10
0
mirror of https://github.com/LCPQ/quantum_package synced 2024-11-05 13:43:57 +01:00
quantum_package/src/Utils/util.irp.f

397 lines
8.0 KiB
Fortran
Raw Normal View History

double precision function binom_func(i,j)
2014-04-07 20:01:30 +02:00
implicit none
BEGIN_DOC
!.. math ::
!
! \frac{i!}{j!(i-j)!}
!
END_DOC
integer,intent(in) :: i,j
2014-10-16 23:13:38 +02:00
double precision :: logfact
2014-04-07 20:01:30 +02:00
integer, save :: ifirst
double precision, save :: memo(0:15,0:15)
!DEC$ ATTRIBUTES ALIGN : $IRP_ALIGN :: memo
integer :: k,l
if (ifirst == 0) then
ifirst = 1
do k=0,15
do l=0,15
2014-10-16 23:13:38 +02:00
memo(k,l) = dexp( logfact(k)-logfact(l)-logfact(k-l) )
2014-04-07 20:01:30 +02:00
enddo
enddo
2014-04-07 20:01:30 +02:00
endif
if ( (i<=15).and.(j<=15) ) then
binom_func = memo(i,j)
else
2014-10-16 23:13:38 +02:00
binom_func = dexp( logfact(i)-logfact(j)-logfact(i-j) )
2014-04-07 20:01:30 +02:00
endif
end
2014-10-06 15:49:16 +02:00
BEGIN_PROVIDER [ double precision, binom, (0:40,0:40) ]
&BEGIN_PROVIDER [ double precision, binom_transp, (0:40,0:40) ]
2014-04-07 20:01:30 +02:00
implicit none
BEGIN_DOC
! Binomial coefficients
END_DOC
integer :: k,l
2014-10-16 23:13:38 +02:00
double precision :: logfact
2014-10-06 15:49:16 +02:00
do k=0,40
do l=0,40
2014-10-16 23:13:38 +02:00
binom(k,l) = dexp( logfact(k)-logfact(l)-logfact(k-l) )
2014-04-07 20:01:30 +02:00
binom_transp(l,k) = binom(k,l)
enddo
enddo
END_PROVIDER
integer function align_double(n)
2014-04-07 20:01:30 +02:00
implicit none
BEGIN_DOC
! Compute 1st dimension such that it is aligned for vectorization.
END_DOC
integer :: n
2015-06-04 16:40:00 +02:00
include 'constants.include.F'
2014-04-07 20:01:30 +02:00
if (mod(n,SIMD_vector/4) /= 0) then
align_double= n + SIMD_vector/4 - mod(n,SIMD_vector/4)
else
align_double= n
endif
end
double precision function fact(n)
implicit none
2014-04-07 20:01:30 +02:00
BEGIN_DOC
! n!
END_DOC
integer :: n
double precision, save :: memo(1:100)
integer, save :: memomax = 1
if (n<=memomax) then
if (n<2) then
fact = 1.d0
else
fact = memo(n)
endif
return
endif
2014-04-07 20:01:30 +02:00
integer :: i
memo(1) = 1.d0
do i=memomax+1,min(n,100)
2014-10-16 23:13:38 +02:00
memo(i) = memo(i-1)*dble(i)
enddo
memomax = min(n,100)
2016-09-30 15:34:59 +02:00
double precision :: logfact
fact = dexp(logfact(n))
2014-10-16 23:13:38 +02:00
end function
double precision function logfact(n)
implicit none
BEGIN_DOC
! n!
END_DOC
integer :: n
double precision, save :: memo(1:100)
integer, save :: memomax = 1
if (n<=memomax) then
if (n<2) then
logfact = 0.d0
else
logfact = memo(n)
endif
return
endif
integer :: i
memo(1) = 0.d0
do i=memomax+1,min(n,100)
memo(i) = memo(i-1)+dlog(dble(i))
enddo
memomax = min(n,100)
logfact = memo(memomax)
do i=101,n
logfact += dlog(dble(i))
enddo
end function
BEGIN_PROVIDER [ double precision, fact_inv, (128) ]
2014-04-07 20:01:30 +02:00
implicit none
BEGIN_DOC
! 1/n!
