aidge_backend_cpu/unit_tests/test_scheduler.py

@@ -13,9 +13,10 @@ class test_scheduler(unittest.TestCase):
         pass
 
     def test_relu_forward(self):
-        values = np.arange(6) - 3
 
-        input_node = aidge_core.Producer(aidge_core.Tensor(values), "Input")
+        t = aidge_core.Tensor(np.arange(6, dtype=np.int32) - 3)
+
+        input_node = aidge_core.Producer(t)
         relu = aidge_core.ReLU()
         input_node.add_child(relu)
 
@@ -34,7 +35,7 @@ class test_scheduler(unittest.TestCase):
         out_tensor = relu.get_operator().get_output(0)
         expected_out = [0,0,0,0,1,2]
         for i in range(len(expected_out)):
-            self.assertEqual(expected_out[i], out_tensor[i])
+            self.assertEqual(expected_out[i], out_tensor[i], f"On idx {i}")
 
     def test_sequential_scheduling(self):
         input_data =  np.array([0]).astype(np.float32)
@@ -69,7 +70,7 @@ class test_scheduler(unittest.TestCase):
             aidge_core.Producer(input_tensor, "X"),
             aidge_core.FC(1, 50, name='0'),
             aidge_core.parallel([aidge_core.FC(50, 50, name='1'), aidge_core.FC(50, 50, name='3')]),
-            aidge_core.Add(2, name='2'),
+            aidge_core.Add(name='2'),
         ])
 
         EXPECTED_SCHEDULE = [['0', '1', '3', '2'],  ['0', '3', '1', '2']] # Both scheduling are valid !

include/aidge/backend/cpu.hpp

@@ -15,6 +15,8 @@
 #include "aidge/backend/cpu/operator/AbsImpl.hpp"
 #include "aidge/backend/cpu/operator/AddImpl.hpp"
 #include "aidge/backend/cpu/operator/AndImpl.hpp"
+#include "aidge/backend/cpu/operator/AtanImpl.hpp"
+
 #include "aidge/backend/cpu/operator/ArgMaxImpl.hpp"
 #include "aidge/backend/cpu/operator/AvgPoolingImpl.hpp"
 #include "aidge/backend/cpu/operator/MaxPoolingImpl.hpp"

include/aidge/backend/cpu/operator/AtanImpl.hpp

+/********************************************************************************
+ * Copyright (c) 2023 CEA-List
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Eclipse Public License 2.0 which is available at
+ * http://www.eclipse.org/legal/epl-2.0.
+ *
+ * SPDX-License-Identifier: EPL-2.0
+ *
+ ********************************************************************************/
+
+#ifndef AIDGE_CPU_OPERATOR_ATAN_H_
+#define AIDGE_CPU_OPERATOR_ATAN_H_
+
+#include "aidge/backend/cpu/operator/OperatorImpl.hpp"
+#include "aidge/operator/Atan.hpp"
+#include "aidge/utils/Registrar.hpp"
+#include "aidge/utils/Types.h"
+#include "aidge/backend/cpu/data/GetCPUPtr.h"
+#include <memory>
+#include <vector>
+
+namespace Aidge {
+// Operator implementation entry point for the backend
+using AtanImpl_cpu = OperatorImpl_cpu<Atan_Op,
+    void(const std::size_t, const void*, void*),
+    void(const std::size_t, const void*, const void*, void*)>;
+
+// Implementation entry point registration to Operator
+REGISTRAR(Atan_Op, "cpu", Aidge::AtanImpl_cpu::create);
+}  // namespace Aidge
+
+#endif /* AIDGE_CPU_OPERATOR_ATAN_H_ */

