oliverbooth/X10D
oliverbooth
/
X10D
mirror of https://github.com/oliverbooth/X10D
1
0
Fork 0
Code Issues

Fix source validator's code reports, remove 95% of AdvSimd implementation to prevent future consequences

This commit is contained in:
RealityProgrammer 2023-03-14 19:17:19 +07:00
parent ec8e60c6dc
commit 8b8aeb3f56
6 changed files with 125 additions and 122 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
X10D/src/Collections/BoolListExtensions.cs
View File

@ -26,7 +26,7 @@ public static class BoolListExtensions
throw new ArgumentNullException(nameof(source));
}
#endif
if (source.Count > 8)
{
throw new ArgumentException("Source cannot contain more than than 8 elements.", nameof(source));

2
X10D/src/Collections/ByteExtensions.cs
View File

@ -47,7 +47,7 @@ public static class ByteExtensions
return;
}
#endif
FallbackImplementation(value, destination);
#if NETCOREAPP3_0_OR_GREATER

20
X10D/src/Collections/Int32Extensions.cs
View File

@ -42,7 +42,7 @@ public static class Int32Extensions
#if NETCOREAPP3_0_OR_GREATER
// TODO: AdvSimd support.
// https://stackoverflow.com/questions/24225786/fastest-way-to-unpack-32-bits-to-a-32-byte-simd-vector
if (Avx2.IsSupported)
{
@ -64,15 +64,15 @@ public static class Int32Extensions
fixed (bool* pDestination = destination)
{
var mask1 = Vector256.Create(
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01,
0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01,
0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02,
0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03
).AsByte();
var mask2 = Vector256.Create(
0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80
);
@ -81,7 +81,7 @@ public static class Int32Extensions
var and = Avx2.AndNot(shuffle, mask2);
var cmp = Avx2.CompareEqual(and, Vector256<byte>.Zero);
var correctness = Avx2.And(cmp, Vector256.Create((byte)0x01));
Avx.Store((byte*)pDestination, correctness);
}
}
@ -103,9 +103,9 @@ public static class Int32Extensions
var and = Sse2.AndNot(shuffle, mask2);
var cmp = Sse2.CompareEqual(and, Vector128<byte>.Zero);
var correctness = Sse2.And(cmp, one);
Sse2.Store((byte*)pDestination, correctness);
shuffle = Ssse3.Shuffle(vec, mask1Hi);
and = Sse2.AndNot(shuffle, mask2);
cmp = Sse2.CompareEqual(and, Vector128<byte>.Zero);

3
X10D/src/Core/IntrinsicExtensions.cs
View File

@ -5,7 +5,8 @@ using System.Runtime.Intrinsics;
namespace X10D.Core;
/// <summary>
/// Extension methods for SIMD vectors, namely <see cref="Vector64{T}"/>, <see cref="Vector128{T}"/> and <see cref="Vector256{T}"/>.
/// Extension methods for SIMD vectors, namely <see cref="Vector64{T}"/>, <see cref="Vector128{T}"/> and
/// <see cref="Vector256{T}"/>.
/// </summary>
public static class IntrinsicExtensions
{

