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
+    GpuWorkloadContext gpu_ctx{ &cl_compile_ctx };
+    GpuWorkloadSketch sketch{ &gpu_ctx };
+
+    // Fuse sigmoid
+    const TensorInfo src_info = sketch.create_tensor_info(input_info);
+
+    const bool res = static_cast<bool>(GpuSigmoid::validate_op(sketch, &src_info));
+    ARM_COMPUTE_EXPECT(res == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
+template <typename T>
+using DynamicFusionSigmoidOpFixture = DynamicFusionSigmoidValidationFixture<CLTensor, CLAccessor, GpuSigmoid, T>;
+
+TEST_SUITE(Float)
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(RunSmallOneOp,
+                       DynamicFusionSigmoidOpFixture<half>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallShapes(),
+                                       framework::dataset::make("Fuse", { false })),
+                               framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall5dOneOp,
+                       DynamicFusionSigmoidOpFixture<half>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::Small5dShapes(),
+                                       framework::dataset::make("Fuse", { false })),
+                               framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmallTwoOps,
+                       DynamicFusionSigmoidOpFixture<half>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallShapes(),
+                                       framework::dataset::make("Fuse", { true })),
+                               framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+TEST_SUITE_END() // FP16
+
+TEST_SUITE(FP32)
+FIXTURE_DATA_TEST_CASE(RunSmallOneOp,
+                       DynamicFusionSigmoidOpFixture<float>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallShapes(),
+                                       framework::dataset::make("Fuse", { false })),
+                               framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall5dOneOp,
+                       DynamicFusionSigmoidOpFixture<float>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::Small5dShapes(),
+                                       framework::dataset::make("Fuse", { false })),
+                               framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmallTwoOps,
+                       DynamicFusionSigmoidOpFixture<float>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallShapes(),
+                                       framework::dataset::make("Fuse", { true })),
+                               framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+TEST_SUITE_END() // FP32
+TEST_SUITE_END() // Float
+
+TEST_SUITE_END() // SIGMOID
+TEST_SUITE_END() // DYNAMIC_FUSION
+TEST_SUITE_END() // CL
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/dynamic_fusion/gpu/cl/Tanh.cpp b/tests/validation/dynamic_fusion/gpu/cl/Tanh.cpp
new file mode 100644
index 0000000000..183cd079a3
--- /dev/null
+++ b/tests/validation/dynamic_fusion/gpu/cl/Tanh.cpp
@@ -0,0 +1,158 @@
+/*
+ * Copyright (c) 2023 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/dynamic_fusion/sketch/gpu/operators/GpuTanh.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
+
+#include "tests/CL/CLAccessor.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/dynamic_fusion/operators/ActivationFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+constexpr AbsoluteTolerance<float> tolerance_f32(0.00001f);
+constexpr AbsoluteTolerance<float> tolerance_f16(0.001f);
+} // namespace
+
+TEST_SUITE(CL)
+TEST_SUITE(DYNAMIC_FUSION)
+TEST_SUITE(TANH)
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(
+                framework::dataset::make("InputInfo", { TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F16),
+                                                        TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::QASYMM8), // Unsupported data type
+                                                    }),
+                framework::dataset::make("Expected", { true, true, false })),
+                input_info, expected)
+{
+    // Create a new workload sketch
+    CLCompileContext cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
+    GpuWorkloadContext gpu_ctx{ &cl_compile_ctx };
+    GpuWorkloadSketch sketch{ &gpu_ctx };
+
+    // Fuse tanh
+    const TensorInfo src_info = sketch.create_tensor_info(input_info);
+
+    const bool res = static_cast<bool>(GpuTanh::validate_op(sketch, &src_info));
+    ARM_COMPUTE_EXPECT(res == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
+template <typename T>
+using DynamicFusionTanhOpFixture = DynamicFusionTanhValidationFixture<CLTensor, CLAccessor, GpuTanh, T>;
+
+TEST_SUITE(Float)
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(RunSmallOneOp,
+                       DynamicFusionTanhOpFixture<half>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallShapes(),
+                                       framework::dataset::make("Fuse", { false })),
+                               framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall5dOneOp,
+                       DynamicFusionTanhOpFixture<half>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::Small5dShapes(),
+                                       framework::dataset::make("Fuse", { false })),
+                               framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmallTwoOps,
+                       DynamicFusionTanhOpFixture<half>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallShapes(),
+                                       framework::dataset::make("Fuse", { true })),
+                               framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+TEST_SUITE_END() // FP16
+
+TEST_SUITE(FP32)
+FIXTURE_DATA_TEST_CASE(RunSmallOneOp,
+                       DynamicFusionTanhOpFixture<float>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallShapes(),
+                                       framework::dataset::make("Fuse", { false })),
+                               framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall5dOneOp,
+                       DynamicFusionTanhOpFixture<float>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::Small5dShapes(),
+                                       framework::dataset::make("Fuse", { false })),
+                               framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmallTwoOps,
+                       DynamicFusionTanhOpFixture<float>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallShapes(),
+                                       framework::dataset::make("Fuse", { true })),
+                               framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+TEST_SUITE_END() // FP32
+TEST_SUITE_END() // Float
+
+TEST_SUITE_END() // TANH
+TEST_SUITE_END() // DYNAMIC_FUSION
+TEST_SUITE_END() // CL
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/dynamic_fusion/operators/ActivationFixture.h b/tests/validation/fixtures/dynamic_fusion/operators/ActivationFixture.h
new file mode 100644
index 0000000000..9656c497ea
--- /dev/null
+++ b/tests/validation/fixtures/dynamic_fusion/operators/ActivationFixture.h
@@ -0,0 +1,204 @@
+/*
+ * Copyright (c) 2023 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 TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_ACTIVATIONFIXTURE
+#define TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_ACTIVATIONFIXTURE
+
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.h"
+#include "arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h"
+#include "arm_compute/dynamic_fusion/sketch/gpu/operators/GpuOutput.h"
+
+#include "tests/framework/Fixture.h"
+#include "tests/validation/reference/ActivationLayer.h"
+
+using namespace arm_compute::experimental::dynamic_fusion;
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T, typename... TArgs>
+class DynamicFusionActivationValidationFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape shape, bool fuse, DataType data_type, ActivationLayerInfo act_info, TArgs... args)
+    {
+        _fuse       = fuse;
+        _data_type  = data_type;
+        _function   = act_info.activation();
+        _target     = compute_target(shape, args...);
+        _reference  = compute_reference(shape, act_info);
+    }
+
+protected:
+    std::vector<T> get_boundary_values(T min, T max)
+    {
+        // This function will return a vector filled with the following values that can
+        // represent two partitions derived from equivalent partitioning.
