[daip] Odd behavior in Y2K-LARGE+ on AMD64 AIPS

[Eric Greisen]

Samuel Conner writes:
 > To Whom it may concern,
 > 
 >    I have installed 31DEC04 AIPS to my AMD64 machine and am now exercising it on the Y2K tests.
 > 
 >    The SMALL and MEDIUM tests terminate normally and the errors are in the range 10-15 bits maximum and several bits better than than rms.
 > 
 >    On the LARGE size, TCODE = 'TEST' operation, CALIB dies at the place that it is seeking a common reference antenna for the 3 different groups of solutions. I noticed that the inputs to CALIB were rather sparse, and that REFANT was not set. An experimental re-execute of CALIB with REFANT set to 15 resulted in a seemingly successful termination (of CALIB).
 > 
 >    I cannot proceed with the Y2K-LARGE test as the procs reset REFANT to 0 during execution.
 > 
 >   The HUGE test dies at the outset of an IMAGR execution.
 > 
 > Q:  Is this a known failure mode on other systems?
 > 
 > Q:  I am concerned that if there are dynamic linkings happening as the .EXEs are invoked, that the wrong (32 bit) libraries may be getting linked. Should I be fiddling with environment variables?
 > 
 >   I did not compile with DEBUG libs, but I could do that easily if I must to sort this out.
 >    
 >   I am getting close; thanks for your help.
 > 

But you will get no closer.  The AMD 64 is, I am nearly certain,
generating dynamic memory at the opposite end of the virtual memory
address space from the source code.  31DEC04 cannot handle this.  I
have fixed 31DEC05 to handle this in most places although FRING will
have limits to the size of fringe fits in this case which will not
apply to machines that issue dynamic memory at addresses close to the
program memory.

Install 31DEC05.

Eric Greisen