home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!spool.mu.edu!news.nd.edu!mentor.cc.purdue.edu!pop.stat.purdue.edu!hrubin
- From: hrubin@pop.stat.purdue.edu (Herman Rubin)
- Newsgroups: comp.arch
- Subject: Re: CISC Microcode (was Re: RISC Mainframe)
- Message-ID: <54638@mentor.cc.purdue.edu>
- Date: 21 Jul 92 13:10:48 GMT
- Article-I.D.: mentor.54638
- References: <TOM.92Jul13121539@amber.ssd.csd.harris.com> <1992Jul20.205209.28863@boole.uucp> <l6mlr0INN7tu@spim.mips.com>
- Sender: news@mentor.cc.purdue.edu
- Organization: Purdue University Statistics Department
- Lines: 52
-
- In article <l6mlr0INN7tu@spim.mips.com> mash@mips.com (John Mashey) writes:
- >In article <1992Jul20.205209.28863@boole.uucp> John@boole.uucp (John Ahlstrom) writes:
-
- >>Did anyone at work do a study of the utility of address modes independent
- >>of the speed of any particular implementation of them? I doubt that an
- >>ex post study of the implementated speed of modes is a good way to
- >>decide which ones are valuable to have. Valuable to use in a give
- >>compiler perhaps, but not valuable to have in an architecture.
-
- >There have been plenty of papers that have addressed this over the years,
- >usually in the form of:
- > a) Take a given CPU architecture+compiler pair, and analyze the
- > distribution of address modes.
-
- The sensibility of this assumes that we have good languages/compilers.
-
- > (Note that this has implicit use of performance, i.e., you might
- > assume that agonizingly-slow modes might be avoided.)
- > On the other hand, if you find modes that are reasonably fast
- > (as fast as any equivalent sequence), and don't get used,
- > either:
-
- > 1) The compilers can't figure out how to use them often.
-
- We keep hearing that compilers can do as well as humans. Now if a
- human can figure out how to use these modes, don't blame the mode,
- blame the compiler if it cannot figure out how to use it efficiently.
-
- > or 2) Most realistic code doesn't lend itself to the mode.
-
- What is realistic code? I will code differently if there is hardware
- auto-increment for addresses. I expect any good programmer to do as
- well; I am not a professional programmer.
-
- > Hennessy & patterson have some discussion of this topic.
-
- > b) Of course, simple analysis of existing modes may tell you ones
- > that are not very useful, but they won't point out ones that
- > aren't present in the example, but would be useful. That is,
- > to decide that auto-increment of some kind might be useful,
- > you have to do more detailed analyses of code sequences to see.
-
- In other words, the programmer has to think about using the machine
- capabilities, or there has to be in the compiler enough so that the
- compiler will try out these capabilities.
-
- ...................
- --
- Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907-1399
- Phone: (317)494-6054
- hrubin@pop.stat.purdue.edu (Internet, bitnet)
- {purdue,pur-ee}!pop.stat!hrubin(UUCP)
-