From c147026abfda196ba7783455ce413fbc56da9ba8 Mon Sep 17 00:00:00 2001
From: Daniel Lemire <daniel@lemire.me>
Date: Tue, 1 Apr 2025 16:56:18 -0400
Subject: [PATCH] slightly faster ARM kernel

---
 src/Base64ARM.cs | 22 +++++++---------------
 1 file changed, 7 insertions(+), 15 deletions(-)

diff --git a/src/Base64ARM.cs b/src/Base64ARM.cs
index dad3362..3a34ad8 100644
--- a/src/Base64ARM.cs
+++ b/src/Base64ARM.cs
@@ -344,23 +344,15 @@ private static unsafe void Base64DecodeBlock(byte* outPtr, byte* srcPtr)
                 // Load 4 vectors from src
                 var (str0, str1, str2, str3) = AdvSimd.Arm64.Load4xVector128AndUnzip(srcPtr);
 
+                // Perform bitwise operations to simulate NEON intrinsics
+                Vector128<byte> outvec0 = AdvSimd.ShiftLeftAndInsert(
+                    AdvSimd.ShiftRightLogical(str1, 4), str0, 2);
 
+                Vector128<byte> outvec1 = AdvSimd.ShiftLeftAndInsert(
+                    AdvSimd.ShiftRightLogical(str2, 2), str1, 4);
 
-                // Perform bitwise operations to simulate NEON intrinsics
-                Vector128<byte> outvec0 = AdvSimd.Or(
-                    AdvSimd.ShiftLeftLogical(str0, 2),
-                    AdvSimd.ShiftRightLogical(str1, 4)
-                );
-
-                Vector128<byte> outvec1 = AdvSimd.Or(
-                    AdvSimd.ShiftLeftLogical(str1, 4),
-                    AdvSimd.ShiftRightLogical(str2, 2)
-                );
-
-                Vector128<byte> outvec2 = AdvSimd.Or(
-                    AdvSimd.ShiftLeftLogical(str2, 6),
-                    str3
-                );
+                Vector128<byte> outvec2 = AdvSimd.ShiftLeftAndInsert(
+                    str3, str2, 6);
 
                 // Store the result in outData
                 AdvSimd.Arm64.StoreVectorAndZip(outPtr, (outvec0, outvec1, outvec2));