I was finally able to build AmberTools 1.5 and then Amber using the
built-in libnetcdf source. Then, I successfully built pmemd.cuda. This
is what I did:
1) AmberTools 1.5 setup
=======================
- extract, set AMBERHOME, bugfix
- ./configure intel
- make serial
2) Amber 11 SERIAL setup
========================
- extracted to the same directory tree as AmberTools 1.5 from above
- bugfix
- AT15_Amber.py
- cd $AMBERHOME/src
- make serial
- cd $AMBERHOME/test; make test
-> all relevant tests passed (cf.
http://archive.ambermd.org/201107/0510.html)
3) Amber 11 PARALLEL setup
==========================
- cd $AMBERHOME/AmberTools/src
-wget
http://www.open-mpi.org/software/ompi/v1.5/downloads/openmpi-1.5.4.tar.gz
- setup of openmpi 1.5.4 using AmberTools/src/configure_openmpi
(modified version check from 1.5.2 to 1.5.4)
- export PATH=$AMBERHOME/bin:$PATH
- export LD_LIBRARY_PATH=$AMBERHOME/lib:$LD_LIBRARY_PATH
- cd $AMBERHOME/AmberTools/src
- ./configure -mpi intel
- AT15_Amber.py
- cd $AMBERHOME/src
- make clean
- make parallel
- cd $AMBERHOME/test; export DO_PARALLEL='mpirun -np 4'
- make test.parallel
-> 297 tests passed / 1 failed (due to too many processes)
4) Amber 11 CUDA setup
======================
- cd $AMBERHOME/AmberTools/src
- make clean
- ./configure -cuda intel
- AT15_Amber11.py
- cd $AMBERHOME/src
- make clean
- make cuda
- cd $AMBERHOME/test; ./test_amber_cuda.sh -1 SPDP
54 file comparisons passed
3 file comparisons failed:
- possible FAILURE: check mdout.tip4pew_oct_nvt.dif
- possible FAILURE: check mdout.tip4pew_box_npt.dif
- possible FAILURE: check mdout.dhfr.ntb2.dif
Is that fine?
Thanks.
On 11/02/2011 11:48 AM, Jan-Philip Gehrcke wrote:
> Thanks for your answer, Dan. It looks like extracting Amber11 and
> AmberTools 1.5 into the same directory tree before building AmberTools
> triggered the problem. Detailed comments below.
>
> On 11/02/2011 03:41 AM, Daniel Roe wrote:
>> Hi,
>>
>> Just a quick question; were you configuring/building AmberTools from a
>> clean directory? If not, did you do a 'make clean' before 'make
>> install'?
>
> I was "playing around" a lot, but always invoked `make clean` before
> `make serial` (which is the suggested make command by the configure
> script).
>
> In fact -- and this paragraph is a totally different topic now --
> inbetween I ran into some trouble with a "wrong"
> AmberTools/src/sff/lex.mm_options.c file as described here:
> http://archive.ambermd.org/201110/0165.html. That happened after the
> first successful build of netcdf and I am pretty sure that I cleaned
> before. Renaming the file solved the problem.
>
> The wrong libnetcdf build definitely had something to do with the
> extraction of AmberTools 1.5 and Amber 11 in the same directory tree
> *before* building AmberTools. I just did a fresh AmberTools directory:
>
> 1) extract AmberTools
> 2) set AMBERHOME
> 3) bugfix.all
> 4) configure gnu
> 5) make serial
>
> This builds libnetcdf correctly.
>
> During this build using GNU compilers, I experienced the safemem.f issue
> described here: http://archive.ambermd.org/201110/0418.html. To answer
> Jason's question in this mail: In my opinion this is worth a bugfix.
> Including that fix, AmberTools 1.5 was built properly.
>
>
>
>> It seems that (as you say) the library is not being generated
>> correctly on your system. I've been able to successfully build the
>> netcdf version that comes with AmberTools using both gnu 4.6 and intel
>> 12 compilers. It may be that the issue has to do with the tools used
>> to generate the library itself. What versions of 'ar' and 'ranlib' do
>> you have?
