fdurastante
/
psblas3
mirror of https://github.com/sfilippone/psblas3.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

psblas3-integer8:

Browse Source 
base/psblas/psb_cdot.f90
 base/psblas/psb_ddot.f90
 base/psblas/psb_sdot.f90
 base/psblas/psb_zdot.f90

Keep moving.
psblas3-type-indexed
Salvatore Filippone 13 years ago
parent baa0d54a03
commit 934759b828
4 changed files with 471 additions and 576 deletions
Show Stats Download Patch File Download Diff File

225
base/psblas/psb_cdot.f90
Unescape Escape View File

@ -57,19 +57,18 @@ function psb_cdot_vect(x, y, desc_a,info) result(res)
use psb_c_vect_mod use psb_c_vect_mod
use psb_c_psblas_mod, psb_protect_name => psb_cdot_vect use psb_c_psblas_mod, psb_protect_name => psb_cdot_vect
implicit none implicit none
complex(psb_spk_) :: res complex(psb_spk_) :: res
type(psb_c_vect_type), intent(inout) :: x, y type(psb_c_vect_type), intent(inout) :: x, y
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr & err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr
complex(psb_spk_) :: dot_local
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_sdot' name='psb_sdot'
res = szero res = czero
if (psb_errstatus_fatal()) return if (psb_errstatus_fatal()) return
info=psb_success_ info=psb_success_
call psb_erractionsave(err_act) call psb_erractionsave(err_act)
@ -118,27 +117,22 @@ function psb_cdot_vect(x, y, desc_a,info) result(res)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
nr = desc_a%get_local_rows() if(nr > 0) then
if(nr > 0) then res = x%dot(nr,y)
dot_local = x%dot(nr,y) ! FIXME
!!$ ! adjust dot_local because overlapped elements are computed more than once ! adjust dot_local because overlapped elements are computed more than once
!!$ do i=1,size(desc_a%ovrlap_elem,1) !!$ do i=1,size(desc_a%ovrlap_elem,1)
!!$ idx = desc_a%ovrlap_elem(i,1) !!$ idx = desc_a%ovrlap_elem(i,1)
!!$ ndm = desc_a%ovrlap_elem(i,2) !!$ ndm = desc_a%ovrlap_elem(i,2)
!!$ dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx)) !!$ dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx))
!!$ end do !!$ end do
else
dot_local=czero
end if
else else
dot_local=czero res = czero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
res = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -154,20 +148,20 @@ function psb_cdot_vect(x, y, desc_a,info) result(res)
end function psb_cdot_vect end function psb_cdot_vect
function psb_cdot(x, y,desc_a, info, jx, jy) function psb_cdot(x, y,desc_a, info, jx, jy) result(res)
use psb_base_mod, psb_protect_name => psb_cdot use psb_base_mod, psb_protect_name => psb_cdot
implicit none implicit none
complex(psb_spk_), intent(in) :: x(:,:), y(:,:) complex(psb_spk_), intent(in) :: x(:,:), y(:,:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(in), optional :: jx, jy integer(psb_ipk_), intent(in), optional :: jx, jy
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
complex(psb_spk_) :: psb_cdot complex(psb_spk_) :: res
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr & err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr, &
complex(psb_spk_) :: dot_local & lldx, lldy
complex(psb_spk_) :: cdotc complex(psb_spk_) :: cdotc
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
@ -205,11 +199,13 @@ function psb_cdot(x, y,desc_a, info, jx, jy)
end if end if
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ijx,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ijx,desc_a,info,iix,jjx)
if (info == psb_success_) & if (info == psb_success_) &
& call psb_chkvect(m,ione,size(y,1),iy,ijy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,ijy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -223,27 +219,21 @@ function psb_cdot(x, y,desc_a, info, jx, jy)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
nr = desc_a%get_local_rows() if(nr > 0) then
if(nr > 0) then res = cdotc(int(nr,kind=psb_mpik_), x(iix:,jjx),1,y(iiy:,jjy),1)
dot_local = cdotc(nr, x(iix:,jjx),ione,y(iiy:,jjy),ione) ! adjust dot_local because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*(conjg(x(idx,jjx))*y(idx,jjy))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(conjg(x(idx,jjx))*y(idx,jjy)) end do
end do
else
dot_local=0.0
end if
else else
dot_local=0.0 res = czero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
psb_cdot = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -304,19 +294,19 @@ end function psb_cdot
! desc_a - type(psb_desc_type). The communication descriptor. ! desc_a - type(psb_desc_type). The communication descriptor.
! info - integer. Return code ! info - integer. Return code
! !
function psb_cdotv(x, y,desc_a, info) function psb_cdotv(x, y,desc_a, info) result(res)
use psb_base_mod, psb_protect_name => psb_cdotv use psb_base_mod, psb_protect_name => psb_cdotv
implicit none implicit none
complex(psb_spk_), intent(in) :: x(:), y(:) complex(psb_spk_), intent(in) :: x(:), y(:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
complex(psb_spk_) :: psb_cdotv complex(psb_spk_) :: res
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, jx, iy, jy, iiy, jjy, i, m, nr & err_act, iix, jjx, ix, jx, iy, jy, iiy, jjy, i, m, nr, &
complex(psb_spk_) :: dot_local & lldx, lldy
complex(psb_spk_) :: cdotc complex(psb_spk_) :: cdotc
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
@ -339,11 +329,12 @@ function psb_cdotv(x, y,desc_a, info)
jx = ione jx = ione
jy = ione jy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,jx,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,jx,desc_a,info,iix,jjx)
if (info == psb_success_)& if (info == psb_success_)&
& call psb_chkvect(m,ione,size(y,1),iy,jy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,jy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -357,27 +348,22 @@ function psb_cdotv(x, y,desc_a, info)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
