On Oct 23, 2015, at 10:36 AM, Scott Hess <shess at google.com> wrote: > > Indeed, and the cost was the need to have two completely independent math > systems, one precise and one fast. For obvious reasons over time people > who did a lot of math just figured out how to make the fast one precise > enough for their needs, so mostly nobody wanted the precise one. The > obvious choice at that point was to reclaim that silicon space to make > everything else faster, and we ended up here.
Also the fact that most older CISC chips did BCD instructions in microcode, in a time when a non-trivial amount of software was written in assembly or with very crude compilers. With modern, RISC-ish systems, it is often almost as fast to do such things in application code, especially with a good compiler that can optimize the lower-level instructions for a specific core design. -j -- Jay A. Kreibich < J A Y @ K R E I B I.C H > "Intelligence is like underwear: it is important that you have it, but showing it to the wrong people has the tendency to make them feel uncomfortable." -- Angela Johnson
