home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.arch:10839 comp.lang.misc:3742
- Path: sparky!uunet!snorkelwacker.mit.edu!ai-lab!life.ai.mit.edu!tmb
- From: tmb@arolla.idiap.ch (Thomas M. Breuel)
- Newsgroups: comp.arch,comp.lang.misc
- Subject: Re: how to advocate new software/hardware features (Re: Hardware Support for Numeric Algorithms)
- Date: 17 Nov 92 22:38:28
- Organization: IDIAP (Institut Dalle Molle d'Intelligence Artificielle
- Perceptive)
- Lines: 34
- Message-ID: <TMB.92Nov17223828@arolla.idiap.ch>
- References: <1992Nov13.155126.3660@linus.mitre.org>
- <1992Nov17.112551.10920@uklirb.informatik.uni-kl.de>
- Reply-To: tmb@idiap.ch
- NNTP-Posting-Host: arolla.idiap.ch
- In-reply-to: kirchner@uklira.informatik.uni-kl.de's message of Tue, 17 Nov 1992 11:25:51 GMT
-
- In article <1992Nov17.112551.10920@uklirb.informatik.uni-kl.de> kirchner@uklira.informatik.uni-kl.de (Reinhard Kirchner) writes:
-
- There are often VERY USEFULL things in the hardware, but now
- programming language supports them. This is mostly the case in the
- field of numerics, like
- - long integer product, as shown by Bob,
- - the carry, the major incredient to build a 'long' arithmetic
- - rounding control, needed to generate verified results ( not a mode bit! )
-
- True, they are occasionally useful.
-
- What bothers me even more is the fact that modern processors are
- designed to support only things that can be used by HLL, mainly C,
- because statistics show no use of other features ( what else ), and
- so we find ourselves severley limited by a chicken and egg problem.
-
- I suspect most people (including myself) are actually perfectly happy
- with 32bit integers and 32bit/64bit floating point numbers with any
- reasonable kind of rounding mode. The main limits that users are
- coming up against are probably 32bit address limits (of course, if you
- go to 64bit addresses, you need 64bit integers for indexing).
-
- It is not very hard to provide access to such features from high-level
- languages, through things that look and behave like library calls but
- are actually recognized and inlined by the compiler ("sincos(a,s,c)",
- "add_with_carry(result,arg1,arg2)", ...). If you are really interested
- in this, I recommend that you work within NCEG (or other language
- standards committees) to make proposals for what library primitives
- might be useful. On the other hand, please do spare the rest of us
- complex syntactic extensions ("!<>", "?)~", etc.) for all those
- special purpose features.
-
- Thomas.
-
-