Re: amber-developers: AIX test case failures

From: Scott Brozell <sbrozell.scripps.edu>
Date: Fri, 11 Apr 2008 19:36:52 -0700 (PDT)

Hi,

On Fri, 11 Apr 2008, Ross Walker wrote:

> AIX problems:
>
> 1) Many problems compiling ambertools on AIX. - no xlCC / xlf90 options etc.

Yes, non gnu building of AmberTools seems to have been an afterthought.
(Do any AIX's have gnu ?)
Dave should articulate the plan here.
If we go non gnu then configure_at should be made consistent
with configure_amber.

> 2) DCQTP is broken on AIX - I have checked in a new configure_amber script
> that sets -nosanderidc as the default for AIX as a workaround.
>
> 3) Multiple test case problems as follows:
>
> There are multiple test case failures on xlf90_aix. A large proportion of
> these are due to test cases having "&end" instead of "/".
>
> grep -R "&end" $AMBERHOME/test/*
>
> returns a HUGE number of entries (including lots of save files) that will
> take a while to update. Note a lot of the NCSU test cases also seem to be
> lacking the initial space in front of &cntrl.
>
> I can try and update all these but it will take a while.

I can fix these in short order and shall do so now.

It seems that aix is not standard conforming.?
Here is what fortran expert Ron Shepard said in 2005:

The f90 namelist termination character is "/", and it can occur in any
column. The defacto standard f77 namelist terminations were "$end" or
"&end", with the first character in the second column. In the 90's
when we were using a mix of f77 and f90, a common "trick" that worked
for both types of compilers was to end the namelist data with "/&end"
with the first character in column 1. That ended namelist in f90
because of the "/" character, and it ended (most) namelist input in f77
compilers because the "/" was ignored and the "&end" was found
beginning in column 2.
These days, this should not be necessary. At least in the very near
future, there should be no need for any f77 artifacts in the codes.

Scott

> Several other test cases fail for unexplained reasons reading namelists -
> mainly when the namelist is in an external file and quite large. There are
> also numerous problems with allocation that "may" go away if compiled with
> -q64 - I will attempt to try this over the weekend.
>
> Test case problems:
>
> rdc: ./Run.dip: Program error
> Chemical shifts will be read from file: gcg10.shf
> namelist reports error reading &shf
>
> pcsa: ./Run.dip: Program error
> namelist reports error reading &align
>
> pcsa: ./Run.csa: Program error
> namelist reports error reading &align
>
> pcsa: ./Run.csa: Program error
> namelist reports error reading &align
>
> pcsa: ./Run.csa_min: Program error
> namelist reports error reading &align
>
> LES: ./Run.PME_LES
>
> addles:
> exec(): 0509-036 Cannot load program ../../exe/addles because of the
> following errors:
> 0509-026 System error: There is not enough memory available now.
> ./Run.PME_LES: Program error
> make: 1254-004 The error code from the last command is 1.
>
> LES_GB test cases: All &end need to be replaced with /
>
> pbsa_trx: ./Run.trxox.min: Program error
> SA Bomb in sa_arc(): Allocation aborted 0 1 1 0 0
>
> pbsa_radi: All fail because of &end
>
> pbsa_tsr:
> SA Bomb in sa_arc(): Allocation aborted 0 0 1 1 1
>
>
> All other test cases including QMMM and DFTB pass.
Received on Fri Apr 18 2008 - 21:18:43 PDT
Custom Search