home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!utcsri!torn!cunews!nrcnet0!bnrgate!bmerh85!bcars64a!bqneh23!schow
- From: schow@bqneh23.bnr.ca (Stanley T.H. Chow)
- Subject: Re: trapping speculative ops (LONG)
- Message-ID: <1992Sep11.143705.19214@bcars64a.bnr.ca>
- Sender: news@bcars64a.bnr.ca (Usenet News)
- Organization: Bell Northern Research Ltd, Ottawa
- References: <id.G_3T.SH1@ferranti.com> <1992Sep11.074838.26488@email.tuwien.ac.at>
- Date: Fri, 11 Sep 1992 14:37:05 GMT
- Lines: 27
-
- In article <1992Sep11.074838.26488@email.tuwien.ac.at> anton@mips.complang.tuwien.ac.at (Anton Martin Ertl) writes:
- >In article <id.G_3T.SH1@ferranti.com>, peter@ferranti.com (peter da silva) writes:
- >|> Isn't this (extra bits for exceptions) basically the whole point of NaNs and
- >|> Infinities in IEEE FP? Is the IEEE method inadequate, or would it be reasonable
- >|> to simply ignore FP interrupts and check for NaNs (assuming integer arithmetic
- >|> has low enough latency it isn't a problem)?
-
- Yes, the ideas are basically the same. IEEE method is inadequate for
- integer - which value would you like to use for NaI? :-) The IEEE stuff
- is wider in scope in that infinities are propagated, etc. while the
- trap bits are more narrowly focus.
-
- >On a superscalar machine even a one-cycle latency is a problem.
- >However, integer arithmetic causes no exceptions (in C), so what's the
- >real problem? Loads.
-
- Surely you jest! C is not the whole world. Many languages and
- applications want to see interger overflow/underflow. Come to think
- of it, your paper mentions this (but I havn't had time to read it yet).
- Loads are certainly a problem.
-
-
- --
- Stanley Chow InterNet: schow@BNR.CA
- Bell Northern Research UUCP: ..!uunet!bnrgate!bqneh3!schow
- (613) 763-2831
- Me? Represent other people? Don't make them laugh so hard.
-