include/aidge/backend/cpu/operator/AtanImpl_kernels.hpp

+/********************************************************************************
+ * Copyright (c) 2023 CEA-List
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Eclipse Public License 2.0 which is available at
+ * http://www.eclipse.org/legal/epl-2.0.
+ *
+ * SPDX-License-Identifier: EPL-2.0
+ *
+ ********************************************************************************/
+
+#ifndef AIDGE_CPU_OPERATOR_ATANIMPL_KERNELS_H_
+#define AIDGE_CPU_OPERATOR_ATANIMPL_KERNELS_H_
+
+#include "aidge/utils/Registrar.hpp"
+
+#include "aidge/backend/cpu/operator/AtanImpl.hpp"
+#include <cmath>  // For atan()
+
+
+namespace Aidge {
+template <class I, class O>
+void AtanImpl_cpu_forward_kernel(std::size_t inputLenght,
+                                    const void* input_,
+                                    void* output_) {
+    const I* input = static_cast<const I*>(input_);
+    O* output = static_cast<O*>(output_);
+
+    for (size_t i = 0; i < inputLenght; ++i) {
+        output[i] = static_cast<O>(atan(input[i]));
+    }
+
+}
+
+template <class O, class GI, class GO>
+void AtanImpl_cpu_backward_kernel(const std::size_t inputLenght,
+                                     const void* output_, const void* grad_output_,
+				     void* grad_input_) {
+    const O* output = static_cast<const O*>(output_);
+    const GO* grad_output = static_cast<const GO*>(grad_output_);
+    GI* grad_input = static_cast<GI*>(grad_input_);
+
+    // Apply the derivative of atan for each element in the input array
+    for (size_t i = 0; i < inputLenght; ++i) {
+        // dx = dy * (1 / (1 + x^2))
+        grad_input[i] = grad_output[i] * static_cast<O>(1.0 / (1.0 + output[i] * output[i]));
+    }
+}
+
+
+// Kernels registration to implementation entry point
+REGISTRAR(AtanImpl_cpu,
+    {DataType::Float32},
+    {ProdConso::inPlaceModel, Aidge::AtanImpl_cpu_forward_kernel<float, float>, Aidge::AtanImpl_cpu_backward_kernel<float, float, float>});
+REGISTRAR(AtanImpl_cpu,
+    {DataType::Float64},
+    {ProdConso::inPlaceModel, Aidge::AtanImpl_cpu_forward_kernel<double, double>, Aidge::AtanImpl_cpu_backward_kernel<double, double, double>});
+}  // namespace Aidge
+
+#endif /* AIDGE_CPU_OPERATOR_ATANIMPL_KERNELS_H_ */

include/aidge/backend/cpu/operator/ConvDepthWiseImpl_kernels.hpp

@@ -137,6 +137,7 @@ void ConvDepthWiseImpl2D_cpu_forward_kernel(const std::array<DimSize_t, 2>& stri
     const std::size_t oxSize =
             static_cast<std::size_t>(std::floor(static_cast<float>(inputDims[2] - dilated_kernel_x + strideDims[0]) /
                                 static_cast<float>(strideDims[0])));
+
     // output W size
     const DimSize_t dilated_kernel_y = dilationDims[1]*(kernelDims[1] - 1) + 1;
     const std::size_t oySize =

include/aidge/backend/cpu/operator/GridSampleImpl_kernels.hpp

@@ -96,11 +96,11 @@ void GridSampleImpl1D_cpu_forward_kernel(const GridSample_Op& op,
                             const std::shared_ptr<Tensor>& in1,
                             const std::shared_ptr<Tensor>& out)
 {
-    const I* const input = static_cast<const I * const>(in0->getImpl()->rawPtr());
+    const I* const input = static_cast<const I *>(in0->getImpl()->rawPtr());
     const I* input_ptr = input;
-    float* const grid = static_cast<float* const>(in1->getImpl()->rawPtr());
+    float* const grid = static_cast<float*>(in1->getImpl()->rawPtr());
     float* grid_ptr = grid;
-    O* const output = static_cast<O* const>(out->getImpl()->rawPtr());
+    O* const output = static_cast<O*>(out->getImpl()->rawPtr());
     O* output_ptr = output;
 
     const std::size_t N = in0->dim(0);
@@ -243,9 +243,9 @@ void GridSampleImpl2D_cpu_forward_kernel(const GridSample_Op& op,
 {
     const I* input = static_cast<const I *>(in0->getImpl()->rawPtr());
     const I* input_ptr = input;
-    float* const grid = static_cast<float* const>(in0->getImpl()->rawPtr());
+    float* const grid = static_cast<float*>(in0->getImpl()->rawPtr());
     float* grid_ptr = grid;
-    O* const output = static_cast<O* const>(out->getImpl()->rawPtr());
+    O* const output = static_cast<O*>(out->getImpl()->rawPtr());
 
