[daip] Re: Question about AIPS results
Eric Greisen
egreisen at nrao.edu
Fri Dec 16 11:06:54 EST 2005
Bryan Follis writes:
> Hello,
>
> I'm building and testing AIPS on a Dual 2.0Ghz G5 system running OS X
> 10.4.3. What I'm looking to do is look at performance across
> different compilers. I am using XLF 8.1 as my FORTRAN compiler the
> entire time and am varying the C compilers. First of all, I build it
> using GCC 3.3, which builds without any problem. However, when I go
> to execute it in an X11 environment, I get these following errors:
There is so little C code in AIPS that you will find almost no
difference between compilers. The Fortran compiler is all that
matters.
> XAS: Using screen width height 1270 860, max grey level 255
> Shared memory id failure: Invalid argument
Your maximum shared memory segment is too small. See the aips
manager FAQ page for information
http://www.aoc.nrao.edu/aips/aipsmgr/
> DISPLAY :0
> XASERVERS: Start message server MSGSRV on pegs-power-mac-g5-3,
> DISPLAY :0
> TVSERVER: could not get ssslock service by name: Invalid argument
> TVSERVER: Check your /etc/services or NIS/YP map
> STARTPMON: [PEGS-POWER-MAC-G5-3] Starting TPMON1 with output SUPPRESSED
> ZVTPO3: tcp/aipsmt0 is not a service
> ZVTPO3: check /etc/services or NIS map!
Your system services need to be updated - also see the
aforementioned page.
>
> This causes the msgsrv and teksrv to both crash immediately. Is there
> some underlying problem with my system that I should know of? Also, I
> ran the Y2K large test and got the following results:
>
> 1 0 15-DEC-2005 17:36:53 AIPS YES
> 1 0 15-DEC-2005 17:36:53 AIPS YES
> 1 5 15-DEC-2005 17:36:53 AIPS 'IMAGE'
> 12.1318769 17.872406
> 1 5 15-DEC-2005 17:36:53 AIPS 'IMBEAM'
> 14.4524546 17.9524899
> 1 5 15-DEC-2005 17:36:53 AIPS 'APCLN'
> 12.1562166 16.6016254
> 1 5 15-DEC-2005 17:36:53 AIPS 'APRES'
> 15.2675838 22.8680859
> 1 5 15-DEC-2005 17:36:53 AIPS 'IMCLN'
> 10.6569357 16.4775925
> 1 5 15-DEC-2005 17:36:53 AIPS 'VTESS'
> 3.5372751 11.5847349
> 1 5 15-DEC-2005 17:36:53 AIPS 'VTESSC'
> 3.4322975 11.1895304
> 1 5 15-DEC-2005 17:36:53 AIPS 'IMAGE' .
> 00006104 0 .00001526
> 1 5 15-DEC-2005 17:36:53 AIPS 'IMBEAM'
> 0 0 .000000477
> 1 5 15-DEC-2005 17:36:53 AIPS 'APCLN' .
> 0001221 .0003662 -.0002141
> 1 5 15-DEC-2005 17:36:53 AIPS 'APRES' -.
> 00001526 0 -.000000224
> 1 5 15-DEC-2005 17:36:53 AIPS 'IMCLN' .
> 0029907 -.0005493 .0004678
> 1 5 15-DEC-2005 17:36:53 AIPS 'VTESS' -.
> 0122681 -.0023804 -.0470033
> 1 5 15-DEC-2005 17:36:53 AIPS 'VTESSC' .
> 0049438 .0019531 -.6733589
>
>
> The values for VTESS and VTESSC are cause for concern. Aren't the BIT
> value for these supposed to be in the 10 to 20 range?
You probably do not have the most current master images. They
are at ftp://ftp.aoc.nrao.edu/pub/software/FITS/31DEC04/
VTESS changed and the master images needed updating.
>
> That is the issue I'm having with GCC 3.3, but, as I said before, I'm
> varying the compiler to also use XLC 6.1 and GCC 4.0. I had to make
> minor changes for both of these in order to have AIPS compile in
> general. But I do get it to the point where it says it compiles
> successfully, however, what is strange is that when I look throughout
> the log file of INSTEP2 on either GCC 4.0 or XLC 6.0, there are
> messages printed mentioning Undefined symbols. However, no error
> messages follow. I'll attach an exert from INSTEP2 from when I
> compiled using XLC 6.0:
>
> COMRPL : Date Thu Dec 15 18:23:00 EST 2005
> PP : Preprocess /Users/bfollis/Projects/AIPS/31DEC05/APL/DEV/
> UNIX/ZSSSX2.C
> PP : into /Users/bfollis/Projects/AIPS/31DEC05/MACPPC/
> PREP/ZSSSX2.c
> AIPSCC : Date Thu Dec 15 18:23:00 EST 2005
> AIPSCC : Interpret AIPSCC \
> AIPSCC : /Users/bfollis/Projects/AIPS/31DEC05/MACPPC/
> PREP/ZSSSX2.c
> AIPSCC : as LIST=FALSE PURGE=TRUE
> AIPSCC : plus /opt/ibmcmp/vacpp/6.0/bin/xlc -O3 -qstrict -
> qarch=g5 -qtune=g5 -qcache=auto -I/Users/bfollis/Projects/AIPS/
> 31DEC05/INC -D_FILE_OFFSET_BITS=64 \
This line is missing the -c option which got lost as you edited the
CCOPTS.SH file. Without that option it tries to link each of the
routines. They are only subroutines and should not be linked.