176
X10D/src/Core/IntrinsicUtility.cs
View File

@ -18,7 +18,9 @@ public static class IntrinsicUtility
// FOR API CONSISTENCY.
/// <summary>
/// <br>Correcting <see cref="Vector64{T}"/> of <see langword="byte"/> into 0 and 1 depend on their boolean truthiness.</br>
/// <br>
/// Correcting <see cref="Vector64{T}"/> of <see langword="byte"/> into 0 and 1 depend on their boolean truthiness.
/// </br>
/// <br>Operation (raw):</br>
/// <code>
/// for (int i = 0; i &lt; 8; i++) {
@ -33,19 +35,15 @@ public static class IntrinsicUtility
/// </code>
/// </summary>
/// <param name="vector">Vector of byte to correct.</param>
/// <returns>A <see cref="Vector64{T}"/> of <see langword="byte"/> which remapped back to 0 and 1 based on boolean truthiness.</returns>
/// <returns>
/// A <see cref="Vector64{T}"/> of <see langword="byte"/> which remapped back to 0 and 1 based on boolean truthiness.
/// </returns>
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static Vector64<byte> CorrectBoolean(Vector64<byte> vector)
{
if (AdvSimd.IsSupported)
{
// Haven't tested since March 6th 2023 (Reason: Unavailable hardware).
var cmp = AdvSimd.CompareEqual(vector, Vector64<byte>.Zero);
var result = AdvSimd.BitwiseSelect(cmp, vector, Vector64<byte>.Zero);
return result;
}
// TODO: AdvSimd implementation.
// TODO: WasmSimd implementation. (?)
var output = GetUninitializedVector64<byte>();
@ -64,7 +62,9 @@ public static class IntrinsicUtility
}
/// <summary>
/// <br>Correcting <see cref="Vector128{T}"/> of <see langword="byte"/> into 0 and 1 depend on their boolean truthiness.</br>
/// <br>
/// Correcting <see cref="Vector128{T}"/> of <see langword="byte"/> into 0 and 1 depend on their boolean truthiness.
/// </br>
/// <br>Operation (raw):</br>
/// <code>
/// for (int i = 0; i &lt; 16; i++) {
@ -79,7 +79,9 @@ public static class IntrinsicUtility
/// </code>
/// </summary>
/// <param name="vector">Vector of byte to correct.</param>
/// <returns>A <see cref="Vector128{T}"/> of <see langword="byte"/> which remapped back to 0 and 1 based on boolean truthiness.</returns>
/// <returns>
/// A <see cref="Vector128{T}"/> of <see langword="byte"/> which remapped back to 0 and 1 based on boolean truthiness.
/// </returns>
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static Vector128<byte> CorrectBoolean(Vector128<byte> vector)
@ -91,33 +93,25 @@ public static class IntrinsicUtility
return result;
}
if (AdvSimd.IsSupported)
{
// Haven't tested since March 6th 2023 (Reason: Unavailable hardware).
var cmp = AdvSimd.CompareEqual(vector, Vector128<byte>.Zero);
var result = AdvSimd.BitwiseSelect(cmp, vector, Vector128<byte>.Zero);
return result;
}
// TODO: AdvSimd implementation.
// TODO: WasmSimd implementation.
var output = GetUninitializedVector128<byte>();
for (int i = 0; i < Vector128<byte>.Count; i++)
{
ref var writeElement = ref Unsafe.Add(ref Unsafe.As<Vector128<byte>, byte>(ref output), i);
#if NET7_0_OR_GREATER
writeElement = vector[i] == 0 ? (byte)0 : (byte)1;
#else
var element = Unsafe.Add(ref Unsafe.As<Vector128<byte>, byte>(ref vector), i);
writeElement = element == 0 ? (byte)0 : (byte)1;
#endif
Unsafe.Add(ref Unsafe.As<Vector128<byte>, byte>(ref output), i) =
Unsafe.Add(ref Unsafe.As<Vector128<byte>, byte>(ref vector), i) == 0 ? (byte)0 : (byte)1;
}
return output;
}
/// <summary>
/// <br>Correcting <see cref="Vector256{T}"/> of <see langword="byte"/> into 0 and 1 depend on their boolean truthiness.</br>
/// <br>
/// Correcting <see cref="Vector256{T}"/> of <see langword="byte"/> into 0 and 1 depend on their boolean truthiness.
/// </br>
/// <br>Operation (raw):</br>
/// <code>
/// for (int i = 0; i &lt; 16; i++) {
@ -132,7 +126,9 @@ public static class IntrinsicUtility
/// </code>
/// </summary>
/// <param name="vector">Vector of byte to correct.</param>
/// <returns>A <see cref="Vector256{T}"/> of <see langword="byte"/> which remapped back to 0 and 1 based on boolean truthiness.</returns>
/// <returns>
/// A <see cref="Vector256{T}"/> of <see langword="byte"/> which remapped back to 0 and 1 based on boolean truthiness.
/// </returns>
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static Vector256<byte> CorrectBoolean(Vector256<byte> vector)
@ -149,20 +145,17 @@ public static class IntrinsicUtility
for (int i = 0; i < Vector256<byte>.Count; i++)
{
ref var writeElement = ref Unsafe.Add(ref Unsafe.As<Vector256<byte>, byte>(ref output), i);
#if NET7_0_OR_GREATER
writeElement = vector[i] == 0 ? (byte)0 : (byte)1;
