Using today's trunk with http://gcc.gnu.org/ml/fortran/2010-08/msg00003.html
applied.
The following example were extracted from FLEUR (www.flapw.de), where
-Warray-temporaries showed that a temporary has been generated.
!-- ONE ----------------------
subroutine t1(n1,n2, gfft, ufft)
implicit none
integer :: n1, n2, i
real :: gfft(n1,n2), ufft(n2)
DO i=1, n1
gfft(i,:)=gfft(i,:)*ufft(i) ! No temporary needed
END DO
end subroutine
!-- TWO ----------------------
we(1:noccbd) = we(n_start:n_end)
If "1" is the (known) lower bound, no temporary is needed (if one copies from
the lower to the upper bound). Using memmove also no temporary is needed (as it
allows for this) - using memcpy this is not possible (memory needs to be
disjunct). For this case of having on the RHS only a variable, one should
probably do a middle-end assignment with ARRAY_RANGE_REF. Cf. PR 40598.
!-- THREE ----------------------
wannint(pw1,pw2,pw3,:)=wannint(pw1,pw2,pw3,:)+wannz2(j,:)
gets a temporary generated (no pointer - and no target with argument aliasing
involved.)
--
Summary: Unneccessary temporaries
Product: gcc
Version: 4.6.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: fortran
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: burnus at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45159
