public void RunFldScenario()
{
var result = Avx.ConvertToVector128Int32WithTruncation(_fld);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_fld, _dataTable.outArrayPtr);
}
public void RunLclFldScenario()
{
var test = new SimpleUnaryOpTest__ConvertToVector128Int32WithTruncationDouble();
var result = Avx.ConvertToVector128Int32WithTruncation(test._fld);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(test._fld, _dataTable.outArrayPtr);
}
public void RunLclVarScenario_UnsafeRead()
{
var firstOp = Unsafe.Read <Vector256 <Double> >(_dataTable.inArrayPtr);
var result = Avx.ConvertToVector128Int32WithTruncation(firstOp);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(firstOp, _dataTable.outArrayPtr);
}
public void RunLclVarScenario_LoadAligned()
{
var firstOp = Avx.LoadAlignedVector256((Double *)(_dataTable.inArrayPtr));
var result = Avx.ConvertToVector128Int32WithTruncation(firstOp);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(firstOp, _dataTable.outArrayPtr);
}
// Used by trig. Low quality because it narrows to Int32.
internal static Vector256 <long> ConvertToInt64(Vector256 <double> vector)
{
if (Avx2.IsSupported)
{
return(Avx2.ConvertToVector256Int64(Avx.ConvertToVector128Int32WithTruncation(vector)));
}
return(SoftwareFallback(vector));
public void RunBasicScenario_Load()
{
var result = Avx.ConvertToVector128Int32WithTruncation(
Avx.LoadVector256((Double *)(_dataTable.inArrayPtr))
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_dataTable.inArrayPtr, _dataTable.outArrayPtr);
}
public void RunBasicScenario_UnsafeRead()
{
var result = Avx.ConvertToVector128Int32WithTruncation(
Unsafe.Read <Vector256 <Double> >(_dataTable.inArrayPtr)
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_dataTable.inArrayPtr, _dataTable.outArrayPtr);
}
public void ResizeSIMD2(FastBitmap rtnImage)
{
float scaleX = (float)this.width / rtnImage.width;
float scaleY = (float)this.height / rtnImage.height;
if (rtnImage.width % 4 == 0)
{
Parallel.For(0, rtnImage.height, (y) =>
{
var basePos = (uint *)(rtnImage._ptr + (rtnImage._stride * y));
var rtnPos = (uint *)(this._ptr + (this._stride * (int)(y * scaleY)));
Vector128 <float> indexf = Vector128.Create(0.0f, 1, 2, 3);
Vector128 <float> iterf = Vector128.Create(4f, 4, 4, 4);
Vector128 <float> scalef = Vector128.Create(scaleX, scaleX, scaleX, scaleX);
for (int x = 0; x < rtnImage.width; x += 4)
{
Vector128 <int> index = Avx.ConvertToVector128Int32WithTruncation(Avx.Multiply(indexf, scalef));
Avx.Store(basePos, Avx2.GatherVector128(rtnPos, index, 4));
indexf = Avx.Add(indexf, iterf);
basePos += 4;
}
});
}
else
{
Parallel.For(0, height, (y) =>
{
var basePos = (uint *)(rtnImage._ptr + (rtnImage._stride * y));
var rtnPos = (uint *)(this._ptr + (this._stride * (int)(y * scaleY)));
for (int x = 0; x < width; x += 4)
{
*(uint *)(basePos + x) = *(uint *)(rtnPos + ((int)(x * scaleX)));
}
});
}
}
private static unsafe double[] BilinearInterpol_AVX(
double[] x,
double[] A,
double minXA,
double maxXA,
double[] B,
double minXB,
double maxXB,
double weightB)
{
double[] z = new double[outputVectorSize];
fixed(double *pX = &x[0], pA = &A[0], pB = &B[0], pZ = &z[0])
{
Vector256 <double> vWeightB = Vector256.Create(weightB);
Vector256 <double> vWeightA = Vector256.Create(1 - weightB);
Vector256 <double> vMinXA = Vector256.Create(minXA);
Vector256 <double> vMaxXA = Vector256.Create(maxXA);
Vector256 <double> vMinXB = Vector256.Create(minXB);
Vector256 <double> vMaxXB = Vector256.Create(maxXB);
double deltaA = (maxXA - minXA) / (double)(A.Length - 1);
double deltaB = (maxXB - minXB) / (double)(B.Length - 1);
Vector256 <double> vDeltaA = Vector256.Create(deltaA);
Vector256 <double> vDeltaB = Vector256.Create(deltaB);
double invDeltaA = 1.0 / deltaA;
double invDeltaB = 1.0 / deltaB;
Vector256 <double> vInvDeltaA = Vector256.Create(invDeltaA);
Vector256 <double> vInvDeltaB = Vector256.Create(invDeltaB);
Vector128 <int> ALengthMinusOne = Vector128.Create(A.Length - 1);
Vector128 <int> BLengthMinusOne = Vector128.Create(B.Length - 1);
Vector128 <int> One = Vector128.Create(1);
for (var i = 0; i < x.Length; i += Vector256 <double> .Count)
{
Vector256 <double> currentX = Avx.LoadVector256(pX + i);
// Determine the largest a, such that A[i] = f(xA) and xA <= x[i].
