A deep dive into the internals of SpiderMonkey's parallel execution primitive `%ParallelBuildArray()`, examining its design flaws and proposing improvements. The post covers how the current intrinsic works — including the warmup phase, IonMonkey JIT compilation for parallel execution, and thread-safety hacks — then proposes replacing it with a more general `%ForkJoin()` primitive paired with `%UnsafeSetElement()`. A key improvement eliminates the warmup phase entirely by using sequential fallback iterations as implicit warmup, with the parallel attempt retried after each chunk. Additional enhancements like avoiding array initialization overhead and detecting inherently non-parallelizable functions are also sketched out.
