Build fail with GCC and OpenMPI (FAST_BYTESWAP)

[slacalle]

Build is failing compiling against:

• GCC 11.3.0
• Openmpi 4.1.4
• netCDF 4.9.0
• using "linux_amd64_gfortran" build_options file.

make[1]:` Entering directory '/home/user/L2_MITgcm/experiments/v8/build'
mpif77 -fconvert=big-endian -fimplicit-none -fallow-argument-mismatch -mcmodel=medium  -O3 -funroll-loops  -c mds_byteswapi4.f
mds_byteswapi4.f:268:27:

  268 |      &         ishft(i.and. z'000000ff',  24)
      |                           1
Error: Operands of logical operator ‘.and.’ at (1) are INTEGER(4)/BOZ
make[1]: *** [Makefile:1547: mds_byteswapi4.o] Error 1
make[1]: Leaving directory '/home/user/L2_MITgcm/experiments/v8/build'
make: *** [Makefile:1460: fwd_exe_target] Error 2

The section in question for that file is:

      subroutine MDS_BYTESWAPI4( n, arr )
C IN:
C   n      integer    :: Number of 4-byte words in arr
C IN/OUT:
C   arr    integer*4  :: Array declared as integer*4(n)
C
C Created: 05/05/99 adcroft@mit.edu (This is an unfortunate hack!!)

      implicit none


C Arguments
      integer n
      integer(kind=4) arr(n), i32

C Local
      integer i

      i32(i) = ishft(i.and. z'ff000000', -24) .or.
     &         ishft(i.and. z'00ff0000',  -8) .or.
     &         ishft(i.and. z'0000ff00',   8) .or.
     &         ishft(i.and. z'000000ff',  24)
      do i = 1,n
      arr(i) = i32(arr(i))
      enddo


      return
      end

[mjlosch]

This is the FAST_BYTESWAP part of the code. When compiling with -fconvert=big-endian, the byte swap code (written at a time, when some compilers could do the byteswapping on their own) is not used (just compiled), so defining the flag FAST_BYTESWAP has no effect and should not be changed from the undef-default.

Other than that, I do not know, if this option was ever used anywhere else than on the (now retired) Altix (@menemenlis ?).

[menemenlis]

Indeed, Altix: https://www.nasa.gov/centers/ames/news/releases/2003/03_91AR.html <https://www.nasa.gov/centers/ames/news/releases/2003/03_91AR.html> which became Columbia: https://www.nas.nasa.gov/hecc/resources/columbia.html <https://www.nas.nasa.gov/hecc/resources/columbia.html> Are both long gone. I have not used FAST_BYTESWAP in quite some time. D.

…

On Oct 14, 2022, at 4:00 AM, Martin Losch ***@***.***> wrote: This is the FAST_BYTESWAP part of the code. When compiling with -fconvert=big-endian, the byte swap code (written at a time, when some compilers could do the byteswapping on their own) is not used (just compiled), so defining the flag FAST_BYTESWAP has no effect and should not be changed from the undef-default. Other than that, I do not know, if this option was ever used anywhere else than on the (now retired) Altix ***@***.*** <https://github.com/menemenlis> ?). — Reply to this email directly, view it on GitHub <#658 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABH7CSN43LF6YM5SBPD42ODWDE4NLANCNFSM6AAAAAARFDGSKM>. You are receiving this because you were mentioned.

[mjlosch]

@menemenlis
I suggest that we remove this code, as it does not seem to work (anymore).

[menemenlis]

Hi Martin, removing is fine. Will you open the request for removal or shall I request this? Cheers, Dimitris

…

On Oct 17, 2022, at 6:33 PM, Martin Losch ***@***.***> wrote:  @menemenlis I suggest that we remove this code, as it does not seem to work (anymore). — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.

[mjlosch]

@menemenlis I'd be happy, if you can do this. I'll review (-:

[jm-c]

As we are waiting for @menemenlis PR, I will update the title of this issue by adding the magic word FAST_BYTESWAP