[Bug fortran/66158] Use DO loop constraints for optimizing bounds checking

[tkoenig at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66158

Thomas Koenig <tkoenig at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW

--- Comment #2 from Thomas Koenig <tkoenig at gcc dot gnu.org> ---
(In reply to Dominique d'Humieres from comment #1)
> Bounds checking is rather cheap. Does the proposed optimization worth the
> work?

It can be quite significant.  An example with recent trunk:

ig25@linux-fd1f:~/Krempel/Matmul/Bounds> cat bounds.f90
program main
  use foo
  implicit none
  real, dimension(10,10) :: a, b, c
  integer :: i

  call random_number(a)
  call random_number(b)
  do i=1,1000000
     call my_matmul(a,b,c)
  end do
end program main
ig25@linux-fd1f:~/Krempel/Matmul/Bounds> cat foo.f90
module foo
  implicit none
contains
  subroutine my_matmul(a,b,c)
    real, dimension(:,:), intent(in) :: a, b
    real, dimension(:,:), intent(inout) :: c
    c = matmul(a,b)
  end subroutine my_matmul
end module foo
ig25@linux-fd1f:~/Krempel/Matmul/Bounds> gfortran -O3 -fcheck=bounds foo.f90
bounds.f90 && time ./a.out

real    0m3.281s
user    0m3.277s
sys     0m0.002s
ig25@linux-fd1f:~/Krempel/Matmul/Bounds> gfortran -O3 foo.f90 bounds.f90 &&
time ./a.out

real    0m1.911s
user    0m1.909s
sys     0m0.002s