mirror of
https://github.com/QuantumPackage/qp2.git
synced 2024-11-09 06:53:38 +01:00
Use integer*8 in cholesky
This commit is contained in:
parent
0a3d462510
commit
b743201efe
@ -27,19 +27,22 @@ END_PROVIDER
|
|||||||
! https://doi.org/10.1016/j.apnum.2011.10.001 : Page 4, Algorithm 1
|
! https://doi.org/10.1016/j.apnum.2011.10.001 : Page 4, Algorithm 1
|
||||||
END_DOC
|
END_DOC
|
||||||
|
|
||||||
integer :: rank, ndim
|
integer*8 :: ndim8
|
||||||
|
integer :: rank
|
||||||
double precision :: tau
|
double precision :: tau
|
||||||
double precision, pointer :: L(:,:), L_old(:,:)
|
double precision, pointer :: L(:,:), L_old(:,:)
|
||||||
|
|
||||||
|
|
||||||
double precision :: s
|
double precision :: s
|
||||||
double precision, parameter :: dscale = 1.d0
|
double precision :: dscale
|
||||||
|
|
||||||
double precision, allocatable :: D(:), Delta(:,:), Ltmp_p(:,:), Ltmp_q(:,:)
|
double precision, allocatable :: D(:), Delta(:,:), Ltmp_p(:,:), Ltmp_q(:,:)
|
||||||
integer, allocatable :: Lset(:), Dset(:), addr(:,:)
|
integer, allocatable :: addr1(:,:), addr2(:,:)
|
||||||
|
integer**, allocatable :: Lset(:), Dset(:), addr3(:,:)
|
||||||
logical, allocatable :: computed(:)
|
logical, allocatable :: computed(:)
|
||||||
|
|
||||||
integer :: i,j,k,m,p,q, qj, dj, p2, q2
|
integer :: i,j,k,m,p,q, dj, p2, q2
|
||||||
|
integer*8 :: i8, j8, p8, qj8
|
||||||
integer :: N, np, nq
|
integer :: N, np, nq
|
||||||
|
|
||||||
double precision :: Dmax, Dmin, Qmax, f
|
double precision :: Dmax, Dmin, Qmax, f
|
||||||
@ -47,15 +50,15 @@ END_PROVIDER
|
|||||||
logical, external :: ao_two_e_integral_zero
|
logical, external :: ao_two_e_integral_zero
|
||||||
|
|
||||||
double precision, external :: ao_two_e_integral
|
double precision, external :: ao_two_e_integral
|
||||||
integer :: block_size, iblock, ierr
|
integer :: block_size, iblock
|
||||||
|
|
||||||
double precision :: mem
|
double precision :: mem
|
||||||
double precision, external :: memory_of_double, memory_of_int
|
double precision, external :: memory_of_double8, memory_of_int8
|
||||||
|
|
||||||
integer, external :: getUnitAndOpen
|
integer, external :: getUnitAndOpen
|
||||||
integer :: iunit
|
integer :: iunit, ierr
|
||||||
|
|
||||||
ndim = ao_num*ao_num
|
ndim8 = ao_num*ao_num*1_8
|
||||||
deallocate(cholesky_ao)
|
deallocate(cholesky_ao)
|
||||||
|
|
||||||
if (read_ao_cholesky) then
|
if (read_ao_cholesky) then
|
||||||
@ -83,13 +86,13 @@ END_PROVIDER
|
|||||||
|
|
||||||
tau = ao_cholesky_threshold
|
tau = ao_cholesky_threshold
|
||||||
|
|
||||||
mem = 6.d0 * memory_of_double(ndim) + 6.d0 * memory_of_int(ndim)
|
mem = 6.d0 * memory_of_double8(ndim8) + 6.d0 * memory_of_int8(ndim8)
|
||||||
call check_mem(mem, irp_here)
|
call check_mem(mem, irp_here)
|
||||||
|
|
||||||
call print_memory_usage()
|
call print_memory_usage()
|
||||||
|
|
||||||
allocate(L(ndim,1))
|
allocate(L(ndim8,1))
|
||||||
!print *, 'allocate : (L(ndim,1))', memory_of_double(ndim)
|
print *, 'allocate : (L(ndim8,1))', memory_of_double8(ndim8)
|
||||||
|
|
||||||
print *, ''
|
print *, ''
|
||||||
print *, 'Cholesky decomposition of AO integrals'
|
print *, 'Cholesky decomposition of AO integrals'
|
||||||
@ -102,36 +105,36 @@ END_PROVIDER
|
|||||||
|
|
||||||
rank = 0
|
rank = 0
|
||||||
|
|
||||||
allocate( D(ndim), Lset(ndim), Dset(ndim) )
|
allocate( D(ndim8), Lset(ndim8), Dset(ndim8) )
|
||||||
allocate( addr(3,ndim) )
|
allocate( addr1(ndim8), addr2(ndim8), addr3(ndim8), )
|
||||||
!print *, 'allocate : (D(ndim))', memory_of_int(ndim)
|
print *, 'allocate : (D(ndim8))', memory_of_int8(ndim8)
|
||||||
!print *, 'allocate : (Lset(ndim))', memory_of_int(ndim)
|
print *, 'allocate : (Lset(ndim8))', memory_of_int8(ndim8)
|
||||||
!print *, 'allocate : (Dset(ndim))', memory_of_int(ndim)
|
print *, 'allocate : (Dset(ndim8))', memory_of_int8(ndim8)
|
||||||
!print *, 'allocate : (3,addr(ndim))', memory_of_int(3*ndim)
|
print *, 'allocate : (4,addr(ndim8))', memory_of_int8(4_8*ndim8)
|
||||||
|
|
||||||
! 1.
