Not that compilers don't already make these optimizations. But it's pretty hard to map sequential languages like C to a many-core processor with really large SI

Not that compilers don't already make these optimizations. But it's pretty hard to map sequential languages like C to a many-core processor with really large SIMD units. Any abstraction that tries to make code still look sequential even if it is being executed on many cores is going to be leaky.