Author: Keith Evans
Date: 18:45:41 06/21/02
Go up one level in this thread
On June 21, 2002 at 21:12:17, Dann Corbit wrote: >On June 21, 2002 at 20:44:55, Keith Evans wrote: > >>On June 21, 2002 at 20:15:43, Dann Corbit wrote: >> >>>On June 21, 2002 at 18:48:53, Keith Evans wrote: >>>[snip] >>>>Just a quick note - when I looked into Cilk a little while back it looked like >>>>they were more into SMP systems now (probably because they were more available >>>>to them and performed better) and weren't supporting AMP in their latest >>>>releases. I thought that Cilk was pretty slick though, and even ran a few >>>>programs on an SMP box running Linux. Unfortunately as far as I know they don't >>>>make the source to their chess program available. If I'm wrong about any of >>>>this, then please post details. >>> >>>From: >>>http://supertech.lcs.mit.edu/cilk/index.html >>> >>>We find this: >>>"1999 World Computer Chess Championship >>>Cilkchess competed in the 1999 World Computer Chess Championship, June 14-20, >>>1999 in Paderborn, Germany. Cilkchess was written in Cilk-5, and ran on a >>>256-processor SGI Origin 2000 at NASA Ames." >>> >>> >>>From: >>>http://supertech.lcs.mit.edu/cilk/FAQ/section1.html#whatiscilk >>> >>>We find this: >>>"Question 1.1. What is Cilk? >>>Cilk is a language for multithreaded parallel programming based on ANSI C. Cilk >>>is designed for general-purpose parallel programming, but it is especially >>>effective for exploiting dynamic, highly asynchronous parallelism, which can be >>>difficult to write in data-parallel or message-passing style. Cilk has been >>>developed since 1994 by the Supercomputing Technologies Group at the MIT >>>Laboratory for Computer Science. Cilk has been used for research, teaching, and >>>for coding applications such as a virus shell assembly simulator and three chess >>>programs." >> >>But how about? >> >>"The 256-processor SGI Origin 2000 is based on breakthrough ccNUMA >>(cache-coherent non-uniform memory access) architecture" >> >>"Question 1.6. Does Cilk run on networks of workstations? >> Cilk-5.3 does not run on distributed-memory machines" >> >>This sounds a lot different (and more expensive) than clustering to me. Maybe I >>made a terminolgy error with AMP? Isn't ccNUMA basically a form of SMP? > >No. >http://searchsystemsmanagement.techtarget.com/sDefinition/0,,sid20_gci212678,00.html > >Allow me to highlight a salient part of the quote above: >"... but it is especially effective for exploiting dynamic, highly asynchronous >parallelism ..." On <http://www.symbio.jst.go.jp/~tino/Html/beowulf.html> I found the following quote: "CILK a package that contains the cilk programming language which is a C extension. Simple commands like cilk, spawn, and sync should be added to the serial C code to obtain parallelism. Unfortunately the current package is for SMP machines only, i.e. a single machine with multiple processors. A alpha release of a distributed version of CILK is available, but is slightly outdated and has severe limitations." I'm personally interested with what somebody could do today with a publicly available Cilk. Last time I looked I got the impression that it was now really targeted at SMP machines, and that latter quote reinforces that belief. I wonder why there aren't more game programs written in Cilk? I guess that once you figure out Posix threads there's no compelling reason to look back? It does have some neat profiling features, and it actually works out of the box on Linux machines. -Keith
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.