SIMPLE: A language-independent tree IR

[Diego Novillo]

On Wed, 02 Jan 2002, Per Bothner wrote:

> Richard Henderson wrote:
> 
> > I don't care.  For the purposes of the language-independant
> > tree nodes, expressions shall have no side effects.  Transfer
> > of control is a side effect, and thus shall be a statement.
> 
> Thus you would also remove COND_EXPR and PREINCREMENT_EXPR and
> CALL_EXPR.
> 
Yes to the first two.  COND_EXPR is turned into its equivalent
if-then-else.  PREINCREMENT_EXPR is explicitly expanded.

No to the last one.  We will model function calls.  Except that a
CALL_EXPR will have all its parameters simplified to be simple
variables or constants.

> Perhaps we are talking about different "levels"?  If we're
>
Yes.

> a radical break with Gcc current practice.  If you're talking
> about some lower-level intermediate format (which could of
> course use the same 'tree' types), then I have no objection.
>
We are talking about lower-level trees with no implicit
semantics.

Diego.