Re: A83: Ti8xcc [82/83/83+/85/86]

[Prev][Next][Index][Thread]

Re: A83: Ti8xcc [82/83/83+/85/86]


> > We can also invoke the optimizer (copt) which is used in Ti8xcc to add
> > new optimizations (if we find any).
> 
> Have you actually tried this?  I'm just slightly worried that the format
> of TISCO output might be different enought that it wouldn't work.  But,
> probably it will.  Anyway, there certainly are some optimizations that
> could be added; I can even think of a few myself.  But, the hard part is
> doing optimizations on a larger scale, rather than just replacing small
> sequences of instructions with better ones.  That's why I recommended a
> Z80 port of GCC.

copt is just a peephole optimizer, you say to it "change ..this.. to
..that.." and it will do it. No special code opimizing-pattern checking or
something...
But the optimizations that are made for Ti8xcc are not always 'directly'
portable to TISCO (different use of (compiler-specific) system-routines).

> > I chose Ti8xcc because it was the most complete 'C'-compiler I could
find,
> > and it was already for 2 calcs (portability). But this lack of structs
is
> > really anoying (just try to make decent file-handling routines without
> > them...).
> 
> When I think of a programming language for the calculator, my main
concern
> is of course how well it can be used to program games.  And if code is
> huge and slow, that doesn't make for good games.

Okay, not all arcade games are possible to make on Ti8xcc (nor TISCO), but
RPG's, puzzel-games, etc. are possible to make. BTW. The speedloss isn't
always that tremendous.
And what about a compiler that can't even set a sprite nor a line to the
screen, what's it name... ûhm... ... . .  .
   ...TISCO maybe???
That helps for gaming purposes, yay, text only arcade games...
   Sorry for being nasty...

> Of course, even though (since it structs and better output) I do consider
> the TISCO compiler the best, I am of course painfully aware of its
> incompleteness as well.

Okay, we will just have to work on those things, line, sprites etc.
But I'm rather thinking of totaly rebuilding the includes, with the ones
of Ti8xcc as basement.

> > [in reference to a GCC port to the Z80]
> > Okay, anybody who wants to help, I would suggest?
> 
> I might be willing to work on this task.  However, I'm not at all certain
> that I have the necessary skills to do it.

Me neither...
BTW: we than will have to find people who wan't to learn 3 Assaembly
languages:
1. Z80, the target one (should be 'know' by everybody around here)
2. the Asm language of the processor that has already been implemented in
GCC
3. The 'GCC-generic-asm-language'. GCC first compiles to a generic
language, which is then optimized, and converted to the target language.
And its then again optimized for the target language.

> > [referring to obtaining TISCO source code]
> >
> > I've mailed to the adresses mentioned in the docs, but these adresses
don't
> > exist anymore. I've searched the web and I found 5 other adress from
> > persons with the same name. Two have already bounced back, maybe I've
some
> > luck with the others.
> 
> Well, good luck.  I would expect that someone who programs a C compiler
> would have enough interest in electronic communication to remain
> available.

I'm not so sure, till now have 'the other 3' not responded, so I think they
are not the guy we're looking for, but I'll mail them again. I've found
some new ones with 'Copernic 2000' too (I think, just have to compare them
with the ones I got).
And if we still don't get any response, I'll mail the adresses to the
archive. So others can try it. Why?, you say. I think Hotmail has added my
mailserver to their list of possible spammers. So most of the time my mails
won't come true. About half of the adresses are Hotmail...

> It would be nice if the compiler did allow "normal" floating-point
> math, but I don't think any of the Z80 calculator ones do (even TI-GCC
> doesn't!)

TI-GCC uses some sort of structure (real + mantissa) to convert the GCC
type of float to a sort of 'Ti float', where after you can do just the
'normal' operations (at least they are named a bit more understandible...)

> [list of other compilers]
> 
> This is probably the most important thing in the message.  Quite likely,
> swithcing to one of these compilers is a good idea.  I will write more on
> this subject soon.
> 
> In particular, I think the z88dk is probably something we should consider
> using.  It is (at least according to the documentation) almost a complete
> compiler, hopefully without the limitations of either TI8XCC or
> TISCO.  Even though you said there was no DOS version, it seems that
there
> is one now (the download page has a link to another page with both Win32
> and DOS versions).

I've downloaded the DOS version too (the DOS version that can run under
protected mode ... = Win32). But it's a older version of the compiler
(don't know if there are any major differnces). The site also told me what
was changed to the code in order to make it work with DOS, that were about
4 lines of code, so 'maybe' we can do it ourselfs with the new version, and
compile it. (we don't have to recompile the assembler, It didn't change for
a
while, 1988 or something...)

> Last, I don't think a gameboy compiler could be used as-is; the gameboy
> CPU is *not* identical to the calculator's.

I asked the guy who made the GBDK/ SDCC, about compatibility with the
real Z80, this was his answer:

+----
| Hi Henk.  sdcc already supports the z80 (as opposed to the gbz80 used in
| the Gameboy) - try "sdcc -mz80 blah.c".  It also uses a librarified
| version of copt as the peepholer.  I'm definatly interested in having
| people who have more z80 knowledge than me working on the code, though.
+---
Follow-Ups: