[Rd] Using a different compiler when creating a package on Windows

Erin Hodgess er|nm@hodge@@ @end|ng |rom gm@||@com
Sat Mar 2 14:11:50 CET 2019


Hello!

I am updating to the R-devel version as of 2/28 and recompiling with
OpenBLAS.   I got that to compile nicely.

I am now updating a package that uses the OpenBLAS and the PGI compiler,
which has changed too.

I just changed the path names in Makevars.  Here is the Makevars file.

FC= d:/PGI/win64/18.10/bin/pgf90
F77= d:/PGI/win64/18.10/bin/pgf90
CC = d:/PGI/win64/18.10/bin/pgcc
FFLAGS= -Ld:/PGI/win64/18.10/bin/pgf90.dll
-Ld:/PGI/win64/18.10/bin/pgc14.dll
FLIBS= -Ld:/PGI/win64/18.10/bin/pgf90.dll
-Ld:/PGI/win64/18.10/bin/pgc14.dll
PKG_LIBS = $(LAPACK_LIBS) $(BLAS_LIBS) $(FLIBS)
MPI_FFLAGS= -Mmpi=msmpi
SHLIB_FFLAGS= -Mmakedll
CUDA_FFLAGS=-acc -Minfo=accel -Mlarge_arrays
-Ld:/PGI/win64/18.10/bin/pgf90.dll -Ld:/PGI/win64/18.10/bin/pgc14.dll
DLLCUDA_FFLAGS=-ta=tesla:nordc
 -Ld:/PGI/win64/18.10/lib/acc_init_link_cuda.obj
-Ld:/PGI/win64/18.10/lib/libaccapi.lib
 -Ld:/PGI/win64/18.10/lib/libaccg.lib
 -Ld:/PGI/win64/18.10/lib/libaccn.lib
 -Ld:/PGI/win64/18.10/lib/libaccg2.lib
 -Ld:/PGI/win64/18.10/lib/libcudadevice.lib
 -Ld:/PGI/win64/18.10/lib/pgc14.lib
 -Ld:/PGI/win64/18.10/lib/libnspgc.lib
-defaultlib:legacy_stdio_definitions -defaultlib:oldnames
-Ld:/PGI/win64/18.10/bin/pgf90.dll -Ld:/PGI/win64/18.10/bin/pgc14.dll


rmpiFort.obj: rmpiFort.f90
$(FC) $(FFLAGS) $(MPI_FFLAGS) -c rmpiFort.f90 -o rmpiFort.obj  -m64
$(FC) $(CUDA_FFLAGS) $(DLLCUDA_FFLAGS) -c test4.f90 -o test4.obj -m64
$(FC) $(SHLIB_FFLAGS) $(FFLAGS) $(MPI_FFLAGS) $(CUDA_FFLAGS) -o
rmpiFort.dll test4.obj rmpiFort.obj  -m64


When I do Make check, I get the following:
* installing *source* package 'rmpiFort' ...

** libs

d:/Rtools/mingw_64/bin/gfortran  -O2  -march=native -pipe -Mmpi=msmpi -c
rmpiFort.f90 -o rmpiFort.obj  -m64
gfortran.exe: error: unrecognized command line option '-Mmpi=msmpi'

make: *** [Makevars:14: rmpiFort.obj] Error 1
ERROR: compilation failed for package 'rmpiFort'

* removing 'D:/rnew/rmpiFort.Rcheck/rmpiFort'

So it is still trying to use the gfortran compiler rather than the PGI
compiler.

The PGI libraries are first in the PATH.

Since it worked fine before, I am at a loss as to why it doesn't work now.
Any suggestions much appreciated.

Thanks,
Erin



Erin Hodgess, PhD
mailto: erinm.hodgess using gmail.com

	[[alternative HTML version deleted]]



More information about the R-devel mailing list