Call vs. JMPL in SPARC

Question

What is the difference between Call and JMPL in SPARC? I know that both instructions can be used to implement function calls.

I also want to know how the return mechanism gets affected depending on the choice of Call or JMPL. I know that if we use JMPL, we should use the RETURN instruction to return from a function call but I do not know the corresponding instruction for the Call instruction.

Answer 1

CALL only takes a relative address (a label) and automatically saves the current PC (later used to calculate the return address) into %o7. JMPL on the other hand only takes an absolute address in a register, and can store the current PC in an arbitrary register.

The function calling convention is the same in both cases. CALL is more convenient for labels and JMPL for function pointers. The latter can be used to emulate the former by loading the target address into a register first.

As for returning, both RET and RETL are synthetic instructions that will be turned into a JMPL by the assembler, using the correct return register: RET becomes JMPL %i7+8,%g0 and RETL becomes JMPL %o7+8,%g0. RET assumes you have set up a new register window (thus the original %o7 is now %i7) and you should follow it with a RESTORE instruction in the delay slot.

Update: The RETURN instruction has apparently been introduced in sparc v9, and it combines a JMPL (having second operand %g0) and a RESTORE (with no operands). As such it can be conveniently used to end a non-leaf function, where you would normally use a RET; RESTORE pair.