I know, but it's not like distutils is alone in making errors. Though, I will admit, I wasn't really aware of what that cling thing was until now. It seems to me that this is cling running scripts (perhaps a test-phase of the build?) so it doesn't really seem like a good idea to drop all (icpc) flags that aren't compatible with their LLVM based Cling plugin, as it would affect the rest of the build as well. I couldn't spot any obvious way where cling would select a linker command, so it's probably builtin somehow, so cleaning up LDFLAGS for this step seems like the only obvious way forward.
On Tue, Feb 4, 2020 at 1:07 PM Åke Sandgren <[email protected]> wrote: > But ROOT is not a pythonpackage it is cmakemake, so that "should" not be > the problem. > > On 2/4/20 10:11 AM, Mikael Öhman wrote: > > I recognize this type of error; it's exactly what you get in > > pythonpackages under intel with Python @ GCCcore where it picks up GCC > > as the linker command, but tries to use the LDFLAGS for intel. > > If this also is the case here, the proper fix would be to fix the linker > > command (maybe the suggested patches fixes that though) > > > > / Micke > > > > > > On Tue, Feb 4, 2020, 06:24 Jack Perdue <[email protected] > > <mailto:[email protected]>> wrote: > > > > Howdy all, > > > > I'm trying to build a ROOT/6.14.06/intel-2018b-Python-2.7.15 > > based on EB's included ROOT-6.14.06-foss-2018b-Python-2.7.15.eb > > (with a simple s/foss/intel/). > > > > It all goes well until the final link where I get: > > > > error: unknown argument: '-fp-model' > > error: unknown argument: '-ftz' > > error: unknown argument: '-fp-speculation=safe' > > <<< cling interactive line includer >>>:1:10: fatal error: 'source' > > file > > not found > > #include "source" > > ^~~~~~~~ > > Error: Parsing Linkdef file etc/dictpch/allLinkDefs.h > > > > It isn't clear why it is waiting until the end to > > choke on these valid Intel flags. > > > > In any case, I tracked those settings down to > > the CFLAGS/FFLAGS/etcFLAGS that EB sets for cmake. > > After digging around, I found line 64 in > > toolchains/compiler/inteliccifort.py which has > > > > 'defaultprec': ['ftz', 'fp-speculation=safe', 'fp-model source'], > > > > for a toolchain option. > > > > So I tried various options to turn that option off > > (some were just guesses). > > > > toolchainopts = {'pic': True, 'defaultprec': False} > > and > > toolchainopts = {'pic': True, 'defaultprec': ''} > > and > > defaultprec = [''] > > > > None of those worked. > > > > If I manually hack inteliccifort.py to disable > > those settings, then my build for ROOT-intel > > works. > > > > Feel to try it yourself (I'm using EB 4.1.1). > > > > Obviously it would be nice to just fix the link, > > but I'm curious about other possible uses. > > > > How does one disable defaultprec when using > > the Intel toolchain? > > > > -- > > Jack Perdue > > Lead Systems Administrator > > High Performance Research Computing > > TAMU Division of Research > > [email protected] <mailto:[email protected]> http://hprc.tamu.edu > > HPRC Helpdesk: [email protected] <mailto:[email protected]> > > > > -- > Ake Sandgren, HPC2N, Umea University, S-90187 Umea, Sweden > Internet: [email protected] Phone: +46 90 7866134 Fax: +46 90-580 14 > Mobile: +46 70 7716134 WWW: http://www.hpc2n.umu.se >
