Add forwardDType to FC operator.

include/aidge/operator/FC.hpp

@@ -29,7 +29,7 @@ namespace Aidge {
  * @brief Description of a Fully Connected (FC) operation on an input Tensor.
  *
  * The Fully Connected (FC) operation applies a linear transformation to the input Tensor
- * by multiplying it with a weight matrix and optionally adding a bias vector: 
+ * by multiplying it with a weight matrix and optionally adding a bias vector:
  * - If `bias` is included:
  *   f(x) = x × weights^T + bias
  * - If `bias` is omitted:
@@ -74,7 +74,7 @@ public:
      *
      * Copies the attributes and output tensor(s) of the operator, but does not
      * copy input tensors. The new operator instance has no associated inputs.
-     * 
+     *
      * @param op The `FC_Op` instance to copy.
      */
     FC_Op(const FC_Op& op)
@@ -114,6 +114,12 @@ public:
      */
     bool forwardDims(bool allowDataDependency = false) override final;
 
+    /**
+     * @brief Forward the data type.
+     * @return True if successful, false otherwise.
+     */
+    bool forwardDType() override final;
+
     /**
      * @brief Sets the backend for the operator.
      *

src/operator/FC.cpp

@@ -40,6 +40,19 @@ void Aidge::FC_Op::associateInput(const Aidge::IOIndex_t inputIdx, const std::sh
         mInputs[inputIdx]->resize({1, getInput(inputIdx)->size()});
 }
 
+bool Aidge::FC_Op::forwardDType(){
+    // Current naive forwarDType based on bias.
+    // Bias is optional so this will not always work
+    // But is good enough for now.
+    // Feel free to upgrade the function!
+    if (getInput(2)) {
+        mOutputs[0]->setDataType(getInput(2)->dataType());
+        return true;
+    }
+    Log::notice("FC_Op: No bias associated, failed to forward data type.");
+    return false;
+}
+
 bool Aidge::FC_Op::forwardDims(bool /*allowDataDependency*/) {
     if (inputsAssociated()) {
         // first check weight since it defines inChannels and outChannels