1
0
mirror of https://github.com/oliverbooth/X10D synced 2024-11-23 00:38:47 +00:00

test: 100% coverage on IntrinsicUtility (#73)

This commit is contained in:
Oliver Booth 2023-04-03 13:41:41 +01:00
parent 783c4b0f8e
commit 6ef48fc3b9
No known key found for this signature in database
GPG Key ID: 20BEB9DC87961025
2 changed files with 262 additions and 126 deletions

View File

@ -78,6 +78,124 @@ public class IntrinsicTests
Assert.AreEqual(expectedResult, result); Assert.AreEqual(expectedResult, result);
} }
[TestMethod]
public void HorizontalOr_ShouldReturnCombinedVector_GivenInputVector128OfUInt32()
{
Vector128<uint> left = Vector128.Create(1U, 2U, 3U, 4U);
Vector128<uint> right = Vector128.Create(5U, 6U, 7U, 8U);
Vector128<uint> expected = Vector128.Create(3U, 7U, 7U, 15U);
Vector128<uint> actual = IntrinsicUtility.HorizontalOr(left, right);
Assert.AreEqual(expected, actual);
}
[TestMethod]
public void HorizontalOrInternal_Sse_ShouldReturnCombinedVector_GivenInputVector128OfInt32()
{
Vector128<int> left = Vector128.Create(1, 2, 3, 4);
Vector128<int> right = Vector128.Create(5, 6, 7, 8);
Vector128<int> expected = Vector128.Create(3, 7, 7, 15);
Vector128<int> actual = IntrinsicUtility.HorizontalOr_Sse(left, right);
Assert.AreEqual(expected, actual);
}
[TestMethod]
public void HorizontalOrInternal_Fallback_ShouldReturnCombinedVector_GivenInputVector128OfInt32()
{
Vector128<int> left = Vector128.Create(1, 2, 3, 4);
Vector128<int> right = Vector128.Create(5, 6, 7, 8);
Vector128<int> expected = Vector128.Create(3, 7, 7, 15);
Vector128<int> actual = IntrinsicUtility.HorizontalOrInternal_Fallback(left, right);
Assert.AreEqual(expected, actual);
}
[TestMethod]
public void Multiply_ShouldReturnMultipliedVector_GivenInputVector128OfInt64()
{
Vector128<long> left = Vector128.Create(6L, 4L);
Vector128<long> right = Vector128.Create(2L, 3L);
Vector128<long> expected = Vector128.Create(12L, 12L);
Vector128<long> actual = IntrinsicUtility.Multiply(left, right);
Assert.AreEqual(expected, actual);
}
[TestMethod]
public void MultiplyInternal_Sse2_ShouldReturnMultipliedVector_GivenInputVector128OfUInt64()
{
if (!Sse2.IsSupported)
{
return;
}
Vector128<ulong> left = Vector128.Create(6UL, 4UL);
Vector128<ulong> right = Vector128.Create(2UL, 3UL);
Vector128<ulong> expected = Vector128.Create(12UL, 12UL);
Vector128<ulong> actual = IntrinsicUtility.MultiplyInternal_Sse2(left, right);
Assert.AreEqual(expected, actual);
}
[TestMethod]
public void MultiplyInternal_Fallback_ShouldReturnMultipliedVector_GivenInputVector128OfUInt64()
{
Vector128<ulong> left = Vector128.Create(6UL, 4UL);
Vector128<ulong> right = Vector128.Create(2UL, 3UL);
Vector128<ulong> expected = Vector128.Create(12UL, 12UL);
Vector128<ulong> actual = IntrinsicUtility.MultiplyInternal_Fallback(left, right);
Assert.AreEqual(expected, actual);
}
[TestMethod]
public void Multiply_ShouldReturnMultipliedVector_GivenInputVector256OfInt64()
{
Vector256<long> left = Vector256.Create(4L, 6L, 8L, 10L);
Vector256<long> right = Vector256.Create(2L, 3L, 4L, 5L);
Vector256<long> expected = Vector256.Create(8L, 18L, 32L, 50L);
Vector256<long> actual = IntrinsicUtility.Multiply(left, right);
Assert.AreEqual(expected, actual);
}
[TestMethod]
