psblas2-dev:

base/modules/psb_tools_mod.f90
 base/serial/psb_dcoins.f90
 base/serial/psb_update_mod.f90
 base/serial/psb_zcoins.f90
 docs/pdf/toolsrout.tex
 docs/userguide.pdf
 test/pargen/Makefile
 test/pargen/ppde.f90

Defined new is_owned, is_local routines.
Merged performance fix for SP5.
This commit is contained in:
Salvatore Filippone
2008-04-15 12:18:04 +00:00
parent 486a011f4b
commit 93b9ebbcf0
8 changed files with 6931 additions and 5371 deletions
+4
View File
@@ -19,6 +19,10 @@ ppde: ppde.o
$(F90LINK) ppde.o -o ppde $(PSBLAS_LIB) $(LDLIBS)
/bin/mv ppde $(EXEDIR)
tpde: tpde.o
$(F90LINK) tpde.o -pg -o tpde $(PSBLAS_LIB) $(LDLIBS)
/bin/mv tpde $(EXEDIR)
.f90.o:
$(MPF90) $(F90COPT) $(FINCLUDES) $(FDEFINES) -c $<
+127 -132
View File
@@ -361,14 +361,12 @@ contains
end interface ! local variables
type(psb_dspmat_type) :: a
real(psb_dpk_) :: zt(nbmax),glob_x,glob_y,glob_z
integer :: m,n,nnz,glob_row
integer :: m,n,nnz,glob_row,loc_row
integer :: x,y,z,ia,indx_owner
integer :: np, iam
integer :: element
integer :: nv, inv
integer, allocatable :: irow(:),icol(:)
real(psb_dpk_), allocatable :: val(:)
integer, allocatable :: prv(:)
! deltah dimension of each grid cell
! deltat discretization time
real(psb_dpk_) :: deltah
@@ -414,7 +412,7 @@ contains
! a bunch of rows per call.
!
allocate(val(20*nbmax),irow(20*nbmax),&
&icol(20*nbmax),prv(np),stat=info)
&icol(20*nbmax),stat=info)
if (info /= 0 ) then
info=4000
call psb_errpush(info,name)
@@ -430,138 +428,135 @@ contains
! icol(1)=1
do glob_row = 1, n
call parts(glob_row,n,np,prv,nv)
do inv = 1, nv
indx_owner = prv(inv)
if (indx_owner == iam) then
! local matrix pointer
element=1
! compute gridpoint coordinates
if (mod(glob_row,(idim*idim)) == 0) then
x = glob_row/(idim*idim)
else
x = glob_row/(idim*idim)+1
endif
if (mod((glob_row-(x-1)*idim*idim),idim) == 0) then
y = (glob_row-(x-1)*idim*idim)/idim
else
y = (glob_row-(x-1)*idim*idim)/idim+1
endif
z = glob_row-(x-1)*idim*idim-(y-1)*idim
! glob_x, glob_y, glob_x coordinates
glob_x=x*deltah
glob_y=y*deltah
glob_z=z*deltah
! Figure out which rows are local to the current process:
if (psb_is_owned(glob_row,desc_a)) then
! local matrix pointer
element=1
! compute gridpoint coordinates
if (mod(glob_row,(idim*idim)) == 0) then
x = glob_row/(idim*idim)
else
x = glob_row/(idim*idim)+1
endif
if (mod((glob_row-(x-1)*idim*idim),idim) == 0) then
y = (glob_row-(x-1)*idim*idim)/idim
else
y = (glob_row-(x-1)*idim*idim)/idim+1
endif
z = glob_row-(x-1)*idim*idim-(y-1)*idim
! glob_x, glob_y, glob_x coordinates
glob_x=x*deltah
glob_y=y*deltah
glob_z=z*deltah
! check on boundary points
zt(1) = 0.d0
! internal point: build discretization
!
! term depending on (x-1,y,z)
!
if (x==1) then
val(element)=-b1(glob_x,glob_y,glob_z)&
& -a1(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
zt(1) = exp(-glob_y**2-glob_z**2)*(-val(element))
else
val(element)=-b1(glob_x,glob_y,glob_z)&
& -a1(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-2)*idim*idim+(y-1)*idim+(z)
element=element+1
endif
! term depending on (x,y-1,z)
if (y==1) then
val(element)=-b2(glob_x,glob_y,glob_z)&
& -a2(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
zt(1) = exp(-glob_y**2-glob_z**2)*exp(-glob_x)*(-val(element))
else
val(element)=-b2(glob_x,glob_y,glob_z)&
& -a2(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-1)*idim*idim+(y-2)*idim+(z)
element=element+1
endif
! term depending on (x,y,z-1)
if (z==1) then
val(element)=-b3(glob_x,glob_y,glob_z)&
& -a3(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
zt(1) = exp(-glob_y**2-glob_z**2)*exp(-glob_x)*(-val(element))
else
val(element)=-b3(glob_x,glob_y,glob_z)&
& -a3(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-1)*idim*idim+(y-1)*idim+(z-1)
element=element+1
endif
! term depending on (x,y,z)
val(element)=2*b1(glob_x,glob_y,glob_z)&
& +2*b2(glob_x,glob_y,glob_z)&
& +2*b3(glob_x,glob_y,glob_z)&
& +a1(glob_x,glob_y,glob_z)&
& +a2(glob_x,glob_y,glob_z)&
& +a3(glob_x,glob_y,glob_z)
! check on boundary points
zt(1) = 0.d0
! internal point: build discretization
!
