Re: [GEN] Random Thoughts, Worth What You Paid For Them.
On 15 Oct 2000, at 16:20, R. W. D. wrote:
> 4. OX is interesting; I -think- I see where Mr. Cuny is going.
> I'd been wishing for yacc & lex in Euphoria too. But I wonder
> how difficult it would have been to implement precedence rules?
> ("...left as an exercise for the reader." ?)
I wrote an additional 32K of machine code for the C64, back in 1987 or so, over
100
new instructions,, including some real fancy goto/gosub options. 
> Apropos of Nothing: For the longest time, I thought Mr. Cuny was
> the author of Euphoria. He certainly seems to contribute... quite
> a bit.
I did too! I even thanked him for doing it!
> 6. There is no number six.
Number six.... number six... number six.....
Hmm, you're right, no answer!
(Beatles reference)
> 7. There is (or was) a language called FORTH. FORTH's creator
> eventually built a microprocessor that can be directly coded in
> FORTH ... that is, the processor's assembly language -IS- FORTH.
>
> I assume that some hardware guy somewhere is slapping together
> microprocessor to execute Java bytecode. If Euphoria could emit
> this bytecode, it would have a jump on other languages when such
> a microprocessor becomes available.
You know you can buy several processors today that can be told to run any
bytecode
construct you want? They are all based on PLDs, and not only run fast as hell,
but
they can be told to run another bytecode block while running. So you could run
68000
code, then tell the same cpu to run 80xxx code, then 6502 or 603 code, no
rebooting,
no errors. You can do native multiprocessor schemes more easily this way than
with
cisc cpus too. And they can be extended width-wise,, tired of 64bit
proccessing?, slab
a couple cpu's together, add the 256bit code, and run it.
Combining #4 and #7 above, i also came up with a multitasking asynchronous
reconfigureable cpu design that was optimized for string handling, most 1ns
"clocks"
could exec several instructions at once, for instance, there were instructions
for using
the 128bit datapath (needed 128bits to keep the cpu fed with instructions fast
enough)
with all the cpu regs at once to do string searches. This had less overhead than
you'd
think, cause, for instance, stuffing all the regs to the stack was a single
500ps
instruction. And cause it was async with multiport registers, you could be
loading the
regs at the same time they were being saved out. I got as far as board layout
for it,
and money calculations (i needed $1200) and marketing feasablity (none),, and
dropped it.
Kat
|
Not Categorized, Please Help
|
|