|
! 1.
|
||||||
k=0
|
k=0
|
||||||
do j=1,ao_num
|
do j=1,ao_num
|
||||||
do i=1,ao_num
|
do i=1,ao_num
|
||||||
k = k+1
|
k = k+1
|
||||||
addr(1,k) = i
|
addr1(k) = i
|
||||||
addr(2,k) = j
|
addr2(k) = j
|
||||||
addr(3,k) = (i-1)*ao_num + j
|
addr3(k) = (i-1)*ao_num + j
|
||||||
enddo
|
enddo
|
||||||
enddo
|
enddo
|
||||||
|
|
||||||
if (do_direct_integrals) then
|
if (do_direct_integrals) then
|
||||||
!$OMP PARALLEL DO DEFAULT(SHARED) PRIVATE(i) SCHEDULE(guided)
|
!$OMP PARALLEL DO DEFAULT(SHARED) PRIVATE(i8) SCHEDULE(guided)
|
||||||
do i=1,ndim
|
do i8=1,ndim8
|
||||||
D(i) = ao_two_e_integral(addr(1,i), addr(2,i), &
|
D(i8) = ao_two_e_integral(addr1(i8), addr2(i8), &
|
||||||
addr(1,i), addr(2,i))
|
addr1(i8), addr2(i8))
|
||||||
enddo
|
enddo
|
||||||
!$OMP END PARALLEL DO
|
!$OMP END PARALLEL DO
|
||||||
else
|
else
|
||||||
!$OMP PARALLEL DO DEFAULT(SHARED) PRIVATE(i) SCHEDULE(guided)
|
!$OMP PARALLEL DO DEFAULT(SHARED) PRIVATE(i8) SCHEDULE(guided)
|
||||||
do i=1,ndim
|
do i8=1,ndim8
|
||||||
D(i) = get_ao_two_e_integral(addr(1,i), addr(1,i), &
|
D(i8) = get_ao_two_e_integral(addr1(i8), addr1(i8), &
|
||||||
addr(2,i), addr(2,i), &
|
addr2(i8), addr2(i8), &
|
||||||
ao_integrals_map)
|
ao_integrals_map)
|
||||||
enddo
|
enddo
|
||||||
!$OMP END PARALLEL DO
|
!$OMP END PARALLEL DO
|
||||||
@ -140,12 +143,21 @@ END_PROVIDER
|
|||||||
Dmax = maxval(D)
|
Dmax = maxval(D)
|
||||||
|
|
||||||
! 2.
|
! 2.
|
||||||
|
np = huge(1_4)
|
||||||
|
dscale = 1.d0
|
||||||
|
do while (np == huge(1_4))
|
||||||
np=0
|
np=0
|
||||||
do p=1,ndim
|
do p8=1,ndim8
|
||||||
if ( dscale*dscale*Dmax*D(p) > tau*tau ) then
|
if ( dscale*dscale*Dmax*D(p8) > tau*tau ) then
|
||||||
np = np+1
|
np = np+1
|
||||||
Lset(np) = p
|
Lset(np) = p8
|
||||||
|
if (np == huge(1_4)) then
|
||||||
|
! Overflow detected
|
||||||
|
dscale = dscale*0.5d0
|
||||||
|
exit
|
||||||
endif
|
endif
|
||||||
|
endif
|
||||||
|
enddo
|
||||||
enddo
|
enddo
|
||||||
|
|
||||||
! 3.