// This involves casting from double to int; here we use a Vector conversion.
Vector256 <double> aDouble = Avx.Multiply(Avx.Subtract(currentX, vMinXA), vInvDeltaA);
Vector128 <int> a = Avx.ConvertToVector128Int32WithTruncation(aDouble);
a = Sse41.Min(Sse41.Max(a, Vector128 <int> .Zero), ALengthMinusOne);
Vector128 <int> aPlusOne = Sse41.Min(Sse2.Add(a, One), ALengthMinusOne);
// Now, get the reference input, xA, for our index a.
// This involves casting from int to double.
Vector256 <double> xA = Avx.Add(Avx.Multiply(Avx.ConvertToVector256Double(a), vDeltaA), vMinXA);
// Now, compute the lambda for our A reference point.
Vector256 <double> currentXNormA = Avx.Max(vMinXA, Avx.Min(currentX, vMaxXA));
Vector256 <double> lambdaA = Avx.Multiply(Avx.Subtract(currentXNormA, xA), vInvDeltaA);
// Now, we need to load up our reference points using Vector Gather operations.
Vector256 <double> AVector = Avx2.GatherVector256(pA, a, 8);
Vector256 <double> AVectorPlusOne = Avx2.GatherVector256(pA, aPlusOne, 8);
// Now, do the all of the above for our B reference point.
Vector256 <double> bDouble = Avx.Multiply(Avx.Subtract(currentX, vMinXB), vInvDeltaB);
Vector128 <int> b = Avx.ConvertToVector128Int32WithTruncation(bDouble);
b = Sse41.Min(Sse41.Max(b, Vector128 <int> .Zero), BLengthMinusOne);
Vector128 <int> bPlusOne = Sse41.Min(Sse2.Add(b, One), BLengthMinusOne);
Vector256 <double> xB = Avx.Add(Avx.Multiply(Avx.ConvertToVector256Double(b), vDeltaB), vMinXB);
Vector256 <double> currentXNormB = Avx.Max(vMinXB, Avx.Min(currentX, vMaxXB));
Vector256 <double> lambdaB = Avx.Multiply(Avx.Subtract(currentXNormB, xB), vInvDeltaB);
Vector256 <double> BVector = Avx2.GatherVector256(pB, b, 8);
Vector256 <double> BVectorPlusOne = Avx2.GatherVector256(pB, bPlusOne, 8);
Vector256 <double> newZ = Avx.Add(Avx.Multiply(vWeightA, Avx.Add(AVector, Avx.Multiply(lambdaA, Avx.Subtract(AVectorPlusOne, AVector)))),
Avx.Multiply(vWeightB, Avx.Add(BVector, Avx.Multiply(lambdaB, Avx.Subtract(BVectorPlusOne, BVector)))));
Avx.Store(pZ + i, newZ);
}
}
return(z);
}