Soham Chakraborty | Fundamentals and Advancements in Relaxed Memory Concurrency

Concurrent programs are notoriously difficult to write and analyze correctly, even for the experts. In most analysis techniques, the behavior of concurrent programs is understood in terms of thread interleavings, a model formally known as sequential consistency (SC). However, for performance reasons, modern hardware implementations allow some executions that cannot be explained by thread interleavings. Similarly, standard compiler optimizations can affect the outcome of a concurrent program in ways that cannot be understood by SC. These concurrency paradigms are known as relaxed memory concurrency. Thus, it is important to reason about concurrent programs under the lens of relaxed memory concurrency models. In this talk, I will introduce relaxed memory concurrency, and provide an overview of my present research, particularly on CPU-GPU heterogeneous shared memory concurrent systems, with future directions.