nr = desc_a%get_local_rows() if(nr > 0) then
if(nr > 0) then res = cdotc(int(nr,kind=psb_mpik_), x,1,y,1)
dot_local = cdotc(nr, x,ione,y,ione) ! adjust res because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*(conjg(x(idx))*y(idx))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(conjg(x(idx))*y(idx)) end do
end do
else
dot_local=0.0
end if
else else
dot_local=0.0 res = czero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
psb_cdotv = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -442,16 +428,16 @@ subroutine psb_cdotvs(res, x, y,desc_a, info)
use psb_base_mod, psb_protect_name => psb_cdotvs use psb_base_mod, psb_protect_name => psb_cdotvs
implicit none implicit none
complex(psb_spk_), intent(in) :: x(:), y(:) complex(psb_spk_), intent(in) :: x(:), y(:)
complex(psb_spk_), intent(out) :: res complex(psb_spk_), intent(out) :: res
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, iy, iiy, jjy, i, m,nr & err_act, iix, jjx, ix, iy, iiy, jjy, i, m,nr, &
complex(psb_spk_) :: dot_local & lldx, lldy
complex(psb_spk_) :: cdotc complex(psb_spk_) :: cdotc
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_cdot' name='psb_cdot'
@ -471,10 +457,12 @@ subroutine psb_cdotvs(res, x, y,desc_a, info)
ix = ione ix = ione
iy = ione iy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ix,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ix,desc_a,info,iix,jjx)
if (info == psb_success_) & if (info == psb_success_) &
& call psb_chkvect(m,ione,size(y,1),iy,iy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,iy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -488,27 +476,21 @@ subroutine psb_cdotvs(res, x, y,desc_a, info)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
nr = desc_a%get_local_rows() if(nr > 0) then
if(nr > 0) then res = cdotc(int(nr,kind=psb_mpik_), x,1,y,1)
dot_local = cdotc(nr, x,ione,y,ione) ! adjust res because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*(conjg(x(idx))*y(idx))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(conjg(x(idx))*y(idx)) end do
end do
else
dot_local=0.0
end if
else else
dot_local=0.0 res = czero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
res = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -574,16 +556,16 @@ subroutine psb_cmdots(res, x, y, desc_a, info)
use psb_base_mod, psb_protect_name => psb_cmdots use psb_base_mod, psb_protect_name => psb_cmdots
implicit none implicit none
complex(psb_spk_), intent(in) :: x(:,:), y(:,:) complex(psb_spk_), intent(in) :: x(:,:), y(:,:)
complex(psb_spk_), intent(out) :: res(:) complex(psb_spk_), intent(out) :: res(:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, iy, iiy, jjy, i, m, j, k, nr & err_act, iix, jjx, ix, iy, iiy, jjy, i, m, j, k, nr, &
complex(psb_spk_),allocatable :: dot_local(:) & lldx, lldy
complex(psb_spk_) :: cdotc complex(psb_spk_) :: cdotc
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_cmdots' name='psb_cmdots'
@ -604,16 +586,18 @@ subroutine psb_cmdots(res, x, y, desc_a, info)
iy = ione iy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ix,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ix,desc_a,info,iix,jjx)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
call psb_errpush(info,name,a_err=ch_err) call psb_errpush(info,name,a_err=ch_err)
goto 9999 goto 9999
end if end if
call psb_chkvect(m,ione,size(y,1),iy,iy,desc_a,info,iiy,jjy) call psb_chkvect(m,ione,lldy,iy,iy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -628,31 +612,26 @@ subroutine psb_cmdots(res, x, y, desc_a, info)
end if end if
k = min(size(x,2),size(y,2)) k = min(size(x,2),size(y,2))
allocate(dot_local(k))
nr = desc_a%get_local_rows()
if(m /= 0) then if(nr > 0) then
nr = desc_a%get_local_rows() do j=1,k
if(nr > 0) then res(j) = cdotc(int(nr,kind=psb_mpik_),x(1:,j),1,y(1:,j),1)
do j=1,k ! adjust res because overlapped elements are computed more than once
dot_local(j) = cdotc(nr, x(1:,j),ione,y(1:,j),ione) end do
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
end do idx = desc_a%ovrlap_elem(i,1)
do i=1,size(desc_a%ovrlap_elem,1) ndm = desc_a%ovrlap_elem(i,2)
idx = desc_a%ovrlap_elem(i,1) res(1:k) = res(1:k) - &
ndm = desc_a%ovrlap_elem(i,2) & (real(ndm-1)/real(ndm))*(conjg(x(idx,1:k))*y(idx,1:k))
dot_local(1:k) = dot_local(1:k) - (real(ndm-1)/real(ndm))*(conjg(x(idx,1:k))*y(idx,1:k)) end do
end do
else
dot_local(:)=0.d0
end if
else else
dot_local(:)=0.d0 res(:) = czero
end if end if
! compute global sum
call psb_sum(ictxt, dot_local(1:k))
res(1:k) = dot_local(1:k) ! compute global sum
call psb_sum(ictxt, res(1:k))
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return

289
base/psblas/psb_ddot.f90
Unescape Escape View File

@ -34,19 +34,19 @@
! Function: psb_ddot ! Function: psb_ddot
! psb_ddot forms the dot product of two distributed vectors, ! psb_ddot forms the dot product of two distributed vectors,
! !
! dot := sub( X )**T * sub( Y ) ! dot := sub( X )**C * sub( Y )
! !
! where sub( X ) denotes X(:,JX) ! where sub( X ) denotes X(:,JX)
! !
! sub( Y ) denotes Y(:,JY). ! sub( Y ) denotes Y(:,JY).
! !
! Arguments: ! Arguments:
! x(:,:) - real The input vector containing the entries of ( X ). ! x(:,:) - complex The input vector containing the entries of sub( X ).
! y(:,:) - real The input vector containing the entries of ( Y ). ! y(:,:) - complex The input vector containing the entries of sub( Y ).
! desc_a - type(psb_desc_type). The communication descriptor. ! desc_a - type(psb_desc_type). The communication descriptor.
! info - integer. Return code ! info - integer. Return code
! jx - integer(optional). The column offset for sub( X ). ! jx - integer(optional). The column offset for sub( X ).
! jy - integer(optional). The column offset for sub( Y ). ! jy - integer(optional). The column offset for sub( Y ).