     const std::size_t N = in0->dim(0);
     const std::size_t C = in0->dim(1);

include/aidge/backend/cpu/operator/OperatorImpl.hpp

@@ -38,8 +38,10 @@ public:
         return impl.prodConso(mOp);
     }
 
-    virtual std::set<ImplSpec> getAvailableImplSpecs() const override {
-        return Registrar<OperatorImpl_cpu>::getKeys();
+    virtual std::vector<ImplSpec> getAvailableImplSpecs() const override {
+        // return Registrar<OperatorImpl_cpu>::getKeys(); // Note: cannot return set due to python binding 
+        std::set<ImplSpec> implSpecsSet = Registrar<OperatorImpl_cpu>::getKeys();
+        return std::vector<ImplSpec>(implSpecsSet.begin(), implSpecsSet.end());
     }
 
     void forward() override;

src/operator/AtanImpl.cpp

+/********************************************************************************
+ * Copyright (c) 2023 CEA-List
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Eclipse Public License 2.0 which is available at
+ * http://www.eclipse.org/legal/epl-2.0.
+ *
+ * SPDX-License-Identifier: EPL-2.0
+ *
+ ********************************************************************************/
+
+#include <cassert>
+#include <chrono>  // std::chrono::milliseconds
+#include <numeric> // std::accumulate
+#include <thread>  // std::this_thread::sleep_for
+#include <vector>
+
+#include "aidge/operator/Atan.hpp"
+#include "aidge/utils/Types.h"
+#include "aidge/backend/cpu/data/GetCPUPtr.h"
+
+#include "aidge/backend/cpu/operator/AtanImpl.hpp"
+#include "aidge/backend/cpu/operator/AtanImpl_kernels.hpp"
+
+template <>
+void Aidge::AtanImpl_cpu::forward() {
+	const Atan_Op& op_ = dynamic_cast<const Atan_Op&>(mOp);
+    std::shared_ptr<Tensor> in0 = op_.getInput(0);
+    std::shared_ptr<Tensor> out0 = op_.getOutput(0);
+    AIDGE_ASSERT(in0, "missing input #0");
+
+    // Find the correct kernel type
+    const auto impl = Registrar<AtanImpl_cpu>::create(getBestMatch(getRequiredSpec()));
+
+    // Call kernel
+    impl.forward(in0->size(),
+        getCPUPtr(mOp.getRawInput(0)),
+        getCPUPtr(mOp.getRawOutput(0)));
+}
+
+template <>
+void Aidge::AtanImpl_cpu::backward() {
+    const Atan_Op& op_ = dynamic_cast<const Atan_Op&>(mOp);
+    std::shared_ptr<Tensor> out0  = op_.getOutput(0);
+    std::shared_ptr<Tensor> gra_int0 = op_.getInput(0)->grad();
+    std::shared_ptr<Tensor> gra_out0 = op_.getOutput(0)->grad();
+    AIDGE_ASSERT(out0, "missing output #0 for current {} operator", op_.type());
+
+    // Find the correct kernel type
+    const auto impl = Registrar<AtanImpl_cpu>::create(getBestMatch(getRequiredSpec()));
+
+    // Call kernel
+    impl.backward(gra_int0->size(), getCPUPtr(out0), getCPUPtr(gra_out0), getCPUPtr(gra_int0));
+}

unit_tests/data/Test_TensorImpl.cpp

@@ -35,7 +35,7 @@ TEST_CASE("Test addition of Tensors","[TensorImpl][Add]") {
     std::uniform_int_distribution<int> boolDist(0,1);
 
     // Create MatMul Operator
-    std::shared_ptr<Node> mySub = Add(2);
+    std::shared_ptr<Node> mySub = Add();
     auto op = std::static_pointer_cast<OperatorTensor>(mySub-> getOperator());
     op->setDataType(DataType::Float32);
     op->setBackend("cpu");

unit_tests/operator/Test_AddImpl.cpp

@@ -39,17 +39,6 @@ TEST_CASE("[cpu/operator] Add(forward)", "[Add][CPU]") {
         }                                       //
     });                                         //
 
