02-179

To: J3
From: John Reid
Subject: Unresolved issue 363
Date: 12 April 2002

Unresolved issue 363 (page 373) is

   Item 33 of paper 02-138r2 says to delete the "unless the result is
   exact" from the above note.  John Reid says this deletion is
   incorrect. Resolution deferred to J3.

The note is note 14.11 on page 372. I think it is OK as it is. The last
para of 7.4 in the IEEE standard says

   When an underflow trap is not implemented or is not enabled (the default
   case) underflow shall be signaled (via the underflow flag) only when
   both tininess and loss of accuracy have been detected.

In an email, Malcolm Cohen commented:

   The note is, I think, ok, even if it is not quite what the IEEE
   standard says.  It is correct if "an underflow trap ... is not enabled".
   This is a case where IEEE were too clever for their own good, because if
   underflow trapping *is* enabled, underflow is raised regardless of
   exactness.  One of the things the 754r committee is considering is a
   substantial simplication in this area.

   On a "real" machine, whether turning HALTING on for IEEE_UNDERFLOW is
   going to alter the underflow signalling, is a matter of interpretation.
   (Since the program stops completely, not invoking a trap handler).

   However, deleting the "unless" clause is DEFINITELY wrong.

Edit

373:0 Delete J3 internal note.