public void MultiplyInternal_Avx2_ShouldReturnMultipliedVector_GivenInputVector256OfUInt64()
{
if (!Avx2.IsSupported)
{
return;
}
Vector256<ulong> left = Vector256.Create(4UL, 6UL, 8UL, 10UL);
Vector256<ulong> right = Vector256.Create(2UL, 3UL, 4UL, 5UL);
Vector256<ulong> expected = Vector256.Create(8UL, 18UL, 32UL, 50UL);
Vector256<ulong> actual = IntrinsicUtility.MultiplyInternal_Avx2(left, right);
Assert.AreEqual(expected, actual);
}
[TestMethod]
public void MultiplyInternal_Fallback_ShouldReturnMultipliedVector_GivenInputVector256OfUInt64()
{
Vector256<ulong> left = Vector256.Create(4UL, 6UL, 8UL, 10UL);
Vector256<ulong> right = Vector256.Create(2UL, 3UL, 4UL, 5UL);
Vector256<ulong> expected = Vector256.Create(8UL, 18UL, 32UL, 50UL);
Vector256<ulong> actual = IntrinsicUtility.MultiplyInternal_Fallback(left, right);
Assert.AreEqual(expected, actual);
}
[TestMethod] [TestMethod]
public void ReverseElementsInternal_Fallback_ShouldReturnExpectedVector128Result_GivenInputVector() public void ReverseElementsInternal_Fallback_ShouldReturnExpectedVector128Result_GivenInputVector()
{ {

View File

@ -1,5 +1,6 @@
#if NETCOREAPP3_0_OR_GREATER #if NETCOREAPP3_0_OR_GREATER
using System.Diagnostics.CodeAnalysis;
using System.Diagnostics.Contracts; using System.Diagnostics.Contracts;
using System.Runtime.CompilerServices; using System.Runtime.CompilerServices;
using System.Runtime.Intrinsics; using System.Runtime.Intrinsics;
@ -22,48 +23,25 @@ public static class IntrinsicUtility
/// </para> /// </para>
/// Operation:<br/> /// Operation:<br/>
/// <code> /// <code>
/// dest[0] = lhs[0] * rhs[0]; /// dest[0] = left[0] * right[0];
/// dest[1] = lhs[1] * rhs[1]; /// dest[1] = left[1] * right[1];
/// </code> /// </code>
/// </summary> /// </summary>
/// <param name="lhs">Left vector.</param> /// <param name="left">Left vector.</param>
/// <param name="rhs">Right vector.</param> /// <param name="right">Right vector.</param>
/// <returns> /// <returns>The truncated product vector.</returns>
/// A <see cref="Vector128{T}"/> of <see langword="ulong"/> whose elements is 64-bit truncated product of lhs and rhs.
/// </returns>
[Pure] [Pure]
[CLSCompliant(false)] [CLSCompliant(false)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)] [MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static Vector128<ulong> Multiply(Vector128<ulong> lhs, Vector128<ulong> rhs) [ExcludeFromCodeCoverage]
public static Vector128<ulong> Multiply(Vector128<ulong> left, Vector128<ulong> right)
{ {
if (Sse2.IsSupported) if (Sse2.IsSupported)
{ {
// https://stackoverflow.com/questions/17863411/sse-multiplication-of-2-64-bit-integers return MultiplyInternal_Sse2(left, right);
Vector128<ulong> ac = Sse2.Multiply(lhs.AsUInt32(), rhs.AsUInt32());
Vector128<uint> b = Sse2.ShiftRightLogical(lhs, 32).AsUInt32();
Vector128<ulong> bc = Sse2.Multiply(b, rhs.AsUInt32());
Vector128<uint> d = Sse2.ShiftRightLogical(rhs, 32).AsUInt32();
Vector128<ulong> ad = Sse2.Multiply(lhs.AsUInt32(), d);
Vector128<ulong> high = Sse2.Add(bc, ad);
high = Sse2.ShiftLeftLogical(high, 32);
return Sse2.Add(high, ac);
} }
// TODO: AdvSimd implementation. return MultiplyInternal_Fallback(left, right);
// TODO: WasmSimd implementation.
