/// <summary>
/// Calculate Cholesky step
/// </summary>
/// <param name="data">Factor matrix</param>
/// <param name="rowDim">Number of rows</param>
/// <param name="firstCol">Column start</param>
/// <param name="colLimit">Total columns</param>
/// <param name="multipliers">Multipliers calculated previously</param>
/// <param name="availableCores">Number of available processors</param>
static void DoCholeskyStep(Matrix <Complex32> data, int rowDim, int firstCol, int colLimit, Complex32[] multipliers, int availableCores)
{
var tmpColCount = colLimit - firstCol;
if ((availableCores > 1) && (tmpColCount > 200))
{
var tmpSplit = firstCol + (tmpColCount / 3);
var tmpCores = availableCores / 2;
CommonParallel.Invoke(
() => DoCholeskyStep(data, rowDim, firstCol, tmpSplit, multipliers, tmpCores),
() => DoCholeskyStep(data, rowDim, tmpSplit, colLimit, multipliers, tmpCores));
}
else
{
for (var j = firstCol; j < colLimit; j++)
{
var tmpVal = multipliers[j];
for (var i = j; i < rowDim; i++)
{
data.At(i, j, data.At(i, j) - (multipliers[i] * tmpVal.Conjugate()));
}
}
}
}
/// <summary>
/// Perform calculation of Q or R
/// </summary>
/// <param name="u">Work array</param>
/// <param name="a">Q or R matrices</param>
/// <param name="rowStart">The first row</param>
/// <param name="rowDim">The last row</param>
/// <param name="columnStart">The first column</param>
/// <param name="columnDim">The last column</param>
/// <param name="availableCores">Number of available CPUs</param>
private static void ComputeQR(double[] u, Matrix <double> a, int rowStart, int rowDim, int columnStart, int columnDim, int availableCores)
{
if (rowDim < rowStart || columnDim < columnStart)
{
return;
}
var tmpColCount = columnDim - columnStart;
if ((availableCores > 1) && (tmpColCount > 200))
{
var tmpSplit = columnStart + (tmpColCount / 2);
var tmpCores = availableCores / 2;
CommonParallel.Invoke(
() => ComputeQR(u, a, rowStart, rowDim, columnStart, tmpSplit, tmpCores),
() => ComputeQR(u, a, rowStart, rowDim, tmpSplit, columnDim, tmpCores));
}
else
{
for (var j = columnStart; j < columnDim; j++)
{
var scale = 0.0;
for (var i = rowStart; i < rowDim; i++)
{
scale += u[i - rowStart] * a.At(i, j);
}
for (var i = rowStart; i < rowDim; i++)
{
a.At(i, j, a.At(i, j) - (u[i - rowStart] * scale));
}
}
}
}
/// <summary>
/// Calculate Cholesky step
/// </summary>
/// <param name="data">Factor matrix</param>
/// <param name="rowDim">Number of rows</param>
/// <param name="firstCol">Column start</param>
/// <param name="colLimit">Total columns</param>
/// <param name="multipliers">Multipliers calculated previously</param>
/// <param name="availableCores">Number of available processors</param>
private static void DoCholeskyStep(Matrix <double> data, int rowDim, int firstCol, int colLimit,
double[] multipliers, int availableCores)
{
var tmpColCount = colLimit - firstCol;
if (availableCores > 1 && tmpColCount > 200)
{
var tmpSplit = firstCol + tmpColCount / 3;
var tmpCores = availableCores / 2;
CommonParallel.Invoke(
() => DoCholeskyStep(data, rowDim, firstCol, tmpSplit, multipliers, tmpCores),
() => DoCholeskyStep(data, rowDim, tmpSplit, colLimit, multipliers, tmpCores));
}
else
{
for (var j = firstCol; j < colLimit; j++)
{
var tmpVal = multipliers[j];
for (var i = j; i < rowDim; i++)
{
data.At(i, j, data.At(i, j) - multipliers[i] * tmpVal);
}
}
}
}
/// <summary>
/// Calculate Cholesky step
/// </summary>
/// <param name="data">Factor matrix</param>
/// <param name="rowDim">Number of rows</param>
/// <param name="firstCol">Column start</param>
