Reply to thread

Message

[QUOTE="glen herrmannsfeldt, post: 1691761"]
Chris Torek wrote:
(snip)

At some point, one of the thinks I look at in a new architecture
is its implementation of (n)*(n)=(2n) multiply, and (2n)/(n)=(n) divide.

I am not so convinced anymore, though.  They are primarily used to
write multiple precision arithmetic routines.  It isn't that much
harder to write them with the 32 bit versions as the 64 bit versions.

The amount of work necessary to make the hardware is not necessarily
worthwhile for the small advantage that it supplies.

When IBM first added the extended precision (128 bit) floating point
instructions to S/360 and S/370 they did not include a divide 
instruction.  They found that it was not worth the cost, but instead
supplied a software simulation of it.  They also supplied a simulator
for all the extended precision instructions for those processors that
needed it.

If Sun provided either a simulator or subroutine that would do those
operations, that would seem useful enough to me.

-- glen
[/QUOTE]

Verification