pearce@calgary.UUCP (07/09/86)
Does anyone out there know of any research into multi-processor ray tracing? I know of the LINKS hardware in Japan, Micheal Ullner's thesis, Dippe and Swensen's SIGGRAPH paper, Apollo's "Long ray's journey into light" (but have not seen anything written about this), Fujimoto's "ARTS" in CG&A, and, of course, what John Cleary and I are doing here. But I've not heard of anyone else. Although any information would be greatly appreciated, I'm not interested so much in vectorization or pipelining the algorithm as in arrays of processors running identical code with differing data spaces. For example, while simply dividing the screen into sub-sections and having each sub-section assigned to a distinct processor with global scene space knowledge is multi-processing, the processors in such a case do not require inter-processor communication durring the ray tracing procedure. Any discussion on the matter would be welcome as well. Andrew Pearce Dept. Computer Science, The University of Calgary, 2500 University Dr. N. W. Calgary, Alberta, CANADA T2N 1N4. Ph. (403)220-3536 Usenet: ...{ubc-vision,ihnp4}!alberta!calgary!pearce [and thanks to all who responded to my request for the Rubin&Whitted paper]
metzger@heathcliff.columbia.edu.UUCP (07/14/86)
In article <249@vaxb.calgary.UUCP> pearce@calgary.UUCP (Andrew Pearce) writes: >Does anyone out there know of any research into multi-processor ray tracing? >I know of the LINKS hardware in Japan, Micheal Ullner's thesis, Dippe and >Swensen's SIGGRAPH paper, Apollo's "Long ray's journey into light" (but have >not seen anything written about this), Fujimoto's "ARTS" in CG&A, and, of >course, what John Cleary and I are doing here. But I've not heard of >anyone else. I am working on a ray tracer to run on the DADO computer we are building here at Columbia. Dado 2 (the current iteration) is basically 1023 processors in a complete binary tree. The machine has a great deal of flexibility in how you can manage the processors, although communication is restricted to Parent-child communication (as one would expect in a tree machine.) >Although any information would be greatly appreciated, I'm not interested >so much in vectorization or pipelining the algorithm as in arrays of >processors running identical code with differing data spaces. >For example, while simply dividing the screen into sub-sections and having >each sub-section assigned to a distinct processor with global scene space >knowledge is multi-processing, the processors in such a case do not require >inter-processor communication durring the ray tracing procedure. We am trying to exploit the machine's flexibility to explore several strategies for ray tracing. The one currently being worked on uses a one object/processor strategy, and the machine is used as a dedicated intersection finder. But other techniques also look promising. One idea floating about is to use the machine partitioning to effect space partitioning. Since DADO is fairly medium grained a lot of exploring seems to be possible. Well, anyone else out there involved in anything in the way of multi processor ray tracing? It looks like a promising area of research, and a discussion here can only benifit everyone. Perry Metzger UUCP: ...![seismo|topaz]!columbia!heathcliff!metzger ARPA: metzger@heathcliff.columbia.edu
liam@qmc-cs.UUCP (07/17/86)
Expires: Sender: Followup-To: Xpath: ukc eagle We have a group in the Computer Science Department at Queen Mary College (University of London, England) which does ray-tracing (amongst) other things. The hardware is an ICL DAP: a 64 x 64 array of single-bit processors with separate memory spaces and a common instruction stream. The total memory is 8 Mbytes and it lives inside an ICL mainframe as a memory-mapped peripheral. All the processors are connected to their NSEW orthogonal neighbours and each instruction can be "masked" so that only processors with a 1 in a selected memory location will obey it. Pictures are usually processed on this machine in one of two layouts: either split into tiles of 64x64 pixels, "stacked" in the memory, or split into 64x64 tiles of nxn pixels and stored one tile per processor. We also have at QMC a 32x32 version which sits inside a PERQ workstation and has sensible IO - the molecule ray-tracing man at the DAP support unit here assures me he will be able to generate ray-traced colour pictures of insulin moelecules in near real-time with this device! Contact addresses for the people doing this work are Hilary Buxton ( hilary@cs.qmc.ac.uk, hilary@qmc-cs.UUCP ) John Quinn ( c/o liam@cs.qmc.ac.uk, liam@qmc-cs.UUCP ) (When I can get John Quinn's real login name I will post it - in any case I will try to prod them into giving some more details: it's a hard life being the Departmental electronic gossip!) -- William Roberts ARPA: liam@cs.qmc.ac.uk (gw: cs.ucl.edu) Queen Mary College UUCP: liam@qmc-cs.UUCP LONDON, UK Tel: 01-980 4811 ext 3900