! !
function psb_ddot_vect(x, y, desc_a,info) result(res) function psb_ddot_vect(x, y, desc_a,info) result(res)
use psb_descriptor_type use psb_descriptor_type
@ -57,19 +57,17 @@ function psb_ddot_vect(x, y, desc_a,info) result(res)
use psb_d_vect_mod use psb_d_vect_mod
use psb_d_psblas_mod, psb_protect_name => psb_ddot_vect use psb_d_psblas_mod, psb_protect_name => psb_ddot_vect
implicit none implicit none
real(psb_dpk_) :: res real(psb_dpk_) :: res
type(psb_d_vect_type), intent(inout) :: x, y type(psb_d_vect_type), intent(inout) :: x, y
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr & err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr
real(psb_dpk_) :: dot_local
real(psb_dpk_) :: ddot
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_ddot' name='psb_sdot'
res = dzero res = dzero
if (psb_errstatus_fatal()) return if (psb_errstatus_fatal()) return
info=psb_success_ info=psb_success_
@ -119,27 +117,22 @@ function psb_ddot_vect(x, y, desc_a,info) result(res)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
nr = desc_a%get_local_rows() if(nr > 0) then
if(nr > 0) then res = x%dot(nr,y)
dot_local = x%dot(nr,y) ! FIXME
!!$ ! adjust dot_local because overlapped elements are computed more than once ! adjust dot_local because overlapped elements are computed more than once
!!$ do i=1,size(desc_a%ovrlap_elem,1) !!$ do i=1,size(desc_a%ovrlap_elem,1)
!!$ idx = desc_a%ovrlap_elem(i,1) !!$ idx = desc_a%ovrlap_elem(i,1)
!!$ ndm = desc_a%ovrlap_elem(i,2) !!$ ndm = desc_a%ovrlap_elem(i,2)
!!$ dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx)) !!$ dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx))
!!$ end do !!$ end do
else
dot_local=dzero
end if
else else
dot_local=dzero res = dzero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
res = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -155,28 +148,25 @@ function psb_ddot_vect(x, y, desc_a,info) result(res)
end function psb_ddot_vect end function psb_ddot_vect
function psb_ddot(x, y,desc_a, info, jx, jy) function psb_ddot(x, y,desc_a, info, jx, jy) result(res)
use psb_descriptor_type use psb_base_mod, psb_protect_name => psb_ddot
use psb_check_mod
use psb_error_mod
use psb_penv_mod
implicit none implicit none
real(psb_dpk_), intent(in) :: x(:,:), y(:,:) real(psb_dpk_), intent(in) :: x(:,:), y(:,:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(in), optional :: jx, jy integer(psb_ipk_), intent(in), optional :: jx, jy
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
real(psb_dpk_) :: psb_ddot real(psb_dpk_) :: res
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr & err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr, &
real(psb_dpk_) :: dot_local & lldx, lldy
real(psb_dpk_) :: ddot real(psb_dpk_) :: ddot
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_ddot' name='psb_ddot'
if (psb_errstatus_fatal()) return if(psb_get_errstatus() /= 0) return
info=psb_success_ info=psb_success_
call psb_erractionsave(err_act) call psb_erractionsave(err_act)
@ -209,11 +199,13 @@ function psb_ddot(x, y,desc_a, info, jx, jy)
end if end if
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ijx,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ijx,desc_a,info,iix,jjx)
if (info == psb_success_) & if (info == psb_success_) &
& call psb_chkvect(m,ione,size(y,1),iy,ijy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,ijy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -227,27 +219,21 @@ function psb_ddot(x, y,desc_a, info, jx, jy)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
nr = desc_a%get_local_rows() if(nr > 0) then
if(nr > 0) then res = ddot(int(nr,kind=psb_mpik_), x(iix:,jjx),1,y(iiy:,jjy),1)
dot_local = ddot(nr, x(iix:,jjx),ione,y(iiy:,jjy),ione) ! adjust dot_local because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*((x(idx,jjx))*y(idx,jjy))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx,jjx)*y(idx,jjy)) end do
end do
else
dot_local=0.d0
end if
else else
dot_local=0.d0 res = dzero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
psb_ddot = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -300,35 +286,32 @@ end function psb_ddot
! Function: psb_ddotv ! Function: psb_ddotv
! psb_ddotv forms the dot product of two distributed vectors, ! psb_ddotv forms the dot product of two distributed vectors,
! !
! dot := X**T * Y ! dot := X**C * Y
! !
! Arguments: ! Arguments:
! x(:) - real The input vector containing the entries of X. ! x(:) - real The input vector containing the entries of X.
! y(:) - real The input vector containing the entries of Y. ! y(:) - real The input vector containing the entries of Y.
! desc_a - type(psb_desc_type). The communication descriptor. ! desc_a - type(psb_desc_type). The communication descriptor.
! info - integer. Return code ! info - integer. Return code
! !
function psb_ddotv(x, y,desc_a, info) function psb_ddotv(x, y,desc_a, info) result(res)
use psb_descriptor_type use psb_base_mod, psb_protect_name => psb_ddotv
use psb_check_mod
use psb_error_mod
use psb_penv_mod
implicit none implicit none
real(psb_dpk_), intent(in) :: x(:), y(:) real(psb_dpk_), intent(in) :: x(:), y(:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
real(psb_dpk_) :: psb_ddotv real(psb_dpk_) :: res
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, jx, iy, jy, iiy, jjy, i, m, nr & err_act, iix, jjx, ix, jx, iy, jy, iiy, jjy, i, m, nr, &
real(psb_dpk_) :: dot_local & lldx, lldy
real(psb_dpk_) :: ddot real(psb_dpk_) :: ddot
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_ddot' name='psb_ddot'
if (psb_errstatus_fatal()) return if(psb_get_errstatus() /= 0) return
info=psb_success_ info=psb_success_
call psb_erractionsave(err_act) call psb_erractionsave(err_act)
@ -346,11 +329,12 @@ function psb_ddotv(x, y,desc_a, info)
jx = ione jx = ione
jy = ione jy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,jx,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,jx,desc_a,info,iix,jjx)
if (info == psb_success_) & if (info == psb_success_)&
& call psb_chkvect(m,ione,size(y,1),iy,jy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,jy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -364,28 +348,22 @@ function psb_ddotv(x, y,desc_a, info)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
nr = desc_a%get_local_rows() if(nr > 0) then
if(nr > 0) then res = ddot(int(nr,kind=psb_mpik_), x,1,y,1)
dot_local = ddot(nr, x,ione,y,ione) ! adjust res because overlapped elements are computed more than once
do i=1,size(desc_a%ovrlap_elem,1)
! adjust dot_local because overlapped elements are computed more than once idx = desc_a%ovrlap_elem(i,1)
do i=1,size(desc_a%ovrlap_elem,1) ndm = desc_a%ovrlap_elem(i,2)
idx = desc_a%ovrlap_elem(i,1) res = res - (real(ndm-1)/real(ndm))*((x(idx))*y(idx))
ndm = desc_a%ovrlap_elem(i,2) end do
dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx))
end do
else
dot_local=0.d0
end if
else else
dot_local=0.d0 res = dzero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
psb_ddotv = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -437,33 +415,33 @@ end function psb_ddotv
! Subroutine: psb_ddotvs ! Subroutine: psb_ddotvs
! psb_ddotvs forms the dot product of two distributed vectors, ! psb_ddotvs forms the dot product of two distributed vectors,
! !
