SHARC Project

The SHARC work was done in 2000 and 2001 as a subcontractor for a communications subsystem. The network was a token ring using ARCnet and some USART devices. The communicattions was byte based, but the SHARC only accessed 32-bit words. The address decoding was on integer boundaries, but packing the stream involved plenty of shifting, masking and OR’ing.

The software was written in C. Debugging was via four bits that the processor could write to quickly, and these lines were connected to a logic analyzer. During a software upgrade to VisualDSP++, we could not compile previous projects and libraries were full of bugs. We lost well over a month trying to pioneer. That was honestly the worst software release I have ever worked on. Analog Devices had done much the same on a previous upgrade according to some of the permanent staff, but they still insisted using the processors for communications.

The project was a small subsection of a much larger system, which included radar and other boards using the SHARC. It made sense to use existing skills, however, there was a lot of movement between projects and we were also involved in a processor selection study with a budget of 150 hours. Management had identified the problems of using DSP devices for communications, but with military projects, the writing of specifications takes precedence over implementation. All those proposed architectures were not going to be overturned lightly, plus budgets and milestones were set in concrete.

Long-term projects with small teams end up in working on obsolete hardware. The fortunate part of this is that as the team jumps ship, there are plenty of opportunities for contractors! The money was good, but we are unlikely to take up similar work.

Analog Devices Inc., has multi-core versions of the SHARC and prices are low in 2011. The TigerSHARC was being designed in at the time of the SHARC work, but we have not looked at newer offerings.