Documentation indicates that if two warps are attempting to execute say a floating point instruction and a global memory instruction, both can execute in parallel. Does this hold true when one warp is executing a floating point instruction and another warp is executing a shared memory load?