On Tue, Mar 04, 2008, Andreas Svrcek-Seiler wrote:
> here's what I observed while building amber10 (on x86_64 linux)
>
> 1) The "INSTALL" file misses a
> "cd ${AMBERHOME}/src"
> before step (2). This is trivial but bound to confuse some.
fixed in CVS ...thanks
>
> 0x000000000044fe05 in ptrajSetupIO
> (argumentStack=0x8351e0,type=TRANSFORM_TRAJIN) at ptraj.c:5737
> 5737 info->remdtrajtemp=boxtmp;
Tom knows about this, and promises to fix soon.
> 3) while building serial sander, ifort (and gfortran) give up with:
> ifort -c -w95 -vec_report0 -mp1 -ip -O3 -axWP -FR -I../../sander
> -I../src/include -DHAS_LAPACK -DQMMM -o ../obj/kinds.o ../src/main/kinds.F90
> error: can't open file ../obj/kinds.o for write
>
> This can be fixed by "mkdir ${AMBERHOME}/src/dcqtp/obj" (no one else
> else have a problem here ?)
I think this is a CVS glitch, and would not be in the tar files sent to users.
I've added a README file to the obj directory to force CVS to create this.
>
> 4) Possible compiler bug in ifort 10.1.013: Only with this compiler I get
> a lot of errors like:
> ifort -c -w95 -vec_report0 -mp1 -ip -O3 -axWP -FR -I../../sander
> -I../src/include -DHAS_LAPACK -DQMMM -o ../obj/qm_mm_div.o
> ../src/qmmm/qm_mm_div.F90
> fortcom: Error: qm_mm_div.F90, line 100: The rank of the
> allocate-shape-spec-list differs from the rank of the allocate-object.
> [LINK_INDEX]
> allocate(link_index(qmmm_struct%nlink+1),stat=ier)
> ---------------------^
> <truncated here>
> and the build fails of course. Being nearly f90-illiterate I can't tell
> whether the trouble stem from ifort or from not so standard-conforming
> code.
Ross or I will look at this.
>
>
> double precision, parameter, private :: TINY = dble(0.000001)
Should be fixed in this morning's CVS updates. I don't know if this fixes
all pathscale problems or not.
>
> ...but this is a <4 letter word here>-compiler anyway.
...agreed.
...thanks for the reports!....dave
Received on Wed Mar 05 2008 - 06:07:35 PST
