| From: | mef |
| Date: | 13 Sep 2000 at 19:07:56 |
| Subject: | RE: AMIOPEN: Re: |
g'o'tz ohnesorge wrote:
>Jim Peters wrote:
>
>> The power of this VP thing is that Amiga can change the processor in
>> their boxes every year (they have expressed this intention). No sane
>> software company is going to lock themselves into this year's model,
>> especially when it means more work for them.
>
>Exactly, and that'll show soon enough in hard-to-ignore ways.
>Umph, can't talk about that yet. ;)
That's great to hear.
Btw., as I've been mentioning all along, I don't believe that regular C/C++
code compiled to VP and then translated to the native machine needs to be
any slower compared to compiling the same code directly. Without benchmarks
this is all just academic, but then again I am in academia. :)
The one data point that I have is the SRC Modula-3 compiler, which compiled
M3 code to an intermediate stack-machine language (IL). This IL was then
read in by a special IL front-end to GCC, which then compiled the IL code to
the native machine. Several years back I worked on an operating systems
project at the University of Washington, called SPIN, which was written
primarily in Modula-3 and of course our initial concern was performance. We
found that the performance of many critical routines written in Modula-3
were nearly identical (i.e., with a few % slower and in some cases a few %
faster) to equivalent routines written in C and compiled with the same GCC
compiler.
So there... that is my data point. I believe something similar can be done
for VP with the result of obtaining performance nearly identical to natively
compiled code. That is, the work would involve writing a front-end for GCC
that reads in VP code, which then recompiles/translates it to the native
backend. In fact, there is no reason why this particular translator
couldn't be used by Ami/Tao in addition to one that comes with the system
now. That is, it complements the one shipped by Tao. Tao's translator
could still be used for applications where performance isn't that important,
or on resource constrained devices (such as cell phones and the like) where
running a ~ 1MB translator isn't feasible. However, on workstations, it
might be better to use the beefier GCC translator.
It is highly likely that I will begin such a project in October or early
November. If there is anyone out there willing to participate, please
contact me at mef@cs.washington.edu.
Cheers,
Marc
Subscribe/Unsubscribe: open-request@amiga.com
Amiga FAQ: http://www.amiga.com/faq.html