diff --git a/mem_arena/src/lib.rs b/mem_arena/src/lib.rs
index 97ebe50..bb1a44a 100644
--- a/mem_arena/src/lib.rs
+++ b/mem_arena/src/lib.rs
@@ -1,6 +1,7 @@
 use std::slice;
 use std::cell::{Cell, RefCell};
 use std::mem::{size_of, align_of};
+use std::cmp::max;
 
 const GROWTH_FRACTION: usize = 8; // 1/N  (smaller number leads to bigger allocations)
 const DEFAULT_MIN_BLOCK_SIZE: usize = 1 << 10; // 1 KiB
@@ -120,6 +121,22 @@ impl MemArena {
         memory.as_mut().unwrap()
     }
 
+    /// Allocates memory for a type `T`, returning a mutable reference to it.
+    ///
+    /// Additionally, the allocation will be made with the given byte alignment or
+    /// the type's inherent alignment, whichever is greater.
+    ///
+    /// CAUTION: the memory returned is uninitialized.  Make sure to initalize before using!
+    pub unsafe fn alloc_uninitialized_with_alignment<'a, T: Copy>(&'a self,
+                                                                  align: usize)
+                                                                  -> &'a mut T {
+        assert!(size_of::<T>() > 0);
+
+        let memory = self.alloc_raw(size_of::<T>(), max(align, align_of::<T>())) as *mut T;
+
+        memory.as_mut().unwrap()
+    }
+
     /// Allocates memory for `len` values of type `T`, returning a mutable slice to it.
     /// All elements are initialized to the given `value`.
     pub fn alloc_array<'a, T: Copy>(&'a self, len: usize, value: T) -> &'a mut [T] {
@@ -145,7 +162,8 @@ impl MemArena {
     }
 
     /// Allocates memory for `len` values of type `T`, returning a mutable slice to it.
-    /// All elements are initialized to the given `value`.
+    ///
+    /// CAUTION: the memory returned is uninitialized.  Make sure to initalize before using!
     pub unsafe fn alloc_array_uninitialized<'a, T: Copy>(&'a self, len: usize) -> &'a mut [T] {
         assert!(size_of::<T>() > 0);
 
@@ -160,6 +178,29 @@ impl MemArena {
         slice::from_raw_parts_mut(memory, len)
     }
 
+    /// Allocates memory for `len` values of type `T`, returning a mutable slice to it.
+    ///
+    /// Additionally, the allocation will be made with the given byte alignment or
+    /// the type's inherent alignment, whichever is greater.
+    ///
+    /// CAUTION: the memory returned is uninitialized.  Make sure to initalize before using!
+    pub unsafe fn alloc_array_uninitialized_with_alignment<'a, T: Copy>(&'a self,
+                                                                        len: usize,
+                                                                        align: usize)
+                                                                        -> &'a mut [T] {
+        assert!(size_of::<T>() > 0);
+
+        let array_mem_size = {
+            let alignment_padding = alignment_offset(size_of::<T>(), align_of::<T>());
+            let aligned_type_size = size_of::<T>() + alignment_padding;
+            aligned_type_size * len
+        };
+
+        let memory = self.alloc_raw(array_mem_size, max(align, align_of::<T>())) as *mut T;
+
+        slice::from_raw_parts_mut(memory, len)
+    }
+
     /// Allocates space with a given size and alignment.
     ///
     /// This is the work-horse code of the MemArena.
diff --git a/src/accel/bvh4.rs b/src/accel/bvh4.rs
index c5311f7..9bee85a 100644
--- a/src/accel/bvh4.rs
+++ b/src/accel/bvh4.rs
@@ -56,7 +56,7 @@ impl<'a> BVH4<'a> {
         } else {
             let base = BVHBase::from_objects(objects, objects_per_leaf, bounder);
 
-            let mut fill_node = unsafe { arena.alloc_uninitialized::<BVH4Node>() };
+            let mut fill_node = unsafe { arena.alloc_uninitialized_with_alignment::<BVH4Node>(32) };
             BVH4::construct_from_base(arena, &base, &base.nodes[base.root_node_index()], fill_node);
 
             BVH4 {
@@ -247,7 +247,8 @@ impl<'a> BVH4<'a> {
                         })
                         .max()
                         .unwrap();
-                    let mut bounds = unsafe { arena.alloc_array_uninitialized(bounds_len) };
+                    let mut bounds =
+                        unsafe { arena.alloc_array_uninitialized_with_alignment(bounds_len, 32) };
                     for (i, b) in bounds.iter_mut().enumerate() {
                         let time = i as f32 / (bounds_len - 1) as f32;
 
@@ -274,7 +275,9 @@ impl<'a> BVH4<'a> {
 
                 // Construct child nodes
                 let mut child_nodes =
-                    unsafe { arena.alloc_array_uninitialized::<BVH4Node>(child_count) };
+                    unsafe {
+                        arena.alloc_array_uninitialized_with_alignment::<BVH4Node>(child_count, 32)
+                    };
                 for (i, c) in children[0..child_count].iter().enumerate() {
                     BVH4::construct_from_base(arena, base, c.unwrap(), &mut child_nodes[i]);
                 }