Chapter 4

Exploring AArch64 assembler – Chapter 4

October 23, 2016 Roger Ferrer Ibáñez, 0

In this chapter we will see some instructions that will allow us to compute things.

Arithmetic instructions

Since a computer is nothing but a glorified calculator (or a calculator it is nothing but a tiny computer), it has to be possible to perform basic arithmetic. For now we will restrict to integer operations. In later chapters we will see how we can manipulate other kinds of numbers.

Addition and subtraction

We can perform addition and subtraction using add and sub instructions. These instructions are pretty flexible in that they allow using many forms.

add Rdest, Rsource1, #immediate          // Rdest ← Rsource1 + immediate
add Rdest, Rsource1, Rsource2            // Rdest ← Rsource1 + Rsource2
add Xdest, Xsource1, Xsource2, shiftop   // Xdest ← Xsource1 + shiftop(Xsource2)
add Xdest, Xsource1, Xsource2, extop     // Xdest ← Xsource1 + extop(Xsource2)
add Wdest, Wsource1, Wsource2, shiftop   // Wdest ← Wsource1 + shiftop(Wsource2)
add Wdest, Wsource1, Wsource2, extop     // Wdest ← Wsource1 + extop(Wsource2)
add Xdest, Xsource1, Wsource2, extop     // Xdest ← Xsource1 + extop(Wsource2)

In the forms above Rx means either Xx or Wx (but without mixing them in the same instruction), shiftop and extop are the shift and extension operands described in chapter 3. In this case, shiftop does not include ROR. All the forms shown for add can be used for sub as well.

Multiplication and division

Compared to addition and subtraction, multiplication and division are harder operations. And there are a few different instructions for this purpose.

Due to the nature of multiplication, multiplying two values of 32/64 bits may end requiring 64/128 bits to be able to fully encode the mathematical result. If we know this will not happen (i.e. the result of the value can be encoded in 32/64 bits) or we do not care we can use the mul instruction. If there are excess bits, they will be dropped.

mul Rdest, Rsource1, Rsource2 // Rdest ← Rsource1 * Rsource2
                              // but be careful with overflow

If we do care about the excess bits, then we have a bunch of instructions we can use. For 32-bit multiplications, umull and smull can be used. The latter is required when multiplying numbers in two’s complement so the sign bit is correctly handled.

umull Xdest, Wsource1, Wsource2  // Xdest ← Wsource1 * Wsource2
smull Xdest, Wsource1, Wsource2  // Xdest ← Wsource1 * Wsource2 
                                 // for two's complement numbers
                                 // in Wsource1 and Wsource2

For the less common case where we multiply two 64-bit registers and still we care about the upper 64-bits of the 128-bit result, then we can use umulh and smulh. For this case we will need two 64-bit registers (named Xlower and Xupper in the example below).

mul Xlower, Xsource1, Xsource2   // Xlower ← Lower64Bits(Xsource1 * Xsource2)
smulh Xupper, Xsource1, Xsource2 // Xupper ← Upper64Bits(Xsource1 * Xsource2)

Division is a bit simpler as only two instructions are necessary: udiv and sdiv. Again, the latter is for integer numbers encoded in two’s complement.

udiv Rdest, Rsource1, Rsource2  // Rdest ← Rsource1 / Rsource2
sdiv Rdest, Rsource1, Rsource2  // Rdest ← Rsource1 / Rsource2
                                // when Rsource1, Rsource2 are
                                // in two's complement

These two instructions will compute the quotient of the division rounded towards zero.

Bitwise instructions

Bitwise instructions directly operate in the bits of the registers, without assuming any encoding in them.

Instruction mvn performs a bitwise not on its operand.

mvn Rdest, Rsource    // Rdest ← ~Rsource

Most of the bitwise instructions use two source registers. The basic ones are and, orr and eorr (exclusive orr). They perform a bitwise and, a bitwise or and a bitwise xor respectively.

and Rdest, Rsource1, #immediate          // Rdest ← Rsource1 & immediate
and Rdest, Rsource1, Rsource2            // Rdest ← Rsource1 & Rsource2
and Xdest, Xsource1, Xsource2, shiftop   // Xdest ← Xsource1 & shiftop(Xsource2)
and Wdest, Wsource1, Wsource2, shiftop   // Wdest ← Wsource1 & shiftop(Wsource2)

Similar forms as the shown above exist for orr and eorr. For bitwise instructions shiftop does include ROR.

There are combined versions of mvn plus and, orr and eor called bic (bit clear), orn (or not) and eon (exclusive or not) respectively. In this case, the second operand is first applied a NOT operation. They have slightly more limited forms.

orn Rdest, Rsource1, Rsource2            // Rdest ← Rsource1 | ~Rsource2
orn Xdest, Xsource1, Xsource2, shiftop   // Xdest ← Xsource1 | ~shiftop(Xsource2)
orn Wdest, Wsource1, Wsource2, shiftop   // Wdest ← Wsource1 | ~shiftop(Wsource2)