/// <param name="colLimit">Total columns</param>
/// <param name="multipliers">Multipliers calculated previously</param>
/// <param name="availableCores">Number of available processors</param>
private static void DoCholeskyStep(double[] data, int rowDim, int firstCol, int colLimit, double[] multipliers,
int availableCores)
{
var tmpColCount = colLimit - firstCol;
if (availableCores > 1 && tmpColCount > Control.ParallelizeElements)
{
var tmpSplit = firstCol + tmpColCount / 3;
var tmpCores = availableCores / 2;
CommonParallel.Invoke(
() => DoCholeskyStep(data, rowDim, firstCol, tmpSplit, multipliers, tmpCores),
() => DoCholeskyStep(data, rowDim, tmpSplit, colLimit, multipliers, tmpCores));
}
else
{
for (var j = firstCol; j < colLimit; j++)
{
var tmpVal = multipliers[j];
for (var i = j; i < rowDim; i++)
{
data[j * rowDim + i] -= multipliers[i] * tmpVal;
}
}
}
}
/// <summary>
/// Perform calculation of Q or R
/// </summary>
/// <param name="u">Work array</param>
/// <param name="a">Q or R matrices</param>
/// <param name="rowStart">The first row</param>
/// <param name="rowDim">The last row</param>
/// <param name="columnStart">The first column</param>
/// <param name="columnDim">The last column</param>
/// <param name="availableCores">Number of available CPUs</param>
static void ComputeQR(Complex32[] u, Matrix <Complex32> a, int rowStart, int rowDim, int columnStart, int columnDim, int availableCores)
{
if ((rowDim < rowStart) || (columnDim < columnStart))
{
return;
}
var tmpColCount = columnDim - columnStart;
if ((availableCores > 1) && (tmpColCount > 200))
{
var tmpSplit = columnStart + (tmpColCount / 2);
var tmpCores = availableCores / 2;
CommonParallel.Invoke(
() => ComputeQR(u, a, rowStart, rowDim, columnStart, tmpSplit, tmpCores),
() => ComputeQR(u, a, rowStart, rowDim, tmpSplit, columnDim, tmpCores));
}
else
{
for (var j = columnStart; j < columnDim; j++)
{
var scale = Complex32.Zero;
for (var i = rowStart; i < rowDim; i++)
{
scale += u[i - rowStart] * a.At(i, j);
}
for (var i = rowStart; i < rowDim; i++)
{
a.At(i, j, a.At(i, j) - (u[i - rowStart].Conjugate() * scale));
}
}
}
}
static void CacheObliviousMatrixMultiply(double[] matrixA, int shiftArow, int shiftAcol, double[] matrixB, int shiftBrow, int shiftBcol, double[] result, int shiftCrow, int shiftCcol, int m, int n, int k, int constM, int constN, int constK, int level)
{
if (m + n <= Control.ParallelizeOrder)
{
for (var m1 = 0; m1 < m; m1++)
{
var matArowPos = m1 + shiftArow;
var matCrowPos = m1 + shiftCrow;
for (var n1 = 0; n1 < n; ++n1)
{
var boffset = ((n1 + shiftBcol) * constK) + shiftBrow;
double sum = 0;
for (var k1 = 0; k1 < k; ++k1)
{
sum += matrixA[((k1 + shiftAcol) * constM) + matArowPos] * matrixB[boffset + k1];
}
result[((n1 + shiftCcol) * constM) + matCrowPos] += sum;
}
}
return;
}
// divide and conquer
int m2 = m / 2, n2 = n / 2, k2 = k / 2;
level++;
if (level <= 2)
{
CommonParallel.Invoke(
() => CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol, matrixB, shiftBrow, shiftBcol, result, shiftCrow, shiftCcol, m2, n2, k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol, matrixB, shiftBrow, shiftBcol + n2, result, shiftCrow, shiftCcol + n2, m2, n - n2, k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol, matrixB, shiftBrow, shiftBcol, result, shiftCrow + m2, shiftCcol, m - m2, n2, k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol, matrixB, shiftBrow, shiftBcol + n2, result, shiftCrow + m2, shiftCcol + n2, m - m2, n - n2, k2, constM, constN, constK, level));
CommonParallel.Invoke(
() => CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol, result, shiftCrow, shiftCcol, m2, n2, k - k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol + n2, result, shiftCrow, shiftCcol + n2, m2, n - n2, k - k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol, result, shiftCrow + m2, shiftCcol, m - m2, n2, k - k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol + n2, result, shiftCrow + m2, shiftCcol + n2, m - m2, n - n2, k - k2, constM, constN, constK, level));
}
else
{
CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol, matrixB, shiftBrow, shiftBcol, result, shiftCrow, shiftCcol, m2, n2, k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol, matrixB, shiftBrow, shiftBcol + n2, result, shiftCrow, shiftCcol + n2, m2, n - n2, k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol, result, shiftCrow, shiftCcol, m2, n2, k - k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol + n2, result, shiftCrow, shiftCcol + n2, m2, n - n2, k - k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol, matrixB, shiftBrow, shiftBcol, result, shiftCrow + m2, shiftCcol, m - m2, n2, k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol, matrixB, shiftBrow, shiftBcol + n2, result, shiftCrow + m2, shiftCcol + n2, m - m2, n - n2, k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol, result, shiftCrow + m2, shiftCcol, m - m2, n2, k - k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol + n2, result, shiftCrow + m2, shiftCcol + n2, m - m2, n - n2, k - k2, constM, constN, constK, level);
}
}
/// <summary>
/// Convolution with the bluestein sequence (Parallel Version).
/// </summary>
/// <param name="samples">Sample Vector.</param>
private static void BluesteinConvolutionParallel(Complex[] samples)
{
int n = samples.Length;
Complex[] sequence = BluesteinSequence(n);
// Padding to power of two >= 2N–1 so we can apply Radix-2 FFT.
int m = ((n << 1) - 1).CeilingToPowerOfTwo();
Complex[] b = new Complex[m];
Complex[] a = new Complex[m];
CommonParallel.Invoke(
() =>
{
// Build and transform padded sequence b_k = exp(I*Pi*k^2/N)
for (int i = 0; i < n; i++)
{
b[i] = sequence[i];
}
for (int i = m - n + 1; i < b.Length; i++)
{
b[i] = sequence[m - i];
}
Radix2(b, -1);
},
() =>
{
// Build and transform padded sequence a_k = x_k * exp(-I*Pi*k^2/N)
for (int i = 0; i < samples.Length; i++)
{
a[i] = sequence[i].Conjugate() * samples[i];
}
Radix2(a, -1);
});
for (int i = 0; i < a.Length; i++)
{
a[i] *= b[i];
}
Radix2Parallel(a, 1);
var nbinv = 1.0 / m;
for (int i = 0; i < samples.Length; i++)
{
samples[i] = nbinv * sequence[i].Conjugate() * a[i];
}
}
/// <summary>
/// Perform calculation of Q or R
/// </summary>
/// <param name="work">Work array</param>
/// <param name="workIndex">Index of column in work array</param>
/// <param name="a">Q or R matrices</param>
/// <param name="rowStart">The first row in </param>
/// <param name="rowCount">The last row</param>
/// <param name="columnStart">The first column</param>
/// <param name="columnCount">The last column</param>
/// <param name="availableCores">Number of available CPUs</param>
private static void ComputeQR(double[] work, int workIndex, double[] a, int rowStart, int rowCount,
int columnStart, int columnCount, int availableCores)
{
if (rowStart > rowCount || columnStart > columnCount)
{
return;
}
var tmpColCount = columnCount - columnStart;
if (availableCores > 1 && tmpColCount > 200)
{
var tmpSplit = columnStart + tmpColCount / 2;
var tmpCores = availableCores / 2;
CommonParallel.Invoke(
() => ComputeQR(work, workIndex, a, rowStart, rowCount, columnStart, tmpSplit, tmpCores),
