compile error on ARM Mac

[boccio-john]

Fortran 90 code - compile works with gfortran 10.2 on intel Mac.
However, when I try to do same compile using gfortran 11.0 on ARM Mac (installed using simple package over 10.2 version),
it fails with message:

ld:library not found for -lm
collect2: error: ld returned 1 exit value

compile command is:

/usr/local/gfortran/bin/gfortran treecode.f90 treegrav.f90 treeio.f90 treeload.f90 -o treecode

[fxcoudert]

Do you have the command-line tools installed (with xcode-select --install)?

[boccio-john]

Yes, they are installed. Thanks.

…

On Dec 2, 2020, at 4:27 PM, FX Coudert ***@***.***> wrote: xcode-select --install

[martin-frbg]

Quoting isuruf from OpenMathLib/OpenBLAS#3032 (comment)

It looks like gfortran was compiled with /Library/Developer/CommandLineTools/SDKs/MacOSX11.1.sdk and gfortran saves this location into its binary. You have /Library/Developer/CommandLineTools/SDKs/MacOSX11.0.sdk which becomes a problem.
A workaround is to do export SDKROOT=/Library/Developer/CommandLineTools/SDKs/MacOSX11.0.sdk

and that suggestion later refined to
export MACOSX_DEPLOYMENT_TARGET=11.0

[boccio-john]

Thanks. That worked. I am a physicist compiling fortran 90 code to do galaxy-galaxy collisions. I am not a UNIX expert at all. I just issued the export command in the terminal window before I ran the compile operation. Is there somewhere I should enter the export command so that is always active? Thanks for any help. By the way it is running extremely fast on the MacBook Air M1 system.

[sholmas]

macOS 11.1 is now available to all users, it solved this issue for me after reinstalling the command line tools (using xcode-select --install).

[boccio-john]

Thanks. I just found that out also this evening. Sent from John Boccio's iPad

…

On Dec 17, 2020, at 9:51 PM, Stian Holmas ***@***.***> wrote:  macOS 11.1 is now available to all users, it solved this issue for me after reinstalling the command line tools (using xcode-select --install). — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[fxcoudert]

macOS 11.1 and the command-line tools 12.3 fix this issue. I will try to make my next build more robust.

[iains]

This issue was referenced in a discussion of static linking:

• we should also fix libgfortran's spec to omit "-lm" which has not been needed on macOS "forever" and
• put a translation spec in for libquadmath to be static for static-libgfortran.

e.g
*lib: %{!static-libgfortran: -lquadmath } %{static-libgfortran: libquadmath.a%s} %(libgcc) %(liborig)

[beausoleilmo]

About lquadmath, when trying to install RcppArmadillo in R, I get this message:

ld: library not found for -lquadmath
clang: error: linker command failed with exit code 1 (use -v to see invocation)
make: *** [RcppArmadillo.so] Error 1
ERROR: compilation failed for package ‘RcppArmadillo’
* removing ‘/Library/Frameworks/R.framework/Versions/4.0/Resources/library/RcppArmadillo’

I already modified the line in Makeconf
FLIBS = -L/usr/local/gfortran/lib/gcc/aarch64-apple-darwin20.2.0/11.0.0 -L/usr/local/gfortran/lib -lgfortran -lquadmath -lm
(see here https://stackoverflow.com/questions/23916219/os-x-package-installation-depends-on-gfortran-4-8)

to specify the "proper" location of the gcc, but lquadmath seems it's not working at all. Do you have a quick workaround?

[iains]

for Arm64, there is no libquadmath - theres no REAL*16 equivalent on the platform at present. This means that we need a configure change to omit the "-lquadmath" from the spec.

You should be able to hack around it in the short-term by editing /path/to/compiler/lib/libgfortran.spec and removing the "-lquadmath" from the last line.

[iains]

(for Arm64 on Darwin)

• the size of long double is set to 64 bits, since that's what the current platform ABI supports.
• The configuration for libgfortran checks for the availability of __float128 before enabling libquadmath.
• I took a scan through the aarch64 backend and can't see anywhere where that's explicitly enabled - so somehow either the libgfortran configuration is failing to detect that the type is not available, or the compiler is publishing its availability when that isn't really true.
• A workaround would be to configure with --disable-libquadmath-support , I guess