Description
Description
SLARAN and SLARUV handle the case of X(I) = 1.0 differently and so they may produce a different sequence of random numbers. They should be the same.
In SLARAN, a value of 1.0 is handled by advancing the seed. In SLARUV, the seed values, which are stored in I1, I2, I3, and I4, are advanced by 2:
I1 = I1 + 2
I2 = I2 + 2
I3 = I3 + 2
I4 = I4 + 2
To be consistent with SLARAN, they should be given the values
I1 = IT1
I2 = IT2
I3 = IT3
I4 = IT4
The same issue affects DLARUV as well.
A seed that triggers regenerating the random number is
(2129, 3241, 0, 1). The attached test program with the current
LAPACK code generates
% ./slarantst
Enter n
10
Enter ISEED:
2129 3241 0 1
i SLARAN SLARNV
1 0.3970 0.8683
2 0.9163 0.4626
3 0.4606 0.7070
4 0.8368 0.7019
5 0.4997 0.8701
6 0.3226 0.8627
7 0.2293 0.3829
8 0.6391 0.7521
9 0.8207 0.3230
10 0.6012 0.6774
The expected result is
Enter ISEED:
2129 3241 0 1
i SLARAN SLARNV
1 0.3970 0.3970
2 0.9163 0.9163
3 0.4606 0.4606
4 0.8368 0.8368
5 0.4997 0.4997
6 0.3226 0.3226
7 0.2293 0.2293
8 0.6391 0.6391
9 0.8207 0.8207
10 0.6012 0.6012
program rantst
integer i, j, k, l, n
integer iseed(4)
real x(128), y(128)
real SLARAN
iseed(1) = 0
iseed(2) = 0
iseed(3) = 0
iseed(4) = 1
write(,) 'Enter n'
read(,) n
n = min(128,n)
do while (n > 0)
write(,) 'Enter ISEED:'
read(,) i, j, k, l
!
! Set x from SLARAN
!
iseed(1) = i
iseed(2) = j
iseed(3) = k
iseed(4) = l
do idx = 1, n
x(idx) = SLARAN( iseed )
end do
!
! Set y from SLARNV
!
iseed(1) = i
iseed(2) = j
iseed(3) = k
iseed(4) = l
call SLARNV( 1, iseed, n, y )
!
! Print the results
!
write(,) ' i SLARAN SLARNV'
do idx = 1, n
write(*,'(i3,3x,f6.4,3x,f6.4)') idx, x(idx), y(idx)
end do
end do
end program
Checklist
- I've included a minimal example to reproduce the issue
- I'd be willing to make a PR to solve this issue