Skip to content

BLAS 1::scal

Jennifer Loe edited this page May 8, 2020 · 2 revisions

KokkosBlas::scal()

Header File: KokkosBlas1_scal.hpp

Usage: KokkosBlas::scal(y,alpha,x);

Multiply each value of x(i) or x(i,j) with a scalar alpha(or alpha(j)), and assigns it to y(i) or y(i,j) respectively.

Interface

template<class OutputVector, class ScalarType, class InputVector>
void scal (const OutputVector& Y, const ScalarType& alpha, const InputVector& X);

Parameters:

  • OutputVector: A rank-1 or rank-2 Kokkos::View with non-const data type.
  • ScalarType: A scalar type or rank-1 Kokkos::View.
  • InputVector: A rank-1 or rank-2 Kokkos::View

Requirements:

  • OutputVector::value_type == OutputVector::non_const_value_type
  • Y.rank == X.rank
  • Y.rank == 1 or Y.rank == 2
  • Y.extent(0) == X.extent(0)
  • Y.extent(1) == X.extent(1)
  • If Y.rank == 1 then ScalarType is a scalar type.
  • If Y.rank == 2 then ScalarType is a Kokkos::View and alpha.rank == 1

Example

#include<Kokkos_Core.hpp>
#include<KokkosBlas1_scal.hpp>

int main(int argc, char* argv[]) {
   Kokkos::initialize();

   int N = atoi(argv[1]);

   Kokkos::View<double*> x("X",N);
   Kokkos::View<double*> y("Y",N);
   Kokkos::deep_copy(x,3.0);

   KokkosBlas::scal(y,1.5,x);

   double sum = 0.0;
   Kokkos::parallel_reduce("CheckValue", N, KOKKOS_LAMBDA (const int& i, double& lsum) {
     lsum += y(i);
   },sum);

   printf("Sum: %lf Expected: %lf Diff: %e\n",sum,1.0*N*4.5,sum-1.0*N*4.5);

   Kokkos::finalize();
}
Clone this wiki locally