home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!wupost!uwm.edu!caen!nic.umass.edu!m2c!crackers!frog!rfm
- From: rfm@frog.CRDS.COM (X)
- Newsgroups: comp.sys.m68k
- Subject: 68040 floating-point traps
- Summary: How to resume after repairing X underflow?
- Keywords: 68040 underflow
- Message-ID: <1992Dec9.163838.2436@frog.CRDS.COM>
- Date: 9 Dec 92 16:38:38 GMT
- Sender: Bob Mabee
- Organization: Charles River Data Systems
- Lines: 18
-
- Here's another case that doesn't work as (feebly) documented:
- Underflow (and overflow) of the extended format is detected late
- in the course of an arithmetic operation, when the next FP instruction
- is already beginning and the PC has advanced even further, to a third
- FP instruction. FSAVE gives the long ($60 -- BUSY) format with E3 set
- and a variety of info about the two instructions in progress. The target
- register of the failed instruction has already been updated (contrary
- to the manual), with a wrapped-around exponent.
-
- I can turn the *flowed result (WBT) into 0 or infinity, as appropriate,
- but FRESTORE doesn't continue properly. The trap merely recurs as soon
- as I try another FP instruction (such as what RTE jumps to). Turning E3
- off seems like an obvious thing to try, but then the CPU wedges.
-
- What other mutations should I make to the saved state? Does it vary with
- mask revisions?
-
- email to frog!rfm@eddie.mit.edu.
-