From: Peter Freddolino (petefred_at_ks.uiuc.edu)
Date: Mon Apr 20 2009 - 14:47:09 CDT
Hrm, see if you can compile with -DNAMD_DISABLE_SSE.
If it works you'll probably want to check against gcc performance to see
which works better...
Peter
BIN ZHANG wrote:
> Dear peter:
>
> Thanks for your reply. I just modified the arch file according to your
> suggestion, but the exactly same error showed up again.
>
> Now my .arch file looks like:
>
> =======================================================
> NAMD_ARCH = CRAY-XT3
> CHARMARCH = mpi-crayxt3
> CHARMOPTS =
> # need -lgmalloc because normal Catamount malloc is horribly slow
>
> CXX = CC -I/usr/include -DMEMUSAGE_USE_SBRK -DNO_SOCKET -DDUMMY_VMDSOCK
> -DNOHOSTNAME -DNO_GETPWUID -DNAMD_NO_STDOUT_FLUSH -DNAMD_NO_O_EXCL
> -DMISSING_mm_cvtsd_f64
> # added the -DMISSING_mm_cvtsd_f64 flag to avoid pgi inline problem.
> CXXOPTS = -fast
> CXXNOALIASOPTS = -Msafeptr=arg,global,local,static -Minfo=all
> -Mneginfo=loop
> #CXXNOALIASOPTS = -fast -Mnodepchk -Msafeptr=arg,global,local,static
> -Minfo=all -Mneginfo=loop
> # -fastsse with -Mnodepchk -Msafeptr=arg,global,local,static produces
> bad code
>
> CC = cc
> COPTS = -fast
>
> =======================================================
>
> Thanks,
> Bin
>
>
> On Apr 20, 2009, at 11:12 AM, Peter Freddolino wrote:
>
>> You should be able to build using PGI if you add -DMISSING_mm_cvtsd_f64
>> to your build flags. The current version of the PGI compiler doesn't
>> fully support SSE2; we've already reported this as a bug to them, and it
>> will be fixed in the next release of their compiler.
>>
>> I've never compared the speed of PGI vs. gcc, certainly not on that
>> platform. The frequent comment that gcc hurts performance is in
>> comparison to icc on x86-related architectures.
>>
>> Best,
>> Peter
>>
>> BIN ZHANG wrote:
>>> Hi, Joachim:
>>>
>>> I will definitely try this. But I heard before that gcc might affect the
>>> efficiency of NAMD significantly, did you notice this?
>>>
>>> Another thing that might worth mention is that, I can compile NAMD2.6
>>> with PGI on the same machine without any problem.
>>>
>>> Thanks,
>>> Bin
>>>
>>>
>>> On Apr 20, 2009, at 4:05 AM, Joachim Hein wrote:
>>>
>>>>> zhngbn_at_gmail.com wrote:
>>>>>> Dear all:
>>>>>>
>>>>>> I was trying to compile the NAMD2.7b1 version on a CRAY-XT3
>>>>>> supercomputer, and get the following error that I do not understand.
>>>>>>
>>>>>> PGCC-W-0277-Cannot inline function _mm_set_pd - data type mismatch
>>>>>> (/opt/pgi/7.0.7/linux86-64/7.0-7/include/CC/iostream: 882)
>>>>>> PGCC-W-0277-Cannot inline function _mm_set_pd - data type mismatch
>>>>>> (/opt/pgi/7.0.7/linux86-64/7.0-7/include/CC/iostream: 883)
>>>>>> PGCC-W-0277-Cannot inline function _mm_set_pd - data type mismatch
>>>>>> (/opt/pgi/7.0.7/linux86-64/7.0-7/include/CC/iostream: 884)
>>>>>>
>>>>>> Does this look familiar to you guys? Any idea what would be the
>>>>>> problem?
>>>>>>
>>>>>> Thanks a lot.
>>>>>> Bin
>>>> Hi Bin,
>>>>
>>>> try using the gcc compiles on a Cray XT.
>>>>
>>>> module swap PrgEnv-pgi PrgEnv-gnu
>>>>
>>>> You need to do this for all: charm++, FFTW, TCL and namd
>>>>
>>>> I have no problems building XT executables with gcc and have not
>>>> received any problem reports for the ones I build (knock on wood).
>>>>
>>>> Hope this helps.
>>>>
>>>> Best wishes
>>>> Joachim
>>>>
>>>>
>>>>
>>>> --
>>>> The University of Edinburgh is a charitable body, registered in
>>>> Scotland, with registration number SC005336.
>>>>
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:52:38 CST