public static Array <Real> Mul(this Array <Real> a, Array <Real> b, Array <Real> result = null)
{
return(Array_.ElementwiseOp(a, b, result,
(n, x, offsetx, incx, y, offsety, incy, z, offsetz, incz) =>
{
if (incx == 1 && incy == 1 && incz == 1)
{
Blas.vmul(n, x, offsetx, y, offsety, z, offsetz);
}
else if (incx == 0) // x[offsetx] is broadcast: y[:] * x[offsetx]
{
Blas.gemv(Order.RowMajor, Transpose.NoTrans, n, 1, 1, y, offsety, incy, x, offsetx, 1, 0, z, offsetz, incz);
}
//else if (incy == 0) // y[offsety] is broadcast: x[:] * y[offsety]
// Blas.gemv(Order.RowMajor, Transpose.NoTrans, n, 1, 1, x, offsetx, incx, y, offsety, 1, 0, z, offsetz, incz);
else // when everything else fails, fallback to slow version
{
for (int i = 0; i < n; i++)
{
z[offsetz] = x[offsetx] * y[offsety];
offsetx += incx;
offsety += incy;
offsetz += incz;
}
}
}));
}
private static void _EinsteinSum(int n, Einstein[] einstein,
Array <float> x, int offX,
Array <float> y, int offY,
Array <float> z, int offZ
)
{
if (n == einstein.Length)
{
z.Values[offZ] += x.Values[offX] * y.Values[offY];
return;
}
var e = einstein[n];
if (n == einstein.Length - 1)
{
switch (e.mode)
{
case EinsteinMode.ELEMENTWISE:
int axisX = (int)e.axisX, axisY = (int)e.axisY, axisZ = (int)e.axisZ;
if (x.Stride[axisX] == 1 && y.Stride[axisY] == 1 && z.Stride[axisZ] == 1)
{
Blas.vmul(x.Shape[axisX], x.Values, offX, y.Values, offY, z.Values, offZ);
return;
}
break;
case EinsteinMode.OUTERX:
var axis = (int)e.axisX;
Blas.axpy(x.Shape[axis], y.Values[offY], x.Values, offX, x.Stride[axis], z.Values, offZ, z.Stride[(int)e.axisZ]);
return;
case EinsteinMode.OUTERY:
axis = (int)e.axisY;
Blas.axpy(y.Shape[axis], x.Values[offX], y.Values, offY, y.Stride[axis], z.Values, offZ, z.Stride[(int)e.axisZ]);
return;
}
}
switch (e.mode)
{
case EinsteinMode.INNER:
for (int i = 0; i < x.Shape[(int)e.axisX]; ++i)
{
_EinsteinSum(n + 1, einstein, x, offX, y, offY, z, offZ);
offX += x.Stride[(int)e.axisX];
offY += y.Stride[(int)e.axisY];
}
return;
case EinsteinMode.ELEMENTWISE:
for (int i = 0; i < x.Shape[(int)e.axisX]; ++i)
{
_EinsteinSum(n + 1, einstein, x, offX, y, offY, z, offZ);
offX += x.Stride[(int)e.axisX];
offY += y.Stride[(int)e.axisY];
offZ += z.Stride[(int)e.axisZ];
}
return;
case EinsteinMode.OUTERX:
var axis = (int)e.axisX;
for (int i = 0; i < x.Shape[axis]; ++i)
{
_EinsteinSum(n + 1, einstein, x, offX, y, offY, z, offZ);
offX += x.Stride[axis];
offZ += z.Stride[(int)e.axisZ];
}
return;
case EinsteinMode.OUTERY:
axis = (int)e.axisY;
for (int i = 0; i < y.Shape[axis]; ++i)
{
_EinsteinSum(n + 1, einstein, x, offX, y, offY, z, offZ);
offY += y.Stride[axis];
offZ += z.Stride[(int)e.axisZ];
}
return;
case EinsteinMode.SUMX:
axis = (int)e.axisX;
for (int i = 0; i < x.Shape[axis]; ++i)
{
_EinsteinSum(n + 1, einstein, x, offX, y, offY, z, offZ);
offX += x.Stride[axis];
}
return;
case EinsteinMode.SUMY:
axis = (int)e.axisY;
for (int i = 0; i < y.Shape[axis]; ++i)
{
_EinsteinSum(n + 1, einstein, x, offX, y, offY, z, offZ);
offY += y.Stride[axis];
}
return;
}
}