WIP

Author: serge-sans-paille

include/xsimd/arch/xsimd_neon.hpp

@@ -20,6 +20,7 @@
 
 #include "../types/xsimd_neon_register.hpp"
 #include "../types/xsimd_utils.hpp"
+#include "./common/xsimd_common_cast.hpp"
 
 // Wrap intrinsics so we can pass them as function pointers
 // - OP: intrinsics name prefix, e.g., vorrq
@@ -3211,6 +3212,19 @@ namespace xsimd
         {
             return vreinterpretq_f32_u32(swizzle(batch<uint32_t, A>(vreinterpretq_u32_f32(self)), mask, A {}));
         }
+
+        /*********
+         * widen *
+         *********/
+        template <class A, class T, detail::enable_sized_signed_t<T, 1> = 0>
+        XSIMD_INLINE std::array<batch<widen_t<T>, A>, 2> widen(batch<T, A> const& x, requires_arch<neon>) noexcept {
+          return {batch<widen_t<T>, A>(vaddl_s8(vget_low_s8(x), vdup_n_s8(0))), batch<widen_t<T>, A>(vaddl_s8(vget_high_s8(x), vdup_n_s8(0)))};
+        }
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 1> = 0>
+        XSIMD_INLINE std::array<batch<widen_t<T>, A>, 2> widen(batch<T, A> const& x, requires_arch<neon>) noexcept {
+          return {batch<widen_t<T>, A>(vaddl_u8(vget_low_u8(x), vdup_n_u8(0))), batch<widen_t<T>, A>(vaddl_u8(vget_high_u8(x), vdup_n_u8(0)))};
+        }
+
     }
 
 }

include/xsimd/arch/xsimd_neon64.hpp

@@ -1552,7 +1552,7 @@ namespace xsimd
          * widen *
          *********/
         template <class A, class T>
-        XSIMD_INLINE std::array<batch<double, A>, 2> widen(batch<float, A> const& x, requires_arch<neon>) noexcept
+        XSIMD_INLINE std::array<batch<double, A>, 2> widen(batch<float, A> const& x, requires_arch<neon64>) noexcept
         {
             return { batch<double, A>(vcvt_f64_f32(vget_low_f32(x))), batch<double, A>(vcvt_high_f64_f32(x)) };
         }