public static bool Equal(Vector128 <float> vector1, Vector128 <float> vector2)
{
// This implementation is based on the DirectX Math Library XMVector4Equal method
// https://github.com/microsoft/DirectXMath/blob/master/Inc/DirectXMathVector.inl
if (AdvSimd.Arm64.IsSupported)
{
Vector128 <uint> vResult = AdvSimd.CompareEqual(vector1, vector2).AsUInt32();
Vector64 <byte> vResult0 = vResult.GetLower().AsByte();
Vector64 <byte> vResult1 = vResult.GetUpper().AsByte();
Vector64 <byte> vTemp10 = AdvSimd.Arm64.ZipLow(vResult0, vResult1);
Vector64 <byte> vTemp11 = AdvSimd.Arm64.ZipHigh(vResult0, vResult1);
Vector64 <ushort> vTemp21 = AdvSimd.Arm64.ZipHigh(vTemp10.AsUInt16(), vTemp11.AsUInt16());
return(vTemp21.AsUInt32().GetElement(1) == 0xFFFFFFFF);
}
else if (Sse.IsSupported)
{
return(Sse.MoveMask(Sse.CompareNotEqual(vector1, vector2)) == 0);
}
else
{
// Redundant test so we won't prejit remainder of this method on platforms without AdvSimd.
throw new PlatformNotSupportedException();
}
}
private static void HorizontalUnfilter(Span <byte> prev, Span <byte> input, Span <byte> dst, int width)
{
#if SUPPORTS_RUNTIME_INTRINSICS
if (Sse2.IsSupported)
{
dst[0] = (byte)(input[0] + (prev.IsEmpty ? 0 : prev[0]));
if (width <= 1)
{
return;
}
nint i;
Vector128 <int> last = Vector128 <int> .Zero.WithElement(0, dst[0]);
ref byte srcRef = ref MemoryMarshal.GetReference(input);
for (i = 1; i + 8 <= width; i += 8)
{
var a0 = Vector128.Create(Unsafe.As <byte, long>(ref Unsafe.Add(ref srcRef, i)), 0);
Vector128 <byte> a1 = Sse2.Add(a0.AsByte(), last.AsByte());
Vector128 <byte> a2 = Sse2.ShiftLeftLogical128BitLane(a1, 1);
Vector128 <byte> a3 = Sse2.Add(a1, a2);
Vector128 <byte> a4 = Sse2.ShiftLeftLogical128BitLane(a3, 2);
Vector128 <byte> a5 = Sse2.Add(a3, a4);
Vector128 <byte> a6 = Sse2.ShiftLeftLogical128BitLane(a5, 4);
Vector128 <byte> a7 = Sse2.Add(a5, a6);
ref byte outputRef = ref Unsafe.Add(ref MemoryMarshal.GetReference(dst), i);
Unsafe.As <byte, Vector64 <byte> >(ref outputRef) = a7.GetLower();
last = Sse2.ShiftRightLogical(a7.AsInt64(), 56).AsInt32();
}
public void RunBasicScenario()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunBasicScenario));
Int64[] values = new Int64[ElementCount];
for (int i = 0; i < ElementCount; i++)
{
values[i] = TestLibrary.Generator.GetInt64();
}
Vector128 <Int64> value = Vector128.Create(values[0], values[1]);
Vector64 <Int64> lowerResult = value.GetLower();
Vector64 <Int64> upperResult = value.GetUpper();
ValidateGetResult(lowerResult, upperResult, values);
Vector128 <Int64> result = value.WithLower(upperResult);
result = result.WithUpper(lowerResult);
ValidateWithResult(result, values);
}
public void RunBasicScenario()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunBasicScenario));
Byte[] values = new Byte[ElementCount];
for (int i = 0; i < ElementCount; i++)
{
values[i] = TestLibrary.Generator.GetByte();
}
Vector128 <Byte> value = Vector128.Create(values[0], values[1], values[2], values[3], values[4], values[5], values[6], values[7], values[8], values[9], values[10], values[11], values[12], values[13], values[14], values[15]);
Vector64 <Byte> lowerResult = value.GetLower();
Vector64 <Byte> upperResult = value.GetUpper();
ValidateGetResult(lowerResult, upperResult, values);
Vector128 <Byte> result = value.WithLower(upperResult);
result = result.WithUpper(lowerResult);
ValidateWithResult(result, values);
}
internal static void Step(ref ushort preSum1, ref ushort preSum2, byte[] buf, uint len)
{
/*
* Split Adler-32 into component sums.
*/
uint s1 = preSum1;
uint s2 = preSum2;
int bufPos = 0;
/*
* Process the data in blocks.
*/
uint BLOCK_SIZE = 1 << 5;
uint blocks = len / BLOCK_SIZE;
len -= blocks * BLOCK_SIZE;
while (blocks != 0)
{
uint n = Adler32Context.NMAX / BLOCK_SIZE; /* The NMAX constraint. */
if (n > blocks)
{
n = blocks;
}
blocks -= n;
/*
* Process n blocks of data. At most NMAX data bytes can be
* processed before s2 must be reduced modulo ADLER_MODULE.
*/
Vector128 <uint> v_s2 = Vector128.Create(s1 * n, 0, 0, 0);
Vector128 <uint> v_s1 = Vector128.Create(0u, 0, 0, 0);
Vector128 <ushort> v_column_sum_1 = AdvSimd.DuplicateToVector128((ushort)0);
Vector128 <ushort> v_column_sum_2 = AdvSimd.DuplicateToVector128((ushort)0);
Vector128 <ushort> v_column_sum_3 = AdvSimd.DuplicateToVector128((ushort)0);
Vector128 <ushort> v_column_sum_4 = AdvSimd.DuplicateToVector128((ushort)0);
do
{
/*
* Load 32 input bytes.
*/
Vector128 <byte> bytes1 = Vector128.Create(buf[bufPos], buf[bufPos + 1], buf[bufPos + 2],
buf[bufPos + 3], buf[bufPos + 4], buf[bufPos + 5],
buf[bufPos + 6], buf[bufPos + 7], buf[bufPos + 8],
buf[bufPos + 9], buf[bufPos + 10], buf[bufPos + 11],
buf[bufPos + 12], buf[bufPos + 13], buf[bufPos + 14],
buf[bufPos + 15]);
bufPos += 16;
Vector128 <byte> bytes2 = Vector128.Create(buf[bufPos], buf[bufPos + 1], buf[bufPos + 2],
buf[bufPos + 3], buf[bufPos + 4], buf[bufPos + 5],
buf[bufPos + 6], buf[bufPos + 7], buf[bufPos + 8],
buf[bufPos + 9], buf[bufPos + 10], buf[bufPos + 11],
buf[bufPos + 12], buf[bufPos + 13], buf[bufPos + 14],
buf[bufPos + 15]);
bufPos += 16;
/*
* Add previous block byte sum to v_s2.
*/
v_s2 = AdvSimd.Add(v_s2, v_s1);
/*
* Horizontally add the bytes for s1.
*/
v_s1 =
AdvSimd.AddPairwiseWideningAndAdd(v_s1,
AdvSimd.
AddPairwiseWideningAndAdd(AdvSimd.AddPairwiseWidening(bytes1),
bytes2));
/*
* Vertically add the bytes for s2.
*/
v_column_sum_1 = AdvSimd.AddWideningLower(v_column_sum_1, bytes1.GetLower());
v_column_sum_2 = AdvSimd.AddWideningLower(v_column_sum_2, bytes1.GetUpper());
v_column_sum_3 = AdvSimd.AddWideningLower(v_column_sum_3, bytes2.GetLower());
v_column_sum_4 = AdvSimd.AddWideningLower(v_column_sum_4, bytes2.GetUpper());
} while(--n != 0);
v_s2 = AdvSimd.ShiftLeftLogical(v_s2, 5);
/*
* Multiply-add bytes by [ 32, 31, 30, ... ] for s2.
*/
v_s2 = AdvSimd.MultiplyWideningLowerAndAdd(v_s2, v_column_sum_1.GetLower(),
Vector64.Create((ushort)32, 31, 30, 29));
v_s2 = AdvSimd.MultiplyWideningLowerAndAdd(v_s2, v_column_sum_1.GetUpper(),
Vector64.Create((ushort)28, 27, 26, 25));
v_s2 = AdvSimd.MultiplyWideningLowerAndAdd(v_s2, v_column_sum_2.GetLower(),
Vector64.Create((ushort)24, 23, 22, 21));
v_s2 = AdvSimd.MultiplyWideningLowerAndAdd(v_s2, v_column_sum_2.GetUpper(),
Vector64.Create((ushort)20, 19, 18, 17));
v_s2 = AdvSimd.MultiplyWideningLowerAndAdd(v_s2, v_column_sum_3.GetLower(),
Vector64.Create((ushort)16, 15, 14, 13));
v_s2 = AdvSimd.MultiplyWideningLowerAndAdd(v_s2, v_column_sum_3.GetUpper(),
Vector64.Create((ushort)12, 11, 10, 9));
v_s2 = AdvSimd.MultiplyWideningLowerAndAdd(v_s2, v_column_sum_4.GetLower(),
Vector64.Create((ushort)8, 7, 6, 5));
v_s2 = AdvSimd.MultiplyWideningLowerAndAdd(v_s2, v_column_sum_4.GetUpper(),
Vector64.Create((ushort)4, 3, 2, 1));
/*
* Sum epi32 ints v_s1(s2) and accumulate in s1(s2).
*/
Vector64 <uint> sum1 = AdvSimd.AddPairwise(v_s1.GetLower(), v_s1.GetUpper());
Vector64 <uint> sum2 = AdvSimd.AddPairwise(v_s2.GetLower(), v_s2.GetUpper());
Vector64 <uint> s1s2 = AdvSimd.AddPairwise(sum1, sum2);
s1 += AdvSimd.Extract(s1s2, 0);
s2 += AdvSimd.Extract(s1s2, 1);
/*
* Reduce.
*/
s1 %= Adler32Context.ADLER_MODULE;
s2 %= Adler32Context.ADLER_MODULE;
}
/*
* Handle leftover data.
*/
if (len != 0)
{
if (len >= 16)
{
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
s2 += s1 += buf[bufPos++];
len -= 16;
}
while (len-- != 0)
{
s2 += s1 += buf[bufPos++];
}
if (s1 >= Adler32Context.ADLER_MODULE)
{
s1 -= Adler32Context.ADLER_MODULE;
}
s2 %= Adler32Context.ADLER_MODULE;
}
/*
* Return the recombined sums.
*/
preSum1 = (ushort)(s1 & 0xFFFF);
preSum2 = (ushort)(s2 & 0xFFFF);
}