public void RunReflectionScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunReflectionScenario_LoadAligned));
typeof(Avx).GetMethod(nameof(Avx.MaskStore), new Type[] { typeof(Double *), typeof(Vector256 <Double>), typeof(Vector256 <Double>) })
.Invoke(null, new object[] {
Pointer.Box(_dataTable.outArrayPtr, typeof(Double *)),
Avx.LoadAlignedVector256((Double *)(_dataTable.inArray1Ptr)),
Avx.LoadAlignedVector256((Double *)(_dataTable.inArray2Ptr))
});
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.outArrayPtr);
}
public void RunBasicScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunBasicScenario_LoadAligned));
var result = Avx.Permute2x128(
Avx.LoadAlignedVector256((Double *)(_dataTable.inArray1Ptr)),
Avx.LoadAlignedVector256((Double *)(_dataTable.inArray2Ptr)),
2
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.outArrayPtr);
}
public void RunReflectionScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunReflectionScenario_LoadAligned));
var result = typeof(Avx2).GetMethod(nameof(Avx2.ShiftLeftLogical), new Type[] { typeof(Vector256 <Int64>), typeof(byte) })
.Invoke(null, new object[] {
Avx.LoadAlignedVector256((Int64 *)(_dataTable.inArrayPtr)),
(byte)64
});
Unsafe.Write(_dataTable.outArrayPtr, (Vector256 <Int64>)(result));
ValidateResult(_dataTable.inArrayPtr, _dataTable.outArrayPtr);
}
public void RunReflectionScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunReflectionScenario_LoadAligned));
var result = typeof(Avx2).GetMethod(nameof(Avx2.Subtract), new Type[] { typeof(Vector256 <UInt32>), typeof(Vector256 <UInt32>) })
.Invoke(null, new object[] {
Avx.LoadAlignedVector256((UInt32 *)(_dataTable.inArray1Ptr)),
Avx.LoadAlignedVector256((UInt32 *)(_dataTable.inArray2Ptr))
});
Unsafe.Write(_dataTable.outArrayPtr, (Vector256 <UInt32>)(result));
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.outArrayPtr);
}
public void RunReflectionScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunReflectionScenario_LoadAligned));
var result = typeof(Avx).GetMethod(nameof(Avx.CompareGreaterThanOrEqual), new Type[] { typeof(Vector256 <Single>), typeof(Vector256 <Single>) })
.Invoke(null, new object[] {
Avx.LoadAlignedVector256((Single *)(_dataTable.inArray1Ptr)),
Avx.LoadAlignedVector256((Single *)(_dataTable.inArray2Ptr))
});
Unsafe.Write(_dataTable.outArrayPtr, (Vector256 <Single>)(result));
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.outArrayPtr);
}
public void RunReflectionScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunReflectionScenario_LoadAligned));
typeof(Avx).GetMethod(nameof(Avx.ExtractVector128), new Type[] { typeof(Int32 *), typeof(Vector256 <Int32>), typeof(byte) })
.Invoke(null, new object[] {
Pointer.Box(_dataTable.outArrayPtr, typeof(Int32 *)),
Avx.LoadAlignedVector256((Int32 *)(_dataTable.inArrayPtr)),
(byte)1
});
ValidateResult(_dataTable.inArrayPtr, _dataTable.outArrayPtr);
}
public void RunBasicScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunBasicScenario_LoadAligned));
var result = Fma.MultiplySubtractNegated(
Avx.LoadAlignedVector256((Double *)(_dataTable.inArray1Ptr)),
Avx.LoadAlignedVector256((Double *)(_dataTable.inArray2Ptr)),
Avx.LoadAlignedVector256((Double *)(_dataTable.inArray3Ptr))
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.inArray3Ptr, _dataTable.outArrayPtr);
}
public void RunReflectionScenario_LoadAligned()
{
var result = typeof(Avx).GetMethods(BindingFlags.Public | BindingFlags.Static)
.FirstOrDefault(mi => mi.Name == nameof(Avx.ExtractVector128) && mi.IsGenericMethod)
.MakeGenericMethod(new[] { typeof(Single) })
.Invoke(null, new object[] {
Avx.LoadAlignedVector256((Single *)(_dataTable.inArrayPtr)),
(byte)1
});
Unsafe.Write(_dataTable.outArrayPtr, (Vector128 <Single>)(result));
