fix: resolve phi swap cycles in parallel copy lowering #140

[BibhabenduMukherjee]

Introduce a topological sort and cycle-breaking algorithm for phi move lowerings in both the textual assembly (Asm) and binary machine code (Rcu) generators. This resolves silent value corruption when two or more phi nodes form a circular swap dependency. Cycle-breaking is achieved by pushing the cycle-start value onto the machine stack and restoring it after walking the copy cycle.

Example of Failure

Consider a block transition that requires a register/stack slot swap between two variables a (located at [rbp - 8]) and b (located at [rbp - 16]):

Move 1: a = b (src: b, dst: a)
Move 2: b = a (src: a, dst: b)

Because the previous implementation of EmitPhiMoves iterated sequentially, it generated the following instructions:

; --- Move 1: a = b ---
mov rax, qword [rbp - 16]   ; Load b's value into rax
mov qword [rbp - 8], rax    ; Store rax into a's slot
                            ; [rbp - 8] (a) is now overwritten with b's value. 
                            ; The original value of a is permanently lost!

; --- Move 2: b = a ---
mov rax, qword [rbp - 8]    ; Load a's value (which is now b's value!) into rax
mov qword [rbp - 16], rax   ; Store rax into b's slot

Tests & Verification

We registered a new regression test package, Tests/PhiSwap, containing a loop-based Fibonacci accumulator swap which naturally generates parallel-copy cycles during lowering.

Test Source (Tests/PhiSwap/Src/Main.rux):

func FibLoop(n: int) -> int {
    var a: int = 0;
    var b: int = 1;
    var i: int = 0;
    while i < n {
        let t: int = a + b;
        a = b;   // triggers parallel-copy swap
        b = t;
        i = i + 1;
    }
    return a;
}

func Main() -> int {
    if FibLoop(0) != 0   { return 1; }
    if FibLoop(1) != 1   { return 2; }
    if FibLoop(5) != 5   { return 3; }
    if FibLoop(10) != 55  { return 4; }
    if FibLoop(15) != 610 { return 5; }
    return 0;
}

Verification Results:

CTest Suite: The new test was successfully registered under CMake and ran as part of CTest:

Start 10: PhiSwap
10/14 Test #10: PhiSwap ..........................   Passed    2.73 sec

Executable Exit Code: Running the compiled debug binary directly exits with status code 0, confirming that the Fibonacci numbers computed correctly without any variable corruption: