m5@lynx.UUCP (Mike McNally) (05/28/88)
Has anyone or anything (meaning a benchmark of some kind) noticed any performance problems with 80386/387 bugs? I guess it's also appropriate to ask if the bugs still exist. Two I've heard about involve bad interactions between paging and the 387, like when ESC instructions start in a resident page but end in a paged-out page, and also some kind of deal with the instruction prefetch unit screwing up addresses from the 387. I assume that some sort of nasty workarounds have to be done for these bugs; are hardware solutions present in typical 386 PC's? Can the bugs be dealt with in software? If so, how does that affect floating point (and other kinds of) performance? Just curious.