ValidateResult(_dataTable.inArrayPtr, _dataTable.outArrayPtr);
}
public void RunReflectionScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunReflectionScenario_LoadAligned));
var result = typeof(Avx).GetMethod(nameof(Avx.GetLowerHalf))
.MakeGenericMethod(new Type[] { typeof(UInt16) })
.Invoke(null, new object[] {
Avx.LoadAlignedVector256((UInt16 *)(_dataTable.inArrayPtr))
});
Unsafe.Write(_dataTable.outArrayPtr, (Vector128 <UInt16>)(result));
ValidateResult(_dataTable.inArrayPtr, _dataTable.outArrayPtr);
}
public void RunBasicScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunBasicScenario_LoadAligned));
var result = Avx2.BlendVariable(
Avx.LoadAlignedVector256((Int32 *)(_dataTable.inArray1Ptr)),
Avx.LoadAlignedVector256((Int32 *)(_dataTable.inArray2Ptr)),
Avx.LoadAlignedVector256((Int32 *)(_dataTable.inArray3Ptr))
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.inArray3Ptr, _dataTable.outArrayPtr);
}
public void RunBasicScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunBasicScenario_LoadAligned));
var result = Avx.Insert(
Avx.LoadAlignedVector256((Byte *)(_dataTable.inArrayPtr)),
(byte)2,
1
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_dataTable.inArrayPtr, _dataTable.outArrayPtr);
}
public void RunBasicScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunBasicScenario_LoadAligned));
var result = AvxVnni.MultiplyWideningAndAdd(
Avx.LoadAlignedVector256((Int32 *)(_dataTable.inArray0Ptr)),
Avx.LoadAlignedVector256((Int16 *)(_dataTable.inArray1Ptr)),
Avx.LoadAlignedVector256((Int16 *)(_dataTable.inArray2Ptr))
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_dataTable.inArray0Ptr, _dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.outArrayPtr);
}
public void RunReflectionScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunReflectionScenario_LoadAligned));
var result = typeof(AvxVnni).GetMethod(nameof(AvxVnni.MultiplyWideningAndAddSaturate), new Type[] { typeof(Vector256 <Int32>), typeof(Vector256 <Int16>), typeof(Vector256 <Int16>) })
.Invoke(null, new object[] {
Avx.LoadAlignedVector256((Int32 *)(_dataTable.inArray0Ptr)),
Avx.LoadAlignedVector256((Int16 *)(_dataTable.inArray1Ptr)),
Avx.LoadAlignedVector256((Int16 *)(_dataTable.inArray2Ptr))
});
Unsafe.Write(_dataTable.outArrayPtr, (Vector256 <Int32>)(result));
ValidateResult(_dataTable.inArray0Ptr, _dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.outArrayPtr);
}
public void RunReflectionScenario_LoadAligned()
{
var method = typeof(Avx).GetMethod(nameof(Avx.TestZ), new Type[] { typeof(Vector256 <UInt16>), typeof(Vector256 <UInt16>) });
if (method != null)
{
var result = method.Invoke(null, new object[] {
Avx.LoadAlignedVector256((UInt16 *)(_dataTable.inArray1Ptr)),
Avx.LoadAlignedVector256((UInt16 *)(_dataTable.inArray2Ptr))
});
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, (bool)(result));
}
}
public void RunReflectionScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunReflectionScenario_LoadAligned));
var result = typeof(Avx).GetMethod(nameof(Avx.Insert), new Type[] { typeof(Vector256 <UInt16>), typeof(UInt16), typeof(byte) })
.Invoke(null, new object[] {
Avx.LoadAlignedVector256((UInt16 *)(_dataTable.inArrayPtr)),
(ushort)2,
(byte)27
});
Unsafe.Write(_dataTable.outArrayPtr, (Vector256 <UInt16>)(result));
ValidateResult(_dataTable.inArrayPtr, _dataTable.outArrayPtr);
}
public void RunReflectionScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunReflectionScenario_LoadAligned));
var result = typeof(Avx2).GetMethod(nameof(Avx2.InsertVector128), new Type[] { typeof(Vector256 <Byte>), typeof(Vector128 <Byte>), typeof(byte) })
.Invoke(null, new object[] {
Avx.LoadAlignedVector256((Byte *)(_dataTable.inArray1Ptr)),
LoadAlignedVector128((Byte *)(_dataTable.inArray2Ptr)),
(byte)1
});
Unsafe.Write(_dataTable.outArrayPtr, (Vector256 <Byte>)(result));
