Basic Branch Judgment

A branch structure is the basic, practical, and frequent module in a program, that is, it selects different program forward paths according to the judgment condition result to realize different function logics.

• B.cond series

  In the Arm64 assembly instructions, a CPSR register is called a program status register with four flags: N, Z, C, and V. The Arm assembly implements the function of the branch structure through the flags and judgment. The basic usage is as follows:

  CMP/CCMP/ADDS/SUBS/ANDS/TST… … // Instructions for rewriting the status register

  B.cond label // Condition status judgment. If the condition is met, go to the label address.

  B is a jump instruction and cond is a condition code suffix of the jump instruction. Different flags indicate different conditions that need to be met by the jump, as shown in Table 1. If B is not followed by the cond flag, it indicates that the system jumps to the label address unconditionally.

  Table 1 Status flags

  ID	mnemonic	Description	Mark
  0000	EQ	Equal to: 1	Z==1
  0001	NE	Not equal to: 0	Z==0
  0010	HS/CS	Unsigned higher or same carry. Carry: 1	C==1
  0011	LO/CC	Unsigned lower carry clear. Borrow: 0	C==0
  0100	MI	Negative: 1	N==1
  0101	PL	Positive or zero: 0	N==0
  0110	VS	Signed overflow: 1	V==1
  0111	VC	No signed overflow: 0	V==0
  1000	HI	Unsigned >	C==1 && Z==0
  1001	LS	Unsigned <=	!( C==1 && Z==0)
  1010	GE	Signed >=	N==V
  1011	LT	With sign <	N!=V
  1100	GT	With sign >	Z==0 && N==V
  1101	LE	With sign <=	!( Z==0 && N==V)
  1110	AL	Always executed	Any
  1111	NV

  The following table shows how to use the Arm64 assembly to replace the common branch structure of the C program implementation.

  Condition Type	C Program Implementation	Assembly Implementation
  Single-condition judgment	/* Assumptions: Cond A: a > b */ If (Cond A){ Part 1 } else { Part 2 } End	/* Assume the values of a and b are stored in registers x1 and x2. Note that value of b can be a variable loaded from its address or a constant. */ ldr x1, [a_addr] ldr x2, [b_addr] cmp x1, x2 b.le Part2 (Part1:) ... ... b End Part2: ... ... End: ... ...
  Dual-condition judgment	/* Assumptions: Cond A: a > b \ Cond B: a > c */ If (Cond A && Cond B){ Part 1 } else { Part 2 } End	/* Assume the values of a, b and c are stored in registers x1, x2 and x3. */ ldr x1, [a_addr] ldr x2, [b_addr] ldr x3, [c_addr] cmp x1, x2 ccmp x1, x3, 0, hi b.hi Part1 Part2: ... ... b End Part1: ... ... End

• Special judgment instructions

  B.cond can meet all judgment logics. However, sometimes, Arm64 provides some instructions for simple judgment and special judgment, as shown in the following table.

  Instruction	Example	Description
  CBZ	CBZ X1, label	If x1 == 0, go to label.
  CBNZ	CBNZ X1, label	If x1 != 0, go to label.
  TBZ	TBZ X1, #3, label	If the third bit of the x1 register is 0, go to label.
  TBNZ	TBNZ X1, #3, label	If the third bit of the x1 register is not 0, go to label.

  Let's look at the application scenarios of TBZ and TBNZ. The instruction function describes only whether a bit of a value is equal to 0. With the minimum unit of data stored by a computer, the relationship between the value and the 2ⁿ value can be further determined.

  C Program Implementation	Assembly Implementation
  /* No assumption */ if (x1 >= 64) { part1 } else if (x1 >= 32) { part2 } else if (x1 >= 16) { part3 } else { part 4 }	tbnz x1, #4, Part3 /* over16 */ ... ... b end Part3: tbnz x1, #5, Part2 /* over32 */ ... ... b end Part2: tbnz x1, #6, Part1 /* over64 */ ... ... b end Part1: ... ... end: ... ...
  /* Assumption: x1 < 128 */ if (x1 >= 64) { part1 } else if (x1 >= 32) { part2 } else if (x1 >= 16) { part3 } else { part 4 }	/* Ensure the x1 is less than 128, which means the bit value higher that 6th bit equals 0. */ cmp x1, 128 b.ge end tbz x1, #6, Part2 /* 64less */ ... ... b end Part2: tbz x1, #5, Part3 /* 32less */ ... ... b end Part3: tbz x1, #4, Part4 /* 16less */ ... ... b end Part4: ... ... end: ... ...

  As shown in the foregoing code, TBNZ may be used to continuously determine a branch whose entry condition value is greater than or equal to 2ⁿ, and TBZ may be used to continuously determine a branch whose entry condition value is less than 2ⁿ. It should be noted that, when TBZ determines that a bit is 0, it does not mean that the value is necessarily less than 2ⁿ, and there may be a 1 in a high-order bit, so that the entire value is greater. Therefore, when TBZ is used for judgment, the range of the condition value should be limited to be less than 2ⁿ⁺¹ first.

• csel: branch result selection instruction

  The branch structure is added because different processes need to be processed in different cases. Generally, a judgment instruction and a jump instruction are introduced, which may affect pipeline smoothness. However, when the process is relatively simple and ends by assigning values to some registers, the csel instruction may be considered to avoid branch jump. The simplest process is similar to a ternary operator, as shown in the following table.

  C Program Implementation	Assembly Implementation
  int a = 3; int b = 4; int c = (a > b) ? a : b;	/* Assume the values of a and b stored in addr_a and addr_b, and the values of c will be stored in addr_c*/ ... ... ldr vala, [addr_a] ldr valb, [addr_b] cmp vala, valb csel valc, vala, valb, ge str valc, [addr_c] ... ...