---

32.1.3 Unknown values return

The unknown values return convention is always used in full call, and is used in local call when the compiler either can’t prove that a fixed number of values are returned, or decides not to use the fixed values convention to allow tail-recursive XEP calls.

The unknown-values return convention has variants: single value and variable values. We make this distinction to optimize the important case of a returner who knows exactly one value is being returned. Note that it is possible to return a single value using the variable-values convention, but it is less efficient.

We indicate single-value return by returning at the return-pc+4; variable value return is indicated by returning at the return PC.

Single-value return makes only the following guarantees: A0 holds the value returned. CSP has been reset: there is no garbage on the stack.

In variable value return, more information is passed back: A0..A2 hold the first three return values. If fewer than three values are returned, then the unused registers are initialized to NIL.

OCFP points to the frame returned from. Note that because of our tail-recursive implementation of call, the frame receiving the values is always immediately under the frame returning the values. This means that we can use OCFP to index the values when we access them, and to restore CSP when we want to discard them.

NARGS holds the number of values returned.

CSP is always (+ OCFP (* NARGS 4)), i.e. there is room on the stack allocated for all returned values, even if they are all actually passed in registers.