From ead07427834c82aac105d36b8671abbe915c441c Mon Sep 17 00:00:00 2001
From: serge-sans-paille <serge.guelton@telecom-bretagne.eu>
Date: Mon, 5 Feb 2024 07:38:40 +0100
Subject: [PATCH] Use dedicated instructions for zip / unzip on arm64

---
 include/xsimd/arch/xsimd_neon64.hpp | 83 +++++++++++++++++++++++++++++
 1 file changed, 83 insertions(+)

diff --git a/include/xsimd/arch/xsimd_neon64.hpp b/include/xsimd/arch/xsimd_neon64.hpp
index bc982c7ce..c6ecb423f 100644
--- a/include/xsimd/arch/xsimd_neon64.hpp
+++ b/include/xsimd/arch/xsimd_neon64.hpp
@@ -952,6 +952,41 @@ namespace xsimd
         /**********
          * zip_lo *
          **********/
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 1> = 0>
+        inline batch<T, A> zip_lo(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip1q_u8(lhs, rhs);
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 1> = 0>
+        inline batch<T, A> zip_lo(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip1q_s8(lhs, rhs);
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 2> = 0>
+        inline batch<T, A> zip_lo(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip1q_u16(lhs, rhs);
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 2> = 0>
+        inline batch<T, A> zip_lo(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip1q_s16(lhs, rhs);
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 4> = 0>
+        inline batch<T, A> zip_lo(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip1q_u32(lhs, rhs);
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 4> = 0>
+        inline batch<T, A> zip_lo(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip1q_s32(lhs, rhs);
+        }
 
         template <class A, class T, detail::enable_sized_unsigned_t<T, 8> = 0>
         inline batch<T, A> zip_lo(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
@@ -965,6 +1000,12 @@ namespace xsimd
             return vzip1q_s64(lhs, rhs);
         }
 
+        template <class A>
+        inline batch<float, A> zip_lo(batch<float, A> const& lhs, batch<float, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip1q_f32(lhs, rhs);
+        }
+
         template <class A>
         inline batch<double, A> zip_lo(batch<double, A> const& lhs, batch<double, A> const& rhs, requires_arch<neon64>) noexcept
         {
@@ -975,6 +1016,42 @@ namespace xsimd
          * zip_hi *
          **********/
 
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 1> = 0>
+        inline batch<T, A> zip_hi(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip2q_u8(lhs, rhs);
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 1> = 0>
+        inline batch<T, A> zip_hi(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip2q_s8(lhs, rhs);
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 2> = 0>
+        inline batch<T, A> zip_hi(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip2q_u16(lhs, rhs);
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 2> = 0>
+        inline batch<T, A> zip_hi(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip2q_s16(lhs, rhs);
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 4> = 0>
+        inline batch<T, A> zip_hi(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip2q_u32(lhs, rhs);
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 4> = 0>
+        inline batch<T, A> zip_hi(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip2q_s32(lhs, rhs);
+        }
+
         template <class A, class T, detail::enable_sized_unsigned_t<T, 8> = 0>
         inline batch<T, A> zip_hi(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon64>) noexcept
         {
@@ -987,6 +1064,12 @@ namespace xsimd
             return vzip2q_s64(lhs, rhs);
         }
 
+        template <class A>
+        inline batch<float, A> zip_hi(batch<float, A> const& lhs, batch<float, A> const& rhs, requires_arch<neon64>) noexcept
+        {
+            return vzip2q_f32(lhs, rhs);
+        }
+
         template <class A>
         inline batch<double, A> zip_hi(batch<double, A> const& lhs, batch<double, A> const& rhs, requires_arch<neon64>) noexcept
         {