! dot := X**T * Y ! res := X**C * Y
! !
! Arguments: ! Arguments:
! res - real The result. ! res - real. The result.
! x(:) - real The input vector containing the entries of X. ! x(:) - real The input vector containing the entries of X.
! y(:) - real The input vector containing the entries of Y. ! y(:) - real The input vector containing the entries of Y.
! desc_a - type(psb_desc_type). The communication descriptor. ! desc_a - type(psb_desc_type). The communication descriptor.
! info - integer. Return code ! info - integer. Return code
! !
subroutine psb_ddotvs(res, x, y,desc_a, info) subroutine psb_ddotvs(res, x, y,desc_a, info)
use psb_base_mod, psb_protect_name => psb_ddotvs use psb_base_mod, psb_protect_name => psb_ddotvs
implicit none implicit none
real(psb_dpk_), intent(in) :: x(:), y(:) real(psb_dpk_), intent(in) :: x(:), y(:)
real(psb_dpk_), intent(out) :: res real(psb_dpk_), intent(out) :: res
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, iy, iiy, jjy, i, m, nr & err_act, iix, jjx, ix, iy, iiy, jjy, i, m,nr, &
real(psb_dpk_) :: dot_local & lldx, lldy
real(psb_dpk_) :: ddot real(psb_dpk_) :: ddot
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_ddot' name='psb_ddot'
if (psb_errstatus_fatal()) return if(psb_get_errstatus() /= 0) return
info=psb_success_ info=psb_success_
call psb_erractionsave(err_act) call psb_erractionsave(err_act)
@ -479,10 +457,12 @@ subroutine psb_ddotvs(res, x, y,desc_a, info)
ix = ione ix = ione
iy = ione iy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ix,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ix,desc_a,info,iix,jjx)
if (info == psb_success_) & if (info == psb_success_) &
& call psb_chkvect(m,ione,size(y,1),iy,iy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,iy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -496,27 +476,21 @@ subroutine psb_ddotvs(res, x, y,desc_a, info)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
nr = desc_a%get_local_rows() if(nr > 0) then
if(nr > 0) then res = ddot(int(nr,kind=psb_mpik_), x,1,y,1)
dot_local = ddot(nr, x,ione,y,ione) ! adjust res because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*((x(idx))*y(idx))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx)) end do
end do
else
dot_local=0.d0
end if
else else
dot_local=0.d0 res = dzero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
res = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -569,12 +543,12 @@ end subroutine psb_ddotvs
! Subroutine: psb_dmdots ! Subroutine: psb_dmdots
! psb_dmdots forms the dot product of multiple distributed vectors, ! psb_dmdots forms the dot product of multiple distributed vectors,
! !
! res(i) := ( X(:,i) )**T * ( Y(:,i) ) ! res(i) := ( X(:,i) )**C * ( Y(:,i) )
! !
! Arguments: ! Arguments:
! res(:) - real. The result. ! res(:) - real. The result.
! x(:,:) - real The input vector containing the entries of ( X ). ! x(:) - real The input vector containing the entries of sub( X ).
! y(:,:) - real The input vector containing the entries of ( Y ). ! y(:) - real The input vector containing the entries of sub( Y ).
! desc_a - type(psb_desc_type). The communication descriptor. ! desc_a - type(psb_desc_type). The communication descriptor.
! info - integer. Return code ! info - integer. Return code
! !
@ -582,20 +556,20 @@ subroutine psb_dmdots(res, x, y, desc_a, info)
use psb_base_mod, psb_protect_name => psb_dmdots use psb_base_mod, psb_protect_name => psb_dmdots
implicit none implicit none
real(psb_dpk_), intent(in) :: x(:,:), y(:,:) real(psb_dpk_), intent(in) :: x(:,:), y(:,:)
real(psb_dpk_), intent(out) :: res(:) real(psb_dpk_), intent(out) :: res(:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, iy, iiy, jjy, i, m, j, k, nr & err_act, iix, jjx, ix, iy, iiy, jjy, i, m, j, k, nr, &
real(psb_dpk_),allocatable :: dot_local(:) & lldx, lldy
real(psb_dpk_) :: ddot real(psb_dpk_) :: ddot
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_dmdots' name='psb_dmdots'
if (psb_errstatus_fatal()) return if(psb_get_errstatus() /= 0) return
info=psb_success_ info=psb_success_
call psb_erractionsave(err_act) call psb_erractionsave(err_act)
@ -612,16 +586,18 @@ subroutine psb_dmdots(res, x, y, desc_a, info)
iy = ione iy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ix,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ix,desc_a,info,iix,jjx)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
call psb_errpush(info,name,a_err=ch_err) call psb_errpush(info,name,a_err=ch_err)
goto 9999 goto 9999
end if end if
call psb_chkvect(m,ione,size(y,1),iy,iy,desc_a,info,iiy,jjy) call psb_chkvect(m,ione,lldy,iy,iy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -636,31 +612,26 @@ subroutine psb_dmdots(res, x, y, desc_a, info)
end if end if
k = min(size(x,2),size(y,2)) k = min(size(x,2),size(y,2))
allocate(dot_local(k))
nr = desc_a%get_local_rows()
if(m /= 0) then if(nr > 0) then
nr = desc_a%get_local_rows() do j=1,k
if(nr > 0) then res(j) = ddot(int(nr,kind=psb_mpik_),x(1:,j),1,y(1:,j),1)
do j=1,k ! adjust res because overlapped elements are computed more than once
dot_local(j) = ddot(nr,x(1:,j),ione,y(1:,j),ione) end do
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
end do idx = desc_a%ovrlap_elem(i,1)
do i=1,size(desc_a%ovrlap_elem,1) ndm = desc_a%ovrlap_elem(i,2)
idx = desc_a%ovrlap_elem(i,1) res(1:k) = res(1:k) - &
ndm = desc_a%ovrlap_elem(i,2) & (real(ndm-1)/real(ndm))*((x(idx,1:k))*y(idx,1:k))
dot_local(1:k) = dot_local(1:k) - (real(ndm-1)/real(ndm))*(x(idx,1:k)*y(idx,1:k)) end do
end do
else
dot_local(:)=0.d0
end if
else else
dot_local(:)=0.d0 res(:) = dzero
end if end if
! compute global sum
call psb_sum(ictxt, dot_local(1:k))
res(1:k) = dot_local(1:k) ! compute global sum
call psb_sum(ictxt, res(1:k))
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return

278
base/psblas/psb_sdot.f90
Unescape Escape View File

@ -34,19 +34,19 @@
! Function: psb_sdot ! Function: psb_sdot
! psb_sdot forms the dot product of two distributed vectors, ! psb_sdot forms the dot product of two distributed vectors,
! !
