Continue refactor

librapid/include/librapid/array/arrayContainer.hpp

@@ -73,7 +73,8 @@ namespace librapid {
 		struct IsArrayContainer : std::false_type {};
 
 		template<typename ShapeType, typename StorageScalar>
-		struct IsArrayContainer<array::ArrayContainer<ShapeType, StorageScalar>> : std::true_type {};
+		struct IsArrayContainer<array::ArrayContainer<ShapeType, StorageScalar>> : std::true_type {
+		};
 
 		LIBRAPID_DEFINE_AS_TYPE(typename StorageScalar,
 								array::ArrayContainer<Shape COMMA StorageScalar>);
@@ -765,12 +766,10 @@ namespace librapid {
 		ArrayContainer<ShapeType_, StorageType_>::packet(size_t index) const -> Packet {
 			auto ptr = LIBRAPID_ASSUME_ALIGNED(m_storage.begin());
 
-#if defined(LIBRAPID_NATIVE_ARCH) && !defined(LIBRAPID_OSX)
-			// On MacOS (and other platforms??) we cannot use aligned loads in arrays due to one
-			// annoying edge case. Normally, all SIMD loads will be aligned to a 64-byte boundary.
-			// Say, however, this array is a sub-array of a larger array. If the outer dimension
-			// of the larger array does not result in a 64-byte alignment, the data of *this* array
-			// will not be correctly aligned, hence causing a segfault.
+#if defined(LIBRAPID_NATIVE_ARCH)
+			LIBRAIPD_ASSERT(ptr % Packet::size == 0,
+							"ArrayContainer::packet called on unaligned storage");
+
 			return xsimd::load_aligned(ptr + index);
 #else
 			return xsimd::load_unaligned(ptr + index);
@@ -786,10 +785,14 @@ namespace librapid {
 		template<typename ShapeType_, typename StorageType_>
 		LIBRAPID_ALWAYS_INLINE void
 		ArrayContainer<ShapeType_, StorageType_>::writePacket(size_t index, const Packet &value) {
+			auto ptr = LIBRAPID_ASSUME_ALIGNED(m_storage.begin());
+
 #if defined(LIBRAPID_NATIVE_ARCH)
-			value.store_aligned(m_storage.begin() + index);
+			LIBRAIPD_ASSERT(ptr % Packet::size == 0,
+							"ArrayContainer::packet called on unaligned storage");
+			value.store_aligned(ptr + index);
 #else
-			value.store_unaligned(m_storage.begin() + index);
+			value.store_unaligned(ptr + index);
 #endif
 		}
 

librapid/include/librapid/array/assignOps.hpp

@@ -328,7 +328,7 @@ namespace librapid {
                                                          ::librapid::detail::LibRapidType::Scalar,
                                                        int> = 0>
         LIBRAPID_NODISCARD LIBRAPID_ALWAYS_INLINE auto dataSourceExtractor(const T &obj) {
-            return obj.storage().begin().get();
+            return obj.storage().begin();
         }
 
         template<typename T, typename std::enable_if_t<typetraits::TypeInfo<T>::type ==
@@ -518,7 +518,7 @@ namespace librapid {
               std::make_index_sequence<argSize>(),
               filename,
               kernelName,
-              reinterpret_cast<Scalar *>(lhs.storage().begin().get()),
+              reinterpret_cast<Scalar *>(lhs.storage().begin()),
               function);
         }
     }  // namespace detail

librapid/include/librapid/array/linalg/transpose.hpp

@@ -624,8 +624,8 @@ namespace librapid {
 				else {
 					if (m_inputShape.ndim() == 2) {
 						int64_t blockSize		= global::cacheLineSize / sizeof(Scalar);
-						auto *__restrict outPtr = out.storage().begin().get();
-						auto *__restrict inPtr	= m_array.storage().begin().get();
+						auto *__restrict outPtr = out.storage().begin();
+						auto *__restrict inPtr	= m_array.storage().begin();
 						detail::cuda::transposeImpl(
 						  outPtr, inPtr, m_inputShape[0], m_inputShape[1], m_alpha, blockSize);
 					} else {

librapid/include/librapid/array/storage.hpp

@@ -99,10 +99,6 @@ namespace librapid {
 		/// Free a Storage object
 		~Storage();
 
-		/// \brief Set this storage object to reference the same data as \p other
-		/// \param other Storage object to reference
-		void set(const Storage &other);
-
 		/// \brief Return a Storage object on the host with the same data as this Storage object
 		/// (mainly for use with CUDA or OpenCL)
 		/// \return
@@ -174,7 +170,7 @@ namespace librapid {
 		template<typename P>
 		LIBRAPID_ALWAYS_INLINE void initData(P begin, SizeType size);
 
-#if defined(LIBRAPID_NATIVE_ARCH) && !defined(LIBRAPID_APPLE)
+#if defined(LIBRAPID_NATIVE_ARCH)
 		alignas(LIBRAPID_MEM_ALIGN) Pointer m_begin = nullptr;
 #else
 		Pointer m_begin = nullptr; // Pointer to the beginning of the data
@@ -475,7 +471,7 @@ namespace librapid {
 		if (this != &other) {
 			size_t oldSize = m_size;
 			m_size		   = other.m_size;
-			if (other.m_size == m_size) LIBRAPID_UNLIKELY {
+			if (other.m_size != m_size) LIBRAPID_UNLIKELY {
 					if (m_ownsData) LIBRAPID_LIKELY {
 							// Reallocate
 							detail::safeDeallocate(m_begin, oldSize);
@@ -494,7 +490,7 @@ namespace librapid {
 	}
 
 	template<typename T>
-	Storage<T> &Storage<T>::operator=(Storage &&other) noexcept {
+	auto Storage<T>::operator=(Storage &&other) noexcept -> Storage & {
 		if (this != &other) {
 			m_begin	   = std::move(other.m_begin);
 			m_size	   = std::move(other.m_size);
@@ -517,6 +513,7 @@ namespace librapid {
 	void Storage<T>::initData(P begin, P end) {
 		m_size			= static_cast<SizeType>(std::distance(begin, end));
 		m_begin			= detail::safeAllocate<T>(m_size);
+		m_ownsData		= true;
 		auto thisBegin	= LIBRAPID_ASSUME_ALIGNED(m_begin);
 		auto otherBegin = LIBRAPID_ASSUME_ALIGNED(begin);
 		detail::fastCopy(thisBegin, otherBegin, m_size);
@@ -528,14 +525,6 @@ namespace librapid {
 		initData(begin, begin + size);
 	}
 
-	// template<typename T>
-	// void Storage<T>::set(const Storage<T> &other) {
-	// 	// We can simply copy the shared pointers across
-	// 	m_begin	   = other.m_begin;
-	// 	m_size	   = other.m_size;
-	// 	m_ownsData = other.m_ownsData;
-	// }
-
 	template<typename T>
 	auto Storage<T>::toHostStorage() const -> Storage {
 		return copy();