home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!sun-barr!cs.utexas.edu!asuvax!asuacad!atpph
- Organization: Arizona State University
- Date: Wednesday, 12 Aug 1992 15:07:16 MST
- From: <ATPPH@ASUACAD.BITNET>
- Message-ID: <92225.150716ATPPH@ASUACAD.BITNET>
- Newsgroups: comp.unix.aix
- Subject: xlf (2.2.1.3) preprocessor bugs
- Lines: 60
-
-
- I have some problems with the preprocessors of xlf. My machine is a
- RS/6000 530H running AIX 3.2 with xlf version 2.2.1.3. The machine has
- 64MB RAM and 356MB paging space.
-
- The problems are:
-
- Pk preprocessor (Version 02.02.0100.0000 2.12o5r3so3):
- options used: optimize=5,roundoff=3,scalaropt=3,library-calls=essl,agressive=a
-
- One line in the offending subroutine is translated incorrectly.
- original statement: s = 1.s0-dtexp(j-1,i)
- translated into: s = dtexp(ii1-1,i) + 1.d0
- (sign of dtexp() is wrong, dtexp is an array)
- The problem does not occur for optimize=2 and lower. The differences between
- optimize=5 and 2 in the translated code are (a) the problematic line and
- (b) two lines where () are placed around an expression (correctly).
- I suspect a bug in the preprocessor here. I had this problem with an earlier
- version but it seemed to be fixed in an IBM test version. However, the bug
- is still in the xlf release that came with our AIX 3.2 upgrade.
-
-
- Pv preprocessor: (Version 02.02.0000.0000 3.05F16)
- options used: -Pv -Wp,-o,-ea478 (or -Pv -Wp,-ew,-ejr4789,-me,-o,-l,-pacrvz)
- 1. Problem: two subroutines give (many) error messages of the form
- -->
- 51.00 1524-282: (I) Unable to allocate multidimensional temporary arrays.
- 47.00 1524-041: (I) Internal error detected. Call service representative.
- Internal segment WHRSET.
- --<
- the Translation continues and the translated code compiles.
- The code does not work but up to now I have not found the time to
- track down the loaction of the problem.
-
- 2. Problem: FORMAT statement label doubling occurs in another subroutine. The
- prepocessor 'optimizes' (label V) an outer loop and the format statement with
- this loop is simply doubled (tripled) without adjusting the label. This is a
- redundant operation anyway.
- The preprocessor reports no trouble, however, the compiler itself
- generates an 'E' code error message (non-fatal).
- These problems just gives annoying error messages without fatal errors as the
- previous cases.
-
- The first two problems cause the code to crash in obscure places, in particular
- the first one (Pk) is nasty because it is 'quiet'.
- The Pv-problems are indicated during compile time, so I was already suspecting
- problems with the code before starting the run.
- The subroutines are not long and I would be happy to send them to
- somebody from IBM to try them and to fix the bugs.
- My preferred E-Mail address is yeti@sara.la.asu.edu.
-
- Best regards,
- Peter
-
- Peter H. Hauschildt
- Dept. of Physics and Astronomy
- ASU
- Tempe, AZ 85287-1504
-
-
-