Re: [Tinycc-devel] bytecode

[Nick Kelsey]

Hi Chris,

I looked into this for one of our applications - TCC is reasonably good 
for generating VM code but a simple VM interpreter won't be fast to execute.

Instead I have been working on a TCC target that generates architecture, 
position and size independent code (with a fixed register/memory model) 
that can be converted into native instructions by the target processor. 
For a 32-bit processor the byte-codes generally translate to a single 
instruction or two for a literal.

The targets we use are all embedded processors, often with less than 64k 
of RAM - with this approach the bytecode isn't stored, it gets 
translated on the fly as it is downloaded. My main test target is a 
Ubicom ip3k and the translator uses less than 4k of code space... less 
space than a VM would :-)

The translator is that it is written in c-code but with an additional 
custom pre-processor that translates target specific assembler into 
binary constants so the author doesn't need to know or enter any binary 
numbers.

What is your application?

Nick

Chris wrote:

> *This message was transferred with a trial version of CommuniGate(tm) 
> Pro*
> I have seen some brief mention of bytecode targets here and there on 
> this list over the last few years.  Has anything come of any of those?
>
> I really would like to see a C to bytecode compiler and an 
> accompanying lightweight VM to run the bytecode.  I imagine the VM 
> would be extremely simple (relatively speaking at least), lightweight, 
> and fast if the only functionality was similar to standard ANSI C.  
> This would allow easy embedding of the VM in applications.  Especially 
> if it allowed direct calling to/from native (binary) system libraries.
>
> This would be very useful as a lightweight cross-platform system that 
> didn't require recompiling for each target and kept the simplicity of 
> C.  Dynamically loading code off the network and running directly from 
> memory would be much easier as well.
>
> Targeting TCC bytecode to a new processor would be as easy as porting 
> the VM to that processor using pre-existing compiler tools for that 
> hardware; no machine code messiness.
>
> The Quake3 game engine actually has a VM and compiler that is pretty 
> close to exactly what I speak of but unfortunately it's GPL'd so not 
> all that useful for commercial embedding.