Ross,
Bravo! -- mostly. Your diagnosis was right on. I had patched amber11
itself, but not the AmberTools1.4. Patching both as per you
specification did the trick, and I got a completely successful compilation.
But there remains a glitch in the `make test.cuda` part. The first time
I ran it (see output copied below), it complained that
"../../../bin/pmemd.cuda_: Command not found." Indeed, in
$AMBERHOME/bin there is pmemd.cuda_SPDP executable to which pmemd.cuda
is linked, but no pmemd.cuda_. I thought I could out-fox that by
putting another link in that directory, namely of pmemd.cuda_SPDP to
pmemd.cuda_. But simply running make test.cuda again gives the same
error. I don't follow the intricacies of the 'test.cuda' makefile. Do
I need to do some sort of a "make clean" before I can run the 'make
test.cuda' a second time?
Bob W.
******************************************************************************************************************************8
root.winter-linux: ...amber11/src (45%)# cd ../test
cd ../test
root.winter-linux: ...amber11/test (46%)# make test.cuda
make test.cuda
(find . -name '*.dif' -o -name 'profile_mpi' | \
while read dif ;\
do \
rm -f $dif ;\
done ;\
)
rm -f TEST_FAILURES.diff
./test_amber_cuda.sh
[: 35: unexpected operator
[: 42: unexpected operator
make[1]: Entering directory `/usr/local/lib/amber11/test'
cd cuda && make -k test.pmemd.cuda GPU_ID= PREC_MODEL=
make[2]: Entering directory `/usr/local/lib/amber11/test/cuda'
------------------------------------
Running CUDA Implicit solvent tests.
Precision Model =
GPU_ID =
------------------------------------
cd trpcage/ && ./Run_md_trpcage netcdf.mod
../../../bin/pmemd.cuda_: Command not found.0 file comparisons failed
./Run_md_trpcage: Program error
make[2]: *** [test.pmemd.cuda.gb] Error 1
cd gb_ala3/ && ./Run.igb1_ntc1_min netcdf.mod
../../../bin/pmemd.cuda_: Command not found.
./Run.igb1_ntc1_min: Program error
make[2]: *** [test.pmemd.cuda.gb.serial] Error 1
------------------------------------
Running CUDA Explicit solvent tests.
Precision Model =
GPU_ID =
------------------------------------
cd 4096wat/ && ./Run.pure_wat netcdf.mod
../../../bin/pmemd.cuda_: Command not found.
./Run.pure_wat: Program error
make[2]: *** [test.pmemd.cuda.pme] Error 1
make[2]: Target `test.pmemd.cuda' not remade because of errors.
make[2]: Leaving directory `/usr/local/lib/amber11/test/cuda'
make[1]: *** [test.pmemd.cuda] Error 2
make[1]: Target `test.serial.cuda' not remade because of errors.
make[1]: Leaving directory `/usr/local/lib/amber11/test'
0 file comparisons passed
0 file comparisons failed
9 tests experienced errors
Test log file saved as logs/test_amber_cuda/2011-04-12_19-02-35.log
No test diffs to save!
root.winter-linux: ...amber11/test (47%)# pwd
pwd
/usr/local/lib/amber11/test
root.winter-linux: ...amber11/test (48%)#
***************************************************************************************
On 4/12/11 6:27 PM, Ross Walker wrote:
> Hi Robert,
>
> The Radix sort header file was removed by the latest patch however the problem you are seeing look suspiciously to me like the bugfixes were only partially applied. One possibility is that the patching was run with an AMBERHOME environment variable pointing to the wrong place.
>
> I suggest (for sanity) going back to the original tar files for Amber 11 and AmberTools 1.4 and then patching and building cleanly from there. An example would be:
>
> ------------------------------------------
>
> cd $AMBERHOME
> cd ..
> rm -rf $AMBERHOME
>
> (Copy the original tar files here)
>
> tar xvjf AmberTools-1.4.tar.bz2
> tar xvjf Amber11.tar.bz2
>
> cd $AMBERHOME
>
> wget http://ambermd.org/bugfixes/AmberTools/1.4/bugfix.all
> patch -p0<bugfix.all
> rm -f bugfix.all
>
> wget http://ambermd.org/bugfixes/11.0/bugfix.all
> wget http://ambermd.org/bugfixes/11.0/apply_bugfix.x
>
> ./apply_bugfix.x bugfix.all
>
> cd $AMBERHOME/AmberTools/src
>
> ./configure -cuda gnu
>
> cd ../../src
> make cuda
>
> cd ../test
> make test.cuda
>
> ------------------------------------------
>
> Good luck,
>
> Ross
>
>> -----Original Message-----
>> From: Robert Wohlhueter [mailto:bobwohlhueter.earthlink.net]
>> Sent: Tuesday, April 12, 2011 9:19 AM
>> To: amber.ambermd.org
>> Subject: [AMBER] errors compiling amber11 for cuda
>>
>> Compiling AmberTools1.4 and amber11 on an AMD-64 4-core cpu under
>> Ubuntu
>> Linux 10.04 (after running bugfix.all, Apr. 12, 2011), with a GeForce
>> GTX 275 card and with the latest cuda toolkit installed
>> (cudatoolkit_3.2.16_linux_64_ubuntu10.04.run):
>>
>> The output from `configure` and `make` (copied below) seem to document
>> the following successes and failures:
>>
>> 1) configuring for -cuda and gnu successfully generate a config.h file
>> 2) cd'ing to the .../amber11/src dir and typing 'make cuda' initiates a
>> lot of compilation work, apparently successful.
>> 3) Until it get to make in dir
>> /usr/local/lib/amber11/src/pmemd/src/cuda, where fatal erros are
>> reported.
>> 4) The errors seem to stem from the absence of radixsort.h, which is
>> indeed not to be found in my $CUDA_HOME directory tree.
>>
>> Can anyone bail me out of the problem?
>>
>> Thanks, Bob Wohlhueter
>>
>> ***********************************************************************
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Apr 12 2011 - 17:00:03 PDT
