Jecel Assumpcao Jr
jecel at merlintec.com
Fri Aug 25 21:22:10 UTC 2000
On Fri, 25 Aug 2000, C. Ramakrishnan wrote:
> Dobes Vandermeer writes:
> > Any thoughts on making a Self compiler for the new Amiga VP
> > architecture?
> ...and as a follow-up, what the the VM mavens here think about the
> AAmiga/Tao-Group's Java VM technology. I've heard it's "real fast" but
> don't know anything about how they've implemented it.
Here is what the Tao group page says:
If code is in Java Byte Code, then it is converted to VP code by
the intent jcode Translator. Once code is in VP, it is then in the
common binary portable format upon which intent depends. VP
code is executed by using a Tao 'Translator' to translate code
from the VP format to the appropriate native code and then
executing the native code directly. Note that this technique is
radically different from interpretation (in the same way that
JIT-compiled code in a JVM is executed directly, not interpreted).
Translation can occur in a "just in time" fashion (i.e. code is
translated to native when it is required), or VP code can be
pre-translated. Any combination of the two approaches can be
employed in a system depending on the relative priorities of ROM
footprint, latency and other memory considerations.
So, basically, this is their version of Pepe, the Java in Self that was
created at Sun. The advantage is that they have VP to native
translators for Motorola M-Core / PowerPC, ColdFire, ARM® (including
ARM 6 / 7 and StrongARM®), ST40, x86, MIPS R4xxx/3xxx, Hitachi SH3 /
SH4, NEC V850. I think these translators were directly coded in
assembly language, so they are very fast and small.
One option would be to patch the current Self 4.1.2 VM to generate VP
code (instead of Sparc or PowerPC) and recompile the VM itself using
their C++ compiler.
More information about the Self-interest