END_DOC
integer :: i
double precision :: fact
do i=1,size(fact_inv)
fact_inv(i) = 1.d0/fact(i)
enddo
END_PROVIDER
2015-05-05 19:13:32 +02:00
double precision function dble_fact(n)
implicit none
integer :: n
2015-05-11 14:54:43 +02:00
double precision :: dble_fact_even, dble_fact_odd
2015-05-05 19:13:32 +02:00
dble_fact = 1.d0
if(n.lt.0) return
if(iand(n,1).eq.0)then
2015-05-11 14:54:43 +02:00
dble_fact = dble_fact_even(n)
2015-05-05 19:13:32 +02:00
else
dble_fact= dble_fact_odd(n)
endif
end function
2015-05-11 14:54:43 +02:00
double precision function dble_fact_even(n) result(fact2)
2015-05-05 19:13:32 +02:00
implicit none
BEGIN_DOC
! n!!
END_DOC
integer :: n,k
2016-10-06 17:39:15 +02:00
double precision, save :: memo(0:100)
integer, save :: memomax = 0
2015-05-05 19:13:32 +02:00
double precision :: prod
ASSERT (iand(n,1) /= 1)
2016-10-06 17:39:15 +02:00
! prod=1.d0
! do k=2,n,2
! prod=prod*dfloat(k)
! enddo
! fact2=prod
! return
!
2016-09-30 15:34:59 +02:00
if (n <= memomax) then
if (n < 2) then
fact2 = 1.d0
else
fact2 = memo(n)
endif
return
endif
integer :: i
2016-10-06 17:39:15 +02:00
memo(0)=1.d0
2016-09-30 15:34:59 +02:00
memo(1)=1.d0
do i=memomax+2,min(n,100),2
memo(i) = memo(i-2)* dble(i)
2015-05-05 19:13:32 +02:00
enddo
2016-09-30 15:34:59 +02:00
memomax = min(n,100)
fact2 = memo(memomax)
if (n > 100) then
double precision :: dble_logfact
fact2 = dexp(dble_logfact(n))
endif
2015-05-05 19:13:32 +02:00
end function
double precision function dble_fact_odd(n) result(fact2)
implicit none
2014-04-07 20:01:30 +02:00
BEGIN_DOC
! n!!
END_DOC
integer :: n
double precision, save :: memo(1:100)
integer, save :: memomax = 1
ASSERT (iand(n,1) /= 0)
if (n<=memomax) then
if (n<3) then
fact2 = 1.d0
else
fact2 = memo(n)
endif
return
endif
2014-04-07 20:01:30 +02:00
integer :: i
memo(1) = 1.d0
do i=memomax+2,min(n,99),2
2014-10-16 23:13:38 +02:00
memo(i) = memo(i-2)* dble(i)
enddo
memomax = min(n,99)
fact2 = memo(memomax)
2014-04-07 20:01:30 +02:00
2014-10-16 23:13:38 +02:00
if (n > 99) then
double precision :: dble_logfact
fact2 = dexp(dble_logfact(n))
endif
end function
double precision function dble_logfact(n) result(logfact2)
implicit none
BEGIN_DOC
! n!!
END_DOC
integer :: n
2017-04-21 22:59:42 +02:00
integer :: k
double precision :: prod
prod=0.d0
do k=2,n,2
prod=prod+dlog(dfloat(k))
enddo
2017-04-21 22:59:42 +02:00
logfact2=prod
return
2014-04-07 20:01:30 +02:00
end function
subroutine write_git_log(iunit)
2014-04-07 20:01:30 +02:00
implicit none
BEGIN_DOC
! Write the last git commit in file iunit.
END_DOC
integer, intent(in) :: iunit
write(iunit,*) '----------------'
write(iunit,*) 'Last git commit:'
BEGIN_SHELL [ /bin/bash ]
git log -1 2>/dev/null | sed "s/'//g"| sed "s/^/ write(iunit,*) '/g" | sed "s/$/'/g" || echo "Unknown"
END_SHELL
write(iunit,*) '----------------'
end
BEGIN_PROVIDER [ double precision, inv_int, (128) ]
2014-04-07 20:01:30 +02:00
implicit none
BEGIN_DOC
! 1/i
END_DOC
integer :: i
2014-05-21 22:41:26 +02:00
do i=1,128
2014-04-07 20:01:30 +02:00
inv_int(i) = 1.d0/dble(i)
enddo
END_PROVIDER
subroutine wall_time(t)
implicit none
2014-04-07 20:01:30 +02:00
BEGIN_DOC
! The equivalent of cpu_time, but for the wall time.
