Assembler or machinecode. Was Re: [colorforth] New Linux 4word

On Fri, Apr 08, 2005 at 08:39:08AM -0700, Jeff Fox wrote:
<SNIP>
>
> > An assembler is the proper starting point. There is no Metacompiler
> > without assembler. And an assembler may be enough.
> > Unless of course you think you have a nice source if it is sprinkled
> > with
> >         cd45 2,      12 1, 3135 2,
>
> I heard Chuck talking about this a couple of days ago.  He said that
> he had always written a target assembler in Forth because it only
> took a block or two in most cases. Then he would write the assembler
> portion of CODE words or a kernel using the assembler so that it
> would provide more meaningful code representation to the programmer.

I think this is attainable even for the Pentium.

> He said that the Pentium assembler was such a mess, and so complex,
> and that he used so few of the Pentium instructions anyway that it
> just worth the effort to provide a full high level Pentium assembler
> for his needs in colorforth.   He added that there were so many
> instructions and address modes etc. that people would need to have
> a book in front of them anyway to look up opcodes to see what they
> actually did.  So given that he needed only a few Pentium codes,
> and assuming that most readers haven't memorized the entire Pentium
> instruction set, or that the variation of the syntax that Chuck
> would use in a RPN assembler would be completely obvious to readers
> he didn't add a layer of Pentium opcode names.  He said, you are
> going to have to go look them up anyway to understand that level
> of the programming.

Even so. I managed to get the whole of the Intel Pentium Assembler
down to 13 screens. The full content (316 pages) of chapter 25
of the Intel Pentium architecture Manual is under control.
The whole shit: the sib byte, all addressing modes, prefixes,
16 bits stuff in 32 bits mode and floating point.
(This is not a RPN assembler, but a new principle PostIt FixUp,
it is more modular.)

Now if I could trim that to the needs of colorforth, this would
be say 5 screens. Floating point for instance now takes 3 screens,
these could be dropped immediately.
Even a target of 2 or 3 screens seems reasonable.

I have done a conversion of retroforth, that is in many respects
similar to colorforth. The 1, 2, 3, thingies are automatically
replaced by assembler instructions.

> He added that after bootstrapping then he was in a position to
> do the normal thing of coding a Forth version that won't need
> a kernel in assembler to get going.  It was just a matter of
> priorities.

I can understand that it is not Chucks priority to do the kind of
research into Intel that I have done. But I would be glad to
share any useful techniques that I have found.
There is no escape: it is either machine code in the kernel or
assembler. I'm convinced that my assembler is preferable over
machine code in the context of colorforth.

Then there are occasionally useful words in the Pentium
instruction set like RDTSC, (read time stamp counter).
(I use that one quite a lot.)

<SNIP>

> Best Wishes

More information :
google for "Intel Pentium assembler".
My site comes up first!

Groetjes Albert

--
Albert van der Horst,Oranjestr 8,3511 RA UTRECHT,THE NETHERLANDS
Economic growth -- like all pyramid schemes -- ultimately falters.
[email protected] http://home.hccnet.nl/a.w.m.van.der.horst