EE 4720 - What's New

Grading Update 5: Course Grades Ready! Thanks to everyone who worked hard in this course! Good luck to everyone!! A solution to the final exam will be posted either in the next few days or two weeks before the Spring 2007 final exam.

About the final, I'm a bit disappointed that no one clearly answered the Mysterious Programmer question. To refresh everyone's memory: the Mysterious (originally I called him or her mischievous) programmer hacked the OS of a software company's computers so that un-aligned loads and stores, rather than resulting in Bus Error signals, would finish normally. By finish normally I meant the correct data was loaded or stored, as though there was no alignment restriction. How did the programmer do it? Remember SPARC quad precision arithmetic instructions that are omitted from most (all?) SPARC implementations? They would raise an illegal instruction exception but the exception handler would do the quad-precision arithmetic using multiple lower-precision instructions, getting the source operands from the registers specified in the faulting instruction and putting the result in the destination registers (four of them, remember?) specified by the instruction. The handler would then resume execution at the instruction following the quad-precision arithmetic instruction. To the program it would be as though the processor had quad-precision instructions, albeit slow ones. The mysterious programmer did something like that for un-aligned accesses. For an un-aligned load the handler would do two aligned loads, shift-and-mask the loaded values into the faulting load's destination register, then resume execution at the instruction following the faulting load (rather than signaling a Bus Error). That's how he or she did it, is it a good thing? No, for two reasons. The first, which some students got, is that customers would not have this forgiving exception handler and so they would get Bus Errors. That would make company X look bad. The second reason is performance: theses unaligned loads would take much longer than an ordinary load to complete.

Now back to the grades. For those who did better on the final than the midterm, the course average was computed using the final weighted 80%. For this calculation 12 points was added to the final so that this adjusted final exam average matched the midterm exam average. Enough talk, to find your grade click here.

For those that are still not tired of the material might want to take my microarchitecture course next semester. Motivated undergraduates are welcome!

Have a good winter break!