#else
var element = Unsafe.Add(ref Unsafe.As<Vector256<byte>, byte>(ref vector), i);
writeElement = element == 0 ? (byte)0 : (byte)1;
#endif
Unsafe.Add(ref Unsafe.As<Vector256<byte>, byte>(ref output), i) =
Unsafe.Add(ref Unsafe.As<Vector256<byte>, byte>(ref vector), i) == 0 ? (byte)0 : (byte)1;
}
return output;
}
/// <summary>
/// <br>Multiply packed 64-bit unsigned integer elements in a and b and truncate the results to 64-bit integer.</br>
/// <br>
/// Multiply packed 64-bit unsigned integer elements in a and b and truncate the results to 64-bit integer.
/// </br>
/// <br>Operation:</br>
/// <code>
/// dest[0] = lhs[0] * rhs[0];
@ -171,7 +164,9 @@ public static class IntrinsicUtility
/// </summary>
/// <param name="lhs">Left vector.</param>
/// <param name="rhs">Right vector.</param>
/// <returns>A <see cref="Vector128{T}"/> of <see langword="ulong"/> whose elements is 64-bit truncated product of lhs and rhs.</returns>
/// <returns>
/// A <see cref="Vector128{T}"/> of <see langword="ulong"/> whose elements is 64-bit truncated product of lhs and rhs.
/// </returns>
[Pure]
[CLSCompliant(false)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
@ -191,32 +186,26 @@ public static class IntrinsicUtility
return Sse2.Add(high, ac);
}
if (AdvSimd.IsSupported)
{
// https://stackoverflow.com/questions/60236627/facing-problem-in-implementing-multiplication-of-64-bit-variables-using-arm-neon
// Hasn't been tested since March 7th 2023 (Reason: Unavailable hardware)
var a = AdvSimd.ExtractNarrowingLower(lhs);
var b = AdvSimd.ExtractNarrowingLower(rhs);
var mul = AdvSimd.Multiply(rhs.AsUInt32(), AdvSimd.ReverseElement32(lhs).AsUInt32());
return AdvSimd.MultiplyWideningLowerAndAdd(AdvSimd.ShiftLeftLogical(mul.AsUInt64(), 32), a, b);
}
// TODO: AdvSimd implementation.
// 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);
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>
/// <br>Multiply packed 64-bit unsigned integer elements in a and b and truncate the results to 64-bit integer.</br>
/// <br>
/// Multiply packed 64-bit unsigned integer elements in a and b and truncate the results to 64-bit integer.
/// </br>
/// <br>Operation:</br>
/// <code>
/// dest[0] = lhs[0] * rhs[0];
@ -227,7 +216,9 @@ public static class IntrinsicUtility
/// </summary>
/// <param name="lhs">Left vector.</param>
/// <param name="rhs">Right vector.</param>
/// <returns>A <see cref="Vector256{T}"/> of <see langword="ulong"/> whose elements is 64-bit truncated product of lhs and rhs.</returns>
/// <returns>
/// A <see cref="Vector256{T}"/> of <see langword="ulong"/> whose elements is 64-bit truncated product of lhs and rhs.
/// </returns>
[Pure]
[CLSCompliant(false)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
@ -253,14 +244,17 @@ public static class IntrinsicUtility
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);
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>
/// <br>Multiply packed 64-bit signed integer elements in a and b and truncate the results to 64-bit integer.</br>
/// <br>
/// Multiply packed 64-bit signed integer elements in a and b and truncate the results to 64-bit integer.
/// </br>
/// <br>Operation:</br>
/// <code>
/// dest[0] = lhs[0] * rhs[0];
@ -269,7 +263,9 @@ public static class IntrinsicUtility
/// </summary>
/// <param name="lhs">Left vector.</param>
/// <param name="rhs">Right vector.</param>
/// <returns>A <see cref="Vector128{T}"/> of <see langword="long"/> whose elements is 64-bit truncated product of lhs and rhs.</returns>
/// <returns>
/// A <see cref="Vector128{T}"/> of <see langword="long"/> whose elements is 64-bit truncated product of lhs and rhs.
/// </returns>
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static Vector128<long> Multiply(Vector128<long> lhs, Vector128<long> rhs)
@ -278,7 +274,9 @@ public static class IntrinsicUtility
}
/// <summary>
/// <br>Multiply packed 64-bit signed integer elements in a and b and truncate the results to 64-bit integer.</br>
/// <br>
/// Multiply packed 64-bit signed integer elements in a and b and truncate the results to 64-bit integer.
/// </br>
/// <br>Operation:</br>
/// <code>
/// dest[0] = lhs[0] * rhs[0];
@ -289,7 +287,9 @@ public static class IntrinsicUtility
/// </summary>
/// <param name="lhs">Left vector.</param>
/// <param name="rhs">Right vector.</param>
/// <returns>A <see cref="Vector256{T}"/> of <see langword="ulong"/> whose elements is 64-bit truncated product of lhs and rhs.</returns>
