home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!europa.asd.contel.com!darwin.sura.net!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!aplcen.apl.jhu.edu!aplcenmp!mef
- From: mef@aplcenmp.apl.jhu.edu (Marty Fraeman)
- Subject: Re: RTX and SC32
- Message-ID: <BxDo6z.4wy@aplcenmp.apl.jhu.edu>
- Summary: SC32 is sort of 8-10 integer mips
- Organization: Johns Hopkins Continuing Professional Programs
- References: <17102@mindlink.bc.ca> <1992Nov3.144748.21826@sobeco.com>
- Date: Sun, 8 Nov 1992 03:25:46 GMT
- Lines: 60
-
- In article <1992Nov3.144748.21826@sobeco.com> jlee@sobeco.com (j.lee) writes:
- >In <17102@mindlink.bc.ca> Nick_Janow@mindlink.bc.ca (Nick Janow) writes:
- >
- >>10 MIPS may sound slow compared to the latest RISC chips, but that 10 MIPS is
- >>for high-level Forth, so the actual performance is significantly higher than
- >>that of a faster RISC chip running C.
- >
- >Can you offer any support for this claim?
- >--
- The closest evidence I am aware of is some work I did with the SC32 a
- few years back. I'm afraid I feel that my results don't fully support
- Nick's claims. I also do not claim that what I did is at all the final
- work on comparing Forth and C. I think the shortcomings of my work are
- obvious to the readers of comp.arch so please don't flame me about
- them. I'm also not an unbiased observer since I was one of the
- designers of the SC32. None the less, I am trying to be objective
- about things.
-
- With all the qualifiers out of the way, what I did was recode the
- Stanford integer benchmarks from C to Forth. Timings indicated that
- the Forth version ran about 8 times faster on a 10 MHz SC32 then a the
- orignal C compiled with the BSD4.2 compiler on a Vax 11/780. If the
- matrix multiply benchmark was left out -- its totally dominated by
- multiplies which the SC32 did with shift and add instructions -- then
- the SC32 was about 10 times faster than the Vax. This work is
- described in an article in the Proceedings of the 1988 Rochester Forth
- Conference.
-
- The problems and limitations of the Stanford suite are well known but
- it was more effort than I could justify to try and convert a larger
- body of code. Having said all this, in my emotional heart of hearts
- (unfortunately, not the evidence requested by the original poster) on a
- real program coded in Forth from the beginning, I think the SC32 would
- look quite a bit better. The orginal C code, which I believe actually
- started life in Pascal, converted very poorly into Forth. Code written
- in Forth from the beginning that produced the same result would do much
- better -- I had a version of the towers program that was about 10 times
- faster than the transliterated C to Forth code.
-
- At any rate, I am comfortable saying that the SC32 programmed in Forth
- had performance approximately equivalent to CISC processors programmed
- with convetional languages and compilers of its day (fully functional
- silicon at the beginning of 1987). I don't think the comparison holds
- up as well to the RISC machines of that era. But it is clear that a
- RISC of that era would run Forth much slower than the SC32. Also keep
- in mind that a small standalone SC32 system (i.e. a dozen chips or so
- including memory) can support its own interactive and extensible
- program development environment without needing cross-compilers,
- emulators, and so on.
-
-
- 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
-