Re: Simple Value Range Propagation

[Jan Hubicka <jh at suse dot cz>]

> 	I would like GCC to have a VRP optimization, but I still do not
> understand why the proposals always record such detailed, accurate
> information, which requires a slow, heavy-weight implementation.  I do not
> see it in other production compilers.
Josef's VRP is much more simple minded that John's one.  It is
interested mostly in the tests so it is more similar to dataflow done by
usual bounds checking ellimination pass than fully weighted VRP pass.  I
think it can have potential to unify the jump bypassing done currently
in gcse and jump threading and thus I like it (even when my first
feelings about idea of another RTL VRP pass implementation were
oposite).

According to benchmarks I've seen it adds just 3 seconds to SPEC2000
compilation that well can be a noise.  With this performance it is
likely cheaper than the jump threading pass itself. At the moment it
does not obsolette it completely as it can not deal with != tests very
well, but it should not be too dificult to avoid.

Honza
> 
> David