>
> Although this shouldn't matter anymore now:
> GNU ar (GNU Binutils for Ubuntu) 2.20.1-system.20100303
> GNU ranlib (GNU Binutils for Ubuntu) 2.20.1-system.20100303
>
>
>> Other than that, would you mind sending me the
>> $AMBERHOME/AmberTools/src/netcdf_config.log that is generated after
>> you run './configure' on your system with the original AmberTools
>> netcdf (i.e. not the one you downloaded and installed manually).
>> Thanks!
>
> To get this configuration log for the build of libnetcdf that does *not*
> work, I've used the "old" sources of libnetcdf again, `make clean`ed,
> `./configure gnu`ed and `make serial`ed in $AMBERHOME/AmberTools/src,
> while $AMBERHOME was the directory tree containing both AmberTools 1.5
> and Amber 11 (both bugfixed). `make serial` failed again due to an
> invalid libnetcdf.a.
>
> You find the corresponding failed_netcdf_makelog and
> failed_netcdf_config.log attached to this mail. I had checked the
> configure log before and it looked good to me.
>
> It looks like I've done things in the "wrong" order, although I was
> reading (not understanding?) the PDF manuals while setting things up.
> What is the canonical way of building AmberTools and Amber 11? I have
> the feeling that the PDF manual did not tell me that I must not extract
> Amber11 into the AMBERHOME directory tree before building AmberTools.
>
> I did it this way:
>
> 1) extract AT1.5
> 2) set AMBERHOME
> 3) bugfixes to AT1.5
> 4) extract Amber11 to same directory tree
> 5) applied bugfixes to Amber11
> 6) configure/make in $AMBERHOME/AmberTools/src
>
> In step 6, libnetcdf build failed.
>
> Currently, I'm compiling a lonely AmberTools 1.5 with Intel compilers.
> Then, I will add Amber 11 to that directory tree, patch it and proceed..
>
> Jan-Philip
>
>>
>> -Dan
>>
>> On Tue, Nov 1, 2011 at 4:45 PM, Jan-Philip Gehrcke
>> <jgehrcke.googlemail.com> wrote:
>>> Hey,
>>>
>>> on Ubuntu 10.04, I have the same problem building ptraj with NetCDF
>>> support as described in these threads:
>>>
>>> http://archive.ambermd.org/201004/0680.html
>>> http://archive.ambermd.org/201106/0731.html
>>>
>>> The error for me looks like:
>>>
>>> icc -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -o ncgen main.o load.o
>>> ncgentab.o escapes.o getfill.o init.o genlib.o
>>> ../libsrc/.libs/libnetcdf.a
>>> ../libsrc/.libs/libnetcdf.a(attr.o): In function `nc_put_att_text':
>>> attr.c:(.text+0x179a): undefined reference to `ncx_pad_putn_text'
>>>
>>> and many more of "undefined references".
>>>
>>> > nm libnetcdf.a | grep ncx_pad_putn_text
>>> > U ncx_pad_putn_text
>>> > U ncx_pad_putn_text
>>>
>>> This shows that libnetcdf.a has been built with undefined symbols like
>>> `ncx_pad_putn_text`. I've reproduced this by running configure/make in
>>> the AmberTools/src/netcdf/src directory. `configure --disable-utils`
>>> deactivates the `ncgen` compilation which works around the problem.
>>> However, during make of AmberTools, libnetcdf.a is used in different
>>> contexts and therefore `--disable-utils`` does not help for building
>>> AmberTools.
>>>
>>> I think that libnetcdf.a is somehow built wrong on "newer" systems. What
>>> do you suggest? Could you even switch to a newer version of the NetCDF
>>> library in AmberTools? I will try building a newer version of
>>> libnetcdf.a and see if AmberTools builds against it...
>>>
>>> Any hint is appreciated!
>>>
>>> Jan-Philip
>>>
>>>
>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>
>>
>>
>>
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Nov 02 2011 - 08:30:04 PDT
