R devel - Apr 2017 - Intel MKL compiling issue

Hi r-devel,

a recent install of R-3.4.0 RC (2017-04-13 r72510)?
on Linux (Ubuntu 16.04.1 LTS) x86_64-linux-gnu?
with?g++ (Ubuntu 5.4.0-6ubuntu1~16.04.4) 5.4.0 20160609
(see?http://bioconductor.org/checkResults/devel/bioc-LATEST/malbec2-NodeInfo.html?for
more)?
results in CXXFLAGS not containing "-O2" as optimisation flag,
there is only "	-Wall", while?CFLAGS are happy with "-g -O2
-Wall"

This has an influence in at least one place
https://stat.ethz.ch/pipermail/bioc-devel/2017-April/010733.html
where we have WARNINGS in R CMD check from "Found ?abort?,?
possibly from ?abort? (C)" in packages xcms/mzR.
The abort() call is not coming from XCMS, but rather?
from the C++ code in the STL:

	[...]
	# 1770 "/usr/include/c++/5/bits/stl_list.h"
	void _M_check_equal_allocators(list& __x) {
	?if (_M_get_Node_allocator())
	???__builtin_abort();
	}

If we compile with -O2 optimisation, this getting rid of?
the abort() symbol, as shown?
in https://github.com/sneumann/xcms/issues/150#issuecomment-293545521

Martin Morgan created a minimum example that shows that?
the symbol is indeed deep down in the STL (see below and in:
https://stat.ethz.ch/pipermail/bioc-devel/2017-April/010837.html?)

This raises several questions:

1) is there any way to avoid the WARNING / abort() inside?
? ?the STL list implementation ? Or just live with it ?

2) If not, is there a reason why the Bioconductor build farm?
? ?Ubuntu machine is not using -O2 as default CXXFLAG ?
? ?BioC admins are trying to have a vanilla R installation with defaults.?
? ?According to Herve Pages, CXXFLAGS without -O2 is default?
? ?since R-3.4 beta, but I don't know enough about the package?
? ?build logic to point to a particular R commit.?

3) I thought about cheating the system and add -O2?
? ?in the package CXXFLAGS, but Martin Morgan?
? ?recommends packages shouldn't mess and override system build defaults
? ?to mask and paper over the actual issue having a nasty abort()?
? ?lurking somewhere.
? ?But I couldn't add PKG_CXXFLAGS=-O2 in first place, since that?
? ?triggers the different WARNING that -O2 is not portable.?

=> Any help and input would be highly appreciated.?

Thanks in advance,?
yours,
Steffen


tmp.cpp by Martin Morgan (also in above linked mail thread)
-------
#include <list>

int foo(int argc, const char *argv[]) {
?????std::list<int> l1, l2;
?????std::list<int>::iterator it;

?????it = l1.begin();
?????l1.splice (it, l2); // mylist1: 1 10 20 30 2 3 4

?????return 0;
}
-------

Test with

???rm -f tmp.o && R CMD SHLIB tmp.cpp && nm tmp.o | grep abort

with compiler settings in ~/.R/Makevars with/without -O2 
-------------
CXXFLAGS = -g -O0
-------------





--
IPB Halle????????????????????AG Massenspektrometrie &
Bioinformatik
Dr. Steffen Neumann??????????http://www.IPB-Halle.DE
Weinberg 3 ? ? ? ? ? ? ? ? ? Tel. +49 (0) 345 5582 - 1470
06120 Halle ? ? ? ? ? ? ? ? ? ? ? +49 (0) 345 5582 - 0 ? ? ? ? ??
sneumann(at)IPB-Halle.DE?????Fax. +49 (0) 345 5582 - 1409

This is fixed in R-rc_2017-04-19_r72555.tar.gz

If you are affected by this issue then please test the RC tarball. This
is the last chance to detect problems (including those created by the
last-minute patch) before the release of R 3.4.0.

