Hi Brent,
Thank you for the messages. Now this new error is most likely due to a
bug in your intel compiler (v12.0?), which could not handle compiling
c++ file with <math.h> included after preprocessing. It was documented
in this intel page,
https://software.intel.com/en-us/articles/use-intel-cc-compilers-v1201107-or-higher-version-to-resolve-error-when-preprocessing-and
. This bug, more precisely is discrepancy between gnu and intel.
Unfortunately pbsa.cuda uses <math.h> in c++ file, and we recommend use
gnu or higher version intel. So right now, you can just comment out the
pbsa.cuda compile in AmberTools Makefile as suggested by Dave to work
this out.
Best,
Ruxi
On 4/7/17 18:18, Brent Krueger wrote:
> Ruxi and I have been trying a few ways to try to work around my old Intel
> 12.0 compiler and I'm afraid that we are unable to get anything simple to
> work (Ruxi -- latest error messages below). I think that for things to
> work on my system I'm going to need some kind of no-cuda-AT option.
> Preferably this would skip all of the cuda-related things that are in
> AmberTools but still build pmemd.cuda in amber16 the same way it did
> previously.
>
> This may mean that we can no longer support older compilers with cuda?
> Something we can worry about more with amber18.
>
>
> Thanks,
> Brent
>
>
>
> /usr/local/cuda/bin/nvcc -gencode arch=compute_20,code=sm_20 -gencode
> arch=compute_30,code=sm_30 -gencode arch=compute_50,code=sm_50 -gencode
> arch=compute_52,code=sm_52 -gencode arch=compute_53,code=sm_53
> -use_fast_math -O3 -ccbin icpc -I../cusplibrary-0.6.0-rc -o
> cuda_cg_wrapper.o -c cuda_cg_wrapper.cu -DCUSP -DPCG -DJacobi -DDIA
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(235): error:
> allowing all exceptions is incompatible with previous function "__isinff"
> /usr/local/cuda/bin/..//include/math_functions.h(7279): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(236): error:
> allowing all exceptions is incompatible with previous function "__isinf"
> /usr/local/cuda/bin/..//include/math_functions.h(7294): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(238): error:
> allowing all exceptions is incompatible with previous function "__isinfl"
> /usr/local/cuda/bin/..//include/math_functions.h(7632): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(245): error:
> allowing all exceptions is incompatible with previous function "__isnanf"
> /usr/local/cuda/bin/..//include/math_functions.h(7280): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(246): error:
> allowing all exceptions is incompatible with previous function "__isnan"
> /usr/local/cuda/bin/..//include/math_functions.h(7293): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(248): error:
> allowing all exceptions is incompatible with previous function "__isnanl"
> /usr/local/cuda/bin/..//include/math_functions.h(7633): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(275): error:
> allowing all exceptions is incompatible with previous function "__finitef"
> /usr/local/cuda/bin/..//include/math_functions.h(7291): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(276): error:
> allowing all exceptions is incompatible with previous function "__finite"
> /usr/local/cuda/bin/..//include/math_functions.h(7290): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(278): error:
> allowing all exceptions is incompatible with previous function "__finitel"
> /usr/local/cuda/bin/..//include/math_functions.h(7631): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(285): error:
> allowing all exceptions is incompatible with previous function "__signbitf"
> /usr/local/cuda/bin/..//include/math_functions.h(7297): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(286): error:
> allowing all exceptions is incompatible with previous function "__signbit"
> /usr/local/cuda/bin/..//include/math_functions.h(7292): here
>
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(288): error:
> allowing all exceptions is incompatible with previous function "__signbitl"
> /usr/local/cuda/bin/..//include/math_functions.h(7625): here
>
> /usr/local/cuda/bin/..//include/math_functions.h(8862): warning: exception
> specification is incompatible with that of previous function
> "signbit(double)"
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(281): here
>
> /usr/local/cuda/bin/..//include/math_functions.h(8870): warning: exception
> specification is incompatible with that of previous function
> "isfinite(double)"
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(261): here
>
> /usr/local/cuda/bin/..//include/math_functions.h(8880): warning: exception
> specification is incompatible with that of previous function "isnan(double)"
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(241): here
>
> /usr/local/cuda/bin/..//include/math_functions.h(8888): warning: exception
> specification is incompatible with that of previous function "isinf(double)"
> /opt/intel/composerxe-2011.0.084/compiler/include/math.h(231): here
>
> 12 errors detected in the compilation of
> "/tmp/tmpxft_00002718_00000000-19_cuda_cg_wrapper.compute_53.cpp1.ii".