! dot := sub( X )**T * sub( Y ) ! dot := sub( X )**C * sub( Y )
! !
! where sub( X ) denotes X(:,JX) ! where sub( X ) denotes X(:,JX)
! !
! sub( Y ) denotes Y(:,JY). ! sub( Y ) denotes Y(:,JY).
! !
! Arguments: ! Arguments:
! x(:,:) - real The input vector containing the entries of ( X ). ! x(:,:) - complex The input vector containing the entries of sub( X ).
! y(:,:) - real The input vector containing the entries of ( Y ). ! y(:,:) - complex The input vector containing the entries of sub( Y ).
! desc_a - type(psb_desc_type). The communication descriptor. ! desc_a - type(psb_desc_type). The communication descriptor.
! info - integer. Return code ! info - integer. Return code
! jx - integer(optional). The column offset for sub( X ). ! jx - integer(optional). The column offset for sub( X ).
! jy - integer(optional). The column offset for sub( Y ). ! jy - integer(optional). The column offset for sub( Y ).
! !
function psb_sdot_vect(x, y, desc_a,info) result(res) function psb_sdot_vect(x, y, desc_a,info) result(res)
use psb_descriptor_type use psb_descriptor_type
@ -57,16 +57,14 @@ function psb_sdot_vect(x, y, desc_a,info) result(res)
use psb_s_vect_mod use psb_s_vect_mod
use psb_s_psblas_mod, psb_protect_name => psb_sdot_vect use psb_s_psblas_mod, psb_protect_name => psb_sdot_vect
implicit none implicit none
real(psb_spk_) :: res real(psb_spk_) :: res
type(psb_s_vect_type), intent(inout) :: x, y type(psb_s_vect_type), intent(inout) :: x, y
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr & err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr
real(psb_spk_) :: dot_local
real(psb_spk_) :: sdot
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_sdot' name='psb_sdot'
@ -119,27 +117,22 @@ function psb_sdot_vect(x, y, desc_a,info) result(res)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
nr = desc_a%get_local_rows() if(nr > 0) then
if(nr > 0) then res = x%dot(nr,y)
dot_local = x%dot(nr,y) ! FIXME
!!$ ! adjust dot_local because overlapped elements are computed more than once ! adjust dot_local because overlapped elements are computed more than once
!!$ do i=1,size(desc_a%ovrlap_elem,1) !!$ do i=1,size(desc_a%ovrlap_elem,1)
!!$ idx = desc_a%ovrlap_elem(i,1) !!$ idx = desc_a%ovrlap_elem(i,1)
!!$ ndm = desc_a%ovrlap_elem(i,2) !!$ ndm = desc_a%ovrlap_elem(i,2)
!!$ dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx)) !!$ dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx))
!!$ end do !!$ end do
else
dot_local=szero
end if
else else
dot_local=szero res = szero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
res = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -155,24 +148,21 @@ function psb_sdot_vect(x, y, desc_a,info) result(res)
end function psb_sdot_vect end function psb_sdot_vect
function psb_sdot(x, y,desc_a, info, jx, jy) function psb_sdot(x, y,desc_a, info, jx, jy) result(res)
use psb_descriptor_type use psb_base_mod, psb_protect_name => psb_sdot
use psb_check_mod
use psb_error_mod
use psb_penv_mod
implicit none implicit none
real(psb_spk_), intent(in) :: x(:,:), y(:,:) real(psb_spk_), intent(in) :: x(:,:), y(:,:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(in), optional :: jx, jy integer(psb_ipk_), intent(in), optional :: jx, jy
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
real(psb_spk_) :: psb_sdot real(psb_spk_) :: res
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m & err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr, &
real(psb_spk_) :: dot_local & lldx, lldy
real(psb_spk_) :: sdot real(psb_spk_) :: sdot
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_sdot' name='psb_sdot'
@ -209,11 +199,13 @@ function psb_sdot(x, y,desc_a, info, jx, jy)
end if end if
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ijx,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ijx,desc_a,info,iix,jjx)
if (info == psb_success_) & if (info == psb_success_) &
& call psb_chkvect(m,ione,size(y,1),iy,ijy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,ijy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -227,27 +219,21 @@ function psb_sdot(x, y,desc_a, info, jx, jy)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
if(desc_a%get_local_rows() > 0) then if(nr > 0) then
dot_local = sdot(desc_a%get_local_rows(),& res = sdot(int(nr,kind=psb_mpik_), x(iix:,jjx),1,y(iiy:,jjy),1)
& x(iix:,jjx),ione,y(iiy:,jjy),ione) ! adjust dot_local because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*((x(idx,jjx))*y(idx,jjy))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx,jjx)*y(idx,jjy)) end do
end do
else
dot_local=0.0
end if
else else
dot_local=0.0 res = szero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
psb_sdot = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -300,30 +286,27 @@ end function psb_sdot
! Function: psb_sdotv ! Function: psb_sdotv
! psb_sdotv forms the dot product of two distributed vectors, ! psb_sdotv forms the dot product of two distributed vectors,
! !
! dot := X**T * Y ! dot := X**C * Y
! !
! Arguments: ! Arguments:
! x(:) - real The input vector containing the entries of X. ! x(:) - real The input vector containing the entries of X.
! y(:) - real The input vector containing the entries of Y. ! y(:) - real The input vector containing the entries of Y.
! desc_a - type(psb_desc_type). The communication descriptor. ! desc_a - type(psb_desc_type). The communication descriptor.
