home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.arch:10481 comp.lang.forth:3452
- Newsgroups: comp.arch,comp.lang.forth
- Path: sparky!uunet!gatech!news.byu.edu!hamblin.math.byu.edu!sol.ctr.columbia.edu!zaphod.mps.ohio-state.edu!darwin.sura.net!aplcen.apl.jhu.edu!aplcenmp!mef
- From: mef@aplcenmp.apl.jhu.edu (Marty Fraeman)
- Subject: Re: What's RIGHT with stack machines
- Message-ID: <BxBFqu.Cu8@aplcenmp.apl.jhu.edu>
- Keywords: stack computers, embedded control
- Organization: Johns Hopkins Continuing Professional Programs
- References: <Bx5AIr.EAy.2@cs.cmu.edu> <1992Nov4.103008.2641@Informatik.TU-Muenchen.DE>
- Date: Fri, 6 Nov 1992 22:28:05 GMT
- Lines: 36
-
- In article <1992Nov4.103008.2641@Informatik.TU-Muenchen.DE> pazsan@Informatik.TU-Muenchen.DE (Bernd Paysan) writes:
- >
- ...
- >Spill&Fill: did anyone thought about automatic spill&fill-buffers? (if the
- >available stack cells decreases to a certain level, a number of stack items are
- >load (4 or 8 would be great), and if it increases over a certain level, the same
- >amount of stack items is spilled (have a hysterese of about 8 stack entries).
- >
- The SC32 does exactly this in hardware. John Hayes has extensively
- measured lots of programs and usually sees far less than 1% overhead
- for this. Our first two Forth processors transfered half the on-chip
- stack on over/underflow. Measurements later firmly convinced us this
- was not a good strategy. So our third chip, the SC32, just transfers
- one register. Our fourth chip will do the same.
-
- >Are there any researches how many stack cells should be directly accessed by
- ><const> pick or <const> roll? Having only three is messy, I think, 4 or 5 would
- >be nice;
- The SC32 has access to the top four elements of both return and data stack.
- Four was choosen after inspecting our applications and finding that picks
- always used a constant known at compile time and in no case was access deeper
- than four needed. This is another decision we don't regret, from a
- software perspective, even on our fourth version of the machine. This
- feature does have a tendency to creep into the critcal path of the machine
- if your not careful.
-
-
- Marty Fraeman
-
- mef@glinda.jhuapl.edu (<--- not the machine this is posted from!)
- 301-953-6000, x8360
-
- Room 13-s587
- Johns Hopkins University/Applied Physics Laboratory
- Johns Hopkins Road
- Laurel, Md. 20723
-