END_DOC
double precision, intent(out) :: t
integer :: c
integer, save :: rate = 0
if (rate == 0) then
CALL SYSTEM_CLOCK(count_rate=rate)
endif
CALL SYSTEM_CLOCK(count=c)
t = dble(c)/dble(rate)
end
BEGIN_PROVIDER [ integer, nproc ]
2014-04-07 20:01:30 +02:00
implicit none
BEGIN_DOC
! Number of current OpenMP threads
END_DOC
integer :: omp_get_num_threads
nproc = 1
!$OMP PARALLEL
!$OMP MASTER
!$ nproc = omp_get_num_threads()
!$OMP END MASTER
!$OMP END PARALLEL
END_PROVIDER
2014-05-14 00:01:31 +02:00
double precision function u_dot_v(u,v,sze)
implicit none
BEGIN_DOC
! Compute <u|v>
END_DOC
integer, intent(in) :: sze
double precision, intent(in) :: u(sze),v(sze)
2016-09-24 02:03:22 +02:00
double precision, external :: ddot
2014-05-14 00:01:31 +02:00
2016-09-24 02:03:22 +02:00
!DIR$ FORCEINLINE
u_dot_v = ddot(sze,u,1,v,1)
2014-05-14 00:01:31 +02:00
end
double precision function u_dot_u(u,sze)
implicit none
BEGIN_DOC
! Compute <u|u>
END_DOC
integer, intent(in) :: sze
double precision, intent(in) :: u(sze)
2016-09-24 02:03:22 +02:00
double precision, external :: ddot
2014-05-14 00:01:31 +02:00
2016-09-24 02:03:22 +02:00
!DIR$ FORCEINLINE
u_dot_u = ddot(sze,u,1,u,1)
2014-05-14 00:01:31 +02:00
end
subroutine normalize(u,sze)
implicit none
BEGIN_DOC
! Normalizes vector u
! u is expected to be aligned in memory.
END_DOC
integer, intent(in) :: sze
double precision, intent(inout):: u(sze)
double precision :: d
2016-09-24 02:03:22 +02:00
double precision, external :: dnrm2
2014-05-14 00:01:31 +02:00
integer :: i
!DIR$ FORCEINLINE
2016-09-24 02:03:22 +02:00
d = dnrm2(sze,u,1)
if (d /= 0.d0) then
2016-09-24 02:03:22 +02:00
d = 1.d0/d
endif
2014-05-14 00:01:31 +02:00
if (d /= 1.d0) then
2016-09-24 02:03:22 +02:00
!DIR$ FORCEINLINE
call dscal(sze,d,u,1)
2014-05-14 00:01:31 +02:00
endif
end
2014-07-09 22:44:42 +02:00
double precision function approx_dble(a,n)
implicit none
integer, intent(in) :: n
double precision, intent(in) :: a
double precision :: f
integer :: i
if (a == 0.d0) then
approx_dble = 0.d0
return
endif
f = 1.d0
do i=1,-int(dlog10(dabs(a)))+n
f = f*.1d0
enddo
do i=1,int(dlog10(dabs(a)))-n
f = f*10.d0
enddo
approx_dble = dnint(a/f)*f
end
2014-05-14 00:01:31 +02:00
2016-02-19 00:20:28 +01:00
subroutine lowercase(txt,n)
implicit none
BEGIN_DOC
! Transform to lower case
END_DOC
character*(*), intent(inout) :: txt
integer, intent(in) :: n
character( * ), PARAMETER :: LOWER_CASE = 'abcdefghijklmnopqrstuvwxyz'
character( * ), PARAMETER :: UPPER_CASE = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
integer :: i, ic
do i=1,n
ic = index( UPPER_CASE, txt(i:i) )
if (ic /= 0) then
txt(i:i) = LOWER_CASE(ic:ic)
endif
enddo
end
2014-05-14 00:01:31 +02:00