This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa]: Constant propagation into PHI's screws up IVR and PRE
On Wed, Jul 30, 2003 at 02:33:57PM -0400, Daniel Berlin wrote:
>
> >the evolution part. The initial condition is the only argument that
> >could contain a constant (after CCP),
>
> If you are positive about this, it's good enough for me.
Yes.
This even simplifies the work of the IV analyzer.
The IVA has to follow the branch out of the loop for computing the
initial condition. But in my opinion this is not to the IVA to approximate
the initial conditions, but this has to be computed by a classic constant
propagation slightly adapted to propagate intervals instead of just constants.
(call it CCPI for example).
If the initial condition is not computable, or hard to compute, then
the CCPI approximates it with the unknown element of the lattice ie. [-oo, +oo].