Secure Data-Dependent Branching via Speculative Parallelism

We propose a mechanism to securely execute code containing data-dependent branches by speculatively executing both sides of the branch. These branches must be specially constructed, but we use this construction to induce performance. Attention is given to the on-chip infrastructure needed for these changes as well as the new mechanisms required in existing instrumentation and simulation software such as SMTSIM. Also provided are several example code segments where this technique could be applied. The increase in security and performance on these example segments is discussed and estimated. Finally, we explore the limitations in our work at the present point.