diff --git a/libsrc/core/simd.hpp b/libsrc/core/simd.hpp
index 3459e66d..4a646dca 100644
--- a/libsrc/core/simd.hpp
+++ b/libsrc/core/simd.hpp
@@ -46,19 +46,27 @@ namespace ngcore
   }
 #endif
 
-
   NETGEN_INLINE void SIMDTranspose (SIMD<double,4> a1, SIMD<double,4> a2, SIMD <double,4> a3, SIMD<double,4> a4,
-                             SIMD<double,4> & b1, SIMD<double,4> & b2, SIMD<double,4> & b3, SIMD<double,4> & b4)
+                                    SIMD<double,4> & b1, SIMD<double,4> & b2, SIMD<double,4> & b3, SIMD<double,4> & b4)
   {
-    SIMD<double,4> h1,h2,h3,h4;
-    std::tie(h1,h2) = Unpack(a1,a2);
-    std::tie(h3,h4) = Unpack(a3,a4);
-    b1 = SIMD<double,4> (h1.Lo(), h3.Lo());
-    b2 = SIMD<double,4> (h2.Lo(), h4.Lo());
-    b3 = SIMD<double,4> (h1.Hi(), h3.Hi());
-    b4 = SIMD<double,4> (h2.Hi(), h4.Hi());
+    if constexpr (sizeof(a1.Lo()) == 16)
+      {
+        auto [h1,h2] = Unpack(a1,a2);
+        auto [h3,h4] = Unpack(a3,a4);
+        b1 = SIMD<double,4> (h1.Lo(), h3.Lo());
+        b2 = SIMD<double,4> (h2.Lo(), h4.Lo());
+        b3 = SIMD<double,4> (h1.Hi(), h3.Hi());
+        b4 = SIMD<double,4> (h2.Hi(), h4.Hi());
+      }
+    else
+      {
+        b1 = SIMD<double,4> (a1[0], a2[0], a3[0], a4[0]);
+        b2 = SIMD<double,4> (a1[1], a2[1], a3[1], a4[1]);
+        b3 = SIMD<double,4> (a1[2], a2[2], a3[2], a4[2]);
+        b4 = SIMD<double,4> (a1[3], a2[3], a3[3], a4[3]);
+      }
   }
-
+  
   template<int N>
   NETGEN_INLINE auto HSum (SIMD<double,N> s1, SIMD<double,N> s2)
   {