[SOLVED] How to handle function calls in a control flow graph with basic blocks

How to handle function calls in a control flow graph with basic blocks

I have been building the middle end of a compiler and I was wondering how I should handle function calls inside the flow graphs.

So as an example lets just look at the following three address code

LABEL label0
a = 1
b = 2
c = a + b
IF c == g GOTO label1 ELSE GOTO label2
LABEL label1
d = c + 0
CALL func0 (d -> y)
GOTO label0
LABEL label2
e = 8
f = 600
g = e + f
CALL func0 (a -> y)
CALL func1
END
FUNC LABEL func0
PRINT y
RETURN
FUNC LABEL func1
result = "PROGRAM IS COMPLETE"
PRINT result
RETURN

The first step is to break this into basic blocks like the following

block1:
LABEL label0
a = 1
b = 2
c = a + b
IF c == g GOTO label1 ELSE GOTO label2

block2:
LABEL label1
d = c + 0
CALL func0 (d -> y)

block3:
GOTO label0

block4:
LABEL label2
e = 8
f = 600
g = e + f
CALL func0 (a -> y)

block5:
CALL func1

block6:
END

block7:
FUNC LABEL func0
PRINT y
RETURN

block8:
FUNC LABEL func1
result = "PROGRAM IS COMPLETE"
PRINT result
RETURN

Here I assumed function calls should be treated as Jumps and therefore they should force a new block to start just like the GOTO or the if-jump statement does.

Afterwards I nest the basic block structure inside of my flow graph structure. And in that structure, I define all the connections. Like block1 should be connected to block2 and block4. Also block3 should connect to block1. I also connect block5 to block8 because of the Call. I connect block4 to block7 and block2 to block7 because of their respective calls. For the return jumps I connect block7 to block5 and block3 and I connect block8 to block6.

I guess my question is am I on the right track by integrating function calls and returns into the flow graph and treating them as jumps or am I supposed to ignore CALL statements when breaking up the instructions into basic blocks and just process the function body blocks as their own sub flow graphs.

Solution

For languages without exceptions calls are usually not treated as jumps when you construct a control flow graph. From the callees perspective a function call is an atomic single-entry, single-exit regions as any other instruction. The call itself can not force a change in control flow.