Actual source code: fmult.F90

  1: !
  2: !
  3: !    Fortran kernel for sparse matrix-vector product in the AIJ matrix format
  4: !
  5: #include <petsc/finclude/petscsys.h>
  6: !
  7: pure subroutine FortranMultTransposeAddAIJ(n, x, ii, jj, a, y)
  8:   use, intrinsic :: ISO_C_binding
  9:   implicit none(type, external)
 10:   PetscScalar, intent(in) :: x(0:*), a(0:*)
 11:   PetscScalar, intent(inout) :: y(0:*)
 12:   PetscInt, intent(in) :: n, ii(*), jj(0:*)

 14:   PetscInt :: i, jstart, jend

 16:   jend = ii(1)
 17:   do i = 1, n
 18:     jstart = jend
 19:     jend = ii(i + 1)
 20:     y(jj(jstart:jend - 1)) = y(jj(jstart:jend - 1)) + x(i - 1)*a(jstart:jend - 1)
 21:   end do
 22: end subroutine FortranMultTransposeAddAIJ

 24: pure subroutine FortranMultAIJ(n, x, ii, jj, a, y)
 25:   use, intrinsic :: ISO_C_binding
 26:   implicit none(type, external)
 27:   PetscScalar, intent(in) :: x(0:*), a(0:*)
 28:   PetscScalar, intent(inout) :: y(*)
 29:   PetscInt, intent(in) :: n, ii(*), jj(0:*)

 31:   PetscInt :: i, jstart, jend

 33: #ifdef PETSC_USE_OPENMP_KERNELS
 34:   !omp parallel do private(jstart,jend)
 35: #endif
 36:   do i = 1, n
 37:     jstart = ii(i)
 38:     jend = ii(i + 1)
 39:     y(i) = sum(a(jstart:jend - 1)*x(jj(jstart:jend - 1)))
 40:   end do
 41: end subroutine FortranMultAIJ