[qcc] Any sljit experience?

[Rob Landley]

On 05/05/2015 09:53 PM, Bruce Ewing wrote:
> I received a suggestion today that sljit might provide a better compiler
> intermediate language (they call theirs a LIR) than TCG does.

There are over a dozen backends out there. The qemu guys actively
maintain support for a large range of processor types, so leveraging
their work interested me.

Also, TCG doesn't particularly have a compiler intermediate language,
it's just a code generator, so this also seemed like a good fit.

> sljit has
> a very similar set of backends that it supports for output.

Let's see... sljit has 5 architectures, of which 4 have both 64 and 32
bit variants. The word "endian" does not appear on their main page, but
let's give 'em that one.

Using tab completion on "qemu-system-" I see alpha, cris, lm32, m68k,
microblaze, moxie, or32, s390, superh, tricore, unicore, and xtensa.

> TCG is
> obviously much more fully tested -- but at first glance, it looks like
> sljit may produce much more optimized/efficient code.

Almost anything produces more optimized code than tinycc. That was never
the point of tinycc.

> I think it's a lot
> smaller than TCG, too, which I like.

If you say so...

> It may also require less modding to
> provide full C99 functionality.
> 
> Maybe we should encourage the QEMU guys to dump TCG and adopt sljit, eh? :D

I wouldn't, and they won't.

> Thoughts?

Have fun with your project. It doesn't sound like it has anything to do
with what I planned, but may wind up interesting in its own right.

> http://sljit.sourceforge.net/

Rob

 1430931110.0