Re: [ALPS-users] Alps 1.3.1 spinmc compile problem
Does anyone have an experience on Ubuntsu Linux? Synge From: "Miles Stoudenmire" <miles.stoudenmire@gmail.com> Date: Fri, 30 May 2008 08:55:53 -0700
Hi Synge - I'm using Ubuntu Linux on a machine with an AMD processor. I tried installing the lapack 3.0 package through apt-get and saying --with-lapack-dir=/usr/lib --with-lapack=-llapack because the lapack library files got installed in /usr/lib (I have no idea if those arguments are sensible, they were just a guess). When that didn't work, I tried installing the ACML library from AMD and using a similar approach but the configure script still said that it couldn't find any LAPACK support.
On Fri, May 30, 2008 at 1:08 AM, Synge Todo <wistaria@comp-phys.org> wrote:
Dear Miles Stoudenmire,
From: "Miles Stoudenmire" <miles@physics.ucsb.edu> Date: Thu, 29 May 2008 22:38:58 -0700
It looks like it's referring to some funny lapack function from what I can tell, so I think I should mention that I did not build the libraries with lapack support (though the configure script claimed that I'd be ok to compile classical monte carlo).
Thank you for your report. It seems that the classical Monte Carlo also uses a SVD routine from lapack. So, this is a bug of configure script. I will fixed it in the next version.
If it does turn out to be because I need to add lapack support, I will have to ask some questions about how to do that.
What kind of platform are you using?
Best, Synge
-- -=Miles Stoudenmire=- miles@physics.ucsb.edu miles.stoudenmire@gmail.com http://www.crazytheory.com/
Hi Synge and Miles, the correct sequence for linking acml with a gcc4 is: -lacml -lacml_mv -lgfortran and with a gcc3: -lacml -lacml_mv -lg2c The libraries are usually not found in /usr/lib, but in /opt/acml- version/type_of_cpu/lib. Miles: please check if they are really in / usr/lib. If so, please run configure --with-blas="-lacml -lacml_mv - lgfortran" --with-lapack="-lacml -lacml_mv -lgfortran" (on a gcc4). That should help. Best, Emanuel On May 31, 2008, at 4:24 AM, Synge Todo wrote:
Does anyone have an experience on Ubuntsu Linux?
Synge
From: "Miles Stoudenmire" <miles.stoudenmire@gmail.com> Date: Fri, 30 May 2008 08:55:53 -0700
Hi Synge - I'm using Ubuntu Linux on a machine with an AMD processor. I tried installing the lapack 3.0 package through apt-get and saying --with-lapack-dir=/usr/lib --with-lapack=-llapack because the lapack library files got installed in /usr/lib (I have no idea if those arguments are sensible, they were just a guess). When that didn't work, I tried installing the ACML library from AMD and using a similar approach but the configure script still said that it couldn't find any LAPACK support.
On Fri, May 30, 2008 at 1:08 AM, Synge Todo <wistaria@comp- phys.org> wrote:
Dear Miles Stoudenmire,
From: "Miles Stoudenmire" <miles@physics.ucsb.edu> Date: Thu, 29 May 2008 22:38:58 -0700
It looks like it's referring to some funny lapack function from what I can tell, so I think I should mention that I did not build the libraries with lapack support (though the configure script claimed that I'd be ok to compile classical monte carlo).
Thank you for your report. It seems that the classical Monte Carlo also uses a SVD routine from lapack. So, this is a bug of configure script. I will fixed it in the next version.
If it does turn out to be because I need to add lapack support, I will have to ask some questions about how to do that.
What kind of platform are you using?