! info - integer. Return code ! info - integer. Return code
! !
function psb_sdotv(x, y,desc_a, info) function psb_sdotv(x, y,desc_a, info) result(res)
use psb_descriptor_type use psb_base_mod, psb_protect_name => psb_sdotv
use psb_check_mod
use psb_error_mod
use psb_penv_mod
implicit none implicit none
real(psb_spk_), intent(in) :: x(:), y(:) real(psb_spk_), intent(in) :: x(:), y(:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
real(psb_spk_) :: psb_sdotv real(psb_spk_) :: res
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, jx, iy, jy, iiy, jjy, i, m & err_act, iix, jjx, ix, jx, iy, jy, iiy, jjy, i, m, nr, &
real(psb_spk_) :: dot_local & lldx, lldy
real(psb_spk_) :: sdot real(psb_spk_) :: sdot
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
@ -346,11 +329,12 @@ function psb_sdotv(x, y,desc_a, info)
jx = ione jx = ione
jy = ione jy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,jx,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,jx,desc_a,info,iix,jjx)
if (info == psb_success_) & if (info == psb_success_)&
& call psb_chkvect(m,ione,size(y,1),iy,jy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,jy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -364,27 +348,22 @@ function psb_sdotv(x, y,desc_a, info)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
if(desc_a%get_local_rows() > 0) then if(nr > 0) then
dot_local = sdot(desc_a%get_local_rows(),& res = sdot(int(nr,kind=psb_mpik_), x,1,y,1)
& x,ione,y,ione) ! adjust res because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*((x(idx))*y(idx))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx)) end do
end do
else
dot_local=0.0
end if
else else
dot_local=0.0 res = szero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
psb_sdotv = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -436,29 +415,29 @@ end function psb_sdotv
! Subroutine: psb_sdotvs ! Subroutine: psb_sdotvs
! psb_sdotvs forms the dot product of two distributed vectors, ! psb_sdotvs forms the dot product of two distributed vectors,
! !
! dot := X**T * Y ! res := X**C * Y
! !
! Arguments: ! Arguments:
! res - real The result. ! res - real. The result.
! x(:) - real The input vector containing the entries of X. ! x(:) - real The input vector containing the entries of X.
! y(:) - real The input vector containing the entries of Y. ! y(:) - real The input vector containing the entries of Y.
! desc_a - type(psb_desc_type). The communication descriptor. ! desc_a - type(psb_desc_type). The communication descriptor.
! info - integer. Return code ! info - integer. Return code
! !
subroutine psb_sdotvs(res, x, y,desc_a, info) subroutine psb_sdotvs(res, x, y,desc_a, info)
use psb_base_mod, psb_protect_name => psb_sdotvs use psb_base_mod, psb_protect_name => psb_sdotvs
implicit none implicit none
real(psb_spk_), intent(in) :: x(:), y(:) real(psb_spk_), intent(in) :: x(:), y(:)
real(psb_spk_), intent(out) :: res real(psb_spk_), intent(out) :: res
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, iy, iiy, jjy, i, m & err_act, iix, jjx, ix, iy, iiy, jjy, i, m,nr, &
real(psb_spk_) :: dot_local & lldx, lldy
real(psb_spk_) :: sdot real(psb_spk_) :: sdot
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_sdot' name='psb_sdot'
@ -478,10 +457,12 @@ subroutine psb_sdotvs(res, x, y,desc_a, info)
ix = ione ix = ione
iy = ione iy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ix,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ix,desc_a,info,iix,jjx)
if (info == psb_success_) & if (info == psb_success_) &
& call psb_chkvect(m,ione,size(y,1),iy,iy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,iy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -495,27 +476,21 @@ subroutine psb_sdotvs(res, x, y,desc_a, info)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
if(desc_a%get_local_rows() > 0) then if(nr > 0) then
dot_local = sdot(desc_a%get_local_rows(),& res = sdot(int(nr,kind=psb_mpik_), x,1,y,1)
& x,ione,y,ione) ! adjust res because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*((x(idx))*y(idx))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx)) end do
end do
else
dot_local=0.0
end if
else else
dot_local=0.0 res = szero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
res = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -568,12 +543,12 @@ end subroutine psb_sdotvs
! Subroutine: psb_smdots ! Subroutine: psb_smdots
! psb_smdots forms the dot product of multiple distributed vectors, ! psb_smdots forms the dot product of multiple distributed vectors,
! !
! res(i) := ( X(:,i) )**T * ( Y(:,i) ) ! res(i) := ( X(:,i) )**C * ( Y(:,i) )
! !
! Arguments: ! Arguments:
! res(:) - real. The result. ! res(:) - real. The result.
! x(:,:) - real The input vector containing the entries of ( X ). ! x(:) - real The input vector containing the entries of sub( X ).
! y(:,:) - real The input vector containing the entries of ( Y ). ! y(:) - real The input vector containing the entries of sub( Y ).
! desc_a - type(psb_desc_type). The communication descriptor. ! desc_a - type(psb_desc_type). The communication descriptor.
! info - integer. Return code ! info - integer. Return code
! !
@ -581,16 +556,16 @@ subroutine psb_smdots(res, x, y, desc_a, info)
use psb_base_mod, psb_protect_name => psb_smdots use psb_base_mod, psb_protect_name => psb_smdots
implicit none implicit none
real(psb_spk_), intent(in) :: x(:,:), y(:,:) real(psb_spk_), intent(in) :: x(:,:), y(:,:)
real(psb_spk_), intent(out) :: res(:) real(psb_spk_), intent(out) :: res(:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, iy, iiy, jjy, i, m, j, k & err_act, iix, jjx, ix, iy, iiy, jjy, i, m, j, k, nr, &
real(psb_spk_),allocatable :: dot_local(:) & lldx, lldy
real(psb_spk_) :: sdot real(psb_spk_) :: sdot
