WIP

Author: serge-sans-paille

include/xsimd/arch/xsimd_neon.hpp

@@ -1039,10 +1039,18 @@ namespace xsimd
             return dispatcher.apply(register_type(lhs), register_type(rhs));
         }
 
-        template <class A, class T, detail::enable_sized_integral_t<T, 8> = 0>
+        template <class A, class T, detail::enable_sized_signed_t<T, 8> = 0>
         XSIMD_INLINE batch_bool<T, A> gt(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
         {
-            return batch_bool<T, A>({ lhs.get(0) > rhs.get(0), lhs.get(1) > rhs.get(1) });
+            using register_type = typename batch<T, A>::register_type;
+            return vshrq_n_s64(vqsubq_s64(register_type(lhs), register_type(rhs)), 63);
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 8> = 0>
+        XSIMD_INLINE batch_bool<T, A> gt(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
+        {
+            using register_type = typename batch<T, A>::register_type;
+            return vshrq_n_s64(vreinterpretq_s64_u64(vqsubq_u64(register_type(lhs), register_type(rhs))), 63);
         }
 
         /******