Changed mul_low algorithm to radix-16-1 algorithm with swap

src/imath.asm

@@ -4,29 +4,51 @@
 ; r2 - The second value
 ; Result:
 ; r1 - The lower 32 bits of the result
-; Clobbers: r2, r3, r4, r5
+; Clobbers: r2, r3, r4, r5, r6
 pub mul_low:
-	add r3, r2, r2 ; r3 = r2 * 2
+	cmp r1, r2
+	jbe mul_low_noswap
+	xor r1, r2, r1
+	xor r2, r1, r2
+	xor r1, r2, r1
+	; Passthrough
+
+; Multiplies r1 and r2, returning the lower part of the result
+; This method assumes r1 is smaller than r2, which results in faster execution
+; Arguments:
+; r1 - The first value
+; r2 - The second value
+; Result:
+; r1 - The lower 32 bits of the result
+; Clobbers: r2, r3, r4, r5, r6
+pub mul_low_noswap:
 	mov r4, 0 ; r4 has the result
+	mov r6, mul_loop_end
+	add r3, r2, r2
 	
-	mul_low_loop:
+	mul_loop:
+		and r5, r1, 14 ; Taking the first four bits, discarding the odd
+		lsl r5, r5, 1 ; 0000 - 0, 0001 - 2, 0010 - 4, 0011 - 8, etc.
+		sub r5, r6, r5
+		jmp r5
 		
-		and r5, r1, 1 ; a0
+		add r4, r4, r3
-		neg r5, r5 ; mask
+		add r4, r4, r3
-		and r5, r2, r5 ; a0 ? r2 : 0
+		add r4, r4, r3
-		add r4, r4, r5
+		add r4, r4, r3
-
+		add r4, r4, r3
-		and r5, r1, 2 ; a1 is now 0 or 2
+		add r4, r4, r3
-		lsr r5, r5, 1 ; normalize to 0/1
+		add r4, r4, r3
-		neg r5, r5 ; mask
+		mul_loop_end:
-		and r5, r3, r5 ; a1 ? r2 * 2 : 0
+			mov flags, r1
-		add r4, r4, r5
+			jne mul_skip_one
-
+			add r4, r4, r2
-		lsl r2, r2, 2
+			mul_skip_one:
-		lsl r3, r3, 2
+			lsl r2, r2, 4
-		lsr r1, r1, 2
+			lsl r3, r3, 4
+			lsr r1, r1, 4
 			cmp r1, zr
-		jne mul_low_loop
+			jne mul_loop
 
 	mov r1, r4
 	jmp r13