Shared-Memory Parallelism Can Be Simple, Fast, and Scalable