Optimize BitOperations.PopCount() with arm64 intrinsics (dotnet#35636)

* Intrinsicy BitOperations.PopCount() for arm64

Co-authored-by: Tanner Gooding <[email protected]>

src/libraries/System.Private.CoreLib/src/System/Numerics/BitOperations.cs

@@ -4,6 +4,7 @@
 
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
+using System.Runtime.Intrinsics;
 using System.Runtime.Intrinsics.Arm;
 using System.Runtime.Intrinsics.X86;
 
@@ -243,6 +244,19 @@ public static int PopCount(uint value)
                 return (int)Popcnt.PopCount(value);
             }
 
+            if (AdvSimd.Arm64.IsSupported)
+            {
+                // PopCount works on vector so convert input value to vector first.
+
+                // Vector64.CreateScalar(uint) generates suboptimal code by storing and
+                // loading the result to memory.
+                // See https://github.com/dotnet/runtime/issues/35976 for details.
+                // Hence use Vector64.Create(ulong) to create Vector64<ulong> and operate on that.
+                Vector64<ulong> input = Vector64.Create((ulong)value);
+                Vector64<byte> aggregated = AdvSimd.Arm64.AddAcross(AdvSimd.PopCount(input.AsByte()));
+                return AdvSimd.Extract(aggregated, 0);
+            }
+
             return SoftwareFallback(value);
 
             static int SoftwareFallback(uint value)
@@ -274,6 +288,14 @@ public static int PopCount(ulong value)
                 return (int)Popcnt.X64.PopCount(value);
             }
 
+            if (AdvSimd.Arm64.IsSupported)
+            {
+                // PopCount works on vector so convert input value to vector first.
+                Vector64<ulong> input = Vector64.Create(value);
+                Vector64<byte> aggregated = AdvSimd.Arm64.AddAcross(AdvSimd.PopCount(input.AsByte()));
+                return AdvSimd.Extract(aggregated, 0);
+            }
+
 #if TARGET_32BIT
             return PopCount((uint)value) // lo
                 + PopCount((uint)(value >> 32)); // hi

src/libraries/System.Utf8String.Experimental/src/System/Runtime/Intrinsics/Intrinsics.Shims.cs

@@ -4,6 +4,16 @@
 
 namespace System.Runtime.Intrinsics
 {
+    internal static class Vector64
+    {
+        public static Vector64<ulong> Create(ulong value) => throw new PlatformNotSupportedException();
+        public static Vector64<byte> AsByte<T>(this Vector64<T> vector) where T : struct => throw new PlatformNotSupportedException();
+    }
+    internal readonly struct Vector64<T>
+        where T : struct
+    {
+    }
+
     internal static class Vector128
     {
         public static Vector128<short> Create(short value) => throw new PlatformNotSupportedException();
@@ -130,4 +140,14 @@ public abstract class Arm64
         public static int LeadingZeroCount(uint value) => throw new PlatformNotSupportedException();
         public static uint ReverseElementBits(uint value) => throw new PlatformNotSupportedException();
     }
+
+    internal abstract class AdvSimd : ArmBase
+    {
+        public new abstract class Arm64 : ArmBase.Arm64
+        {
+            public static Vector64<byte> AddAcross(Vector64<byte> value) => throw new PlatformNotSupportedException();
+        }
+        public static byte Extract(Vector64<byte> vector, byte index) => throw new PlatformNotSupportedException();
+        public static Vector64<byte> PopCount(Vector64<byte> value) => throw new PlatformNotSupportedException();
+    }
 }