VLIW pre-history
|
From: Eugene Miya on 18 May 2007 19:35 The first obscure keywords tell me that "In the beginning was the pipe. And the pipe was good. And Ken blessed the pipe..... In article <f2kue0$afg$1(a)joe.rice.edu>, jle(a)forest.owlnet.rice.edu (Jason Lee Eckhardt) writes: >|> Rau, due to his 1981 MICRO-14 paper. But clearly the programming >|> ... showing the idea to pre-date Rau's article. >|> I suppose Rau could still be legitimately credited with having made >|> the idea more systematic (the "modulo constraint", etc), and >|> therefore easier to incorporate into a compiler. In article <f2l126$5a0$1(a)gemini.csx.cam.ac.uk>, Nick Maclaren <nmm1(a)cus.cam.ac.uk> wrote: >That, I regret, is because the modern generation of computer scientists >refuse to accept that anything happened before the Unix revolution. >Though the ability to publish old results as new inventions gives them >an incentive. It's called publish or perish. It goes toward tenure. I'm surprised that you even grace to call it a Revolution. Peer review clearly has its limitations. >Bob Rau may well have done important new work, and even been the first >to write it up, but software pipelining as a compilation technique >dates from no later than the 1960s. It was old hat when I came on >the scene - in that context, c. 1971 - and the methodology was well >understood, too. Trace scheduling on horizontally microcoded wide instruction words (say 512 bits) even in the 60s eh? It's a GUI era now. The real benefit to come with app users use pipes, eh Nick? >In those days, we Just Did It. oh. --
From: John Savard on 18 May 2007 20:31 On 18 May 2007 09:21:30 -0800, eugene(a)cse.ucsc.edu (Eugene Miya) wrote, in part: >In article <1179466666.901608.324210(a)o5g2000hsb.googlegroups.com>, >Quadibloc <jsavard(a)ecn.ab.ca> wrote: >>I'll agree that this isn't true VLIW, just "vaguely VLIW-ish", as >>you've put it. But with a dataflow architecture, the program counter >>is a means and not an end, so the relationship may be closer than it >>seems. >What kinds of "dataflow" architectures are you talking about? >The two kinds I am familiar don't have program counters. I'm applying the term Dataflow to the Cyberplus from Control Data. This machine had an instruction that controlled a large number of functional units - and two ports to memory. As a result, to keep the functional units busy, instructions had to largely take the output of one functional unit, and send it to another functional unit as its input rather than storing all results, and fetching all operands, from memory or a register. So if you have a short loop of instructions, it isn't necessary for the instructions to all be identical; instead, a chain involving as many steps as the number of functional units *times* the number of instructions in the loop is possible. John Savard http://www.quadibloc.com/index.html http://www.quadibloc.com/index.html
From: Leif Harcke on 19 May 2007 21:17 On Fri, 18 May 2007 15:27:17 -0800, Eugene Miya wrote: > My thinking is that some place with some obsolete obscure > 120/164/264 processor attached to something like an MRI or CT will > be found. Meanwhile I'm not holding my breath. The VAX 11/780 in the pedestal of the Goldstone tracking station had one attached to its side up through 2002. It was part of the real-time ranging system of the planetary radar. I believe it has since been de-commissioned and surplussed. -Leif
From: Stephen Fuld on 19 May 2007 23:34 Eugene Miya wrote: > "In the beginning was the pipe. And the pipe was good. > And Ken blessed the pipe..... snip > Pipelining goes back to 1964 in my biblio alone, but there are > subtle terminology distnctions between all the different ways > people use it (ask how errors propagate backward, that's a good first > order question). I thought that pipelining went back at least to Richard Feynman's use of pipelining his "computers", the human kind, in the work on the Manhattan Project. :-) -- - Stephen Fuld (e-mail address disguised to prevent spam)
From: Nick Maclaren on 20 May 2007 04:44
In article <dlP3i.16047$p47.5679(a)bgtnsc04-news.ops.worldnet.att.net>, Stephen Fuld <S.Fuld(a)PleaseRemove.att.net> writes: |> Eugene Miya wrote: |> |> > Pipelining goes back to 1964 in my biblio alone, but there are |> > subtle terminology distnctions between all the different ways |> > people use it (ask how errors propagate backward, that's a good first |> > order question). |> |> I thought that pipelining went back at least to Richard Feynman's use of |> pipelining his "computers", the human kind, in the work on the Manhattan |> Project. :-) As an industrial technique, it is not later than mediaeval. In computing, the word seems to date from the mid-1960s, though the technique is older. It was often just lumped in with other forms of instruction scheduling. Regards, Nick Maclaren. |