Lenet operators

Added backend_cuda for the following operators:

• AvgPooling
• FC/Gemm
• MaxPooling
• ReLU

include/aidge/backend/cuda/data/TensorImpl.hpp

@@ -24,8 +24,8 @@ void thrust_copy(const half_float::half* srcData, half_float::half* dstData, siz
 
 /**
  * @brief Abstract class for the TensorImpl_cuda class template.
- * @details Its purpose is to provide access to base methods that are specific 
- * to the implementation (which are therefore not present in the TensorImpl 
+ * @details Its purpose is to provide access to base methods that are specific
+ * to the implementation (which are therefore not present in the TensorImpl
  * class), but whose data type does not need to be known.
  */
 class TensorImpl_cuda_ {
@@ -35,7 +35,7 @@ protected:
 public:
     /**
      * @brief Return the CuDNN tensor descriptor of the tensor.
-     * @details This method uses lazy initialization for the descriptor 
+     * @details This method uses lazy initialization for the descriptor
      * (which is therefore mutable in the derived class).
      * @return cudnnTensorDescriptor_t CuDNN tensor descriptor.
      */
@@ -80,7 +80,7 @@ public:
     // native interface
     const future_std::span<T>& data() const { return mData; }
 
-    std::size_t scalarSize() const override { return sizeof(T); }
+    std::size_t scalarSize() const noexcept override { return sizeof(T); }
 
     void copy(const void *src, NbElts_t length, NbElts_t offset = 0) override {
         AIDGE_ASSERT(length <= mData.size() || length <= mNbElts, "copy length is above capacity");