! term depending on (x-1,y,z)
!
if (x==1) then
val(element)=-b1(glob_x,glob_y,glob_z)&
& -a1(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-1)*idim*idim+(y-1)*idim+(z)
element=element+1
! term depending on (x,y,z+1)
if (z==idim) then
val(element)=-b1(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
zt(1) = exp(-glob_y**2-glob_z**2)*exp(-glob_x)*(-val(element))
else
val(element)=-b1(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-1)*idim*idim+(y-1)*idim+(z+1)
element=element+1
endif
! term depending on (x,y+1,z)
if (y==idim) then
val(element)=-b2(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
zt(1) = exp(-glob_y**2-glob_z**2)*exp(-glob_x)*(-val(element))
else
val(element)=-b2(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-1)*idim*idim+(y)*idim+(z)
element=element+1
endif
! term depending on (x+1,y,z)
if (x<idim) then
val(element)=-b3(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x)*idim*idim+(y-1)*idim+(z)
element=element+1
endif
irow(1:element-1)=glob_row
ia=glob_row
zt(1) = exp(-glob_y**2-glob_z**2)*(-val(element))
else
val(element)=-b1(glob_x,glob_y,glob_z)&
& -a1(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-2)*idim*idim+(y-1)*idim+(z)
element=element+1
endif
! term depending on (x,y-1,z)
if (y==1) then
val(element)=-b2(glob_x,glob_y,glob_z)&
& -a2(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
zt(1) = exp(-glob_y**2-glob_z**2)*exp(-glob_x)*(-val(element))
else
val(element)=-b2(glob_x,glob_y,glob_z)&
& -a2(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-1)*idim*idim+(y-2)*idim+(z)
element=element+1
endif
! term depending on (x,y,z-1)
if (z==1) then
val(element)=-b3(glob_x,glob_y,glob_z)&
& -a3(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
zt(1) = exp(-glob_y**2-glob_z**2)*exp(-glob_x)*(-val(element))
else
val(element)=-b3(glob_x,glob_y,glob_z)&
& -a3(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-1)*idim*idim+(y-1)*idim+(z-1)
element=element+1
endif
! term depending on (x,y,z)
val(element)=2*b1(glob_x,glob_y,glob_z)&
& +2*b2(glob_x,glob_y,glob_z)&
& +2*b3(glob_x,glob_y,glob_z)&
& +a1(glob_x,glob_y,glob_z)&
& +a2(glob_x,glob_y,glob_z)&
& +a3(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-1)*idim*idim+(y-1)*idim+(z)
element=element+1
! term depending on (x,y,z+1)
if (z==idim) then
val(element)=-b1(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
zt(1) = exp(-glob_y**2-glob_z**2)*exp(-glob_x)*(-val(element))
else
val(element)=-b1(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-1)*idim*idim+(y-1)*idim+(z+1)
element=element+1
endif
! term depending on (x,y+1,z)
if (y==idim) then
val(element)=-b2(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
zt(1) = exp(-glob_y**2-glob_z**2)*exp(-glob_x)*(-val(element))
else
val(element)=-b2(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x-1)*idim*idim+(y)*idim+(z)
element=element+1
endif
! term depending on (x+1,y,z)
if (x<idim) then
val(element)=-b3(glob_x,glob_y,glob_z)
val(element) = val(element)/(deltah*&
& deltah)
icol(element)=(x)*idim*idim+(y-1)*idim+(z)
element=element+1
endif
irow(1:element-1)=glob_row
ia=glob_row
t3 = psb_wtime()
call psb_spins(element-1,irow,icol,val,a,desc_a,info)
if(info /= 0) exit
tins = tins + (psb_wtime()-t3)
call psb_geins(1,(/ia/),zt(1:1),b,desc_a,info)
if(info /= 0) exit
zt(1)=0.d0
call psb_geins(1,(/ia/),zt(1:1),xv,desc_a,info)
if(info /= 0) exit
end if
end do
t3 = psb_wtime()
call psb_spins(element-1,irow,icol,val,a,desc_a,info)
if(info /= 0) exit
tins = tins + (psb_wtime()-t3)
call psb_geins(1,(/ia/),zt(1:1),b,desc_a,info)
if(info /= 0) exit
zt(1)=0.d0
call psb_geins(1,(/ia/),zt(1:1),xv,desc_a,info)
if(info /= 0) exit
end if
end do
call psb_barrier(ictxt)