Would profile guided compilation works for identifying potential data dependencies? For instance, data dependency of a program can be gathered at runtime during the serial run. Then, based on the gathered profile information and data dependency information of the program, compiler can then generate parallelized code.

@gohan2021 This seems like a good strategy for programs with relatively few use cases/simple inputs, but for a complex piece of software with extremely broad parameters, profiling could be too cumbersome. That's likely why a parallelizing compiler for general-purpose code doesn't exist.