The ABINIT Users Mailing List ( CLOSED )

[Masayoshi Mikami <mmikami@rc.m-kagaku.co.jp>]

Dear Torben,

I almost forget the details of the compiler options
for HITACHI SR* machines, but I just thought that
why don't we reduce (or remove) the optimization
and stick to the simplest (lowest) options ?

For example, following the makefile macros
Machine_dept_files/HITACHI/makefile_macros.mikami001025.seq
we could remove "-pvfunc=3 "... use -O1(or 2, or no-optimization
as the last resort) instead of -O4,  use -noparallel, ... etc.
(What is the " -W0,'OPT(ROUGHPOW(1))" ? We could remove this as well.)
Such "trial and error" may be needed. Good luck.

In passing, the error message means:
"The value of the actual argument XX in the DLOG or DLOG10
 intrinsic function is zero or negative."
(Could you point out the location where the error occurs in the source  
?)

From my old experience on the HITACHI,
the versions of compiler/OS might also be crucial...
Upgrading the compiler may induce such errors
even without touching  the source codes.

Keep my fingers crossed,
Masayoshi

On 2005/02/24, at 14:07, Torben M. Hansen wrote:

> On Thursday 24 February 2005 01:28, Michel C �é  wrote:
> > The best is to try to run in serial mode to see if the error is  
> > reproduced
> > but I think not and it will take the same amount of time.
>
> Now I tried to run it in serial mode and the error is the same (see  
> log and
> makefile_macros below). I also tried on my PC (Pentium4,Linux), but  
> did not
> get any error (though it never converged as predicted).
>
> Thanks
> Torben
>
> -----------Log file ------
        
        (snipped ...)

> ======================================================================= 
> ==
>  fconv : at Broyd/MD step   1, gradients have not converged yet.
>   max grad (force/stress) = 5.0578E-02 > tolmxf= 5.0000E-04 ha/bohr  
> (free
> atoms)
>
>
>  BROYDEN STEP NUMBER   2
> ------------------------------------------------------
>  KCHF141R argument 0.0000000000000000E+000 for dlog/dlog10 is 0 or  
> less.
>  0x0000000100007d8c  hf_err+0x4c4
>  0x0000000100096ff8  ppc_err+0xa8
>  0x0000000100097198  ._hf_dlog+0x168
>  0x00000001004c4f2c  findmin+0x43c
>  0x00000001005cfe88  brdene+0x624
>  0x00000001005ca7b0  brdmin+0x5fcc
>  0x0000000100403bc4  gstate+0x10708
>  0x0000000100189668  driver+0x159fc
>  0x0000000100004760  abinit+0x4338
>  0x00000001000002d4  __start+0x94
>
>
>
>
> -------------------------------------------------------
> # makefile_macro for Hitachi SR11000
> # Machine type
> MACHINE=sr8k
> #FC=mpif90_r  #parallel
> FC=f90  #serial
> #FFLAGS= -w -O4 -pvfunc=3 -rdma -pvec -parallel=3 -64 -nolimit \(cont)
> -loopdiag -pvdiag -pardiag=2 "-W0,'OPT(ROUGHPOW(1))'" \(cont)
> "-W0,'FORM(FREE)'" #parallel
> FFLAGS= -w -O4 -pvfunc=3 -64 -nolimit -loopdiag \(cont)
> "-W0,'OPT(ROUGHPOW(1))'" "-W0,'FORM(FREE)'"  #serial
> #FFLAGS_LIBS= -w -O4 -pvec -pvfunc=3 -rdma -parallel=3 -64 -nolimit  
> \(cont)
> -loopdiag -pvdiag -pardiag=2 "-W0,'OPT(ROUGHPOW(1))'" \(cont)
> "-W0,'FORM(FIXED(80))'" #parallel
> FFLAGS_LIBS= -w -O4 -pvec -pvfunc=3 -rdma -64 -nolimit -loopdiag  
> \(cont)
> -pvdiag "-W0,'OPT(ROUGHPOW(1))'" "-W0,'FORM(FIXED(80))'"  #serial
> #FLINK= -parallel -64 "-Wl,'-bloadmap:loadmap'" #parallel
> FLINK="-Wl,'-bloadmap:loadmap'" -64 # serial
>
> AR_COMMAND= ar -X64
>
> # C preprocessor, used to preprocess the fortran source.
> CPP=/usr/ccs/lib/cpp
> CPP_FLAGS=-P
>
> # C optimized compilation.
> CC=mpcc_r  #parallel
> #CC=cc  #serial
> CFLAGS=-O3 -q64 -qstrict #This is probably wrong but there is no error
> messager
>
> # Location of perl . Used to generate the script fldiff, in
> ~ABINIT/Utilities .
> PERL=/bin/perl
>
> ####################################################################
> FFLAGS_PAR=$(FFLAGS) -I/usr/mpi/include/mpif.h
> # Location of the MPI library
> LIBS_PAR=$(LIBS) -L/usr/mpi/lib/lib64 -lmpi
> # This is a last line in makefile_macros ----------