gfortran.dg/subref_array_pointer_2.f90 fails on Intel Darwin9 at all level of
optimization. The following reduced case shows the origin of the problem:
! { dg-do run }
! Test the fix for PRs29396, 29606, 30625 and 30871, in which pointers
! to arrays with subreferences did not work.
!
type :: t
real :: r
integer :: i
character(3) :: chr
end type t
type :: s
type(t), pointer :: t(:)
end type s
type(t), target :: tar1(2) = (/t(1.0, 2, "abc"), t(3.0, 4, "efg")/)
type(s), target :: tar3
real, pointer :: ptr3(:)
!_______________something more complicated___________
tar3%t => tar1
ptr3 => tar3%t%r
ptr3 = cos (ptr3)
print *, ptr3
print *, (/cos(1.0_4), cos(3.0_4)/)
print *, any (ptr3 .ne. (/cos(1.0_4), cos(3.0_4)/))
if (any (abs(tr3 - (/cos(1.0_4), cos(3.0_4)/)) < epsilon(1.0_4))) call abort
()
! if (any (ptr3 .ne. (/cos(1.0_4), cos(3.0_4)/))) call abort ()
end
It succeeds as above, but fails with the last if uncommented because the value
of 'ptr3' corresponding to cos(1.0_4) differs by one lsb.
0.54030234 -0.98999250
0.54030228 -0.98999250
T
I have run the following test on OSX 10.4 and 10.5
integer :: i, n
real(4) :: x, y
real(8) :: c
n = 24
x = acos(-1.0)/2.0**n
y = 0
do i = 1, 2**n
c = cos(real(i*x,8))
y =max(y, abs(c-cos(i*x))/spacing(real(c)))
end do
print *, y
end
y time
10.4 0.499..97 2.71s
10.5 0.84559828 1.89s
So the new library seems slightly less accurate, but faster than
the old one. This cold not probably considered as a bug, rather
a feature. This support some whishes to have OS independent libraries.
I think this could be fix by a fuzzy comparison for both real(4) and real(8),
even if the test succeeds in the later case.
--
Summary: gfortran.dg/subref_array_pointer_2.f90 fails on Intel
Darwin9
Product: gcc
Version: 4.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dominiq at lps dot ens dot fr
GCC build triplet: i686-apple-darwin9
GCC host triplet: i686-apple-darwin9
GCC target triplet: i686-apple-darwin9
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34139