Best, Synge
-- -=Miles Stoudenmire=- miles@physics.ucsb.edu miles.stoudenmire@gmail.com http://www.crazytheory.com/
Hi, I've tried a number of approaches (combinations of the suggestions above) and none are working. I installed the gfortran package through apt-get just to make sure it was around, and I downloaded and installed acml 3.6.1 compiled with gfortran (for 32 bit) from the AMD website and installed it. I checked that my system is using gcc4. I then ran ./configure --with-blas="-lacml -lacml_mv -lgfortran" --with-lapack="-lacml -lacml_mv -lgfortran" --with-blas-dir="/opt/acml3.6.1/gfortran32/lib/" in the alps-1.3.1 directory. When it didn't work, I also tried a version of acml "compiled with gnu" (according to the AMD website, I'm not sure what that is supposed to mean though). I even tried the gcc3 version of the library flags. Below is a section of the output of the configure command above (the other tries gave similar results): configure: checking for BLAS Library checking for BLAS library directory... /opt/acml3.6.1/gfortran32/lib checking for dgemm_ in -L/opt/acml3.6.1/gfortran32/lib -lacml -lacml_mv -lgfortran... no checking for dgemm_ in -L/opt/acml3.6.1/gfortran32/lib -lsci... no checking for dgemm_ in -L/opt/acml3.6.1/gfortran32/lib -lacml -lacml_mv -lgfortran... no checking for dgemm_ in -L/opt/acml3.6.1/gfortran32/lib -lblas... no checking for dgemm_ in -L/opt/acml3.6.1/gfortran32/lib -lblas -lf2c... no checking for dgemm_ in -L/opt/acml3.6.1/gfortran32/lib -lblas -lg2c... no checking for dgemm_ in -L/opt/acml3.6.1/gfortran32/lib -lblas -lpgftnrtl -lpgc... no checking for dgemm_ in -L/opt/acml3.6.1/gfortran32/lib -lblas /usr/lib/libg2c.so.0.0.0... no configure: disabling BLAS support configure: disabling LAPACK support Please let me know what I should try next - thanks! On Sat, May 31, 2008 at 5:54 AM, Emanuel Gull <gull@itp.phys.ethz.ch> wrote:
Hi Synge and Miles, the correct sequence for linking acml with a gcc4 is: -lacml -lacml_mv -lgfortran and with a gcc3: -lacml -lacml_mv -lg2c The libraries are usually not found in /usr/lib, but in /opt/acml-version/type_of_cpu/lib. Miles: please check if they are really in /usr/lib. If so, please run configure --with-blas="-lacml -lacml_mv -lgfortran" --with-lapack="-lacml -lacml_mv -lgfortran" (on a gcc4). That should help. Best, Emanuel
On May 31, 2008, at 4:24 AM, Synge Todo wrote:
Does anyone have an experience on Ubuntsu Linux?
Synge
From: "Miles Stoudenmire" <miles.stoudenmire@gmail.com> Date: Fri, 30 May 2008 08:55:53 -0700
Hi Synge - I'm using Ubuntu Linux on a machine with an AMD processor.
I tried installing the lapack 3.0 package through apt-get and saying
--with-lapack-dir=/usr/lib --with-lapack=-llapack because the lapack
library files got installed in /usr/lib (I have no idea if those
arguments are sensible, they were just a guess). When that didn't
work, I tried installing the ACML library from AMD and using a similar
approach but the configure script still said that it couldn't find any
LAPACK support.
On Fri, May 30, 2008 at 1:08 AM, Synge Todo <wistaria@comp-phys.org> wrote:
Dear Miles Stoudenmire,
From: "Miles Stoudenmire" <miles@physics.ucsb.edu>
Date: Thu, 29 May 2008 22:38:58 -0700
It looks like it's referring to some funny lapack function from what I
can tell, so I think I should mention that I did not build the
libraries with lapack support (though the configure script claimed
that I'd be ok to compile classical monte carlo).
Thank you for your report. It seems that the classical Monte Carlo
also uses a SVD routine from lapack. So, this is a bug of configure
script. I will fixed it in the next version.
If it does turn out
to be because I need to add lapack support, I will have to ask some
questions about how to do that.
What kind of platform are you using?
Best,
Synge
--
-=Miles Stoudenmire=-
miles@physics.ucsb.edu
miles.stoudenmire@gmail.com
-- -=Miles Stoudenmire=- miles@physics.ucsb.edu miles.stoudenmire@gmail.com http://www.crazytheory.com/
participants (3)
-
Emanuel Gull
-
Miles Stoudenmire
-
Synge Todo