Possible Fortran 8.2 compiler bug on Itanium

[Jeremy Begg]

HIi,

I am helping a customer port their Fortran applications from OpenVMS Alpha V7.2-2 to OpenVMS IA64 V8.4.

The code (mostly) compiles cleanly but some programs are giving wrong results.  For example, when compiling and running the attached file (TMINUSPLUS.FOR) we get wrong values displayed:

$ sho sym fortran
%DCL-W-UNDSYM, undefined symbol - check validity and spelling
$ fortran tminusplus
$ link tminusplus
$ run tminusplus
         18
         20
         20
         18
         20
         20
$

The value being output by the program should be '20' on every line; '18' is incorrect.

It looks like the compiler is evaluatiing the expression

    result = FILE.INFO.SEG(1).END - FILE.INFO.SEG(1).START + 1

as if it was written

    result = FILE.INFO.SEG(1).END - (FILE.INFO.SEG(1).START + 1)

and in fact the erroneous instruction, as seen in the Debugger, is

    sub         r9 = r9, r10, 1

The trailing ", 1" has the effect of subtracting 1 from the result, whereas the expression requires 1 be added to the result.

I've tried both the Fortran 8.2 compiler and the 8.2 ECO1 compiler and both give the same results.

I am going to try to report this to HP.  If anyone has seen this before, and/or knows a workaround (which hopefully doesn't require extensive code changes!) I'd like to know about it.

Thanks,

Jeremy Begg

[Jeremy Begg]

Thanks for the info, Steven. HP has acknowledged receipt of my support request so we'll see what happens.

[Eberhard Heuser]

There is a newer fortran version available:

$ fortran/version
HP Fortran V8.2-104953-50M2N

Eberhard

[Jeremy Begg]

Hi Eberhad,

That's the Fortran 8.2 ECO1 release, which gives the same results.  In fact that's where we first noticed it.

Jeremy

[Fekko Stubbe]

I tried this on alpha (8.4) with the same result (error)

$fortran/vers

Hp Fortran v8.2-104679-48H9K

If I use the old f77 compiler (fortran/old) the result is correct (all 20s)

$fortran/old/version

HP Fortran 77 V7.6-198-48H9K (this version reports 7.6 while it really is the 8.2 version, another bug)

[abrsvc]

The report of V7.6 is correct as the F77 compiler is a completely different image. F90$main is the F90 compiler while FORT$main is the F77 compiler.

Dan

[Fekko Stubbe]

You could be right, but the image is installed from the fortran 8.2 kit. This kit installs both the fort$main as the f90$main. I find it highly  confusing. If they wat to distribute different version for the f90 and the f77 compiler they should have deliverd 2 kits. As far aas I rmember  the 8.1 kit distributes an 8.1 version for both compilers.

[Mike Kier]

It worked correctly under the Fortran 95 compiler on a much older Alpha version - running OpenVMS Alpha V7.3-2 and Fortran V7.5-1961-48BCD

[John Brodribb]

This was fixed recently. Please contact your Support Centre for a remedial side image.

[John Gillings]

Jeremy,

   If only all "I've found a compiler bug" reports were so well researched, characterised and documented with such a simple and unarguable reproducer! :-)