This work by Girlin provides a working implementation of a simple ALU scheduler and register allocator, while not regressing Piglit. Right now this ALU scheduler and register allocator is just for Evergreen hardware, but he plans to make it work with other generations of Radeon hardware too. Vadim also plans to implement some earlier optimization passes.
See more in his mailing list message. The R600g shader optimization work is currently found in this GitHub repository.