Newsgroups: comp.sys.transputer From: Alec Cawley Subject: Re: Processor sought to replace transputer network Date: Fri, 8 May 1998 19:39:28 +0100 Mime-Version: 1.0 Message-ID: In article <01bd7a60$403c88c0$3713c8c3@imshq.ftech.co.uk>, Peter Ross writes >Hello everyone, > >Due to the imminent demise of the transputer I am being asked to redesign >our embedded controller product using different microprocessor technology. >I am therefore seeking views and experiences (good or bad) on migration. The obvious transputer-like processors are the Texas TMS320 and Analog Devices Sharc processors. Both of these are DSP-type devices with links somewhat similar to those of the transputer. Neither has the integrated link interface/task switcher of the transputer. However, 3L (http:://www.threel.co.uk) have ported the library calls which interfaced their C compiler on the transputer to those of both these processors. It is therefore quite straightforward to produce a distributed C implementation similar to a traditional transputer C implementation on these processors >Our current product uses a 25Mhz T805 to implement a motion control state >machine programmed in C. In addition to this we have a network of T225s for >peripheral interfacing and numerical processing in Occam 2. I am not aware of any implementations of Occam on linked processors; all the "live" implemetations I know of seem to be running on Unix engines. ================ However, having faced this same problem for our own networks of T425s, I have chosen the StrongArm. Unfortunately, we haven't yet got any of these going yet (the first one should go for layout in the next week or so) so I can't answer for results. One thing we have found is that it is quite easily to implement the inmos OS link in Altera Flex devices - simple enough that two of our engineers did so independantly when they needed to communicate with a transputer. Another implementation has been done in VHDL by Professor Brian O'Neill of Nottingham Trent University, and he said that he was intending to post this on the WWWW "soon" - try http://www.ntu.ac.uk/ee, or similar. The reason he is doing this, and one I commend to your attention, is to interwork with his ICR C416 OS-link self routing crosspoint. I have recently switched to using this instead of a C004 connected network, which needed through-routing in order to get to remote processors. The difference is incredible - it literally saved my project. This is the main reason I would use the OS link rather than the TMS320/Sharc links. If your problem is numerically intensive and/or is basically a pipelined process (where data is passed from one processor to an adjacent one in a staged process), the DSP chips are better; they have lots of raw crunch power, and their links are faster between adjacent processors. If your problem is like mine - distributed control with an any-to-any communications requirement - and you need agility rather than floating point crunch, the StrongArm looks to me a better bet. I wouldn now not dream of implementing a multiprocessor network in which I have not got an all-to-all interconection. Brian O'Neill has a project to design an FPGA intended to convert a StrongArm into a transputer replacement. It provides the OS link interface (only one, because it is intended to be used with the C416) with DMA, sdram interface, ability to boot over the link, interrupt controller, support for an inmos-style timer and so on. The microkernel to convert this to a tranputer is, hopefully, small. However, it will probably be programmed in GNU C/C++ only - though there is no reason why the Kent Occam compiler could not be ported to it. With essentially no evidence to prove it, my feeling is that a StrongArm should approximately match a T805 on Floating Point and task switching/channel communications, and beat it about ten to one on integer and miscellaneous code. -- Alec Cawley