var output = GetUninitializedVector128<ulong>();
Unsafe.As<Vector128<ulong>, ulong>(ref output) =
Unsafe.As<Vector128<ulong>, ulong>(ref lhs) * Unsafe.As<Vector128<ulong>, ulong>(ref rhs);
Unsafe.Add(ref Unsafe.As<Vector128<ulong>, ulong>(ref output), 1) =
Unsafe.Add(ref Unsafe.As<Vector128<ulong>, ulong>(ref lhs), 1) *
Unsafe.Add(ref Unsafe.As<Vector128<ulong>, ulong>(ref rhs), 1);
return output;
} }
/// <summary> /// <summary>
@ -72,10 +50,10 @@ public static class IntrinsicUtility
/// </para> /// </para>
/// Operation:<br/> /// Operation:<br/>
/// <code> /// <code>
/// dest[0] = lhs[0] * rhs[0]; /// dest[0] = left[0] * right[0];
/// dest[1] = lhs[1] * rhs[1]; /// dest[1] = left[1] * right[1];
/// dest[2] = lhs[2] * rhs[2]; /// dest[2] = left[2] * right[2];
/// dest[3] = lhs[3] * rhs[3]; /// dest[3] = left[3] * right[3];
/// </code> /// </code>
/// </summary> /// </summary>
/// <param name="lhs">Left vector.</param> /// <param name="lhs">Left vector.</param>
@ -86,33 +64,15 @@ public static class IntrinsicUtility
[Pure] [Pure]
[CLSCompliant(false)] [CLSCompliant(false)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)] [MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
[ExcludeFromCodeCoverage]
public static Vector256<ulong> Multiply(Vector256<ulong> lhs, Vector256<ulong> rhs) public static Vector256<ulong> Multiply(Vector256<ulong> lhs, Vector256<ulong> rhs)
{ {
if (Avx2.IsSupported) if (Avx2.IsSupported)
{ {
// https://stackoverflow.com/questions/17863411/sse-multiplication-of-2-64-bit-integers return MultiplyInternal_Avx2(lhs, rhs);
Vector256<ulong> ac = Avx2.Multiply(lhs.AsUInt32(), rhs.AsUInt32());
Vector256<uint> b = Avx2.ShiftRightLogical(lhs, 32).AsUInt32();
Vector256<ulong> bc = Avx2.Multiply(b, rhs.AsUInt32());
Vector256<uint> d = Avx2.ShiftRightLogical(rhs, 32).AsUInt32();
Vector256<ulong> ad = Avx2.Multiply(lhs.AsUInt32(), d);
Vector256<ulong> high = Avx2.Add(bc, ad);
high = Avx2.ShiftLeftLogical(high, 32);
return Avx2.Add(high, ac);
} }
var output = GetUninitializedVector256<ulong>(); return MultiplyInternal_Fallback(lhs, rhs);
for (int i = 0; i < Vector256<ulong>.Count; i++)
{
Unsafe.Add(ref Unsafe.As<Vector256<ulong>, ulong>(ref output), i) =
Unsafe.Add(ref Unsafe.As<Vector256<ulong>, ulong>(ref lhs), i) *
Unsafe.Add(ref Unsafe.As<Vector256<ulong>, ulong>(ref rhs), i);
}
return output;
} }
/// <summary> /// <summary>
@ -121,8 +81,8 @@ public static class IntrinsicUtility
/// </para> /// </para>
/// Operation:<br/> /// Operation:<br/>
/// <code> /// <code>
/// dest[0] = lhs[0] * rhs[0]; /// dest[0] = left[0] * right[0];
/// dest[1] = lhs[1] * rhs[1]; /// dest[1] = left[1] * right[1];
/// </code> /// </code>
/// </summary> /// </summary>
/// <param name="lhs">Left vector.</param> /// <param name="lhs">Left vector.</param>
@ -143,10 +103,10 @@ public static class IntrinsicUtility
/// </para> /// </para>
/// Operation:<br/> /// Operation:<br/>
/// <code> /// <code>
/// dest[0] = lhs[0] * rhs[0]; /// dest[0] = left[0] * right[0];