-    SECTION("One input") {
-        std::shared_ptr<Node> myAdd = Add(1);
-        auto op = std::static_pointer_cast<OperatorTensor>(myAdd -> getOperator());
-        op->associateInput(0, input1);
-        op->setBackend("cpu");
-        op->setDataType(DataType::Int32);
-        myAdd->forward();
-
-        REQUIRE(*(op->getOutput(0)) == *input1);
-    }
-
     SECTION("Two inputs") {
         std::shared_ptr<Tensor> expectedOutput = std::make_shared<Tensor>(Array4D<int,3,3,3,2> {
             {
@@ -71,7 +60,7 @@ TEST_CASE("[cpu/operator] Add(forward)", "[Add][CPU]") {
             }
         });
 
-        std::shared_ptr<Node> myAdd = Add(2);
+        std::shared_ptr<Node> myAdd = Add();
         auto op = std::static_pointer_cast<OperatorTensor>(myAdd -> getOperator());
         op->associateInput(0, input1);
         op->associateInput(1, input1);
@@ -82,39 +71,6 @@ TEST_CASE("[cpu/operator] Add(forward)", "[Add][CPU]") {
         REQUIRE(*(op->getOutput(0)) == *expectedOutput);
     }
 
-    SECTION("Three inputs") {
-        std::shared_ptr<Tensor> expectedOutput = std::make_shared<Tensor>(Array4D<int,3,3,3,2> {
-            {
-                {
-                    {{ 60, 141},{ 63, 144},{ 66, 147}},
-                    {{ 69, 150},{ 72, 153},{ 75, 156}},
-                    {{ 78, 159},{ 81, 162},{ 84, 165}}
-                },
-                {
-                    {{ 87, 168},{ 90, 171},{ 93, 174}},
-                    {{ 96, 177},{ 99, 180},{102, 183}},
-                    {{105, 186},{108, 189},{111, 192}}
-                },
-                {
-                    {{114, 195},{117, 198},{120, 201}},
-                    {{123, 204},{126, 207},{129, 210}},
-                    {{132, 213},{135, 216},{138, 219}}
-                }
-            }
-        });
-
-        std::shared_ptr<Node> myAdd = Add(3);
-        auto op = std::static_pointer_cast<OperatorTensor>(myAdd -> getOperator());
-        op->associateInput(0, input1);
-        op->associateInput(1, input1);
-        op->associateInput(2, input1);
-        op->setDataType(DataType::Int32);
-        op->setBackend("cpu");
-        myAdd->forward();
-
-        REQUIRE(*op->getOutput(0) == *expectedOutput);
-    }
-
     SECTION("Broadcasting") {
         std::shared_ptr<Tensor> input_0 = std::make_shared<Tensor>(Array4D<int,3,1,3,2> {
         {                                       //
@@ -139,7 +95,7 @@ TEST_CASE("[cpu/operator] Add(forward)", "[Add][CPU]") {
         }                                       //
         });                                     //
 
-        std::shared_ptr<Tensor> input_2 = std::make_shared<Tensor>(Array1D<int,2> {{100,200}});  
+        std::shared_ptr<Tensor> input_2 = std::make_shared<Tensor>(Array1D<int,2> {{100,200}});
         std::shared_ptr<Tensor> expectedOutput = std::make_shared<Tensor>(Array4D<int,3,3,3,2> {
             {                                               //
                 {                                           //
@@ -160,16 +116,23 @@ TEST_CASE("[cpu/operator] Add(forward)", "[Add][CPU]") {
             }                                               //
         });                                                 //
 
