Armv8 Registers

Before introducing operation instructions, let's have a brief look at the register names and data features in the Armv8 architecture.

Figure 1 64-bit register

Figure 2 128-bit register

Figure 3 Vector register

In the Armv8 architecture, registers are classified into 64-bit registers and 128-bit registers. For a 64-bit register, Xn and Wn are used to represent the 64 bits and lower 32 bits of the register, respectively, as shown in Figure 1.

A 128-bit register can be used as a scalar register. Qn, Dn, Sn, Hn, and Bn are used to represent the 128 bits, lower 64 bits, lower 32 bits, lower 16 bits, and lower 8 bits of the register, respectively, as shown in Figure 2. It can also be used as a vector register. Vn.2D, Vn.4S, Vn.8H, and Vn.16B indicate that the 128-bit register is divided into 64-bit, 32-bit, 16-bit, and 8-bit channels, respectively, as shown in Figure 3.

According to the preceding naming rules, the same instruction can be used to process data of different lengths. This is often applied to data truncation. For example, in the following code, eight bytes are loaded to registers x1 and x2 respectively. However, the logic only needs to compare the lower 32-bit data. In this case, the two registers are represented in the form of w, and the 32-bit data does not need to be reloaded to another register.

... ... 
 ldr         x3, [x0] 
 ldr         x4, [x1] 
 cmp      w3, w4  /* compare */ 
 ... ...

It should be noted that a prerequisite for data multiplexing is that the instruction takes effect on a register represented in this form. The Q register (128-bit scalar register) and the V register (128-bit vector register) are the same set of registers using two sets of instructions. That is, data is stored in the same 128-bit register. However, the scalar assembly instruction applies only to the 128-bit register represented in scalar form, and the vector assembly instruction (also called the neon instruction) applies only to the 128-bit register represented in vector form. The two types of instructions cannot be used together. Otherwise, the compiler reports an error.

Defining Registers

If we use digits to identify registers for a large amount of assembly code, the register values will be confusing. In addition, the code will be developer unfriendly. A better way is to define the register function by using define at the header of the code file as shown in the following code:

 #define data1        x3
#define data2        x4
 #define Lowdata1         w3
 #define Lowdata2         w4
 #define src1  x0
 #define src2  x1
 ... ... 
 ldr     data1, [src1] 
 ldr     data2, [src2] 
 /* compare */ 
 cmp  Lowdata1, Lowdata2 
 ... ...