From mboxrd@z Thu Jan 1 00:00:00 1970 Mime-Version: 1.0 (Apple Message framework v728) In-Reply-To: References: <36111140-8D4D-41B7-BCA2-17659EA08E26@telus.net> Content-Type: text/plain; charset=US-ASCII; delsp=yes; format=flowed Message-Id: Content-Transfer-Encoding: 7bit From: Paul Lalonde Subject: Re: [9fans] Plan9 on the Cell... Date: Tue, 24 May 2005 16:04:25 -0700 To: Eric Van Hensbergen , Fans of the OS Plan 9 from Bell Labs <9fans@cse.psu.edu> Cc: Topicbox-Message-UUID: 52d60abc-ead0-11e9-9d60-3106f5b1d025 Yes, I've been reading the compiler code in anticipation. It's been close to 15 years since I last did any real compiler (code generation) work; it will take me some effort to get back up to speed. I've been itching for an excuse to get a power mac, and this might be it. Is the 32 bit power compiler heavilly used by anyone? The SPEs, of course, are the interesting part from the systems point of view. It would be interesting to find a clean way of offering them up (along with the required PU code - that's the hard part) as a cpu-like computing resource. The tricky part is that SPE code seems to like to set up pipelines using multiple SPEs, which makes allocation trickier. Pre-emption looks expensive because of the local memories. Paul On 24-May-05, at 12:26 PM, Eric Van Hensbergen wrote: > On 5/24/05, Paul Lalonde wrote: > >> IBM looks like they are going to open up the Cell architecture: >> http://www.eetimes.com/news/latest/showArticle.jhtml? >> articleID=163106213 >> Which, if someone can start shipping dev boards, would lead me to >> attempt a port :-) >> I've been lucky enough to be working on one of these puppies for a >> bit (demo at http://ps3.ign.com/articles/614/614705p1.html) and the >> thing is so sweet to squeeze performance out of that I'd love to be >> running a CPU server native on it... >> >> > > I have no direct knowledge of these things, but development boards are > going to be scarce for a while. I've heard rumors that IBM will be > releasing the Cell full-systems simulator as part of them opening up > the architecture -- that will likely be your best candidate for a port > until development systems become more widely available. > > In the meantime, porting to the G5 (or any other readily available > ppc64 platform) would probably be your best bet. First step is the > compiler (while you may be able to use the 32-bit power compiler, > you'll really want 64-bit if you are looking to squeeze performance). > Next is working with jmk (who is adding 64-bit support to the kernels > for opteron) to do the same thing for PowerPC. > > That pretty much sums up the hard parts (at least for an initial > support). How to deal with the SPE's is a really big question, but > that can be dealt with after the core port is complete. > > As you can see, I haven't put much thought into this myself ;) > > -eric >