Fix dispatch

Author: AntoinePrv

include/xsimd/arch/utils/shifts.hpp

@@ -27,11 +27,10 @@ namespace xsimd
             template <typename I, I offset, I length, I... Vs>
             struct select_stride
             {
-                static constexpr I values_array[] = { Vs... };
-
                 template <typename K>
                 static constexpr K get(K i, K)
                 {
+                    constexpr I values_array[] = { Vs... };
                     return static_cast<K>(values_array[length * i + offset]);
                 }
             };

include/xsimd/arch/xsimd_avx2.hpp

@@ -337,7 +337,7 @@ namespace xsimd
         // Missing implementations are dispatched to the `batch` overload in xsimd_api.
         // The 1 byte constant implementation calls the 2 bytes constant version, the 2 bytes
         // constant version calls into the 4 bytes version which resolves to the dynamic one above.
-        template <class T, class A, T... Vs,
+        template <class A, class T, T... Vs,
                   std::enable_if_t<std::is_integral<T>::value && (sizeof(T) <= 2), int> = 0>
         XSIMD_INLINE batch<T, A> bitwise_lshift(
             batch<T, A> const& self, batch_constant<T, A, Vs...> shifts, requires_arch<avx2> req) noexcept

include/xsimd/arch/xsimd_sse2.hpp

@@ -329,7 +329,7 @@ namespace xsimd
 
         // bitwise_lshift multiple (constant)
         // Missing implementations are dispacthed to the `batch` overload in xsimd_api.
-        template <class T, class A, T... Vs, detail::enable_sized_integral_t<T, 2> = 0>
+        template <class A, class T, T... Vs, detail::enable_sized_integral_t<T, 2> = 0>
         XSIMD_INLINE batch<T, A> bitwise_lshift(
             batch<T, A> const& self, batch_constant<T, A, Vs...> shifts, requires_arch<sse2> req) noexcept
         {
@@ -341,7 +341,7 @@ namespace xsimd
             return _mm_mullo_epi16(self, mults.as_batch());
         }
 
-        template <class T, class A, T... Vs, detail::enable_sized_integral_t<T, 1> = 0>
+        template <class A, class T, T... Vs, detail::enable_sized_integral_t<T, 1> = 0>
         XSIMD_INLINE batch<T, A> bitwise_lshift(
             batch<T, A> const& self, batch_constant<T, A, Vs...> shifts, requires_arch<sse2> req) noexcept
         {