From ec26662ac9f1e89b2556d81fa311f91ab3abd492 Mon Sep 17 00:00:00 2001
From: Alexandre Mutel <alexandre_mutel@live.com>
Date: Sat, 5 Nov 2022 16:05:12 +0100
Subject: [PATCH] Optimize XxHash3 on ARM platform (#77881)
* Optimize XxHash3 on ARM platform
* Extract code to MultiplyWideningLower
* Update src/libraries/System.IO.Hashing/src/System/IO/Hashing/XxHash3.cs
Co-authored-by: Stephen Toub <stoub@microsoft.com>
Co-authored-by: Stephen Toub <stoub@microsoft.com>
---
.../src/System/IO/Hashing/XxHash3.cs | 24 +++++++++++++++----
1 file changed, 20 insertions(+), 4 deletions(-)
diff --git a/src/libraries/System.IO.Hashing/src/System/IO/Hashing/XxHash3.cs b/src/libraries/System.IO.Hashing/src/System/IO/Hashing/XxHash3.cs
index 4e821618d36e3e..17087d87631036 100644
--- a/src/libraries/System.IO.Hashing/src/System/IO/Hashing/XxHash3.cs
+++ b/src/libraries/System.IO.Hashing/src/System/IO/Hashing/XxHash3.cs
@@ -10,6 +10,7 @@
using System.Runtime.InteropServices;
#if NET7_0_OR_GREATER
using System.Runtime.Intrinsics;
+using System.Runtime.Intrinsics.Arm;
using System.Runtime.Intrinsics.X86;
#endif
@@ -896,16 +897,31 @@ private static Vector128<ulong> Accumulate128(Vector128<ulong> accVec, byte* sou
Vector128<uint> sourceKey = sourceVec ^ secret;
// TODO: Figure out how to unwind this shuffle and just use Vector128.Multiply
- Vector128<uint> sourceKeyLow = Vector128.Shuffle(sourceKey, Vector128.Create(1u, 0, 3, 0));
Vector128<uint> sourceSwap = Vector128.Shuffle(sourceVec, Vector128.Create(2u, 3, 0, 1));
Vector128<ulong> sum = accVec + sourceSwap.AsUInt64();
- Vector128<ulong> product = Sse2.IsSupported ?
- Sse2.Multiply(sourceKey, sourceKeyLow) :
- (sourceKey & Vector128.Create(~0u, 0u, ~0u, 0u)).AsUInt64() * (sourceKeyLow & Vector128.Create(~0u, 0u, ~0u, 0u)).AsUInt64();
+ Vector128<ulong> product = MultiplyWideningLower(sourceKey);
accVec = product + sum;
return accVec;
}
+
+ [MethodImpl(MethodImplOptions.AggressiveInlining)]
+ private static Vector128<ulong> MultiplyWideningLower(Vector128<uint> source)
+ {
+ if (AdvSimd.IsSupported)
+ {
+ Vector64<uint> sourceLow = Vector128.Shuffle(source, Vector128.Create(0u, 2, 0, 0)).GetLower();
+ Vector64<uint> sourceHigh = Vector128.Shuffle(source, Vector128.Create(1u, 3, 0, 0)).GetLower();
+ return AdvSimd.MultiplyWideningLower(sourceLow, sourceHigh);
+ }
+ else
+ {
+ Vector128<uint> sourceLow = Vector128.Shuffle(source, Vector128.Create(1u, 0, 3, 0));
+ return Sse2.IsSupported ?
+ Sse2.Multiply(source, sourceLow) :
+ (source & Vector128.Create(~0u, 0u, ~0u, 0u)).AsUInt64() * (sourceLow & Vector128.Create(~0u, 0u, ~0u, 0u)).AsUInt64();
+ }
+ }
#endif
private static void ScrambleAccumulators(ulong* accumulators, byte* secret)