Re: Prolegomena to any future decimal discussion

[John R Harrison <johnh@xxxxxxxxxxxxxxxx>]

Hi Ivan,

| In a static architecture,  it is irrelevant that the long-latency
| path is rare, because all operations take the time of the longest
| possible path.

Well, even architectures that I would class as "statically scheduled",
such as our very own Intel Itanium architecture, sometimes get hit
with (rare, we hope) special cases where the runtime is much greater,
such as cache misses and computation on binary FP denorms. This
doesn't require scheduling for the worst case, but rather doing
something special when it happens. (Your definition of static
scheduling may be more hawkish than mine, though.)

| To deal with this I expect that general-case performance computation 
| (hardware) will be BCD internally to the functional unit.

That's possible, but it's by no means certain; see John Crawford's
earlier message about binary-based hardware, for example. In any case,
we need to pay due attention to the needs of software implementations on
current platforms. These will play a central role in the acceptance or
otherwise of the decimal standard. If performance on mainstream
platforms is mediocre, there could well be more limited uptake of the
standard. Then we may never get to the point where hardware support for
decimal is generally considered worthwhile.

John.