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_smdots' name='psb_smdots'
@ -611,16 +586,18 @@ subroutine psb_smdots(res, x, y, desc_a, info)
iy = ione iy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ix,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ix,desc_a,info,iix,jjx)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
call psb_errpush(info,name,a_err=ch_err) call psb_errpush(info,name,a_err=ch_err)
goto 9999 goto 9999
end if end if
call psb_chkvect(m,ione,size(y,1),iy,iy,desc_a,info,iiy,jjy) call psb_chkvect(m,ione,lldy,iy,iy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -635,31 +612,26 @@ subroutine psb_smdots(res, x, y, desc_a, info)
end if end if
k = min(size(x,2),size(y,2)) k = min(size(x,2),size(y,2))
allocate(dot_local(k))
nr = desc_a%get_local_rows()
if(m /= 0) then if(nr > 0) then
if(desc_a%get_local_rows() > 0) then do j=1,k
do j=1,k res(j) = sdot(int(nr,kind=psb_mpik_),x(1:,j),1,y(1:,j),1)
dot_local(j) = sdot(desc_a%get_local_rows(),& ! adjust res because overlapped elements are computed more than once
& x(1:,j),ione,y(1:,j),ione) end do
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
end do idx = desc_a%ovrlap_elem(i,1)
do i=1,size(desc_a%ovrlap_elem,1) ndm = desc_a%ovrlap_elem(i,2)
idx = desc_a%ovrlap_elem(i,1) res(1:k) = res(1:k) - &
ndm = desc_a%ovrlap_elem(i,2) & (real(ndm-1)/real(ndm))*((x(idx,1:k))*y(idx,1:k))
dot_local(1:k) = dot_local(1:k) - (real(ndm-1)/real(ndm))*(x(idx,1:k)*y(idx,1:k)) end do
end do
else
dot_local(:)=0.0
end if
else else
dot_local(:)=0.0 res(:) = szero
end if end if
! compute global sum
call psb_sum(ictxt, dot_local(1:k))
res(1:k) = dot_local(1:k) ! compute global sum
call psb_sum(ictxt, res(1:k))
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return

231
base/psblas/psb_zdot.f90
Unescape Escape View File

@ -60,16 +60,15 @@ function psb_zdot_vect(x, y, desc_a,info) result(res)
complex(psb_dpk_) :: res complex(psb_dpk_) :: res
type(psb_z_vect_type), intent(inout) :: x, y type(psb_z_vect_type), intent(inout) :: x, y
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr & err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr
complex(psb_dpk_) :: dot_local
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_sdot' name='psb_sdot'
res = szero res = zzero
if (psb_errstatus_fatal()) return if (psb_errstatus_fatal()) return
info=psb_success_ info=psb_success_
call psb_erractionsave(err_act) call psb_erractionsave(err_act)
@ -118,27 +117,22 @@ function psb_zdot_vect(x, y, desc_a,info) result(res)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
nr = desc_a%get_local_rows() if(nr > 0) then
if(nr > 0) then res = x%dot(nr,y)
dot_local = x%dot(nr,y) ! FIXME
!!$ ! adjust dot_local because overlapped elements are computed more than once ! adjust dot_local because overlapped elements are computed more than once
!!$ do i=1,size(desc_a%ovrlap_elem,1) !!$ do i=1,size(desc_a%ovrlap_elem,1)
!!$ idx = desc_a%ovrlap_elem(i,1) !!$ idx = desc_a%ovrlap_elem(i,1)
!!$ ndm = desc_a%ovrlap_elem(i,2) !!$ ndm = desc_a%ovrlap_elem(i,2)
!!$ dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx)) !!$ dot_local = dot_local - (real(ndm-1)/real(ndm))*(x(idx)*y(idx))
!!$ end do !!$ end do
else
dot_local=zzero
end if
else else
dot_local=zzero res = zzero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
res = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -154,24 +148,21 @@ function psb_zdot_vect(x, y, desc_a,info) result(res)
end function psb_zdot_vect end function psb_zdot_vect
function psb_zdot(x, y,desc_a, info, jx, jy) function psb_zdot(x, y,desc_a, info, jx, jy) result(res)
use psb_descriptor_type use psb_base_mod, psb_protect_name => psb_zdot
use psb_check_mod
use psb_error_mod
use psb_penv_mod
implicit none implicit none
complex(psb_dpk_), intent(in) :: x(:,:), y(:,:) complex(psb_dpk_), intent(in) :: x(:,:), y(:,:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(in), optional :: jx, jy integer(psb_ipk_), intent(in), optional :: jx, jy
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
complex(psb_dpk_) :: psb_zdot complex(psb_dpk_) :: res
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m & err_act, iix, jjx, ix, ijx, iy, ijy, iiy, jjy, i, m, nr, &
complex(psb_dpk_) :: dot_local & lldx, lldy
complex(psb_dpk_) :: zdotc complex(psb_dpk_) :: zdotc
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_zdot' name='psb_zdot'
@ -208,11 +199,13 @@ function psb_zdot(x, y,desc_a, info, jx, jy)
end if end if
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ijx,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ijx,desc_a,info,iix,jjx)
if (info == psb_success_) & if (info == psb_success_) &
& call psb_chkvect(m,ione,size(y,1),iy,ijy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,ijy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -226,27 +219,21 @@ function psb_zdot(x, y,desc_a, info, jx, jy)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
if(desc_a%get_local_rows() > 0) then if(nr > 0) then
dot_local = zdotc(desc_a%get_local_rows(),& res = zdotc(int(nr,kind=psb_mpik_), x(iix:,jjx),1,y(iiy:,jjy),1)
& x(iix:,jjx),ione,y(iiy:,jjy),ione) ! adjust dot_local because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*(conjg(x(idx,jjx))*y(idx,jjy))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(conjg(x(idx,jjx))*y(idx,jjy)) end do
end do
else
dot_local=0.d0
end if
else else
dot_local=0.d0 res = zzero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
psb_zdot = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -307,22 +294,19 @@ end function psb_zdot
! desc_a - type(psb_desc_type). The communication descriptor. ! desc_a - type(psb_desc_type). The communication descriptor.
! info - integer. Return code ! info - integer. Return code
! !
