Implement NEON64, use template argument deduction

Author: onalante-ebay

include/xsimd/arch/xsimd_neon.hpp

@@ -3365,7 +3365,7 @@ namespace xsimd
         template <class A, class T>
         XSIMD_INLINE size_t count(batch_bool<T, A> const& self, requires_arch<neon>) noexcept
         {
-            uint8x16_t popcnts = vcntq_u8(bitwise_cast<uint8_t, T, A>(bitwise_cast<T, A>(self)));
+            uint8x16_t popcnts = vcntq_u8(bitwise_cast<uint8_t>(bitwise_cast(self)));
             uint64x2_t psum = vpaddlq_u32(vpaddlq_u16(vpaddlq_u8(popcnts)));
             uint64x1_t total = vadd_u64(vget_low_u64(psum), vget_high_u64(psum));
 

include/xsimd/arch/xsimd_neon64.hpp

@@ -652,6 +652,15 @@ namespace xsimd
             return vaddvq_u32(positioned);
         }
 
+        /*********
+         * count *
+         *********/
+        template <class A, class T>
+        XSIMD_INLINE size_t count(batch_bool<T, A> const& self, requires_arch<neon64>) noexcept
+        {
+            return vaddlvq_u8(vcntq_u8(bitwise_cast<uint8_t>(bitwise_cast(self)))) / (sizeof(T) * 8);
+        }
+
         /*******
          * abs *
          *******/