Re: [DotGNU]gcc?

[Rhys Weatherley]

Fergus Henderson wrote:

> > Is there a particular IL optimization that Mercury would
> > benefit from?
>
> There's a few that might help: [...]

Most of these are more sophisticated than the assembler
can deal with.  They need knowledge of the program's
parse tree or post-parse intermediate representation.

A separate optimization tool that rebuilds the tree and
optimises that would be a better approach.  Fortunately,
there's quite a bit of code hanging around should someone
want to tackle this.

The bytecode verifier walks the code, verifies it, and
then calls API methods on a "coder" object.  The current
coder object generates CVM bytecode, but it should be
possible to write a tree building coder instead.

>         - Elimination of branches around branches.
>         We sometimes generate code such as this:
>                 br label2
>                 label1:
>                 label2:
>         The `br label2' instruction can be optimized away.
>
>         - Jump threading (elimination of branches to branches).

These may be possible in the assembler.  The first is
really a variant of jump threading, where a jump to the
next instruction is replaced with nothing.

> Now, that said, I don't think any of this is a high priority right now.
> Obviously getting all it working first is much more important than
> worrying about optimizations.

Yep.  However, if some other brave soul wants to take
this on, they can start with the stand-alone verifier
"ilverify", and move forward from there.

Cheers,

Rhys.