/// <returns>
/// A <see cref="Vector256{T}"/> of <see langword="ulong"/> whose elements is 64-bit truncated product of lhs and rhs.
/// </returns>
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static Vector256<long> Multiply(Vector256<long> lhs, Vector256<long> rhs)
@ -298,7 +298,10 @@ public static class IntrinsicUtility
}
/// <summary>
/// <br>Horizontally apply OR operation on adjacent pairs of single-precision (32-bit) floating-point elements in lhs and rhs.</br>
/// <br>
/// Horizontally apply OR operation on adjacent pairs of single-precision (32-bit) floating-point elements in lhs and
/// rhs.
/// </br>
/// <br>Operation:</br>
/// <code>
/// dest[0] = lhs[0] | lhs[1];
@ -309,7 +312,10 @@ public static class IntrinsicUtility
/// </summary>
/// <param name="lhs">Left vector.</param>
/// <param name="rhs">Right vector.</param>
/// <returns>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.</returns>
/// <returns>
/// 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.
/// </returns>
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public static Vector128<float> HorizontalOr(Vector128<float> lhs, Vector128<float> rhs)
@ -321,34 +327,35 @@ public static class IntrinsicUtility
return Sse.Or(s1, s2);
}
if (AdvSimd.Arm64.IsSupported)
{
// Hasn't been tested since March 7th 2023 (Reason: Unavailable hardware).
var s1 = AdvSimd.Arm64.UnzipEven(lhs, rhs);
var s2 = AdvSimd.Arm64.UnzipOdd(lhs, rhs);
return AdvSimd.Or(s1, s2);
}
// TODO: AdvSimd implementation.
// TODO: WasmSimd implementation. (?)
Vector128<float> output = GetUninitializedVector128<float>();
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.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 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.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);
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>
/// <br>Horizontally apply OR operation on adjacent pairs of 32-bit integer elements in lhs and rhs.</br>
/// <br>
/// Horizontally apply OR operation on adjacent pairs of 32-bit integer elements in lhs and rhs.
/// </br>
/// <br>Operation:</br>
/// <code>
/// dest[0] = lhs[0] | lhs[1];
@ -359,9 +366,10 @@ public static class IntrinsicUtility
/// </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>
/// <remarks>API avaliable on SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, AVX, AVX2, ARM64 NEON (untested) hardwares.</remarks>
/// <exception cref="PlatformNotSupportedException">Hardware doesn't support ARM64 NEON or SSE instruction set.</exception>
/// <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)
@ -370,7 +378,9 @@ public static class IntrinsicUtility
}
/// <summary>
/// <br>Horizontally apply OR operation on adjacent pairs of 32-bit unsigned integer elements in lhs and rhs.</br>
/// <br>
/// Horizontally apply OR operation on adjacent pairs of 32-bit unsigned integer elements in lhs and rhs.
/// </br>
/// <br>Operation:</br>
/// <code>
/// dest[0] = lhs[0] | lhs[1];
@ -381,9 +391,10 @@ public static class IntrinsicUtility
/// </summary>
/// <param name="lhs">Left vector.</param>
/// <param name="rhs">Right vector.</param>
/// <returns>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.</returns>
/// <remarks>API avaliable on SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, AVX, AVX2, ARM64 NEON (untested) hardwares.</remarks>
/// <exception cref="PlatformNotSupportedException">Hardware doesn't support ARM64 NEON or SSE2 instruction set.</exception>
/// <returns>
/// 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.
/// </returns>
[Pure]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
[CLSCompliant(false)]
@ -402,9 +413,10 @@ public static class IntrinsicUtility
/// </code>
/// </summary>
/// <param name="vector">Input vector.</param>
/// <returns>A <see cref="Vector128{T}"/> of <see langword="ulong"/> with elements the same as input vector except their positions/indices are reversed.</returns>
/// <remarks>API available on SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, AVX, AVX2 hardwares.</remarks>
/// <exception cref="PlatformNotSupportedException">Hardware doesn't support SSE2 instruction set.</exception>
/// <returns>
/// A <see cref="Vector128{T}"/> of <see langword="ulong"/> with elements the same as input vector except their positions
/// (or indices) are reversed.
/// </returns>
[Pure]
[CLSCompliant(false)]
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]

44
X10D/src/Core/SpanExtensions.cs
View File

@ -72,13 +72,14 @@ public static class SpanExtensions
public static bool Contains<T>(this ReadOnlySpan<T> span, T value) where T : struct, Enum
{
#if NET6_0_OR_GREATER
// Use MemoryMarshal.CreateSpan instead of using creating new Span instance from pointer will trim down a lot of instructions
// on Release mode.
// https://sharplab.io/#v2:EYLgxg9gTgpgtADwGwBYA0AXEBDAzgWwB8ABABgAJiBGAOgCUBXAOwwEt8YaBJFmKCAA4BlPgDdWYGLgDcAWABQZSrUYt2nAMIR8A1gBs+IqOMkyFxAExVzFIQAtsUAQBlsweszYc588wGZyGCYGfHIAFSkMAFFg0JByVhZyAG8FcnTyAEE0cgAhHI0cgBE0BQBfBX9KC3INFLSMgG0AKVYMAHEgvgkACgwATwEYCAAzHojcaNiASmmAXQb0xoBZGAw7CAATLh09HtX1rZ2BPQB5ATYIJlwaTIBzO9hcXFZRGB49RMS78kJyA4221250u11uDyeLzeIPYrAAXthQfNFpQAtQkORmLhsCMYORgBAIHp/mtAVQADxhAB8PSEAmwTEpVPIuHpTByYXIomwegYMGm5AA7nY+HjOfEYiF6vIMrLyLARgkkkEQrhyABeeUwRUAVWuOM4mVwlJyiQwNIVJPw0H6y0cuAcehonQwdG1oqYkh6rIZsx8coyxAA7FabXaoA6eTQNLBETA6QyepaVfhcDkfUwaM4gnd1tNo1cMNhErgenrsbjbsawqaWBbtVyeXy/SiKjKMiiWm1OkxumA+oNhmMJlMQrMFu2lgCjrt9qSZycYVcbvdHlIoe8mJ8mN9fiTDkDFxdWMvwWvnq8YDD8PDESemMjJ6jlBisQb8YTidPNhYmbS2UyLJshyja8vyQoirA4TkBKsTSgG6TBuQvaCuQCaMmaNLlgaVYAAoQGafBJg2qzWlAtr2o6zprG6uKwJ6MDemyszpmyWY5nmBYsMW1xlvqlZGiaSrmsRircmBLZPm2ZRAA===
// Use MemoryMarshal.CreateSpan instead of using creating new Span instance from pointer will trim down a lot of
// instructions on Release mode.
// Also use reference instead of MemoryMarshal.Cast to remove boundary check (or something, it just result in something like that).
// Also use reference instead of MemoryMarshal.Cast to remove boundary check (or something, it just result in something
// like that).
// TODO: Figure out some kind of way to directly pass the Span directly into Contains call, which make method smaller and more prone to inlining...
// TODO: Figure out some kind of way to directly pass the Span directly into Contains call, which make method smaller and
// more prone to inlining...
unsafe
{
#pragma warning disable CS8500 // This takes the address of, gets the size of, or declares a pointer to a managed type
@ -176,6 +177,10 @@ public static class SpanExtensions
return unchecked((byte)(IntegerPackingMagic * correct.AsUInt64().GetElement(0) >> 56));
}
// Probably should remove this piece of code because it is untested, but I see no reason why it should fail
// unless vld1_u8 reverse positions of 8 bytes for some reason.
if (AdvSimd.IsSupported)
{
// Hasn't been tested since March 6th 2023 (Reason: Unavailable hardware).
@ -240,12 +245,12 @@ public static class SpanExtensions
goto default;
}
fixed (bool* pSource = source)
{
// TODO: .NET 8.0 Wasm support.
// TODO: Implement a replacement for UInt64 vector multiplication (there are no instruction for this built-in).
// TODO: AdvSimd implementation.
// TODO: WasmSimd implementation.
if (Sse2.IsSupported)
if (Sse2.IsSupported)
{
fixed (bool* pSource = source)
{
var load = Sse2.LoadVector128((byte*)pSource);
var correct = IntrinsicUtility.CorrectBoolean(load).AsUInt64();
@ -254,21 +259,9 @@ public static class SpanExtensions
return (short)(shift.GetElement(0) | (shift.GetElement(1) << 8));
}
if (AdvSimd.IsSupported)
{
// Hasn't been tested since March 6th 2023 (Reason: Unavailable hardware).
var load = AdvSimd.LoadVector128((byte*)pSource);
var correct = IntrinsicUtility.CorrectBoolean(load).AsUInt64();
var multiply = IntrinsicUtility.Multiply(IntegerPackingMagicV128, correct);
var shift = AdvSimd.ShiftRightLogical(multiply, 56);
return (short)(shift.GetElement(0) | (shift.GetElement(1) << 8));
}
else
{
goto default;
}
}
goto default;
#endif
default:
@ -324,9 +317,6 @@ public static class SpanExtensions
fixed (bool* pSource = source)
{
// TODO: .NET 8.0 Wasm support.
// TODO: Implement a replacement for UInt64 vector multiplication (there are no instruction for this built-in).
if (Avx2.IsSupported)
{
var load = Avx.LoadVector256((byte*)pSource);