Re: [FORTRAN PATCH] PR 30404: Rewrite nested FORALL mask expansion

[Steve Kargl <sgk at troutmask dot apl dot washington dot edu>]

On Sat, Jan 13, 2007 at 04:04:05PM -0700, Roger Sayle wrote:
> 
> If someone could do the usual benchmarking and conformance testing,
> I'd be interested to know if there's a positive or negative impact.
> 

Roger, I scanned my work/ directory where I store the Fortran
code I've snarfed from the internet.  With over million lines of
code, there is exact ZERO ocurrences of FORALL. :(  work/ includes
the Polyhedron Benchmarks, mpfun90, and moware as well as a few
other modern Fortran codes.

I asked for examples on comp.lang.fortran, and Mike Metcalf posted
a small program (that is from his book).  Sorry about the long URL.

http://groups-beta.google.com/group/comp.lang.fortran/browse_thread/thread/9d261ff21ebf683c/fc5599887199f26b?hl=en#fc5599887199f26b

Your patch has no negative impact on runtime with Mike's code.
In fact, this code contains a RESHAPE command, which appears 
to be also quadratic in time and accounts for over 99% of the
compile and execution times.

I think the patch is OK for trunk.  See below for one change.

PS: I'm still trying to come up with additional codes that use FORALL.

> ! PR fortran/30404
> ! Checks that we correctly handle nested masks in nested FORALL blocks.
> ! Contributed by Paul Thomas  <pault@gcc.gnu.org>
> !
> ! { dg-do run }
>   logical :: l1(2,2) = reshape ((/.false.,.true.,.true.,.false./), (/2,2/))
>   integer :: it(2,2) = reshape ((/1,2,3,4/), (/2,2/))

gfortran issues a warning with the above declarations.  You need
to change this to

   logical :: l1(2,2)
   integer :: it(2,2)
   l1 = reshape ((/.false.,.true.,.true.,.false./), (/2,2/))
   it = reshape ((/1,2,3,4/), (/2,2/))

>   forall (i = 1:2, i < 3)
>     forall (j = 1:2, l1(i,j))
>       it(i, j) = 0
>     end forall
>   end forall
> !  print *, l1
> !  print '(4i2)', it
>   if (any (it .ne. reshape ((/1, 0, 0, 4/), (/2, 2/)))) call abort ()
> end