-        std::shared_ptr<Node> myAdd = Add(3);
-        auto op = std::static_pointer_cast<OperatorTensor>(myAdd -> getOperator());
-        op->associateInput(0, input_0);
-        op->associateInput(1, input_1);
-        op->associateInput(2, input_2);
-        op->setDataType(DataType::Int32);
-        op->setBackend("cpu");
-        myAdd->forward();
-        op->getOutput(0)->print();
+        std::shared_ptr<Node> myAdd_0 = Add();
+        std::shared_ptr<Node> myAdd_1 = Add();
+        auto op_0 = std::static_pointer_cast<OperatorTensor>(myAdd_0 -> getOperator());
+        auto op_1 = std::static_pointer_cast<OperatorTensor>(myAdd_1 -> getOperator());
+        op_0->associateInput(0, input_0);
+        op_0->associateInput(1, input_1);
+
+        op_1->associateInput(0, input_2);
+        op_1->associateInput(1, op_0->getOutput(0));
+        op_0->setDataType(DataType::Int32);
+        op_1->setDataType(DataType::Int32);
+        op_0->setBackend("cpu");
+        op_1->setBackend("cpu");
+        myAdd_0->forward();
+        myAdd_1->forward();
+        op_1->getOutput(0)->print();
         expectedOutput->print();
-        REQUIRE(*op->getOutput(0) == *expectedOutput);
+        REQUIRE(*op_1->getOutput(0) == *expectedOutput);
     }
 }
\ No newline at end of file

unit_tests/operator/Test_Atan.cpp

+/********************************************************************************
+ * Copyright (c) 2023 CEA-List
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Eclipse Public License 2.0 which is available at
+ * http://www.eclipse.org/legal/epl-2.0.
+ *
+ * SPDX-License-Identifier: EPL-2.0
+ *
+ ********************************************************************************/
+
+#include <catch2/catch_test_macros.hpp>
+
+#include "aidge/data/Tensor.hpp"
+#include "aidge/operator/Atan.hpp"
+
+#include "aidge/backend/cpu.hpp"
+
+#include <memory>
+
+using namespace Aidge;
+
+TEST_CASE("[cpu/operator] Atan(forward)") {
+  SECTION("1D Tensor") {
+    std::shared_ptr<Tensor> input0 =
+        std::make_shared<Tensor>(Array1D<float, 10>{
+            {0.41384590, 0.43120754, 0.93762982, 0.31049860, 0.77547199,
+             0.09514862, 0.16145366, 0.42776686, 0.43487436, 0.41170865}});
+    std::shared_ptr<Tensor> expectedOutput =
+        std::make_shared<Tensor>(Array1D<float, 10>{
+            {0.39238522, 0.40711672, 0.75322037, 0.30106049, 0.65960488,
+             0.09486303, 0.16007232, 0.40421187, 0.4102045, 0.39055911}});
+
+    std::shared_ptr<Node> myAtan = Atan();
+    auto op = std::static_pointer_cast<OperatorTensor>(myAtan->getOperator());
+    op->associateInput(0, input0);
+    op->setDataType(DataType::Float32);
+    op->setBackend("cpu");
+    myAtan->forward();
+
+    float* resPtr = static_cast<float*>(op->getOutput(0)->getImpl()->rawPtr());
+    float* expectedPtr =
+        static_cast<float*>(expectedOutput->getImpl()->rawPtr());
+    for (std::size_t i = 0; i < expectedOutput->size(); ++i) {
+      REQUIRE(std::abs(resPtr[i] - expectedPtr[i]) < 0.00001);
+    }
+  }
+
+  SECTION("3D Tensor") {
+    std::shared_ptr<Tensor> input0 = std::make_shared<Tensor>(
+        Array3D<float, 2, 2, 3>{{{
+                                     {0.97037154, 0.86208081, 0.77767169},
+                                     {0.38160080, 0.11422747, 0.77284443},
+                                 },
+                                 {{0.51592529, 0.72543722, 0.54641193},
+                                  {0.93866944, 0.97767913, 0.34172094}}}});
+    std::shared_ptr<Tensor> expectedOutput = std::make_shared<Tensor>(
+        Array3D<float, 2, 2, 3>{{{{0.77036231, 0.71146592, 0.66097706},
+                                  {0.36454508, 0.11373451, 0.65796196}},
+                                 {{0.47630652, 0.62759472, 0.50008428},
+                                  {0.75377332, 0.77411225, 0.32928031}}}});
+
+    std::shared_ptr<Node> myAtan = Atan();
+    auto op = std::static_pointer_cast<OperatorTensor>(myAtan->getOperator());
+    op->associateInput(0, input0);
+    op->setDataType(DataType::Float32);
+    op->setBackend("cpu");
+    myAtan->forward();
+
+    float* resPtr = static_cast<float*>(op->getOutput(0)->getImpl()->rawPtr());
+    float* expectedPtr =
+        static_cast<float*>(expectedOutput->getImpl()->rawPtr());
+    for (std::size_t i = 0; i < expectedOutput->size(); ++i) {
+      REQUIRE(std::abs(resPtr[i] - expectedPtr[i]) < 0.00001);
+    }
+  }
+}

