I run a small beowulf cluster, and I’d like to try and use QNX for the
operating system on some of the nodes. I’ve gotten it up and running
on one of the machines, network and rsh work, but I can’t seem to get
LAM(implementation of MPI I use) to compile. I’ve gotten the newest
config.guess and config.sub files, and it configs properly but on compile
I get errors on about 10 source files relating to incomplete structs in
a header file. All of this compiles fine on Linux/Irix. Any thoughts would be appreciated.
Aaron
incl: error and the header file which causes it.
/usr/local/lam-6.3.2/include/mpisys.h:200: parse error before int8' /usr/local/lam-6.3.2/include/mpisys.h:200: warning: no semicolon at end of struct or union /usr/local/lam-6.3.2/include/mpisys.h:221: field w_header’ has incomplete type
mpisys.h:
(too big, see it at: http://wiglaf.cs-i.brandeis.edu/error/mpisys.h
This is probably due to the fact that QNX (unlike all Unices) have those
types defined with front underscore. I.e., they have _int8 instead of
int8 defined in sys/types.h.
You could add your own definitions.
Aaron Macks wrote:
I run a small beowulf cluster, and I’d like to try and use QNX for the
operating system on some of the nodes. I’ve gotten it up and running
on one of the machines, network and rsh work, but I can’t seem to get
LAM(implementation of MPI I use) to compile. I’ve gotten the newest
config.guess and config.sub files, and it configs properly but on compile
I get errors on about 10 source files relating to incomplete structs in
a header file. All of this compiles fine on Linux/Irix. Any thoughts would be appreciated.
Aaron
incl: error and the header file which causes it.
/usr/local/lam-6.3.2/include/mpisys.h:200: parse error before int8' /usr/local/lam-6.3.2/include/mpisys.h:200: warning: no semicolon at end of struct or union /usr/local/lam-6.3.2/include/mpisys.h:221: field w_header’ has incomplete type
mpisys.h:
(too big, see it at: > http://wiglaf.cs-i.brandeis.edu/error/mpisys.h
Previously, Igor Kovalenko wrote in qdn.public.porting:
This is probably due to the fact that QNX (unlike all Unices) have those
types defined with front underscore. I.e., they have _int8 instead of
int8 defined in sys/types.h.
You could add your own definitions.
I thought the POSIX types for these were int8_t, uint8_t, etc. You
find these in <inttypes.h>.
Andrew
It seems you were both right, when I changed the type to int8_t, and
included inttypes.h it has compiled. I also need to link sys/signal.h to
signal.h, but that can be fixed in code. If this seems to work I will
contact the authors and see if I can get them to include this in the
distribution Aaron In article
<Voyager.010410115218.540694A@andrewhome.cogent.ca>, Andrew Thomas
<Andrew@cogent.ca> wrote:
Previously, Igor Kovalenko wrote in qdn.public.porting:
This is probably due to the fact that QNX (unlike all Unices) have
those types defined with front underscore. I.e., they have _int8
instead of int8 defined in sys/types.h.
You could add your own definitions.
I thought the POSIX types for these were int8_t, uint8_t, etc. You find
these in <inttypes.h>.
Andrew
Aaron Macks wrote:
It seems you were both right, when I changed the type to int8_t, and
included inttypes.h it has compiled. I also need to link sys/signal.h to
signal.h, but that can be fixed in code.
Yes … there are several similar quirks in the sources.
You have also to run the configure script with --without-fc and
–without-ROMIO because of there isn’t a FORTRAN compiler available
and QNX6 doesn’t support the POSIX asynchronous IO services for
ROMIO.
If you want to include systems with M$Windows in your heterogenous
computing environment … go with PVM. A PVM port is available at
sourceforge/projects/pyqnx.
If this seems to work I will
contact the authors and see if I can get them to include this in the
distribution Aaron
Good idea …
Armin
In article
Voyager.010410115218.540694A@andrewhome.cogent.ca> >, Andrew Thomas
Andrew@cogent.ca> > wrote:
Previously, Igor Kovalenko wrote in qdn.public.porting:
This is probably due to the fact that QNX (unlike all Unices) have
those types defined with front underscore. I.e., they have _int8
instead of int8 defined in sys/types.h.
You could add your own definitions.
I thought the POSIX types for these were int8_t, uint8_t, etc. You find
these in <inttypes.h>.
Andrew
thanks, but everything else is running MPI(specifically LAM) and I’m not
going to in the work to change it all just to use QNX. Aaron
If you want to include systems with M$Windows in your heterogenous
computing environment … go with PVM. A PVM port is available at
sourceforge/projects/pyqnx.
If this seems to work I will
contact the authors and see if I can get them to include this in the
distribution Aaron
Good idea …
Armin ns.
Aaron Macks wrote:
thanks, but everything else is running MPI(specifically LAM) and I’m not
going to in the work to change it all just to use QNX. Aaron
Hmm … is there a LAM-MPI version for M$Windows ??
Armin
If you want to include systems with M$Windows in your heterogenous
computing environment … go with PVM. A PVM port is available at
sourceforge/projects/pyqnx.
If this seems to work I will
contact the authors and see if I can get them to include this in the
distribution Aaron
Good idea …
Armin ns.