libquad question

[Dominique Dhumieres]

Steve,

I am not sure to understand your problem. With the following
modification of your code:

integer, parameter :: k1 = selected_real_kind (precision (0.0))
integer, parameter :: k2 = max(k1, selected_real_kind (precision (0.0_k1) + 1))
integer, parameter :: k3 = max(k2, selected_real_kind (precision (0.0_k2) + 1))
integer, parameter :: k4 = max(k3, selected_real_kind (precision (0.0_k3) + 1))
integer, parameter :: k5 = max(k4, selected_real_kind (precision (0.0_k4) + 1))
real(k1) x4
real(k2) x8
real(k3) x10
real(k4) x16
real(k5) x32

!print *, k1, k2, k3, k4, k5
print *, 'k1 =', k1, selected_real_kind (precision (0.0))
print *, 'k2 =', k2, selected_real_kind (precision (0.0_k1) + 1)
print *, 'k3 =', k3, selected_real_kind (precision (0.0_k2) + 1)
print *, 'k4 =', k4, selected_real_kind (precision (0.0_k3) + 1)
print *, 'k5 =', k5, selected_real_kind (precision (0.0_k4) + 1)
print *
print *, 'x4 =', precision(x4), range(x4)
print *, 'x8 =', precision(x8), range(x8)
print *, 'xk3=', precision(x10), range(x10)
print *, 'xk4=', precision(x16), range(x16)
print *, 'xk5=', precision(x32), range(x32)

end

without the libquad patch I get for x86_64-apple-darwin10.4.0

[macbook] f90/bug% a.out
 k1 =           4           4
 k2 =           8           8
 k3 =          10          10
 k4 =          10          -1
 k5 =          10          -1

 x4 =           6          37
 x8 =          15         307
 xk3=          18        4931
 xk4=          18        4931
 xk5=          18        4931

and for powerpc-apple-darwin9

[karma] f90/bug% a.out
 k1 =           4           4
 k2 =           8           8
 k3 =          16          16
 k4 =          16          -1
 k5 =          16          -1

 x4 =           6          37
 x8 =          15         307
 xk3=          31         291
 xk4=          31         291
 xk5=          31         291

With the libquad patch on x86_64-apple-darwin10.4.0 I get

[macbook] f90/bug% a.out
 k1 =           4           4
 k2 =           8           8
 k3 =          10          10
 k4 =          16          16
 k5 =          16          -1

 x4 =           6          37
 x8 =          15         307
 xk3=          18        4931
 xk4=          33        4931
 xk5=          33        4931

i.e., the libquad patch adds REAL(16) to REAL(4), REAL(8), and REAL(10),
the later being unavailable on PPC for which REAL(16) is already
provided by the so called IBM format (in fact two REAL(8) glued together).

Cheers,

Dominique