Performance Guidance Despite Optimization Interaction

Improved application performance allows for computational analysis of bigger problems, saves money spent on computation resources, and handles more demanding multimedia applications. Having the compiler make all performance guidance solutions would be ideal. This allows performance to be dealt with separately from functionality, which is useful because it avoids having aspects of performance obfuscate the program's purpose. However, compilers do not su ciently handle the interaction of various performance optimizations which can be performed on programs. This is due not only to the fact that automatic/compiler optimizations interact, but also that program transformations (optimizations) occur at all stages of the application development process: compiler development, machine-speci c library development, portable library development, and speci c application development. We explore how current work guides performance optimization at the various stages of application development, and evaluate these techniques based on where the guidance occurs, applicability, and performance results.