+        // * Lower partition: min, min + delta, lower quarter (nominal), center - delta
+        // * Upper partition: center, center + delta, upper quarter (nominal), max - delta, max
+        const auto delta         = is_data_type_float(_data_type) ? T(0.1f) : T(1);
+        const auto center_value  = (min + max) / 2;
+        const auto lower_quarter = (min + center_value) / 2;
+        const auto upper_quarter = (center_value + max) / 2;
+
+        std::vector<T> boundary_values{};
+
+        // To ensure all the inserted values are within the given range after subtracing/adding delta
+        auto insert_values = [&boundary_values, &min, &max](const std::initializer_list<T> &new_values)
+        {
+            for(auto &v : new_values)
+            {
+                if(v >= min && v <= max)
+                {
+                    boundary_values.emplace_back(v);
+                }
+            }
+        };
+
+        insert_values({ min, static_cast<T>(min + delta), static_cast<T>(lower_quarter), static_cast<T>(center_value - delta) });                               // lower partition
+        insert_values({ static_cast<T>(center_value), static_cast<T>(center_value + delta), static_cast<T>(upper_quarter), static_cast<T>(max - delta), max }); // upper partition
+
+        return boundary_values;
+    }
+
+    template <typename U>
+    void fill(U &&tensor)
+    {
+        float min_bound = 0;
+        float max_bound = 0;
+        std::tie(min_bound, max_bound) = get_activation_layer_test_bounds<T>(_function, _data_type);
+        library->fill_static_values(tensor, get_boundary_values(static_cast<T>(min_bound), static_cast<T>(max_bound)));
+    }
+
+    TensorType compute_target(const TensorShape &shape, TArgs... args)
+    {
+        // Create a new workload sketch
+        CLCompileContext   cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
+        GpuWorkloadContext gpu_ctx{ &cl_compile_ctx };
+        GpuWorkloadSketch  sketch{ &gpu_ctx };
+
+        // Create sketch tensors
+        TensorInfo src_info = sketch.create_tensor_info(TensorInfo(shape, 1, _data_type));
+        TensorInfo dst_info = sketch.create_tensor_info(TensorInfo(shape, 1, _data_type));
+
+        ITensorInfo *ans_0_info = FunctionType::create_op(sketch, &src_info, args...);
+        if(_fuse)
+        {
+            ITensorInfo *ans_1_info = FunctionType::create_op(sketch, ans_0_info, args...);
+            GpuOutput::create_op(sketch, ans_1_info, &dst_info);
+        }
+        else
+        {
+            GpuOutput::create_op(sketch, ans_0_info, &dst_info);
+        }
+
+        // Configure runtime
+        ClWorkloadRuntime runtime;
+        runtime.configure(sketch);
+
+        // Construct user tensors
+        TensorType t_src{};
+        TensorType t_dst{};
+
+        // Initialize user tensors
+        t_src.allocator()->init(src_info);
+        t_dst.allocator()->init(dst_info);
+
+        // Allocate and fill user tensors
+        t_src.allocator()->allocate();
+        t_dst.allocator()->allocate();
+
+        fill(AccessorType(t_src));
+
+        // Run runtime
+        runtime.run({ &t_src, &t_dst });
+
+        return t_dst;
+    }
+
+    SimpleTensor<T> compute_reference(const TensorShape &shape, ActivationLayerInfo act_info)
+    {
+        // Create reference
+        SimpleTensor<T> src{ shape, _data_type, 1 };
+
+        // Fill reference
+        fill(src);
+
+        auto tmp = reference::activation_layer<T>(src, act_info);
+
+        if(_fuse)
+        {
+            auto dst = reference::activation_layer<T>(tmp, act_info);
+            return dst;
+        }
+        else
+        {
+            return tmp;
+        }
+    }
+
+protected:
+    ActivationLayerInfo::ActivationFunction _function{};
+    bool                                    _fuse{ false };
+    DataType                                _data_type{};
+    TensorType                              _target{};
+    SimpleTensor<T>                         _reference{};
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class DynamicFusionSigmoidValidationFixture : public DynamicFusionActivationValidationFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(TensorShape shape, bool fuse, DataType data_type)
+    {
+        ActivationLayerInfo act_info{ ActivationLayerInfo::ActivationFunction::LOGISTIC };
+        DynamicFusionActivationValidationFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, fuse, data_type, act_info);
+    }
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class DynamicFusionTanhValidationFixture : public DynamicFusionActivationValidationFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(TensorShape shape, bool fuse, DataType data_type)
+    {
+        ActivationLayerInfo act_info{ ActivationLayerInfo::ActivationFunction::TANH };
+        DynamicFusionActivationValidationFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, fuse, data_type, act_info);
+    }
+};
+
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+
+#endif /* TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_ACTIVATIONFIXTURE */