8 Mar
2011
8 Mar
'11
4:34 p.m.
Mark Tapley wrote:
At 23:56 -0600 3/7/11, Jules wrote:
I see what you mean - I wonder if the actual add only takes one cycle, but
there's an additional 1-cycle overhead (none of the instructions seem to be
less than 2 cycles) which doesn't scale to 8 shifts/adds needed for an
8-bit multiply? That would make sense, I think - 8 cycles of shifts/adds,
standard 1 cycle overhead, and 2 more cycles eaten up doing housekeeping
"stuff".
cheers
Jules
Yes, that seems to be the 'famous' one
that gets mentioned everywhere. It
seems it was of the shift-add variety. Anyone recall if it would work
with
signed integers? (I'm just trying to work out how the math works for
signed
multiplies at the moment)
I have a scanned .pdf of the 6809 programmer's manual here, happy to
send out if anyone wants it, but it's 12.8 MBytes. (I got it from the
freescale website, since reorganized; it's online at
http://www.maddes.net/m6809pm/ .)
It says the MUL instruction A x B -> D (unsigned) takes 11 MPU
cycles. ADD instructions take a minimum of 2 cycles, as do shift (roll,
etc) instructions. So I think maybe there must have been some silicon
(vs. microcode) involved in the multiply? I don't see how the shift-add
sequence could be done in only 11 cycles.