> AIPSCC : /Users/bfollis/Projects/AIPS/31DEC05/MACPPC/
> PREP/ZSSSX2.c
> /usr/bin/ld: Undefined symbols:
> _main
> _buffered
> _pack_offset
> _Z_sndinis
> _Z_sndbufs
> _packet_int
> _Z_bytflp
> _bufferop
> _Z_rcvbufs
> _Z_rcvinis
> AIPSCC : Compile of /Users/bfollis/Projects/AIPS/31DEC05/MACPPC/
> PREP/ZSSSX2.c
> AIPSCC : ends successfully.
> AIPSCC : Delete /Users/bfollis/Projects/AIPS/31DEC05/MACPPC/
> PREP/ZSSSX2.c
> COMRPL : Copied /Users/bfollis/Projects/AIPS/31DEC05/MACPPC/
> PREP/ZSSSX2.o
> COMRPL : to /Users/bfollis/Projects/AIPS/31DEC05/MACPPC/
> LIBR/APLMAC
> COMRPL : Removed /Users/bfollis/Projects/AIPS/31DEC05/MACPPC/
> PREP/ZSSSX2.o
> COMRPL : Changed /Users/bfollis/Projects/AIPS/31DEC05/APL/DEV/
> UNIX/ZSSSX2.C
> COMRPL : in /Users/bfollis/Projects/AIPS/31DEC05/MACPPC/
> INSTALL/APLGEN.LIS
> COMRPL : to -/Users/bfollis/Projects/AIPS/31DEC05/APL/DEV/
> UNIX/ZSSSX2.C
>
> It says that the compilation of that particular file ends
> successfully (as with all the others that go wrong), but should there
> not be some concern about the undefined symbols? In almost every file
> for the library APLMAC has an undefined symbol error. Usually it is
> just _main, but it can vary to have many other undefined symbols. I'm
> confused as to why this is happening, and even more confused as to
> why it does not seem to effect the overall compilation. As I said,
> this happens for GCC 4.0 and XLC 6.0, any ideas as to why this is?
> And any ideas why it doesn't seem to effect the successful completion
> of the compilation?
>
> Also, to prove that XLC works after the strange messages during
> compile-time, I ran Y2K and go the following output:
>
> 1 5 15-DEC-2005 23:56:23 AIPS 'IMAGE'
> 12.1318769 17.872406
> 1 5 15-DEC-2005 23:56:23 AIPS 'IMBEAM'
> 14.4524546 17.9524899
> 1 5 15-DEC-2005 23:56:23 AIPS 'APCLN'
> 12.1562166 16.6016254
> 1 5 15-DEC-2005 23:56:23 AIPS 'APRES'
> 15.2675838 22.8680859
> 1 5 15-DEC-2005 23:56:23 AIPS 'IMCLN'
> 10.6569357 16.4775925
> 1 5 15-DEC-2005 23:56:23 AIPS 'VTESS'
> 3.5372751 11.5847349
> 1 5 15-DEC-2005 23:56:23 AIPS 'VTESSC'
> 3.4322975 11.1895304
> 1 5 15-DEC-2005 23:56:23 AIPS 'IMAGE' .
> 00006104 0 .00001526
> 1 5 15-DEC-2005 23:56:23 AIPS 'IMBEAM'
> 0 0 .000000477
> 1 5 15-DEC-2005 23:56:23 AIPS 'APCLN' .
> 0001221 .0003662 -.0002141
> 1 5 15-DEC-2005 23:56:23 AIPS 'APRES' -.
> 00001526 0 -.000000224
> 1 5 15-DEC-2005 23:56:23 AIPS 'IMCLN' .
> 0029907 -.0005493 .0004678
> 1 5 15-DEC-2005 23:56:23 AIPS 'VTESS' -.
> 0122681 -.0023804 -.0470033
> 1 5 15-DEC-2005 23:56:23 AIPS 'VTESSC' .
> 0049438 .0019531 -.6733589
>
>
> Regardless of the different compilers and optimization flags, I get
> the exact same results for Y2K, with the exact same questionable
> VTESS and VTESSC values. The same went for the DDT test (except in
> GCC 3.3 and XLC 6.0 all the BIT values in the printed output were
> within the expected range).
>
> I suppose my question is, do you think any of this information
> indicates there being something wrong with the way AIPS is being
> compiled or the system that it is running on? If you desire any more
> information about how I'm compiling/running AIPS or what kind of
> machine I'm running, please let me know.
>
> Thank you very much,
>
> Bryan Follis
I have pointed out the only things I saw wrong. It is interesting
that the -c missing error caused so little trouble but things do seem
to be working more or less.
ERic Greisen
More information about the Daip
mailing list