Re: Errors in compiling Octave 4.2.0

[Olaf Till]

On Wed, Jan 18, 2017 at 05:54:53PM +0000, Cui, Shengting wrote:
> Hi,
> 
> I recently tried to install Octave 4.2.0 using intel compiler 
> intel/xe_2016_update3 on a linux computer. My configuration line was:
> 
> 
> ./configure 
> --prefix=/data/kamiak/shengting.cui/SourceCodes/octave-4.2-install 
> --with-blas="-lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -liomp5 
> -lpthread" --with-lapack="-lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core 
> -liomp5 -lpthread" --with-fftw3="-lmkl_intel_lp64 -lmkl_intel_thread 
> -lmkl_core -liomp5 -lpthread -lm" --with-fftw3f="-lmkl_intel_lp64 
> -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -lm" CC=icc CXX=icpc 
> F77=ifort --enable-fortran-calling-convention=f2c CFLAGS="-O3 -ipo -xAVX 
> -fPIC -DMKL_LP64 -DM_PI=3.1415926535897932384" CPPFLAGS="-I$MKLROOT/include 
> -I$MKLROOT/include/fftw" LDFLAGS="-L$MKLROOT/lib/intel64 
> -L$MKLROOT/../compiler/lib/intel64" JAVA_HOME=/opt/apps/java/oracle_1.8.0_92/
> 
> 
> The MKLROOT variable was exported. The configuration went ok. But the 
> compiling step failed at linking stage, giving the following errors:
> 
> 
> icpc: command line warning #10006: ignoring unknown option '-Wold-style-cast'
> 
> icpc: command line warning #10006: ignoring unknown option '-Wcast-align'
> 
> liboctave/.libs/liboctave.so: undefined reference to `__builtin_add_overflow'
> 
> liboctave/.libs/liboctave.so: undefined reference to `__builtin_mul_overflow'
> 
> liboctave/.libs/liboctave.so: undefined reference to `__builtin_sub_overflow'
> 
> make[2]: *** [src/octave-cli] Error 1
> 
> make[2]: Leaving directory 
> `/data/kamiak/shengting.cui/SourceCodes/octave-4.2.0'
> 
> make[1]: *** [all-recursive] Error 1
> 
> make[1]: Leaving directory 
> `/data/kamiak/shengting.cui/SourceCodes/octave-4.2.0'
> 
> make: *** [all] Error 2
> 
> 
> I searched on the internet and there are people seem to suggest that Intel 
> compiler does not support `__builtin_add_overflow', `__builtin_mul_overflow', 
> and `__builtin_sub_overflow' functions. Do you know if this is a fact? If 
> yes, is there a way to get around it?

Don't know if the Intel compiler has the builtins you name, but they
are referenced by gnulib, and for non-gcc only if the preprocessor
defines the macro __has_builtin() and evaluates e.g.
__has_builtin(__builtin_add_overflow) to non-zero. So it seems that
your preprocessor eveluates e.g.
__has_builtin(__builtin_add_overflow) to non-zero, but linking against
this builtin fails.

I'd try to add `-U __has_builtin' to your CPPFLAGS as a
workaround. But the reason for the problem must be somewhere else in
your configuration.

Olaf

-- 
public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net

signature.asc
Description: Digital signature