I have recently done a clean install of Mandrake 8.1 and am trying to
compile R with an aim to preparing a working RPM (which I managed
perfectly with Mandrake 8.0, before subscribing to this list: the rpm is
available on cran-r-project.org).
When I try, R-base build keeps on failing both with gcc-2.96 and
gcc-3.0.1, no matter what the optimizations. The details vary according to
the choice of compiler and options, but failure is during build of lapack.
Installing the liblapack3 and liblapack3-devel rpms hasn't solved the
problem.
The following examples were run cleanly, i.e. the entire old build directory
was cancelled each time. Result: five different errors in double.f, one in
cmplx.f. I have kept the config.cache and config.log if anyone is
interested.
Example 1:
gcc 3.0.1, no optimizations set in R-base.spec
gcc -I. -I../../../src/include -I../../../src/include -I/usr/local/include
-DHAVE_CONFIG_H -mieee-fp -D__NO_MATH_INLINES -fPIC -g -O2 -c Lapack.c -o
Lapack.lo
g77 -fPIC -g -O2 -c double.f -o double.lo
double.f: In subroutine `dgees':
double.f:2362: Internal error: Segmentation fault.
Example 2:
gcc 3.0.1, CFLAGS and FFLAGS set to $RPM-OPT-FLAGS but
-ffast-math removed (this flag has always caused R build
to fail in the past)
gcc -I. -I../../../src/include -I../../../src/include -I/usr/local/include
-DHAVE_CONFIG_H -mieee-fp -D__NO_MATH_INLINES -fPIC -O3
-fomit-frame-pointer -pipe -mcpu=pentiumpro -march=i586
-fno-strength-reduce -c Lapack.c -o Lapack.lo
g77 -fPIC -O3 -fomit-frame-pointer -pipe -mcpu=pentiumpro -march=i586
-fno-strength-reduce -c double.f -o double.lo
double.f: In subroutine `dbdsqr':
double.f:731: Internal error: Segmentation fault.
Example 3:
gcc 3.0.1, $RPM-OPT-FLAGS enabled
gcc -I. -I../../../src/include -I../../../src/include -I/usr/local/include
-DHAVE_CONFIG_H -mieee-fp -D__NO_MATH_INLINES -fPIC -O3
-fomit-frame-pointer -pipe -mcpu=pentiumpro -march=i586 -ffast-math
-fno-strength-reduce -c Lapack.c -o Lapack.lo
etc.gives a segmentation fault in yet another subroutine of double.f
Example 4:
gcc 2.96, no optimizations set in R-base.spec
make[4]: Entering directory
`/home/mike/rpm/BUILD/R-1.3.1/src/modules/lapack'
gcc -I. -I../../../src/include -I../../../src/include -I/usr/local/include
-DHAVE_CONFIG_H -mieee-fp -D__NO_MATH_INLINES -fPIC -g -O2 -c Lapack.c -o
Lapack.lo
g77 -fPIC -g -O2 -c double.f -o double.lo
double.f: In subroutine `dtrsyl':
double.f:27942: Internal error: Segmentation fault.
Example 5:
gcc 2.96, CFLAGS and FFLAGS set to $RPM-OPT-FLAGS
but -ffast-math removed (this flag has always caused R
build to fail in the past)
gcc -I. -I../../../src/include -I../../../src/include -I/usr/local/include
-DHAVE_CONFIG_H -mieee-fp -D__NO_MATH_INLINES -fPIC -O3
-fomit-frame-pointer -pipe -mcpu=pentiumpro -march=i586
-fno-strength-reduce -c Lapack.c -o Lapack.lo
g77 -fPIC -O3 -fomit-frame-pointer -pipe -mcpu=pentiumpro -march=i586
-fno-strength-reduce -c double.f -o double.lo
double.f: In subroutine `dgecon':
double.f:1931: Internal error: Segmentation fault.
Example 6:
gcc 2.96, $RPM-OPT-FLAGS enabled
mcpu=pentiumpro -march=i586 -ffast-math -fno-strength-reduce -c Lapack.c
-o Lapack.lo
g77 -fPIC -O3 -fomit-frame-pointer -pipe -mcpu=pentiumpro -march=i586
-ffast-math -fno-strength-reduce -c double.f -o double.lo
g77 -fPIC -O3 -fomit-frame-pointer -pipe -mcpu=pentiumpro -march=i586
-ffast-math -fno-strength-reduce -c cmplx.f -o cmplx.lo
cmplx.f: In subroutine `zbdsqr':
cmplx.f:783: Internal error: Segmentation fault.
Any suggestions ?
--
Dr. Michele Alzetta
