mwm@VIOLET.BERKELEY.EDU.UUCP (05/05/87)
Uh, anyone want to tell me what's wrong with the old DEC-10 solution? It allowed arbitrary registers (with the always-exceptoin of 0) for string instructions. If the instruction was interrupted, the micro-registers used to do the counting were pushed back into the registers the user specified. Thus, you couldn't depend on those registers to have any specific value after the instruction, and could tell if you'd been interrupted by checking to see if they changed. No extra stack stores, a little extra logic to cause the micro-registers to get pushed, and _no_ magic registers. The cost (that the values in registers after those instructions is unknown) seems perfectly acceptable to me; but I have a compiler-writers bias. <mike