|
! 3.
|
||||||
@ -155,7 +167,7 @@ END_PROVIDER
|
|||||||
i = 0
|
i = 0
|
||||||
|
|
||||||
! 5.
|
! 5.
|
||||||
do while ( (Dmax > tau).and.(rank < ndim) )
|
do while ( (Dmax > tau).and.(rank < min(ndim8,huge(1_4)) )
|
||||||
! a.
|
! a.
|
||||||
i = i+1
|
i = i+1
|
||||||
|
|
||||||
@ -181,7 +193,8 @@ END_PROVIDER
|
|||||||
call total_memory(mem)
|
call total_memory(mem)
|
||||||
mem = mem &
|
mem = mem &
|
||||||
+ np*memory_of_double(nq) &! Delta(np,nq)
|
+ np*memory_of_double(nq) &! Delta(np,nq)
|
||||||
+ (rank+nq)* memory_of_double(ndim) &! L(ndim,rank+nq)
|
+ (rank+nq)* memory_of_double8(ndim8)
|
||||||
|
&! L(ndim8,rank+nq)
|
||||||
+ (np+nq)*memory_of_double(block_size) ! Ltmp_p(np,block_size) + Ltmp_q(nq,block_size)
|
+ (np+nq)*memory_of_double(block_size) ! Ltmp_p(np,block_size) + Ltmp_q(nq,block_size)
|
||||||
|
|
||||||
if (mem > qp_max_mem) then
|
if (mem > qp_max_mem) then
|
||||||
@ -205,19 +218,19 @@ END_PROVIDER
|
|||||||
! d., e.
|
! d., e.
|
||||||
|
|
||||||
L_old => L
|
L_old => L
|
||||||
allocate(L(ndim,rank+nq), stat=ierr)
|
allocate(L(ndim8,rank+nq), stat=ierr)
|
||||||
!print *, 'allocate : L(ndim,rank+nq)', memory_of_double(ndim*(rank+nq))
|
print *, 'allocate : L(ndim8,rank+nq)', memory_of_double8(ndim8*(rank+nq))
|
||||||
|
|
||||||
if (ierr /= 0) then
|
if (ierr /= 0) then
|
||||||
call print_memory_usage()
|
call print_memory_usage()
|
||||||
print *, irp_here, ': allocation failed : (L(ndim,rank+nq))'
|
print *, irp_here, ': allocation failed : (L(ndim8,rank+nq))'
|
||||||
stop -1
|
stop -1
|
||||||
endif
|
endif
|
||||||
|
|
||||||
!$OMP PARALLEL DO PRIVATE(k,j)
|
!$OMP PARALLEL DO PRIVATE(k,j)
|
||||||
do k=1,rank
|
do k=1,rank
|
||||||
do j=1,ndim
|
do j8=1,ndim8
|
||||||
L(j,k) = L_old(j,k)
|
L(j8,k) = L_old(j8,k)
|
||||||
enddo
|
enddo
|
||||||
enddo
|
enddo
|
||||||
!$OMP END PARALLEL DO
|
!$OMP END PARALLEL DO
|
||||||
@ -225,7 +238,7 @@ END_PROVIDER
|
|||||||
deallocate(L_old)
|
deallocate(L_old)
|
||||||
|
|
||||||
allocate(Delta(np,nq), stat=ierr)
|
allocate(Delta(np,nq), stat=ierr)
|
||||||
!print *, 'allocate : Delta(np,nq)', memory_of_double(np*nq)
|
print *, 'allocate : Delta(np,nq)', memory_of_double8(np*nq*1_8)
|
||||||
|
|
||||||
if (ierr /= 0) then
|
if (ierr /= 0) then
|
||||||
call print_memory_usage()
|
call print_memory_usage()
|
||||||
@ -234,7 +247,7 @@ END_PROVIDER
|
|||||||
endif
|
endif
|
||||||
|
|
||||||
allocate(Ltmp_p(np,block_size), stat=ierr)
|
allocate(Ltmp_p(np,block_size), stat=ierr)
|
||||||
!print *, 'allocate : Ltmp_p(np,block_size)', memory_of_double(np*block_size)
|
print *, 'allocate : Ltmp_p(np,block_size)', memory_of_double8(np*block_size*1_8)
|
||||||
|
|
||||||
if (ierr /= 0) then
|
if (ierr /= 0) then
|
||||||
call print_memory_usage()
|
