3 ms·

Ah. The transputer. Did my doctoral thesis using CSP converted to occam for the transputer. Fun times. There's some of my occam code here: http://blog.jgc.org/

Ah. The transputer. Did my doctoral thesis using CSP converted to occam for the transputer. Fun times. There's some of my occam code here: http://blog.jgc.org/2009/08/in-which-i-resurrect-13-year-old-35.html http://blog.jgc.org/2009/08/in-which-i-resurrect-13-year-old... Oxford had some transputers from INMOS. The story I heard was that they'd been manufactured with the die in the wrong way round meaning the pinouts were all messed up. So a bunch (16 IIRC?) had been wirewrapped together to make some hideous but working machine.

dboreham 10y ago
T212s I seem to remember.
david-given 10y ago
I did some work on a transputer at Tao --- we had a nine-node demo system; eight transputer nodes, and one Pentium node, running as an asymmetric multiprocessing cluster. The demo program drew Mandelbrot images by farming out scanlines to each processor node. Unfortunately, the Pentium was so much faster than the transputers that by the time each transputer node had rendered one scanline, the Pentium had done all the rest... Later, the transputer metamorphosed into the ST20, which was sold as a single-processor embedded microcontroller, used a lot in the TV and set-top-box market. It actually worked pretty well. One exciting thing we discovered: on at least on transputer system (and the details are from memory, so possibly incorrect here), if you asked it to do a shift right by 0 bits, the processor would hang. We eventually decided that the microcode that implemented this was doing: while (--shift) value >>= 1; So, if you passed in 0, in the first iteration the shift rolled round to 0xffffffff, and kept going; with interrupts off. IIRC we calculated that it would probably take about a fortnight to run. I don't think we ever tried it to see, though.

3 ms·

Ah. The transputer. Did my doctoral thesis using CSP converted to occam for the transputer. Fun times. There's some of my occam code here: http://blog.jgc.org/

by jgrahamc 10y ago

dboreham 10y ago
T212s I seem to remember.
david-given 10y ago
I did some work on a transputer at Tao --- we had a nine-node demo system; eight transputer nodes, and one Pentium node, running as an asymmetric multiprocessing cluster. The demo program drew Mandelbrot images by farming out scanlines to each processor node. Unfortunately, the Pentium was so much faster than the transputers that by the time each transputer node had rendered one scanline, the Pentium had done all the rest... Later, the transputer metamorphosed into the ST20, which was sold as a single-processor embedded microcontroller, used a lot in the TV and set-top-box market. It actually worked pretty well. One exciting thing we discovered: on at least on transputer system (and the details are from memory, so possibly incorrect here), if you asked it to do a shift right by 0 bits, the processor would hang. We eventually decided that the microcode that implemented this was doing: while (--shift) value >>= 1; So, if you passed in 0, in the first iteration the shift rolled round to 0xffffffff, and kept going; with interrupts off. IIRC we calculated that it would probably take about a fortnight to run. I don't think we ever tried it to see, though.