Nvfortran 24.3 false compiler error for type-bound operator

Accelerated Computing HPC Compilers nvc, nvc++ and nvfortran
1

I’ve encountered a likely bug with type-bound operators. Here’s the reproducer:

module bug0004_mod

implicit none
private

type, public :: t
    real :: v
contains
    procedure, private :: t_exp
    generic, public :: operator(**) => t_exp
end type t

contains

elemental function t_exp(t_in, int_in)
    class(t), intent(in) :: t_in
    integer, intent(in)  :: int_in
    
    type(t) :: t_exp

    t_exp%v  = t_in%v**int_in
end function t_exp

end module bug0004_mod

program bug0004

use bug0004_mod, only: t
implicit none

type(t) :: x, y

x%v = 5.0

! Commenting out the next two lines eliminates the error. But why?
y = x**7
print *, y%v

y = f(x)
print *, y%v

contains

! If `f` is a module procedure instead of an internal procedure, there are no errors.

function f(x)
    ! Uncommenting this makes no difference.
    ! use bug0004_mod, only: t
    
    type(t), intent(in) :: x
    
    type(t) :: f
    
    f = x**2
end function f

end program bug0004

Here’s the command line log:

$ uname -a
Linux trident 5.15.0-102-generic #112~20.04.1-Ubuntu SMP Thu Mar 14 14:28:24 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux
$ nvfortran --version

nvfortran 24.3-0 64-bit target on x86-64 Linux -tp haswell 
NVIDIA Compilers and Tools
Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES.  All rights reserved.
$ nvfortran bug0004.f90
NVFORTRAN-S-0099-Illegal use of operator ** on a derived type (bug0004.f90: 54)
NVFORTRAN-S-0099-Illegal use of derived type (bug0004.f90: 54)
NVFORTRAN-S-0148-Reference to derived type required (bug0004.f90: 54)
  0 inform,   0 warnings,   3 severes, 0 fatal for f

The use of the ** operator here is not illegal as far as I’m aware. gfortran and ifx both have no problem with this code. And neither does nvfortran if I comment out two lines labeled above! Alternatively, if I make f a module procedure, there is no error.

Also: You should be able to change the operator to *, /, +, or - and get the same false error as with **. I only checked minimal reproducers for ** and *, but in the original code I got errors for all 5 operators.

2

Thanks btrettel2, I’ve added a problem report, TPR #35548, and sent it to engineering for investigation.

-Mat