Failure to simplify some array constructors with SIZE intrinsic

[Harald1]

Hello,

the following code may be using SIZE in a weird way to compute the shape and size of an array but should be valid Fortran:

program p
  implicit none
  integer :: k
  integer, parameter :: x(2,3) = 1
  integer, parameter :: s(*) = [shape(x)]
  integer, parameter :: t(*) = [(size(x, dim=k), k=1,2)]       ! fail
  integer, parameter :: u(*) = [(size(x, dim=k), k=1,rank(x))] ! fail
  integer, parameter :: v = product(shape(x))
  integer, parameter :: w = product([(size(x, dim=k), k=1,2)]) ! fail
  print *, s
  print *,         [(size(x, dim=k), k=1,rank(x))]
  print *, product([(size(x, dim=k), k=1,rank(x))])
end

With current ifort/ifx I get:

% ifort ifort-size-simplify.f90
ifort-size-simplify.f90(6): error #7747: Invalid parameter expression.
  integer, parameter :: t(*) = [(size(x, dim=k), k=1,2)]       ! fail
---------------------------------^
ifort-size-simplify.f90(7): error #7747: Invalid parameter expression.
  integer, parameter :: u(*) = [(size(x, dim=k), k=1,rank(x))] ! fail
---------------------------------^
ifort-size-simplify.f90(9): error #6063: An INTEGER or REAL data type is required in this context.   [K]
  integer, parameter :: w = product([(size(x, dim=k), k=1,2)]) ! fail
--------------------------------------------------^
compilation aborted for ifort-size-simplify.f90 (code 1)

Commenting the marked code lines allows the code to compile and work as expected.  It is only the simplification of the array constructors needed for the constant expressions above that fails.

Thanks,

Harald

[Barbara_P_Intel · Accepted solution]

This reproducer is indeed legal Fortran. I filed a bug report, CMPLRLLVM-48095. I'll let you know when a fix is available.

[Barbara_P_Intel]

@Harald1, no more faulty error messages with ifx or ifort with the releases that were published this week. Please check them out!

[Harald1]

Wow, the ICE is gone, and I get a correct result at runtime....

Thanks!