[Coin-discuss] clp performance much worse with g++4.1 vs. g++4.0
Stuart A Siegel
stus at us.ibm.com
Mon Dec 10 12:27:22 EST 2007
Ted,
I'm now able to confirm your results. The following produces clp binaries
(/tmp/clponly/bin/clp and /tmp/allatonce/bin/clp) having the same (good)
performance characteristics. I think what happened was that about 10 days
ago I compiled the libs with --enable-debug. Although I used a separate
build directory and wipe out the contents of this build directory before
doing a new build, the debugging flag must have become sticky in the Cbc
build configuration somehow. I guess the final step would be to try to
replicate the problem by doing the debugging thing all over and not
removing the Cbc source tree before a rebuild with debugging not enabled.
I'm not sure, however, if this is really a worthy exercise or if confirmed
would constitute a bug in the build configuration scripts or instead a bug
in me :-)
# let's build just clp
OSI=Osi-0.97.0
CLP=Clp-1.6.0
CBC=Cbc-2.0.0
rm -rf $CLP
tar xzvf $CLP.tgz
cd coinbuild
../$CLP/configure --prefix=/tmp/clponly && make -j20 install
# let's build everything
cd ..
rm -rf $CBC $CLP $OSI coinbuild
tar xzvf $CLP.tgz
tar xzvf $OSI.tgz
tar xzvf $CBC.tgz
mkdir coinbuild
cd coinbuild
../$CLP/configure --prefix=/tmp/allatonce && make -j20 install
../$OSI/configure --prefix=/tmp/allatonce && make -j20 install
../$CBC/configure --prefix=/tmp/allatonce && make -j20 install
-----------------------------------------------------------------------------------------
Stuart Siegel - IBM TJ Watson Laboratory 33-238
(M) 914.396.9506 (H) 973.379.1810 (O) 914.945.1674
More information about the Coin-discuss
mailing list