call print_memory_usage()
|
||||||
@ -243,7 +256,7 @@ END_PROVIDER
|
|||||||
endif
|
endif
|
||||||
|
|
||||||
allocate(Ltmp_q(nq,block_size), stat=ierr)
|
allocate(Ltmp_q(nq,block_size), stat=ierr)
|
||||||
!print *, 'allocate : Ltmp_q(nq,block_size)', memory_of_double(nq*block_size)
|
print *, 'allocate : Ltmp_q(nq,block_size)', memory_of_double8(nq*block_size*1_8)
|
||||||
|
|
||||||
if (ierr /= 0) then
|
if (ierr /= 0) then
|
||||||
call print_memory_usage()
|
call print_memory_usage()
|
||||||
@ -253,8 +266,9 @@ END_PROVIDER
|
|||||||
|
|
||||||
|
|
||||||
allocate(computed(nq))
|
allocate(computed(nq))
|
||||||
!print *, 'allocate : computed(nq)', memory_of_int(nq)
|
print *, 'allocate : computed(nq)', memory_of_int(nq)
|
||||||
|
|
||||||
|
print *, 'p1'
|
||||||
!$OMP PARALLEL DEFAULT(SHARED) PRIVATE(k,p,q,j)
|
!$OMP PARALLEL DEFAULT(SHARED) PRIVATE(k,p,q,j)
|
||||||
|
|
||||||
!$OMP DO
|
!$OMP DO
|
||||||
@ -296,7 +310,7 @@ END_PROVIDER
|
|||||||
iblock = 0
|
iblock = 0
|
||||||
do j=1,nq
|
do j=1,nq
|
||||||
|
|
||||||
if ( (Qmax <= Dmin).or.(N+j > ndim) ) exit
|
if ( (Qmax <= Dmin).or.(N+j*1_8 > ndim8) ) exit
|
||||||
! i.
|
! i.
|
||||||
rank = N+j
|
rank = N+j
|
||||||
|
|
||||||
@ -308,28 +322,28 @@ END_PROVIDER
|
|||||||
|
|
||||||
! ii.
|
! ii.
|
||||||
do dj=1,nq
|
do dj=1,nq
|
||||||
qj = Dset(dj)
|
qj8 = Dset(dj)
|
||||||
if (D(qj) == Qmax) then
|
if (D(qj8) == Qmax) then
|
||||||
exit
|
exit
|
||||||
endif
|
endif
|
||||||
enddo
|
enddo
|
||||||
|
|
||||||
L(1:ndim, rank) = 0.d0
|
L(1:ndim8, rank) = 0.d0
|
||||||
|
|
||||||
if (.not.computed(dj)) then
|
if (.not.computed(dj)) then
|
||||||
m = dj
|
m = dj
|
||||||
!$OMP PARALLEL DO PRIVATE(k) SCHEDULE(guided)
|
!$OMP PARALLEL DO PRIVATE(k) SCHEDULE(guided)
|
||||||
do k=np,1,-1
|
do k=np,1,-1
|
||||||
if (.not.ao_two_e_integral_zero( addr(1,Lset(k)), addr(1,Dset(m)),&
|
if (.not.ao_two_e_integral_zero( addr1(Lset(k)), addr1(Dset(m)),&
|
||||||
addr(2,Lset(k)), addr(2,Dset(m)) ) ) then
|
addr2(Lset(k)), addr2(Dset(m)) ) ) then
|
||||||
if (do_direct_integrals) then
|
if (do_direct_integrals) then
|
||||||
Delta(k,m) = Delta(k,m) + &
|
Delta(k,m) = Delta(k,m) + &
|
||||||
ao_two_e_integral(addr(1,Lset(k)), addr(2,Lset(k)),&
|
ao_two_e_integral(addr1(Lset(k)), addr2(Lset(k)),&
|
||||||
addr(1,Dset(m)), addr(2,Dset(m)))
|
addr1(Dset(m)), addr2(Dset(m)))
|
||||||
else
|
else
|
||||||
Delta(k,m) = Delta(k,m) + &
|
Delta(k,m) = Delta(k,m) + &
|
||||||
get_ao_two_e_integral( addr(1,Lset(k)), addr(1,Dset(m)),&
|
get_ao_two_e_integral( addr1(Lset(k)), addr1(Dset(m)),&
|
||||||
addr(2,Lset(k)), addr(2,Dset(m)), ao_integrals_map)
|
addr2(Lset(k)), addr2(Dset(m)), ao_integrals_map)
|
||||||
endif
|
endif
|
||||||
endif
|
endif
|
||||||
enddo
|
enddo
|
||||||
@ -391,18 +405,28 @@ END_PROVIDER