Martyn

On Wed, 2017-04-19 at 12:19 +0000, Neumann, Steffen
wrote:
> Hi r-devel,
> 
> a recent install of R-3.4.0 RC (2017-04-13 r72510)?
> on Linux (Ubuntu 16.04.1 LTS) x86_64-linux-gnu?
> with?g++ (Ubuntu 5.4.0-6ubuntu1~16.04.4) 5.4.0 20160609
> (see?http://bioconductor.org/checkResults/devel/bioc-LATEST/malbec2-N
> odeInfo.html?for more)?
> results in CXXFLAGS not containing "-O2" as optimisation flag,
> there is only "	-Wall", while?CFLAGS are happy with "-g -O2
> -Wall"
> 
> This has an influence in at least one place
> https://stat.ethz.ch/pipermail/bioc-devel/2017-April/010733.html
> where we have WARNINGS in R CMD check from "Found ?abort?,?
> possibly from ?abort? (C)" in packages xcms/mzR.
> The abort() call is not coming from XCMS, but rather?
> from the C++ code in the STL:
> 
> 	[...]
> 	# 1770 "/usr/include/c++/5/bits/stl_list.h"
> 	void _M_check_equal_allocators(list& __x) {
> 	?if (_M_get_Node_allocator())
> 	???__builtin_abort();
> 	}
> 
> If we compile with -O2 optimisation, this getting rid of?
> the abort() symbol, as shown?
> in https://github.com/sneumann/xcms/issues/150#issuecomment-293545521
> 
> Martin Morgan created a minimum example that shows that?
> the symbol is indeed deep down in the STL (see below and in:
> https://stat.ethz.ch/pipermail/bioc-devel/2017-April/010837.html?)
> 
> This raises several questions:
> 
> 1) is there any way to avoid the WARNING / abort() inside?
> ? ?the STL list implementation ? Or just live with it ?
> 
> 2) If not, is there a reason why the Bioconductor build farm?
> ? ?Ubuntu machine is not using -O2 as default CXXFLAG ?
> ? ?BioC admins are trying to have a vanilla R installation with
> defaults.?
> ? ?According to Herve Pages, CXXFLAGS without -O2 is default?
> ? ?since R-3.4 beta, but I don't know enough about the package?
> ? ?build logic to point to a particular R commit.?
> 
> 3) I thought about cheating the system and add -O2?
> ? ?in the package CXXFLAGS, but Martin Morgan?
> ? ?recommends packages shouldn't mess and override system build
> defaults
> ? ?to mask and paper over the actual issue having a nasty abort()?
> ? ?lurking somewhere.
> ? ?But I couldn't add PKG_CXXFLAGS=-O2 in first place, since that?
> ? ?triggers the different WARNING that -O2 is not portable.?
> 
> => Any help and input would be highly appreciated.?
> 
> Thanks in advance,?
> yours,
> Steffen
> 
> 
> tmp.cpp by Martin Morgan (also in above linked mail thread)
> -------
> #include <list>
> 
> int foo(int argc, const char *argv[]) {
> ?????std::list<int> l1, l2;
> ?????std::list<int>::iterator it;
> 
> ?????it = l1.begin();
> ?????l1.splice (it, l2); // mylist1: 1 10 20 30 2 3 4
> 
> ?????return 0;
> }
> -------
> 
> Test with
> 
> ???rm -f tmp.o && R CMD SHLIB tmp.cpp && nm tmp.o | grep
abort
> 
> with compiler settings in ~/.R/Makevars with/without -O2?
> -------------
> CXXFLAGS = -g -O0
> -------------
> 
> 
> 
> 
> 
> --
> IPB Halle????????????????????AG Massenspektrometrie &
> Bioinformatik
> Dr. Steffen Neumann??????????http://www.IPB-Halle.DE
> Weinberg 3 ? ? ? ? ? ? ? ? ? Tel. +49 (0) 345 5582 - 1470
> 06120 Halle ? ? ? ? ? ? ? ? ? ? ? +49 (0) 345 5582 - 0 ? ? ? ? ??
> sneumann(at)IPB-Halle.DE?????Fax. +49 (0) 345 5582 - 1409
> 
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel

Dear R-developers,


I would appreciate any insights over compiling R 3.4 with Intel MKL -- I have
been successful until R 3.3.3 but now it stops complaining about pcre though it
worked without Intel MKL as follows,


./configure LDFLAGS=-L/genetics/data/software/lib CFLAGS=-fPIC
-I/genetics/data/software/include --enable-R-shlib


I have used,


export MKL_NUM_THREADS=15
export MKLROOT=/genetics/data/software/intel/composer_xe_2013.4.183/mkl
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$MKLROOT/lib/intel64
/genetics/data/software/intel/composer_xe_2013.4.183/mkl/bin/mklvars.sh intel64
CPPFLAGS="-I/genetics/data/software/include
-L/genetics/data/software/lib" \
./configure --prefix=/genetics/data/software --enable-R-shlib
--enable-threads=posix --with-lapack \
 --with-blas="-fopenmp -m64 -I$MKLROOT/include -L$MKLROOT/lib/intel64
-lmkl_gf_lp64 -lmkl_gnu_thread -lmkl_core -lpthread -lm"


Many thanks,



Jing Hua

	[[alternative HTML version deleted]]