Fibonacci contest winner

Fibonacci code

ULONG __declspec(naked) CalcFibMT5 (int dwFibNum)
{
	// dwFibNum is passed at esp-4
	// Result must be returned in eax
	__asm 
	{
		// start pushing working registers ecx, ebx. eax is clobbered
		push ecx
		mov ecx, round_finish

		// retrieve dwFibNum into ecx from stack
		mov eax, esp
		add eax, 8		// 4 plus a ULONG push
		mov eax, [eax]	// this is dwFibNum

		// work registers: ebx, eax. Final result always ends up in eax
		push ebx

		// register-register ADD instruction is 2 bytes 
		shl eax, 1		// each round is 2 bytes, so NOW this is how many instructions we want to execute
		sub ecx, eax	// walk backward as many instructions as we need

		// push our initial state for fib(1) and fib(2)
		mov ebx, 1		// load with initial state (eax = 1)
		mov eax, 1		// our initial answer

		jmp ecx			// and jump to our computation

// round_start:

		// 48
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 46
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 44
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 42
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 40
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 38
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 36
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 34
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 32
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 30
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 28
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 26
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 24
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 22
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 20
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 18
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 16
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 14
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 12
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 10
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 8
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 6
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax

		// 4
		add ebx, eax	// result of 1st round stored in ebx
		add eax, ebx	// result of 2nd round stored in eax


// oh, what a tangled web we weave
// when first we practice to deceive

// this is NEVER executed as 4 noops
// it is mov ecx, 0x90908bff
// or mov esi, esi
// depending on where you jump into it

_emit 0xb9
_emit 0x90
_emit 0x90
_emit 0x8b

round_finish:
_emit 0xf6

		// the 'straightforward' version
		// 2
//		nop
//		nop

		// 1
//		nop
//		nop


		// result in eax
		pop ebx
		pop ecx
		ret 4
	}

}