> make[2]: *** [cuda_cg_wrapper.o] Error 2
> make[2]: Leaving directory
> `/extra/krieg/source/amber16_AT17_rc2/AmberTools/src/pbsa'
> make[1]: *** [cuda_serial] Error 2
> make[1]: Leaving directory
> `/extra/krieg/source/amber16_AT17_rc2/AmberTools/src'
> make: *** [install] Error 2
>
>
>
> On Fri, Apr 7, 2017 at 3:03 PM, Ruxi Qi <ruxiq.uci.edu> wrote:
>
>> Hi Brent,
>>
>> Commenting out just #error line in "#if defined(__ICC)" block should do
>> the trick. Or you can remove the whole block from "#if defined(__ICC)"
>> to "#endif /*__ICC */".
>>
>> Thanks!
>>
>> Ruxi
>>
>>
>> On 4/7/17 10:41, Brent Krueger wrote:
>>> Ruxi,
>>>
>>> Sorry, but our cluster is having some issues right now so I can't
>> actually
>>> do anything at the moment from the command line. And class is starting
>> in
>>> a few minutes, so I probably won't be able to actually look around for
>>> quite awhlie.
>>>
>>> I was able to find the file you are talking about. Are you suggesting
>> that
>>> I comment out just the #error line? I'm afraid I have not used this
>> kind
>>> of #if compiler directive -- can I just remove that line -- or does that
>>> screw up the logic of the if above and/or below as well. Should I change
>>> the if statement that goes along with it instead?
>>>
>>>
>>> Thanks,
>>> Brent
>>>
>>>
>>> On Fri, Apr 7, 2017 at 1:33 PM, Ruxi Qi <ruxiq.uci.edu> wrote:
>>>
>>>> Hi Brent,
>>>>
>>>> Sorry for my jumping to the conclusion. I just realized it's impossible
>>>> to fully remove <cuda_runtime.h>. As we used other libraries cublas and
>>>> cusparse which also include that header file. I am afraid this option
>>>> doesn't work, and we have to go back tweaking the configure to fix this
>>>> issue.
>>>>
>>>> Right now If you want, you can comment out the #error line in you
>>>> cuda/include/host_config.h to see whether it gets fixed. Thanks!
>>>>
>>>> Best,
>>>>
>>>> Ruxi
>>>>
>>>> On 04/07/2017 05:32 AM, Brent Krueger wrote:
>>>>> Thank you Ruxi for identifying the source of this issue.
>>>>>
>>>>> I've commented out the cuda_rutime.h line from the file
>>>> cuda_cg_wrapper.cu
>>>>> in AmberTools/src/pbsa, did a make clean, reran configure -cuda intel
>> and
>>>>> that build also seemed to fail in the same spot. I've done a make
>>>>> uninstall and am now building again.
>>>>>
>>>>> I'm afraid I've got to run to a meeting now and I've got meetings/class
>>>>> pretty much all day so I'm not sure when I'll be able to get to
>> testing,
>>>>> but I should at least be able to verify whether this build worked for
>> me
>>>> or
>>>>> not in a couple hours...
