| From: | mef |
| Date: | 13 Sep 2000 at 23:03:24 |
| Subject: | RE: AMIOPEN: Re: |
g'o'tz ohnesorge wrote:
>I see that you're betting on the hard work on the
>GCC part being done for free, ...
Exactly!
>but your way may still be overly complicated, and
>may prove not too efficient in practice.
Overly complicated and inefficient in what way? Could you expand on this
thought a bit.
>Even GCC is probably not the most advanced thing ever..
That is true, but it has a huge following and I believe that it will improve
over time.
>.. are you aware of IBM's DAISY?
Sort of... not sure how one could straightforwardly use this technology to
built a better translator.
>Also HP have done some work on processors emulating
>themselves, and incrementally improving their code
>during that .. might be worth right the next look.
I am aware of this work, but I forgot the name of the project. In essence
it is similar to Transmeta's code morphing, except they apply it solely to
the HP-PA processor. I believe they do runtime trace scheduling, by
figuring out what the most commonly used basic blocks are. Not sure what
other optimizations they apply. However, I believe that their work is
predicated on the fact that the initial compilation was already done by a
very good compiler. So although it is applicable, I think there is still
work at the compilation/translation phase.
I still think that the GCC approach, which reads in VP code and converts it
to one of the many backends, is the most viable and straightforward
approach; and, the licensing agreement is something I can live with. My
expectation is that a prototype of this could be hammered out within a
matter of weeks.
Cheers,
Marc
Subscribe/Unsubscribe: open-request@amiga.com
Amiga FAQ: http://www.amiga.com/faq.html