eliminating null pointer checks on mmu-less machines

Sun Sep 14 05:53:00 GMT 2003

On 13 Sep 2003, Tom Tromey wrote:
> How good a job does the optimizer do at eliminating redundant checks?

The RTL optimizers do pretty well recognizing redundant scalar checks, but
lose on COMPONENT_REF. This is already fixed on tree-ssa.
Gcj still needs to somehow tell the optimizers about methods that never
return null (_Jv_AllocObject etc.) before the null check elimination will
be very effective.
> If it doesn't do so well, then I wonder whether the VRP pass (which,
> as far as I know, still hasn't gone in) would help.

VRP ought to help with bounds checking, but null checks are currently
handled in a separate pass.
Jeff