The quick answer is "both", but the stoichiometry is not 1-to-1. The stoichiometry of the calvin cycle doesn't add up when thinking about a single molecules making one "turn" of the cycle. There is only one carbon (CO2) added, but three carbons (in PGAL) leave, which obviously cannot work.
What's happening is that multiple molecules are processed at the same time. The cycle is of course just a set of enzymes that continually catalyze their reactions, so there are always lots of molecules available for enzymes act on, and the cycle doesn't "get stuck" just because things don't add up in a single turn. The "regenerating" steps between PGAL and RuBP produce 3 molecules of RuBP from 5 molecules of PGAL (15 carbons total). Carboxylating these 3 RuBP molecules adds 3 carbons, so that one PGAL molecule can leave the cycle.
The reason for this odd stoichiometry is that the cycle needs to turn 3-carbon sugars (PGAL) into 5-carbon sugars (RuBP), which is rather complicated. Several enzymes work together in a complicated series of reactions to achieve this. A good resource for learning about the details is the MetaCyc database, where you can explore all the molecular details of the pathway.