Timerix/DTLib.XXHash
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add SSE2

Browse Source
This commit is contained in:
Oleksandr Melnyk 2022-06-05 01:25:50 +03:00
parent 0a27c518a7
commit ab1a2055b7
2 changed files with 100 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
src/Standart.Hash.xxHash/xxHash128.XXH.cs
View File

@ -1,6 +1,7 @@
// ReSharper disable InconsistentNaming // ReSharper disable InconsistentNaming
using System.Runtime.CompilerServices; using System.Runtime.CompilerServices;
using System.Runtime.Intrinsics.X86;
namespace Standart.Hash.xxHash namespace Standart.Hash.xxHash
{ {
@ -86,11 +87,23 @@ namespace Standart.Hash.xxHash
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
private static uint128 XXH_mult64to128(ulong lhs, ulong rhs) private static uint128 XXH_mult64to128(ulong lhs, ulong rhs)
{ {
// TODO: SIMD if (Bmi2.IsSupported)
return XXH_mult64to128_bmi2(lhs, rhs);
return XXH_mult64to128_scalar(lhs, rhs); return XXH_mult64to128_scalar(lhs, rhs);
} }
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static unsafe uint128 XXH_mult64to128_bmi2(ulong lhs, ulong rhs)
{
ulong product_high;
ulong product_low = Bmi2.X64.MultiplyNoFlags(lhs, rhs, &product_high);
uint128 r128;
r128.low64 = product_low;
r128.high64 = product_high;
return r128;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
private static uint128 XXH_mult64to128_scalar(ulong lhs, ulong rhs) private static uint128 XXH_mult64to128_scalar(ulong lhs, ulong rhs)
{ {
@ -114,5 +127,11 @@ namespace Standart.Hash.xxHash
{ {
*(ulong*) dst = v64; *(ulong*) dst = v64;
} }
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static byte _MM_SHUFFLE(byte p3, byte p2, byte p1, byte p0)
{
return (byte)((p3 << 6) | (p2 << 4) | (p1 << 2) | p0);
}
} }
} }

87
src/Standart.Hash.xxHash/xxHash128.XXH3.cs
View File

@ -2,6 +2,8 @@
using System.Diagnostics; using System.Diagnostics;
using System.Runtime.CompilerServices; using System.Runtime.CompilerServices;
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.X86;
namespace Standart.Hash.xxHash namespace Standart.Hash.xxHash
{ {
@ -377,9 +379,33 @@ namespace Standart.Hash.xxHash
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
private static unsafe void XXH3_accumulate_512(ulong* acc, byte* input, byte* secret) private static unsafe void XXH3_accumulate_512(ulong* acc, byte* input, byte* secret)
{ {
// TODO: SIMD if (Sse2.IsSupported)
XXH3_accumulate_512_sse2(acc, input, secret);
else
XXH3_accumulate_512_scalar(acc, input, secret);
}
XXH3_accumulate_512_scalar(acc, input, secret); [MethodImpl(MethodImplOptions.AggressiveInlining)]
private static unsafe void XXH3_accumulate_512_sse2(ulong* acc, byte* input, byte* secret)
{
const int m128i_size = 16;
for (int i = 0; i < XXH_STRIPE_LEN / m128i_size; i++)
{
int uint32_offset = i * 4;
int uint64_offset = i * 2;
var acc_vec = Sse2.LoadVector128(acc + uint64_offset);
var data_vec = Sse2.LoadVector128((uint*) input + uint32_offset);
var key_vec = Sse2.LoadVector128((uint*) secret + uint32_offset);
var data_key = Sse2.Xor(data_vec, key_vec);
var data_key_lo = Sse2.Shuffle(data_key, _MM_SHUFFLE(0, 3, 0, 1));
var product = Sse2.Multiply(data_key, data_key_lo);
var data_swap = Sse2.Shuffle(data_vec, _MM_SHUFFLE(1, 0, 3, 2)).AsUInt64();
var sum = Sse2.Add(acc_vec, data_swap);
var result = Sse2.Add(product, sum);
Sse2.Store(acc + uint64_offset, result);
}
} }
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
@ -407,9 +433,34 @@ namespace Standart.Hash.xxHash
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
private static unsafe void XXH3_scrambleAcc(ulong* acc, byte* secret) private static unsafe void XXH3_scrambleAcc(ulong* acc, byte* secret)
{ {
// TODO: SIMD if (Sse2.IsSupported)
XXH3_scrambleAcc_sse2(acc, secret);
else
XXH3_scrambleAcc_scalar(acc, secret);
}
XXH3_scrambleAcc_scalar(acc, secret); private static unsafe void XXH3_scrambleAcc_sse2(ulong* acc, byte* secret)
{
const int m128i_size = 16;
var prime32 = Vector128.Create(XXH_PRIME32_1);
for (int i = 0; i < XXH_STRIPE_LEN / m128i_size; i++)
{
int uint32_offset = i * 4;
int uint64_offset = i * 2;
var acc_vec = Sse2.LoadVector128(acc + uint64_offset).AsUInt32();
var shifted = Sse2.ShiftRightLogical(acc_vec, 47);
var data_vec = Sse2.Xor(acc_vec, shifted);
var key_vec = Sse2.LoadVector128((uint*) secret + uint32_offset);
var data_key = Sse2.Xor(data_vec, key_vec);
var data_key_hi = Sse2.Shuffle(data_key.AsUInt32(), _MM_SHUFFLE(0, 3, 0, 1));
var prod_lo = Sse2.Multiply(data_key, prime32);
var prod_hi = Sse2.Multiply(data_key_hi, prime32);
var result = Sse2.Add(prod_lo, Sse2.ShiftLeftLogical(prod_hi, 32));
Sse2.Store(acc + uint64_offset, result);
}
} }
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
@ -438,11 +489,33 @@ namespace Standart.Hash.xxHash
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
private static unsafe void XXH3_initCustomSecret(byte* customSecret, ulong seed) private static unsafe void XXH3_initCustomSecret(byte* customSecret, ulong seed)
{ {
// TODO: SIMD if (Sse2.IsSupported)
XXH3_initCustomSecret_sse2(customSecret, seed);
XXH3_initCustomSecret_scalar(customSecret, seed); else
XXH3_initCustomSecret_scalar(customSecret, seed);
} }
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static unsafe void XXH3_initCustomSecret_sse2(byte* customSecret, ulong seed64)
{
const int m128i_size = 16;
var seed = Vector128.Create((long) seed64, (long) (0U - seed64));
fixed (byte* secret = &XXH3_SECRET[0])
{
for (int i = 0; i < XXH_SECRET_DEFAULT_SIZE / m128i_size; ++i)
{
int uint64_offset = i * 2;
var src16 = Sse2.LoadVector128((long*) secret + uint64_offset);
var dst16 = Sse2.Add(src16, seed);
Sse2.Store((long*) customSecret, dst16);
}
}
}
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
private static unsafe void XXH3_initCustomSecret_scalar(byte* customSecret, ulong seed) private static unsafe void XXH3_initCustomSecret_scalar(byte* customSecret, ulong seed)
{ {