[LegNeato]

To be more technically correct, we compile to NVVM IR and then use NVIDIA's NVVM to convert it to PTX.