/// dest[1] = lhs[1] * rhs[1]; /// dest[1] = left[1] * right[1];
/// dest[2] = lhs[2] * rhs[2]; /// dest[2] = left[2] * right[2];
/// dest[3] = lhs[3] * rhs[3]; /// dest[3] = left[3] * right[3];
/// </code> /// </code>
/// </summary> /// </summary>
/// <param name="lhs">Left vector.</param> /// <param name="lhs">Left vector.</param>
@ -168,77 +128,32 @@ public static class IntrinsicUtility
/// </para> /// </para>
/// Operation:<br/> /// Operation:<br/>
/// <code> /// <code>
/// dest[0] = lhs[0] | lhs[1]; /// dest[0] = left[0] | left[1];
/// dest[1] = lhs[2] | lhs[3]; /// dest[1] = left[2] | left[3];
/// dest[2] = rhs[0] | rhs[1]; /// dest[2] = right[0] | right[1];
/// dest[3] = rhs[2] | rhs[3]; /// dest[3] = right[2] | right[3];
/// </code> /// </code>
/// </summary> /// </summary>
/// <param name="lhs">Left vector.</param> /// <param name="left">Left vector.</param>
/// <param name="rhs">Right vector.</param> /// <param name="right">Right vector.</param>
/// <returns> /// <returns>
/// A <see cref="Vector128{T}"/> of <see langword="float"/> with all elements is result of OR operation on adjacent pairs of /// A <see cref="Vector128{T}"/> of <see langword="float"/> with all elements is result of OR operation on adjacent pairs of
/// elements in lhs and rhs. /// elements in lhs and rhs.
/// </returns> /// </returns>
[Pure] [Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)] [MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static Vector128<float> HorizontalOr(Vector128<float> lhs, Vector128<float> rhs) [ExcludeFromCodeCoverage]
public static Vector128<int> HorizontalOr(Vector128<int> left, Vector128<int> right)
{ {
if (Sse.IsSupported) if (Sse.IsSupported)
{ {
var s1 = Sse.Shuffle(lhs, rhs, 0b10_00_10_00); // s1 = { lhs[0] ; lhs[2] ; rhs[0] ; rhs[2] } return HorizontalOr_Sse(left, right);
var s2 = Sse.Shuffle(lhs, rhs, 0b11_01_11_01); // s2 = { lhs[1] ; lhs[3] ; rhs[1] ; rhs[3] }
return Sse.Or(s1, s2);
} }
// TODO: AdvSimd implementation. // TODO: AdvSimd implementation.
// TODO: WasmSimd implementation. (?) // TODO: WasmSimd implementation. (?)
Vector128<float> output = GetUninitializedVector128<float>(); return HorizontalOrInternal_Fallback(left, right);
Unsafe.As<Vector128<float>, uint>(ref output) =
Unsafe.As<Vector128<float>, uint>(ref lhs) |
Unsafe.Add(ref Unsafe.As<Vector128<float>, uint>(ref lhs), 1);
Unsafe.Add(ref Unsafe.As<Vector128<float>, uint>(ref output), 1) =
Unsafe.Add(ref Unsafe.As<Vector128<float>, uint>(ref lhs), 2) |
Unsafe.Add(ref Unsafe.As<Vector128<float>, uint>(ref lhs), 3);
Unsafe.Add(ref Unsafe.As<Vector128<float>, uint>(ref output), 2) =
Unsafe.As<Vector128<float>, uint>(ref rhs) |
Unsafe.Add(ref Unsafe.As<Vector128<float>, uint>(ref rhs), 1);
Unsafe.Add(ref Unsafe.As<Vector128<float>, uint>(ref output), 3) =
Unsafe.Add(ref Unsafe.As<Vector128<float>, uint>(ref rhs), 2) |
Unsafe.Add(ref Unsafe.As<Vector128<float>, uint>(ref rhs), 3);
return output;
}
/// <summary>
/// <para>
/// Horizontally apply OR operation on adjacent pairs of 32-bit integer elements in lhs and rhs.