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.outArrayPtr);
}
public void RunReflectionScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunReflectionScenario_LoadAligned));
var method = typeof(Avx).GetMethod(nameof(Avx.TestNotZAndNotC), new Type[] { typeof(Vector256 <SByte>), typeof(Vector256 <SByte>) });
if (method != null)
{
var result = method.Invoke(null, new object[] {
Avx.LoadAlignedVector256((SByte *)(_dataTable.inArray1Ptr)),
Avx.LoadAlignedVector256((SByte *)(_dataTable.inArray2Ptr))
});
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, (bool)(result));
}
}
public unsafe byte Run()
{
var buffer = _buffer;
byte value = 0;
// begin
var i = 0;
if (Avx2.IsSupported) // assume is true
{
const int vectorSize = 256 / 8;
var tmp = Avx.SetZeroVector256 <byte>();
fixed(byte *pBuf = &buffer[0])
{
var n = buffer.Length / vectorSize * vectorSize;
for (; i < n; i += vectorSize)
{
tmp = Avx2.Xor(tmp, Avx.LoadAlignedVector256(pBuf + i));
}
}
var tmpResult = stackalloc byte[vectorSize];
Avx.Store(tmpResult, tmp);
for (var j = 0; j < vectorSize; ++j)
{
value ^= tmpResult[j];
}
}
for (; i < buffer.Length; i++)
{
value ^= buffer[i];
}
// end
return(value);
}
public void RunLclVarScenario_LoadAligned()
{
var firstOp = Avx.LoadAlignedVector256((Byte *)(_dataTable.inArrayPtr));
Avx2.ExtractVector128((Byte *)_dataTable.outArrayPtr, firstOp, 1);
}
public float SumUsingAvxAlignedPipelined()
{
var values = this.values;
float sum = 0;
if (Avx.IsSupported)
{
unsafe
{
fixed(float *valuesPtr = values)
{
// Determine how many elements we need to sum sequential to reach 256 bit alignment
const int ElementsPerByte = sizeof(float) / sizeof(byte);
var alignmentOffset = (long)(uint)(-(int)valuesPtr / ElementsPerByte) & (Vector256 <float> .Count - 1);
// handle first values sequentially until we hit the 256bit alignment boundary
for (long i = 0; i < alignmentOffset; i++)
{
sum += *(valuesPtr + i);
}
var remainingLength = values.Length - alignmentOffset;
var vectorizableLength = values.Length - remainingLength % (long)Vector256 <float> .Count;
// handle batches of 4 vectors for pipelining benefits
var pipelineVectorizableLength = values.Length - remainingLength % (4 * (long)Vector256 <float> .Count);
var tempVector1 = Vector256 <float> .Zero;
var tempVector2 = Vector256 <float> .Zero;
var tempVector3 = Vector256 <float> .Zero;
var tempVector4 = Vector256 <float> .Zero;
for (long i = alignmentOffset; i < pipelineVectorizableLength; i += 4 * (long)Vector256 <float> .Count)
{
var valuesVector1 = Avx.LoadAlignedVector256(valuesPtr + i);
tempVector1 = Avx.Add(tempVector1, valuesVector1);
var valuesVector2 = Avx.LoadAlignedVector256(valuesPtr + i + Vector256 <float> .Count);
tempVector2 = Avx.Add(tempVector2, valuesVector2);
var valuesVector3 = Avx.LoadAlignedVector256(valuesPtr + i + 2 * Vector256 <float> .Count);
tempVector3 = Avx.Add(tempVector3, valuesVector3);
var valuesVector4 = Avx.LoadAlignedVector256(valuesPtr + i + 3 * Vector256 <float> .Count);
tempVector4 = Avx.Add(tempVector4, valuesVector4);
}
var tempVector = Avx.Add(Avx.Add(Avx.Add(tempVector1, tempVector2), tempVector3), tempVector4);
// handle remaining vectors
for (long i = pipelineVectorizableLength; i < vectorizableLength; i += Vector256 <float> .Count)
{
var valuesVector = Avx.LoadAlignedVector256(valuesPtr + i);
tempVector = Avx.Add(tempVector, valuesVector);
}
// sum the temporary vector
for (int iVector = 0; iVector < Vector256 <float> .Count; iVector++)
{
sum += tempVector.GetElement(iVector);
}
// handle remaining items
for (int i = (int)vectorizableLength; i < values.Length; i++)
{
sum += values[i];
}
}
}
}
else
{
// non-AVX capable machines
for (int i = 0; i < values.Length; i++)
{
sum += values[i];
}
}
return(sum);
}
