Feedback from SHARE to NCEG

[Fred Tydeman]

I presented a talk "Extending C for Numerical Computing (NCEG)" to the
IBM user group SHARE.  SHARE members tend to be scientific or technical,
rather than commercial.  To be a member of SHARE, the user's shop must
have a main frame (System/370), mini (AS/400), or workstation (RISC
System/6000).
 
As part of the talk, there were four questions for the audience.  I told
them that I would feedback the results to NCEG, so here they are.
 
  1) What will be your language of choice for numeric computing:
 
     Fortran 90 - 5
     Fortran 77 - 2
     HPF        - 1      (High Performance Fortran)
     C++        - 1
     Extended C - 0
 
  2) What kind of elementary math library do you want:
 
     Standards compliant - 2    (set errno and/or IEEE-754 flags)
     Call error handler  - 1    (like AT&T SVID matherr)
     Exact answer        - 0
     Fast, but sloppy    - 0
 
  3) What areas should NCEG address:
 
     Look at High Performance Fortran forum with respect to array
     handling.
 
     Treating array like objects as first class object would be a boon
     for using C in numerical computing and should be strongly
     considered.
 
  4) Any message or comments that you want taken back to NCEG or IBM:
 
     Long double should have more precision than double so as to be
     consistent with the requirement that double have more precision
     than float.
 
     Have NCEG get in contact with the end users, in particular, the
     financial (wall street) people since they do a lot of numeric work.
 
     Support IEEE-754 floating-point on the main frame.
 
     Make long double in C be the same as real*16 in Fortran.
 
These are not IBM's views, these are the personal ones of:
Fred Tydeman, IBM, Austin, Texas (512) 838-3322; fax (512) 838-3484
AIX S/6000 Math library architect & IBM's rep to NCEG (X3J11.1)
Internet: tydemanaibmpa.awdpa.ibm.com    uucp: uunet!ibmsupt!tydeman