Remove initialization from aligned_array

Author: stijnh

include/kernel_float/base.h

@@ -22,13 +22,15 @@ struct alignas(Alignment) aligned_array {
         return items_;
     }
 
-    T items_[N] = {};
+    T items_[N];
 };
 
 template<typename T, size_t Alignment>
 struct alignas(Alignment) aligned_array<T, 1, Alignment> {
+    aligned_array() = default;
+
     KERNEL_FLOAT_INLINE
-    aligned_array(T item = {}) : item_(item) {}
+    aligned_array(T item) : item_(item) {}
 
     KERNEL_FLOAT_INLINE
     operator T() const {
@@ -45,7 +47,7 @@ struct alignas(Alignment) aligned_array<T, 1, Alignment> {
         return &item_;
     }
 
-    T item_ = {};
+    T item_;
 };
 
 template<typename T, size_t Alignment>

single_include/kernel_float.h

@@ -16,8 +16,8 @@
 
 //================================================================================
 // this file has been auto-generated, do not modify its contents!
-// date: 2026-03-02 16:31:41.575770
-// git hash: a9bc5a6b4eaaa934175759ad44eb4b3b89f7ded2
+// date: 2026-03-11 14:54:13.201632
+// git hash: 82ce926cbd4572184b5e19b456eda9c03dba1064
 //================================================================================
 
 #ifndef KERNEL_FLOAT_MACROS_H
@@ -424,13 +424,15 @@ struct alignas(Alignment) aligned_array {
         return items_;
     }
 
-    T items_[N] = {};
+    T items_[N];
 };
 
 template<typename T, size_t Alignment>
 struct alignas(Alignment) aligned_array<T, 1, Alignment> {
+    aligned_array() = default;
+
     KERNEL_FLOAT_INLINE
-    aligned_array(T item = {}) : item_(item) {}
+    aligned_array(T item) : item_(item) {}
 
     KERNEL_FLOAT_INLINE
     operator T() const {
@@ -447,7 +449,7 @@ struct alignas(Alignment) aligned_array<T, 1, Alignment> {
         return &item_;
     }
 
-    T item_ = {};
+    T item_;
 };
 
 template<typename T, size_t Alignment>

tests/basics.cu

@@ -146,3 +146,21 @@ struct creation_tests {
 };
 
 REGISTER_TEST_CASE("into_vec and make_vec", creation_tests, int, float)
+
+struct shared_memory_tests {
+    template<typename T, size_t... I, size_t N = sizeof...(I)>
+    __device__ void operator()(generator<T> gen, std::index_sequence<I...>) {
+        // It should be possible to use vector_storage for shared memory.
+        // Using the regular vector type is not possible as it needs initialization.
+        __shared__ kernel_float::vector_storage<T, N> items;
+    }
+};
+
+REGISTER_TEST_CASE_GPU(
+    "shared memory",
+    shared_memory_tests,
+    int,
+    float,
+    double,
+    __half,
+    __nv_bfloat16)