|
|||||||
Dmax = max(Dmax, D(Lset(p)))
|
Dmax = max(Dmax, D(Lset(p)))
|
||||||
enddo
|
enddo
|
||||||
|
|
||||||
|
np = huge(1_4)
|
||||||
|
dscale = 1.d0
|
||||||
|
do while (np == huge(1_4))
|
||||||
np=0
|
np=0
|
||||||
do p=1,ndim
|
do p8=1,ndim8
|
||||||
if ( dscale*dscale*Dmax*D(p) > tau*tau ) then
|
if ( dscale*dscale*Dmax*D(p8) > tau*tau ) then
|
||||||
np = np+1
|
np = np+1
|
||||||
Lset(np) = p
|
Lset(np) = p8
|
||||||
|
if (np == huge(1_4)) then
|
||||||
|
! Overflow detected
|
||||||
|
dscale = dscale*0.5d0
|
||||||
|
exit
|
||||||
|
endif
|
||||||
endif
|
endif
|
||||||
enddo
|
enddo
|
||||||
|
enddo
|
||||||
|
|
||||||
|
|
||||||
enddo
|
enddo
|
||||||
|
|
||||||
allocate(cholesky_ao(ao_num,ao_num,rank), stat=ierr)
|
allocate(cholesky_ao(ao_num,ao_num,rank), stat=ierr)
|
||||||
!print *, 'allocate : cholesky_ao(ao_num,ao_num,rank)', ao_num*ao_num*(rank*1_8) * 8_8 / 1024_8**3
|
print *, 'allocate : cholesky_ao(ao_num,ao_num,rank)', memory_of_double8(ao_num*ao_num*rank*1_8)
|
||||||
|
|
||||||
if (ierr /= 0) then
|
if (ierr /= 0) then
|
||||||
call print_memory_usage()
|
call print_memory_usage()
|
||||||
@ -411,7 +435,9 @@ END_PROVIDER
|
|||||||
endif
|
endif
|
||||||
!$OMP PARALLEL DO PRIVATE(k)
|
!$OMP PARALLEL DO PRIVATE(k)
|
||||||
do k=1,rank
|
do k=1,rank
|
||||||
call dcopy(ndim, L(1,k), 1, cholesky_ao(1,1,k), 1)
|
do j=1,ao_num
|
||||||
|
call dcopy(ao_num, L((j-1)*ao_num+1,k), 1, cholesky_ao(1,j,k), 1)
|
||||||
|
enddo
|
||||||
enddo
|
enddo
|
||||||
!$OMP END PARALLEL DO
|
!$OMP END PARALLEL DO
|
||||||
deallocate(L)
|
deallocate(L)
|
||||||
|
|||||||
@ -79,6 +79,26 @@ IRP_ENDIF
|
|||||||
call unlock_io()
|
call unlock_io()
|
||||||
end function
|
end function
|
||||||
|
|
||||||
|
double precision function memory_of_double8(n)
|
||||||
|
implicit none
|
||||||
|
BEGIN_DOC
|
||||||
|
! Computes the memory required for n double precision elements in gigabytes.
|
||||||
|
END_DOC
|
||||||
|
integer*8, intent(in) :: n
|
||||||
|
double precision, parameter :: f = 8.d0 / (1024.d0*1024.d0*1024.d0)
|
||||||
|
memory_of_double8 = dble(n) * f
|
||||||
|
end function
|
||||||
|
|
||||||
|
double precision function memory_of_int8(n)
|
||||||
|
implicit none
|
||||||
|
BEGIN_DOC
|
||||||
|
! Computes the memory required for n double precision elements in gigabytes.
|
||||||
|
END_DOC
|
||||||
|
integer*8, intent(in) :: n
|
||||||
|
double precision, parameter :: f = 4.d0 / (1024.d0*1024.d0*1024.d0)
|
||||||
|
memory_of_int8 = dble(n) * f
|
||||||
|
end function
|
||||||
|
|
||||||
double precision function memory_of_double(n)
|
double precision function memory_of_double(n)
|
||||||
implicit none
|
implicit none
|
||||||
BEGIN_DOC
|
BEGIN_DOC
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user