Optimizing compilers and high performace programmers desire to make code run as fast as possible given certain time constraints. Computer architects desire to make computer systems run as fast as possible given certain monetary and power contraints. Optimization of architecture may in fact be the primary reason optimization of programs is required.

Is there a point at which program optimization becomes inordintately difficult or impossible? If so, which features of the architecture are to blame? Can we evolve architectures so as to remove these optimization bottlenecks?

In this talk, I will present three constraining architectural features. For each, I will describe the nature and magnitude of the constraints on performance and corresponding architectural evolutions.