() => ComputeQR(work, workIndex, a, rowStart, rowCount, tmpSplit, columnCount, tmpCores));
}
else
{
for (var j = columnStart; j < columnCount; j++)
{
var scale = 0.0;
for (var i = rowStart; i < rowCount; i++)
{
scale += work[workIndex * rowCount + i - rowStart] * a[j * rowCount + i];
}
for (var i = rowStart; i < rowCount; i++)
{
a[j * rowCount + i] -= work[workIndex * rowCount + i - rowStart] * scale;
}
}
}
}
static void CacheObliviousMatrixMultiply(double[] matrixA, int shiftArow, int shiftAcol, double[] matrixB, int shiftBrow, int shiftBcol, double[] result, int shiftCrow, int shiftCcol, int m, int n, int k, int constM, int constN, int constK, int level)
{
if (m + n <= Control.ParallelizeOrder)
{
fixed(double *resultPtr = &result[0])
fixed(double *aPtr = &matrixA[0])
fixed(double *bPtr = &matrixB[0])
{
double *a = aPtr + shiftArow;
double *c = resultPtr + shiftCrow;
for (var m1 = 0; m1 < m; m1++)
{
for (var n1 = 0; n1 < n; ++n1)
{
double *b = bPtr + (n1 + shiftBcol) * constK + shiftBrow;
double sum = 0;
for (var k1 = 0; k1 < k; ++k1)
{
sum += a[((k1 + shiftAcol) * constM)] * b[k1];
}
c[((n1 + shiftCcol) * constM)] += sum;
}
a++;
c++;
}
}
return;
}
// divide and conquer
int m2 = m / 2, n2 = n / 2, k2 = k / 2;
level++;
if (level <= 2)
{
CommonParallel.Invoke(
() => CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol, matrixB, shiftBrow, shiftBcol, result, shiftCrow, shiftCcol, m2, n2, k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol, matrixB, shiftBrow, shiftBcol + n2, result, shiftCrow, shiftCcol + n2, m2, n - n2, k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol, matrixB, shiftBrow, shiftBcol, result, shiftCrow + m2, shiftCcol, m - m2, n2, k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol, matrixB, shiftBrow, shiftBcol + n2, result, shiftCrow + m2, shiftCcol + n2, m - m2, n - n2, k2, constM, constN, constK, level));
CommonParallel.Invoke(
() => CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol, result, shiftCrow, shiftCcol, m2, n2, k - k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol + n2, result, shiftCrow, shiftCcol + n2, m2, n - n2, k - k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol, result, shiftCrow + m2, shiftCcol, m - m2, n2, k - k2, constM, constN, constK, level),
() => CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol + n2, result, shiftCrow + m2, shiftCcol + n2, m - m2, n - n2, k - k2, constM, constN, constK, level));
}
else
{
CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol, matrixB, shiftBrow, shiftBcol, result, shiftCrow, shiftCcol, m2, n2, k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol, matrixB, shiftBrow, shiftBcol + n2, result, shiftCrow, shiftCcol + n2, m2, n - n2, k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol, result, shiftCrow, shiftCcol, m2, n2, k - k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol + n2, result, shiftCrow, shiftCcol + n2, m2, n - n2, k - k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol, matrixB, shiftBrow, shiftBcol, result, shiftCrow + m2, shiftCcol, m - m2, n2, k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol, matrixB, shiftBrow, shiftBcol + n2, result, shiftCrow + m2, shiftCcol + n2, m - m2, n - n2, k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol, result, shiftCrow + m2, shiftCcol, m - m2, n2, k - k2, constM, constN, constK, level);
CacheObliviousMatrixMultiply(matrixA, shiftArow + m2, shiftAcol + k2, matrixB, shiftBrow + k2, shiftBcol + n2, result, shiftCrow + m2, shiftCcol + n2, m - m2, n - n2, k - k2, constM, constN, constK, level);
}
}
