Bryan Parkoff wrote:
What I understand is that Intel has more instructions than 6502
does. For example, 6502 does not have floating registers, but it has
to build routines in order to support floating registers. It can
decrease the performance. Intel does have built-in floating
registers including more floating instructions that may improve
performance. Does it make sense? If we WANT to design
microprocessor from scratch that will add floating instructions to
65816 and 65832, it is impossible to duplicate the 65816
microprocessor's schematic, but we can study using AND/OR/NOT GATES
and boole truth table inside microprocessor's schematic in order to
create binary instructions. If we have to break black cover
before we can see naked IC wires inside microprocessor, it might
provide us more information before we can be able to build one from
scratch. What do you think?
Well, Roy has convinced me to give this group another chance. So here I go
with my attempt to contribute positively. (Watch me wind up putting my foot
in my mouth.)
The original 8086, and its stablemate with the 8-bit data bus, the 8088, did
not have floating point registers -- they were only implemented in the 8087.
This was followed in turn by the 80287 and 80387 (was there an 80187 to go
with the underappreciated 80186?) until the floating point circuitry was
integrated into the 80486. The 80487SX was simply a fully working 80486DX
chip that replaced the functionality of the 80486SX chip, which was just an
80486DX that had a disabled floating point system.
The 68000 series did not include an internal floating point system until the
68040. Motorola floating point accelerators included, if I recall correctly,
the 68851, 68852, 68881, and 68882. I believe, but I'm not sure, that they
were designed to be the companions respectively of the 68000, 68010, 68020,
and 68030.
While the various floating point chips were designed to work best with their
companions, it is theoretically possible for any CPU to use any FPU, given
the right software and interface logic.
That is to say, for a dedicated individual, it would not be impossible to
use a Motorola 68851 or an Intel 8087 with a WDC 65816, although a
speed-matched FPU might be better suited, in order to use the same clock and
not require async operation. I think neither the 8087 nor the 68851 were
that widely sold in speeds greater than about 8MHz. Given that the 65816
seems to run up to about 20MHz, perhaps a 68882 or 80387 would be a better
match. But then, you have to deal with the 32-bit data bus of the 68882 and
80387. (Was there an 80387SX with a 16-bit data bus to go with the 80386SX?)
So then you move your CPU up to the 65832, perhaps?
However, an FPGA implementation of a 65816 or 65832 might consider using an
array large enough to accommodate an FPU core onboard. There are possibly
other IEEE floating point VHDL cores out there that could be used.
I probably haven't contributed any answers to the discussion, but hopefully
I've contributed some points to think about, without too much incorrect or
incomplete information.