[Wien] Bug Report: Wien2k 9.1

[Scott Beardsley]

> Our group at UC Davis recently compiled Wien2k 9.1  I
> employed plain option (-O2 without any sophisticated optimization) and found
> the following bugs:

A few of these (and a couple others that you've missed) were reported on
this list earlier (2/6/9 I believe). You might want to check out the
archives. I'm not sure if there is a place to get all of the latest
patches or not.

> 1. For 64bit compiler, default integer is declared as integer*8,
> not integer*4, which crashes with doreallocate_i4_dX routine. You may
> use option "-i4" at 64 bit compilation.

You didn't say what compiler you were using (I'm assuming gfortran or
Intel). In our case (ie on urd), "-i4" is the default when using
Pathscale per "man pathf90".

BTW, you might want to use the Pathscale compiler on urd, we've found
Intel to be noticeably slower on this hardware. GCC 4.3 is getting close
in some cases. In any case, I'd be interested to see the performance
differences (can you run mpi-benchmark?).

> 2. Some integer operations use "NEQV", which is used for logical variables.
> Change them as "/="

I found this when compiling for urd also. I used NE to fix it which
should be equivalent.

> 3. At SRC_lapw1, service.f implements "real function dtime4(lasttime)"
> but it should be defined as double precision, like "double precision
> function dtime4(lasttime)"

I missed this one. I fixed, recompiled, and just installed it on urd.

> 4. At SRC_trig, change "lattice" => "lattic" at #30 of xyz2struct.f

Yep I found that one also.

Scott