topic Re: Compiler problems. in Operating System - OpenVMS

Compiler problems.

Daniel Fernandez Illan — Tue, 20 Jul 2004 05:00:57 GMT

hi people.
We have a problem with C compiler ( V6.4 and VMS 7.3-1).
When alpha had VMS 7.2.1, compiler works fine but when we upgraded OS appears this error in compilation time:

%SYSTEM-F-HPARITH, high performance arithmetic trap, Imask=00000000, Fmask=00000
001, summary=02, PC=00000000004EDEDC, PS=0000001B
-SYSTEM-F-FLTINV, floating invalid operation, PC=00000000004EDEDC, PS=0000001B
%TRACE-F-TRACEBACK, symbolic stack dump follows
image module routine line rel PC abs PC
DECC$COMPILER GEM_ARCH_HAZARD_ALPHA Contains
33151 00000000000000FC 00000000004EDEDC.

Some idea?.

Re: Compiler problems.

Kris Clippeleyr — Tue, 20 Jul 2004 05:46:55 GMT

Daniel,

Given your information, floating register F0 is affected (Fmask), and according to the "summary" an attempt was made to perform an arithmetic, conversion or comparison operation and one or more of the operands do not represent a valid floating point value.
This could be a latent bug in the compiler.
I suggest two things:
upgrade to V6.5 of the compiler, and/or compile the offending module /NOOPT

Good luck,

Kris

Re: Compiler problems.

Daniel Fernandez Illan — Tue, 20 Jul 2004 06:43:38 GMT

Thank for you help Kris.

Indeed, using the option /NOOPTIMIZE, it works in agreement with the awaited thing.

Best Regards.

Re: Compiler problems.

Craig A Berry — Tue, 20 Jul 2004 11:42:35 GMT

I would recommend installing all recommended OS patches (as well as upgrading the compiler). The latest VMS731_SYS and VMS731_UPDATE should be sufficient. There have been some floating point issues where the compiler was a victim of the support infrastructure in the OS.

Re: Compiler problems.

Ian Miller. — Tue, 20 Jul 2004 12:38:00 GMT

the NFS server in TCPIP was once such program that could corrupt floating point state.
If was a build problem I belive.

Re: Compiler problems.

Daniel Fernandez Illan — Wed, 21 Jul 2004 03:00:21 GMT

All recommended patches were installed, but problem remained, appearing randomly after someone compilations.

At this moment, /NOOPT option works fine (apparently).

Re: Compiler problems.

Antoniov. — Wed, 21 Jul 2004 03:23:01 GMT

Daniel,
can you change floating option (see /iee)?

Antonio Vigliotti

Re: Compiler problems.

Daniel Fernandez Illan — Wed, 21 Jul 2004 03:32:42 GMT

Antoniov, Trap occurs in compilation time. I think that this option only works in execution time.

Re: Compiler problems.

Daniel Fernandez Illan — Wed, 21 Jul 2004 03:39:00 GMT

Antinio, options in compilation are:

CC=/prefix=all/INCLUDE=(SP_FA_INC:,SP_LI_FUENTES:,SP_FA_LIB:,SP_FU_COMPOSITOR:)/define=("__EXPLOTACION__")

and we are add /NOOPT only

Re: Compiler problems.

Antoniov. — Wed, 21 Jul 2004 04:18:14 GMT

Daniel,
I known /iee is a runtime option; because using /noopt seems solve I guess there is soe compiler bug evaluating arithmetic expression. Best way is send source code to HP to solve bug while you could change some qualifier to avoid problem.
If you can manager NaN and infinite values you could use /iee=under because compiler don't need add trap code for theese values.

Antonio Vigliotti

Re: Compiler problems.

Daniel Fernandez Illan — Thu, 22 Jul 2004 04:50:43 GMT

Antonio.

I think that the problem takes place by the compiler and operating system versions, because trap occurs in a automatic script of compilation, and in different sources.

Including /IEE option, trap doesn't occurs at this moment.

Best Regards.

Re: Compiler problems.

Richard W Hunt — Thu, 22 Jul 2004 14:23:25 GMT

Ada is notorious for the analogous problem in which you get a HPARITH trap unless you compile that program with /NOOPT

I wonder if whoever wrote the optimizer for C also worked on Ada?

But to be honest, the old TOPS-10 system for DEC KA/KI/KL-10s had a similar problem for their Pascal compiler. The optimizer optimizes an entire instruction out of existence, then gets lost trying to find what it just threw away.

Re: Compiler problems.

John Gillings — Thu, 22 Jul 2004 17:18:24 GMT

Daniel,

Any SYSTEM-F error from any compiler is a bug and should be reported formally.

That said, the first diagnostic check will be to upgrade to the latest compiler and retry. If the problem persists with the latest compiler, please log a case against your service contract.

Please have the exact command, and exact error message available and, if possible, the complete source code so the service engineer can reproduce the symptom (extra points if you manage to submit a C program with ALL the required include files! ;-)

Re: Compiler problems.

Daniel Fernandez Illan — Fri, 01 Apr 2005 05:42:37 GMT

Thanks to all.
I had a problem with mi old identifier.
Saludos.

Re: Compiler problems.

Daniel Fernandez Illan — Fri, 01 Apr 2005 05:43:14 GMT

Clos this threat