| heinz456 -> RE: Toying around with wcc (Feb. 16, '06, 2:45:26 AM) |
Hi,
and thanx for your fast reply. First up, i went the road you
sugessted(except that consultant thing, man i am into computers since '82).
I started with SFU and got to the point, where all except
one thing was up and could run, if i had a chance to. The problem
was, that gcc seg faulted on one specific file. So i tried another version of gcc(4.0.1)
and had the same outcome (segmentation fault). Then i tried to get the hotfix, mentioned
in one of the other posts. But to get this baby i would have spent to much time on wrestling
around with some accountants here (No offense). But there was plan b (mingw) and plan c (cygwin)
(... and also plan d, but that got dropped due to missing contacts to the russian mafia ...;-) )
Mingw was dropped, because pthreads weren't directly supported the way i wanted it to be, and as
i recall, the posix compability is missing.
After some long and painfull installation sessions i had cygwin up and running.
And i presume a couple of minutes later the whole software package compiled cleanly.
Some minor bugs later even the test examples provided, gave the results i expected.
But the ultimate goal is still, to convert the whole package into a microsoft thing, involving:
a) Correcting the sourcecode, where needed.
(The software package was written in plain C, without any signals and other hardcore Unix stuff)
Only paths, timer and some minor functions (file access and getpid()) had to be corrected (No problem).
The sourcecode compiled cleanly (read : lots of warnings, but no errors, as of the Linux version
i use as reference) Btw there is a pthread implemantation.. .
b) Convert to several visual studio projects for future ms only stuff (wmi).
No problems at all.
c) Convert the test examples and get them up and runnig.
d) Correcting errors. hmm...
e) Building a ms install package.
Result of c) ähh, ähh, ähh, ALARM. This is the point MPI, etc. pp. comes
into play. The problem right now is:
The test examples (C, Fortran77..95, High Performance Fortran, some using MPI)
are organized in a big directory tree, where the
bottom of the subtree are the test programs. To compile this, you
only need cd to the root directory of this subtree, do a "make" and
there you go. To start all the test programs you invoke a "make run".
This is when nmake comes into play and failed misarebly. To convert the
Makefiles on the bottom is not that problem, but the rest is a problem,
because of the number of files etc. pp. ok, that means scripting and converting
to nmake style makefiles. After a lot of reading and testing i decided, that nmake dont have the
capabilities, ok, i didn't find a way for converting, maybe i am to stupid.
On this point i remebered that SFU is able to execute ms commandline programs
within a SFU shell and there was wcc to wrap around lc and do it the way i wanted
it, and so i was back in business. But my first make run was errornous,
clearly indicating that i am still in a unix enviroment. So i have to change
one or more symbols and provide my defined symbol.
My question is where do i have to change the script ?
Btw, i found out that fortran is not needed any more, because it's called within
one of the programs (stupid me).
Thanx in advance
HeinZ |
|
|
|