Re: [AMBER] include/fftw3.h(370): error: identifier "__float128" is undefined

From: Mark Rance <rancemd.gmail.com>
Date: Fri, 11 May 2012 15:51:48 -0400

Hi Tyler,

Dave Case had suggested I could use the -nofftw3 flag to get around the
compilation problem I had reported for the Intel compiler ( Version
12.1.2.273 Build 20111128). I tried this but initially it failed, reporting
the same error message I had gotten before. After poking around the
configure scripts, I found the following line in the configure2 script in
AmberTools/src:

        -nofftw3) rism='no'; pbsaflag=''; ;; #mdgx='no'; ;;

It seemed to me that the # symbol should not be there, as it presumably
comments out the rest of the line. So I removed the # symbol and
immediately preceeding ;; to end up with the following:

        -nofftw3) rism='no'; pbsaflag=''; mdgx='no'; ;;

and then the compilation completed successfully and the tests ran fine.

Regarding point 3 in your message, I had not done any modifications of
anything within the Amber12 code and configure scripts until the change I
mentioned above to get the -nofftw3 flag to work.






On Fri, May 11, 2012 at 11:13 AM, tluchko <tluchko.rutgers.edu> wrote:

> On 05/07/2012 08:10 AM, David A Case wrote:
> > On Sun, May 06, 2012, Mark Rance wrote:
> >> In attempting to compile amber 12 with the intel compilers, I
> encountered
> >> the following error message:
> >>
> >> include/fftw3.h(370): error: identifier "__float128" is undefined
> >> FFTW_DEFINE_API(FFTW_MANGLE_QUAD, __float128, fftwq_complex)
> > This sounds like the following problem, which is fixed in the latest
> version
> > (3.3.2) of fftw3:
> >
> > * Fix incompatibility with Intel icc which pretends to be gcc
> > but does not support quad precision.
> >
> > What I don't quite understand is that we tested the code with Intel
> 12.1.0
> > and 12.1.1; we need to find out if the problem is really unique to
> 12.1.2, or
> > whether some other problem is popping up.
> >
> >
> Hi Mark,
>
> We've tested FFTW 3.3 with Intel compilers 12.1.0 and 12.1.3 and have
> not seen any problems. There are a few possible solutions to your
> problem though.
>
> 1) -nofftw3. If you don't need RISM, MDGX or the FFT solver for PBSA
> you can skip compiling FFTW 3 completely by adding -nofftw3 to the
> configure command. FFTW 3 is not used by any other programs.
>
> 2) If you need FFTW 3, you can consider downloading FFTW 3.3.2 and
> replacing FFTW 3.3 in you AmberTools/src directory. Be sure to rename
> fftw-3.3.2/ to fftw-3.3/. If you are encountering the compiler bug
> identified by the FFTW developers, this should solve the problem.
>
> 3) Tells us more about your configure command and if you have made any
> modifications. AFAIK, quad precision FFTW should not be built by
> default so it is unclear why you are getting that error.
>
> Tyler
>
> _______________________________________________
> 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 Fri May 11 2012 - 13:00:03 PDT
Custom Search