[tree-ssa] patch suggestion: def_op vs vdef_ops

[Diego Novillo]

On Wed, 29 Jan 2003, Diego Novillo wrote:

> On Wed, 29 Jan 2003, Jose Renau wrote:
> 
> > On Wed, 29 Jan 2003 law@redhat.com wrote:
> > 
> > > If you wanted to run ccp, dce, etc etc on memory operands, then you'd
> > > iterate over both.
> > > 
> > > The decision would (of course) be controlled by a user switch :-)
> > 
> >  Do you mean that when "get_stmt_operands (stmt)" is called, it would have
> > an additional flag that would force to store in dep_op?
> > 
> >  Something like:
> > 
> >  get_stmt_operands (stmt, false); /* Current version */
> > 
> >  get_stmt_operands (stmt, true); /* The one sugested in the email */
> > 
> This future flag will just tell get_stmt_operands() to collect
> real or virtual operands.  It will not force it to return real
> operands as if they were virtual and vice-versa.
> 
Sorry.  I replied blindly.  The user switch that Jeff is
referring to has nothing to do with what I said.  I think Jeff is
talking about a potential switch to tell the compiler to do its
best trying to follow all the alias links.

I was referring to passes that are only interested in certain
types of operands.  Have a filter in get_stmt_operands() that
only look for the operands that the caller is interested in.

For instance, when CCP wants to replace uses, it's only interested
in unaliased operands on the RHS of the expression.

Instead, DCE always wants to know both real and virtual operands
to determine usefulness.


Diego.