[Flang] selected_real_kind (33) returns -1

[arrowd]

Building the DQFUN package from https://www.davidhbailey.com/dhbsoftware/ with Flang, I get the following error:

error: Semantic errors in dqfuna.f90
./dqfuna.f90:54:1: error: REAL(KIND=-1) is not a supported type
  real (dqknd), public, parameter :: dqdpw = 34.016389510029875059152495103867712q0, &
  ^

The problematic code block:

module dqfuna
integer, public, parameter :: dqknd = selected_real_kind (33, 4931), dqldb = 6, &
  dqnbt = 113, dqnwx = 2
real (dqknd), public, parameter :: dqdpw = 34.016389510029875059152495103867712q0, &
  dqlogb = 77.632484222713874654729997603315776q0, dqrdfz = 1.q0 / 2.q0**110

I don't know Fortran, so maybe it isn't really a Flang problem.

[llvmbot]

@llvm/issue-subscribers-flang-frontend

Author: Gleb Popov (arrowd)

Building the DQFUN package from https://www.davidhbailey.com/dhbsoftware/ with Flang, I get the following error: ``` error: Semantic errors in dqfuna.f90 ./dqfuna.f90:54:1: error: REAL(KIND=-1) is not a supported type real (dqknd), public, parameter :: dqdpw = 34.016389510029875059152495103867712q0, & ^ ```

The problematic code block:

module dqfuna
integer, public, parameter :: dqknd = selected_real_kind (33, 4931), dqldb = 6, &
  dqnbt = 113, dqnwx = 2
real (dqknd), public, parameter :: dqdpw = 34.016389510029875059152495103867712q0, &
  dqlogb = 77.632484222713874654729997603315776q0, dqrdfz = 1.q0 / 2.q0**110

I don't know Fortran, so maybe it isn't really a Flang problem.

[eugeneepshteyn]

@arrowd , this indicates that flang wasn't built with quad real (real(16)) support. For example, with flang properly built with real(16) support, I get the following output with the following program:

print *, selected_real_kind (33, 4931)
end

Output:

$ flang real16.f90 && ./a.out
 16

Without proper real(16) support, I get this:

$ flang real16.f90 && ./a.out
 -1

Please see here for more information: https://flang.llvm.org/docs/Real16MathSupport.html

[arrowd]

Then it looks like this isn't a Flang problem. Should we close this issue?

[eugeneepshteyn]

Closed

[arrowd]

In another project I get this error:

/wrkdirs/usr/ports/misc/libemos/work/libemos-4.5.9-Source/interpolation/wavexx2.F:462:34: error: Actual argument for 'x=' has bad type or kind 'REAL(4)'
                    CSE_PT = __COS(SE_PT*RAD)
                                   ^^^^^^^^^

Is it the same instance of Flang built without quadmath support or some other problem?

[eugeneepshteyn]

Not clear. Could you please create a small reproducer that includes declaration of __COS()?

[arrowd]

__COS is defined as

#if (defined REAL_8)
#define __COS   DCOS
#endif

Arguments defs:

      REAL    C1, C2, ...
     X        SE_PT, RAD

[eugeneepshteyn]

DCOS expects double precision arguments. It's not the same issue as quadmath support.