Title: Correct and Efficient Run-time Parallelization (aka Australia and New Zealand are Awesome) Abstract: Parallelization of existing serial applications is important to improve performance on current multicore architectures. For applications such as partial differential equation solvers over irregular meshes, the determination of which iterations in a loop can be executed in parallel happens at runtime. Current solutions come in the form of hand-written inspector/executor run-time library functions with some research compilers inserting calls to such functions into code. The correctness of current solutions depend on many assumptions such as correct data dependence analysis and correct run-time library routines. The efficiency depends on hand tuning the code using a number of sparse matrix data structure manipulation tricks. In this talk, we will present preliminary work on abstractions for specifying run-time parallelization inspectors and show examples of how these abstractions enable formal proofs of correctness AND automation of the efficiency tricks. ——— Since Michelle just got back from a nine month sabbatical at Australia National University, she will also be showing some fun little videos and pictures of Australia and New Zealand. The experience has been amazing!