function psb_zdotv(x, y,desc_a, info) function psb_zdotv(x, y,desc_a, info) result(res)
use psb_descriptor_type use psb_base_mod, psb_protect_name => psb_zdotv
use psb_check_mod
use psb_error_mod
use psb_penv_mod
implicit none implicit none
complex(psb_dpk_), intent(in) :: x(:), y(:) complex(psb_dpk_), intent(in) :: x(:), y(:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
complex(psb_dpk_) :: psb_zdotv complex(psb_dpk_) :: res
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, jx, iy, jy, iiy, jjy, i, m & err_act, iix, jjx, ix, jx, iy, jy, iiy, jjy, i, m, nr, &
complex(psb_dpk_) :: dot_local & lldx, lldy
complex(psb_dpk_) :: zdotc complex(psb_dpk_) :: zdotc
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
@ -345,11 +329,12 @@ function psb_zdotv(x, y,desc_a, info)
jx = ione jx = ione
jy = ione jy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,jx,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,jx,desc_a,info,iix,jjx)
if (info == psb_success_)& if (info == psb_success_)&
& call psb_chkvect(m,ione,size(y,1),iy,jy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,jy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -363,27 +348,22 @@ function psb_zdotv(x, y,desc_a, info)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
if(desc_a%get_local_rows() > 0) then if(nr > 0) then
dot_local = zdotc(desc_a%get_local_rows(),& res = zdotc(int(nr,kind=psb_mpik_), x,1,y,1)
& x,ione,y,ione) ! adjust res because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*(conjg(x(idx))*y(idx))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(conjg(x(idx))*y(idx)) end do
end do
else
dot_local=0.d0
end if
else else
dot_local=0.d0 res = zzero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
psb_zdotv = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -448,16 +428,16 @@ subroutine psb_zdotvs(res, x, y,desc_a, info)
use psb_base_mod, psb_protect_name => psb_zdotvs use psb_base_mod, psb_protect_name => psb_zdotvs
implicit none implicit none
complex(psb_dpk_), intent(in) :: x(:), y(:) complex(psb_dpk_), intent(in) :: x(:), y(:)
complex(psb_dpk_), intent(out) :: res complex(psb_dpk_), intent(out) :: res
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, iy, iiy, jjy, i, m & err_act, iix, jjx, ix, iy, iiy, jjy, i, m,nr, &
complex(psb_dpk_) :: dot_local & lldx, lldy
complex(psb_dpk_) :: zdotc complex(psb_dpk_) :: zdotc
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_zdot' name='psb_zdot'
@ -477,10 +457,12 @@ subroutine psb_zdotvs(res, x, y,desc_a, info)
ix = ione ix = ione
iy = ione iy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ix,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ix,desc_a,info,iix,jjx)
if (info == psb_success_) & if (info == psb_success_) &
& call psb_chkvect(m,ione,size(y,1),iy,iy,desc_a,info,iiy,jjy) & call psb_chkvect(m,ione,lldy,iy,iy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -494,27 +476,21 @@ subroutine psb_zdotvs(res, x, y,desc_a, info)
goto 9999 goto 9999
end if end if
if(m /= 0) then nr = desc_a%get_local_rows()
if(desc_a%get_local_rows() > 0) then if(nr > 0) then
dot_local = zdotc(desc_a%get_local_rows(),& res = zdotc(int(nr,kind=psb_mpik_), x,1,y,1)
& x,ione,y,ione) ! adjust res because overlapped elements are computed more than once
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
do i=1,size(desc_a%ovrlap_elem,1) idx = desc_a%ovrlap_elem(i,1)
idx = desc_a%ovrlap_elem(i,1) ndm = desc_a%ovrlap_elem(i,2)
ndm = desc_a%ovrlap_elem(i,2) res = res - (real(ndm-1)/real(ndm))*(conjg(x(idx))*y(idx))
dot_local = dot_local - (real(ndm-1)/real(ndm))*(conjg(x(idx))*y(idx)) end do
end do
else
dot_local=0.d0
end if
else else
dot_local=0.d0 res = zzero
end if end if
! compute global sum ! compute global sum
call psb_sum(ictxt, dot_local) call psb_sum(ictxt, res)
res = dot_local
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return
@ -580,16 +556,16 @@ subroutine psb_zmdots(res, x, y, desc_a, info)
use psb_base_mod, psb_protect_name => psb_zmdots use psb_base_mod, psb_protect_name => psb_zmdots
implicit none implicit none
complex(psb_dpk_), intent(in) :: x(:,:), y(:,:) complex(psb_dpk_), intent(in) :: x(:,:), y(:,:)
complex(psb_dpk_), intent(out) :: res(:) complex(psb_dpk_), intent(out) :: res(:)
type(psb_desc_type), intent(in) :: desc_a type(psb_desc_type), intent(in) :: desc_a
integer(psb_ipk_), intent(out) :: info integer(psb_ipk_), intent(out) :: info
! locals ! locals
integer(psb_ipk_) :: ictxt, np, me, idx, ndm,& integer(psb_ipk_) :: ictxt, np, me, idx, ndm,&
& err_act, iix, jjx, ix, iy, iiy, jjy, i, m, j, k & err_act, iix, jjx, ix, iy, iiy, jjy, i, m, j, k, nr, &
complex(psb_dpk_),allocatable :: dot_local(:) & lldx, lldy
complex(psb_dpk_) :: zdotc complex(psb_dpk_) :: zdotc
character(len=20) :: name, ch_err character(len=20) :: name, ch_err
name='psb_zmdots' name='psb_zmdots'
@ -610,16 +586,18 @@ subroutine psb_zmdots(res, x, y, desc_a, info)
iy = ione iy = ione
m = desc_a%get_global_rows() m = desc_a%get_global_rows()
lldx = size(x,1)
lldy = size(y,1)
! check vector correctness ! check vector correctness
call psb_chkvect(m,ione,size(x,1),ix,ix,desc_a,info,iix,jjx) call psb_chkvect(m,ione,lldx,ix,ix,desc_a,info,iix,jjx)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
call psb_errpush(info,name,a_err=ch_err) call psb_errpush(info,name,a_err=ch_err)
goto 9999 goto 9999
end if end if
call psb_chkvect(m,ione,size(y,1),iy,iy,desc_a,info,iiy,jjy) call psb_chkvect(m,ione,lldy,iy,iy,desc_a,info,iiy,jjy)
if(info /= psb_success_) then if(info /= psb_success_) then
info=psb_err_from_subroutine_ info=psb_err_from_subroutine_
ch_err='psb_chkvect' ch_err='psb_chkvect'
@ -634,31 +612,26 @@ subroutine psb_zmdots(res, x, y, desc_a, info)
end if end if
k = min(size(x,2),size(y,2)) k = min(size(x,2),size(y,2))
allocate(dot_local(k))
nr = desc_a%get_local_rows()
if(m /= 0) then if(nr > 0) then
if(desc_a%get_local_rows() > 0) then do j=1,k
do j=1,k res(j) = zdotc(int(nr,kind=psb_mpik_),x(1:,j),1,y(1:,j),1)
dot_local(j) = zdotc(desc_a%get_local_rows(),& ! adjust res because overlapped elements are computed more than once
& x(1:,j),ione,y(1:,j),ione) end do
! adjust dot_local because overlapped elements are computed more than once do i=1,size(desc_a%ovrlap_elem,1)
end do idx = desc_a%ovrlap_elem(i,1)
do i=1,size(desc_a%ovrlap_elem,1) ndm = desc_a%ovrlap_elem(i,2)
idx = desc_a%ovrlap_elem(i,1) res(1:k) = res(1:k) - &
ndm = desc_a%ovrlap_elem(i,2) & (real(ndm-1)/real(ndm))*(conjg(x(idx,1:k))*y(idx,1:k))
dot_local(1:k) = dot_local(1:k) - (real(ndm-1)/real(ndm))*(conjg(x(idx,1:k))*y(idx,1:k)) end do
end do
else
dot_local(:)=0.d0
end if
else else
dot_local(:)=0.d0 res(:) = zzero
end if end if
! compute global sum
call psb_sum(ictxt, dot_local(1:k))
res(1:k) = dot_local(1:k) ! compute global sum
call psb_sum(ictxt, res(1:k))
call psb_erractionrestore(err_act) call psb_erractionrestore(err_act)
return return

Write Preview
Loading…
Cancel
Save