This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [Fortran] Another round of WHERE statement optimizations
- From: Paul Thomas <paulthomas2 at wanadoo dot fr>
- To: Roger Sayle <roger at eyesopen dot com>, Brooks Moses <bmoses at stanford dot edu>
- Cc: fortran at gcc dot gnu dot org, gcc-patches at gcc dot gnu dot org, Steve Kargl <sgk at troutmask dot apl dot washington dot edu>
- Date: Fri, 17 Feb 2006 18:29:59 +0100
- Subject: Re: [Fortran] Another round of WHERE statement optimizations
- References: <Pine.LNX.4.44.0602161347130.27386-100000@www.eyesopen.com>
Roger (and Brooks),
The patch is OK. All the testscases need their DejaGnu { dg-do run }'s
of course.
Brooks' tests are a useful addition; especially since the last one fails
(and is wrong, I think).
The failure occurs in the 1st half of the test and is present in 4.1:
! Check to ensure result is calculated from unmodified
! version of the right-hand-side in WHERE statements.
program where_19_reduced
integer :: a(4)
integer :: b(3)
integer :: c(3)
equivalence (a(1), b(1)), (a(2), c(1))
a = (/1, 2, 3, 4/)
where (b .gt. 1) ! applies to b(2) and b(3)
c = b ! so a(3) = c(2) = b(2) = 2 & a(4) = c(3) = b(3) = 3
endwhere ! => a = (/1,2,2,3/) which is what ifort gives.
if (any (a .ne. (/1, 1, 2, 3/))) & ! condition is wrong
print *, "#1", a ! gfortran gives a = (/1,2,2,2/) - no temporary for a
! equivalence of c and b, thru' a should be detected.
end program
This is a failure of dependency analysis in gfortran for equivalence
group objects (PR, Andrew?) that is already present and has been reported.
Since this is another problem, I think that the patch is OK for trunk,
with an XFAIL on a corrected where_19.f90
Paul T