Module Nvector.NVECTOR_OPS.Local

dotprod x y returns the dot product of x and y.

maxnorm x returns the maximum absolute value in x.

min x returns the smallest element in x.

l1norm x returns the l1 norm of x.

invtest x z calculates z(i) = 1 / x(i) with prior testing for zero values. This routine returns true if all components of x are nonzero (successful inversion) and false otherwise (not all elements inverted).

constrmask c x m calculates m(i) = Pi x(i) returning the conjunction. The value of Pi depends on c(i): 2: x(i) > 0, 1: x(i) >= 0, 0: true, -1: x(i) <= 0, and -2: x(i) < 0.

minquotient num denom returns the minimum of num(i) / denom(i). Zero denom elements are skipped.

wsqrsum x w calculates the weighted squared sum of x with weight vector w. That is, $s = \sum_{i=0}^{n_\mathit{local} - 1}(x_i w_i)^2$ .

wsqrsummask x w id calculates the weighted squared sum of x with weight vector w for the elements where id is positive. That is, $m = \sum_{i=0}^{n_\mathit{local} - 1}(x_i w_i H(\math{id}_i))^2$ where $H(\alpha) = \begin{cases} 1 & \alpha > 0 \ 0 & \alpha \le 0 \end{cases} $ .

dotprodmulti x y d calculates d(j) = x(0)*y(j)(0) + ... + x(n-1)*y(j)(n-1) for the nl task-local elements in the nvectors and where j ranges over the array elements.

dotprodmulti_allreduce x d combines the task-local portions of the dot product of a vector x with nv vectors.