Fix the BatchNorm operator

EDIT : oudated

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

@@ -94,7 +94,7 @@ public:
     }
 
     void copy(const void *src, NbElts_t length, NbElts_t offset = 0) override {
-        AIDGE_ASSERT(length <= mData.size() || length <= mNbElts, "TensorImpl_cuda<{}>::copy(): copy length ({}) is above capacity ({})", typeid(T).name(), length, mNbElts);
+        AIDGE_ASSERT(offset + length <= mNbElts, "TensorImpl_cuda<{}>::copy(): copy offset ({}) + length ({}) is above capacity ({})", typeid(T).name(), offset, length, mNbElts);
         const T* srcT = static_cast<const T *>(src);
         T* dstT = static_cast<T *>(rawPtr(offset));
 
@@ -107,7 +107,7 @@ public:
             return;
         }
 
-        AIDGE_ASSERT(length <= mData.size() || length <= mNbElts, "TensorImpl_cuda<{}>::copyCast(): copy length ({}) is above capacity ({})", typeid(T).name(), length, mNbElts);
+        AIDGE_ASSERT(offset + length <= mNbElts, "TensorImpl_cuda<{}>::copyCast(): copy offset ({}) + length ({}) is above capacity ({})", typeid(T).name(), offset, length, mNbElts);
         switch (srcDt) {
         case DataType::Float64:
             thrust_copy(static_cast<const double*>(src),
@@ -171,17 +171,17 @@ public:
     }
 
     void copyFromDevice(const void *src, const std::pair<std::string, DeviceIdx_t>& device, NbElts_t length, NbElts_t offset = 0) override {
-        AIDGE_ASSERT(length <= mData.size() || length <= mNbElts, "TensorImpl_cuda<{}>::copyFromDevice(): copy length ({}) is above capacity ({})", typeid(T).name(), length, mNbElts);
+        AIDGE_ASSERT(offset + length <= mNbElts, "TensorImpl_cuda<{}>::copyFromDevice(): copy offset ({}) + length ({}) is above capacity ({})", typeid(T).name(), offset, length, mNbElts);
         CHECK_CUDA_STATUS(cudaMemcpy(rawPtr(offset), src, length * sizeof(T), cudaMemcpyDeviceToDevice));
     }
 
     void copyFromHost(const void *src, NbElts_t length, NbElts_t offset = 0) override {
-        AIDGE_ASSERT(length <= mData.size() || length <= mNbElts, "TensorImpl_cuda<{}>::copyFromHost(): copy length ({}) is above capacity ({})", typeid(T).name(), length, mNbElts);
+        AIDGE_ASSERT(offset + length <= mNbElts, "TensorImpl_cuda<{}>::copyFromHost(): copy offset ({}) + length ({}) is above capacity ({})", typeid(T).name(), offset, length, mNbElts);
         CHECK_CUDA_STATUS(cudaMemcpy(rawPtr(offset), src, length * sizeof(T), cudaMemcpyHostToDevice));
     }
 
     void copyToHost(void *dst, NbElts_t length, NbElts_t offset = 0) const override {
-        AIDGE_ASSERT(length <= mData.size() || length <= mNbElts, "TensorImpl_cuda<{}>::copyToHost(): copy length ({}) is above capacity ({})", typeid(T).name(), length, mNbElts);
+        AIDGE_ASSERT(offset + length <= mNbElts, "TensorImpl_cuda<{}>::copyToHost(): copy offset ({}) + length ({}) is above capacity ({})", typeid(T).name(), offset, length, mNbElts);
         CHECK_CUDA_STATUS(cudaMemcpy(dst, rawPtr(offset), length * sizeof(T), cudaMemcpyDeviceToHost));
     }