unit_tests/recipies/Test_ConstantFolding.cpp

@@ -22,12 +22,12 @@
 
 using namespace Aidge;
 
-TEST_CASE("[ConstantFolding] test") {
+TEST_CASE("[ConstantFolding] forward", "[ConstantFolding][forward][CPU]") {
     // generate the original GraphView
     auto matmul0 = MatMul("matmul0");
-    auto add0 = Add(2, "add0");
+    auto add0 = Add("add0");
     auto matmul1 = MatMul("matmul1");
-    auto add1 = Add(2, "add1");
+    auto add1 = Add("add1");
 
     auto b0 = Producer(std::make_shared<Tensor>(Array1D<float,5>{{1, 2, 3, 4, 5}}), "B0", true);
     auto w0 = Producer(std::make_shared<Tensor>(Array2D<float,5,5>{{{1, 2, 3, 4, 5}, {6, 7, 8, 9, 0}, {1, 2, 3, 4, 5}, {6, 7, 8, 9, 0}, {1, 2, 3, 4, 5}}}), "W0", true);

unit_tests/scheduler/Test_Scheduler.cpp

@@ -147,10 +147,13 @@ TEST_CASE("[cpu/scheduler] SequentialScheduler(forward)") {
         std::shared_ptr<GraphView> g =
                 Sequential({Conv(1, 3, {3, 3}, "inputConv"),
                             Parallel({
-                                Conv(3, 3, {1, 1}, "conv1.1"),
-                                Conv(3, 3, {1, 1}, "conv1.2"),
+                                Sequential({
+                                    Parallel({
+                                        Conv(3, 3, {1, 1}, "conv1.1"),
+                                        Conv(3, 3, {1, 1}, "conv1.2")}),
+                                    Add("add1")}),
                                 Conv(3, 3, {1, 1}, "conv1.3")}),
-                            Add(3, "add1"),
+                            Add("add2"),
                             Conv(3, 2, {1, 1}, "conv2"),
                             FC(18, 5, false, "out")});
 
@@ -216,9 +219,9 @@ TEST_CASE("[cpu/scheduler] SequentialScheduler(forward)") {
         std::shared_ptr<Tensor> biasTensor = std::make_shared<Tensor>(
                 Array2D<int, 2, 3>{{{2, 0, 0}, {1, 0, 0}}});
 
-        auto add1 = Add(2, "add1");
+        auto add1 = Add("add1");
         auto mem = Memorize(3, "mem1");
-        auto add2 = Add(2, "add2");
+        auto add2 = Add("add2");
         auto bias = Producer(biasTensor, "bias");
         auto init = Producer(initTensor, "init");
         auto input = Producer(in, "input");
@@ -260,9 +263,9 @@ TEST_CASE("[cpu/scheduler] SequentialScheduler(forward)") {
         std::shared_ptr<Tensor> biasTensor = std::make_shared<Tensor>(
                 Array2D<int, 2, 3>{{{2, 0, 0}, {1, 0, 0}}});
 
-        auto add1 = Add(2, "add1");
+        auto add1 = Add("add1");
         auto mem = Memorize(3, "mem1");
-        auto add2 = Add(2, "add2");
+        auto add2 = Add("add2");
         auto bias = Producer(biasTensor, "bias");
         auto init = Producer(initTensor, "init");
         auto input = Producer(in, "input");