/// </para>
/// Operation:<br/>
/// <code>
/// dest[0] = lhs[0] | lhs[1];
/// dest[1] = lhs[2] | lhs[3];
/// dest[2] = rhs[0] | rhs[1];
/// dest[3] = rhs[2] | rhs[3];
/// </code>
/// </summary>
/// <param name="lhs">Left vector.</param>
/// <param name="rhs">Right vector.</param>
/// <returns>
/// A <see cref="Vector128{T}"/> of <see langword="int"/> with all elements is result of OR operation on adjacent pairs of
/// elements in lhs and rhs.
/// </returns>
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static Vector128<int> HorizontalOr(Vector128<int> lhs, Vector128<int> rhs)
{
return HorizontalOr(lhs.AsSingle(), rhs.AsSingle()).AsInt32();
} }
/// <summary> /// <summary>
@ -247,14 +162,14 @@ public static class IntrinsicUtility
/// </para> /// </para>
/// Operation:<br/> /// Operation:<br/>
/// <code> /// <code>
/// dest[0] = lhs[0] | lhs[1]; /// dest[0] = left[0] | left[1];
/// dest[1] = lhs[2] | lhs[3]; /// dest[1] = left[2] | left[3];
/// dest[2] = rhs[0] | rhs[1]; /// dest[2] = right[0] | right[1];
/// dest[3] = rhs[2] | rhs[3]; /// dest[3] = right[2] | right[3];
/// </code> /// </code>
/// </summary> /// </summary>
/// <param name="lhs">Left vector.</param> /// <param name="left">Left vector.</param>
/// <param name="rhs">Right vector.</param> /// <param name="right">Right vector.</param>
/// <returns> /// <returns>
/// A <see cref="Vector128{T}"/> of <see langword="uint"/> with all elements is result of OR operation on adjacent pairs of /// A <see cref="Vector128{T}"/> of <see langword="uint"/> with all elements is result of OR operation on adjacent pairs of
/// elements in lhs and rhs. /// elements in lhs and rhs.
@ -262,9 +177,9 @@ public static class IntrinsicUtility
[Pure] [Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)] [MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
[CLSCompliant(false)] [CLSCompliant(false)]
public static Vector128<uint> HorizontalOr(Vector128<uint> lhs, Vector128<uint> rhs) public static Vector128<uint> HorizontalOr(Vector128<uint> left, Vector128<uint> right)
{ {
return HorizontalOr(lhs.AsSingle(), rhs.AsSingle()).AsUInt32(); return HorizontalOr(left.AsInt32(), right.AsInt32()).AsUInt32();
} }
// Helper methods // Helper methods
@ -300,6 +215,109 @@ public static class IntrinsicUtility
return default; return default;
#endif #endif
} }
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static Vector128<int> HorizontalOr_Sse(Vector128<int> left, Vector128<int> right)
{
Vector128<float> leftSingle = left.AsSingle();
Vector128<float> rightSingle = right.AsSingle();
// first = { left[0] ; left[2] ; right[0] ; right[2] }
// second = { left[1] ; left[3] ; right[1] ; right[3] }
Vector128<float> first = Sse.Shuffle(leftSingle, rightSingle, 0b10_00_10_00);
Vector128<float> second = Sse.Shuffle(leftSingle, rightSingle, 0b11_01_11_01);
return Sse.Or(first, second).AsInt32();
}
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static Vector128<int> HorizontalOrInternal_Fallback(Vector128<int> left, Vector128<int> right)
{
Vector128<int> output = GetUninitializedVector128<int>();