>>>>>
>>>>>
>>>>> Cheers,
>>>>> Brent
>>>>>
>>>>>
>>>>> On Fri, Apr 7, 2017 at 2:42 AM, Ruxi Qi <ruxiq.uci.edu> wrote:
>>>>>
>>>>>> Just found a similar story with GCC in pmemd.cuda of Amber12. See
>>>>>> http://archive.ambermd.org/201307/0456.html
>>>>>>
>>>>>> Best,
>>>>>>
>>>>>> Ruxi
>>>>>>
>>>>>> On 04/06/2017 11:38 PM, Ruxi Qi wrote:
>>>>>>> Hi All,
>>>>>>>
>>>>>>> I found the real cause. It is the <cuda_runtime.h> included in our
>>>>>>> cuda wrapper code (not cusp) that triggered the ICC version checking
>>>>>>> in host_config.h. In contrast, pmemd.cuda doesn't use this header,
>>>>>>> thus won't trigger the runtime compiler checking. A quick "fix" is
>>>>>>> just removing the <cuda_runtime.h> inclusion in our code as long as
>> we
>>>>>>> use nvcc to compile.
>>>>>>>
>>>>>>> Two concerns:
>>>>>>>
>>>>>>> 1) As shown by those OS/Compiler/CUDA compatibility matrices from
>>>>>>> Nvidia, allowing the use of the compiler not officially supported
>>>>>>> (tested) by CUDA is of risk. I don't know to what extent this effect
>>>>>>> will be, maybe not a big deal?
>>>>>>>
>>>>>>> 2) For now, we can fix this compiler error either through a) Add a
>>>>>>> logic checking on the icc version to configure as suggested
>>>>>>> previously, or b) I remove the <cuda_rutime.h> inclusion in our code
>> -
>>>>>>> this will need another round combination tests. Also just mention
>>>>>>> again, user manually commenting out the #error line in host_config.h
>>>>>>> will also be a workaround (not recommended?).
>>>>>>>
>>>>>>> Thanks!
>>>>>>>
>>>>>>> Ruxi
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> On 04/06/2017 07:55 PM, Daniel Roe wrote:
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> On Thu, Apr 6, 2017 at 6:59 PM, Ruxi Qi <ruxiq.uci.edu> wrote:
>>>>>>>>> 1. I once encountered the same error. Indeed, this error is not
>>>> because
>>>>>>>>> of linking CUSP library. It's nvcc that complains the outdated ICC
>>>>>>>>> version. Per the Nvidia official documents, CUDA 7.5 only supports
>>>> ICC
>>>>>>>>> 15.0.0, as shown bellow:
>>>>>>>> This is not really the case, at least for Intel compilers version
>> 14.
>>>>>>>> I was just able to compile pmemd.cuda from Amber 16 and
>>>>>>>> cpptraj/pmemd.cuda from amber17-with-patches with Intel V 14
>> compilers
>>>>>>>> and CUDA 7.5. I think this issue is really CUSP-related since as
>> Brent
>>>>>>>> stated he was previously able to compile pmemd.cuda from Amber 16
>> with
>>>>>>>> Intel V12 compilers and CUDA 7.5.
>>>>>>>>
>>>>>>>>> 3. Adding a --no-pbsa.cuda option is good for safety's sake in this
>>>>>>>>> release. I agree with Hai that we may put it built as default,
>> since
>>>> it
>>>>>>>>> generally follows the Nvidia CUDA installation rule.
>>>>>>>> Probably the best solution would be to add logic to 'configure' that
>>>>>>>> will disable pbsa.cuda if the Intel compiler version is not high
>>>>>>>> enough. May need to do the same thing for GNU compilers if there is
>> a
>>>>>>>> similar issue.
>>>>>>>>
>>>>>>>> -Dan
>>>>>>>>
>>>>>> _______________________________________________
>>>>>> AMBER-Developers mailing list
>>>>>> AMBER-Developers.ambermd.org
>>>>>> http://lists.ambermd.org/mailman/listinfo/amber-developers
>>>>>>
>>>> _______________________________________________
>>>> AMBER-Developers mailing list
>>>> AMBER-Developers.ambermd.org
>>>> http://lists.ambermd.org/mailman/listinfo/amber-developers
>>>>
>>>
>>
>> _______________________________________________
>> AMBER-Developers mailing list
>> AMBER-Developers.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber-developers
>>
>
>
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Fri Apr 07 2017 - 19:30:02 PDT
