makeContiguous optimization (single copy for contiguous chuncks of memory)

src/data/Tensor.cpp

@@ -45,10 +45,24 @@ void Aidge::Tensor::makeContiguous() {
     {
         // Create a new storage that will be contiguous
         std::shared_ptr<TensorImpl> newImpl = Registrar<Tensor>::create({mImpl->backend(), mDataType})(mImpl->device().second, mSize);
-        // Copy elements one by one from old to new storage
-        for (size_t idx = 0; idx < mSize; ++idx) {
+        // Copy elements from old to new storage
+        size_t idx = 0;
+        while (idx < mSize) {
             const size_t storageIdx = getStorageIdx(getCoord(idx));
-            newImpl->copy(mImpl->rawPtr(mImplOffset + storageIdx), 1, idx);
+
+            // Determine the size of the contiguous chunk
+            size_t copySize = 1;
+            while (idx + copySize < mSize && 
+                getStorageIdx(getCoord(idx + copySize)) == storageIdx + copySize)
+            {
+                ++copySize;
+            }
+
+            // Perform a single copy for the contiguous chunk
+            newImpl->copy(mImpl->rawPtr(mImplOffset + storageIdx), copySize, idx);
+
+            // Move to the next index after the contiguous chunk
+            idx += copySize;
         }
         // Replace old storage by new, contiguous, storage
         setImpl(newImpl);