ref int outputInteger = ref Unsafe.As<Vector128<int>, int>(ref output);
ref int leftInteger = ref Unsafe.As<Vector128<int>, int>(ref left);
ref int rightInteger = ref Unsafe.As<Vector128<int>, int>(ref right);
outputInteger = leftInteger | Unsafe.Add(ref leftInteger, 1);
Unsafe.Add(ref outputInteger, 1) = Unsafe.Add(ref leftInteger, 2) | Unsafe.Add(ref leftInteger, 3);
Unsafe.Add(ref outputInteger, 2) = rightInteger | Unsafe.Add(ref rightInteger, 1);
Unsafe.Add(ref outputInteger, 3) = Unsafe.Add(ref rightInteger, 2) | Unsafe.Add(ref rightInteger, 3);
return output;
}
[Pure]
[CLSCompliant(false)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static Vector128<ulong> MultiplyInternal_Fallback(Vector128<ulong> left, Vector128<ulong> right)
{
ulong leftInteger1 = Unsafe.As<Vector128<ulong>, ulong>(ref left);
ulong rightInteger1 = Unsafe.As<Vector128<ulong>, ulong>(ref right);
ulong result1 = leftInteger1 * rightInteger1;
ulong leftInteger2 = Unsafe.Add(ref Unsafe.As<Vector128<ulong>, ulong>(ref left), 1);
ulong rightInteger2 = Unsafe.Add(ref Unsafe.As<Vector128<ulong>, ulong>(ref right), 1);
ulong result2 = leftInteger2 * rightInteger2;
Vector128<ulong> output = Vector128.Create(result1, result2);
return output;
}
[Pure]
[CLSCompliant(false)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static Vector128<ulong> MultiplyInternal_Sse2(Vector128<ulong> left, Vector128<ulong> right)
{
// https://stackoverflow.com/questions/17863411/sse-multiplication-of-2-64-bit-integers
Vector128<ulong> ac = Sse2.Multiply(left.AsUInt32(), right.AsUInt32());
Vector128<uint> b = Sse2.ShiftRightLogical(left, 32).AsUInt32();
Vector128<ulong> bc = Sse2.Multiply(b, right.AsUInt32());
Vector128<uint> d = Sse2.ShiftRightLogical(right, 32).AsUInt32();
Vector128<ulong> ad = Sse2.Multiply(left.AsUInt32(), d);
Vector128<ulong> high = Sse2.Add(bc, ad);
high = Sse2.ShiftLeftLogical(high, 32);
return Sse2.Add(high, ac);
}
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static Vector256<ulong> MultiplyInternal_Fallback(Vector256<ulong> left, Vector256<ulong> right)
{
Vector256<ulong> output = GetUninitializedVector256<ulong>();
for (var index = 0; index < Vector256<ulong>.Count; index++)
{
Unsafe.Add(ref Unsafe.As<Vector256<ulong>, ulong>(ref output), index) =
Unsafe.Add(ref Unsafe.As<Vector256<ulong>, ulong>(ref left), index) *
Unsafe.Add(ref Unsafe.As<Vector256<ulong>, ulong>(ref right), index);
}
return output;
}
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
internal static Vector256<ulong> MultiplyInternal_Avx2(Vector256<ulong> left, Vector256<ulong> right)
{
// https://stackoverflow.com/questions/17863411/sse-multiplication-of-2-64-bit-integers
Vector256<ulong> ac = Avx2.Multiply(left.AsUInt32(), right.AsUInt32());
Vector256<uint> b = Avx2.ShiftRightLogical(left, 32).AsUInt32();
Vector256<ulong> bc = Avx2.Multiply(b, right.AsUInt32());
Vector256<uint> d = Avx2.ShiftRightLogical(right, 32).AsUInt32();
Vector256<ulong> ad = Avx2.Multiply(left.AsUInt32(), d);
Vector256<ulong> high = Avx2.Add(bc, ad);
high = Avx2.ShiftLeftLogical(high, 32);
return Avx2.Add(high, ac);
}
} }
#endif #endif