Assembly Language for Beginners

.1. X86

Listing 2: Intel C++ 2011

_f1 PROC NEAR

mov ecx, DWORD PTR [4+esp] ; ecx = a

lea edx, DWORD PTR [ecx+ecx*8] ; edx = a*9

lea eax, DWORD PTR [edx+ecx*4] ; eax = a*9 + a*4 = a*13

ret

These two instructions performs faster than one IMUL.

MOVSB/MOVSW/MOVSD/MOVSQcopy byte/ 16-bit word/ 32-bit word/ 64-bit word from the address
which is in SI/ESI/RSI into the address which is in DI/EDI/RDI.

Togetherwith the REP prefix, it is to be repeatedin a loop, the count is to be storedin the CX/ECX/RCX

register: it works like memcpy() in C. If the block size is known to the compiler in the compile stage,

memcpy() is often inlined into a short code fragment using REP MOVSx, sometimes even as several

instructions.

The memcpy(EDI, ESI, 15) equivalent is:

; copy 15 bytes from ESI to EDI

CLD ; set direction toforward

MOV ECX, 3

REP MOVSD ; copy 12 bytes

MOVSW ; copy 2 more bytes

MOVSB ; copy remaining byte

( Supposedly, it works faster than copying 15 bytes using just one REP MOVSB).

MOVSXload with sign extension see also: (1.17.1 on page 201)

MOVZXload and clear all other bits see also: (1.17.1 on page 202)

MOVload value. this instruction name is misnomer, resulting in some confusion (data is not moved
but copied), in other architectures the same instructions is usually named “LOAD” and/or “STORE”
or something like that.

One important thing: if you set the low 16-bit part of a 32-bit register in 32-bit mode, the high 16

bits remains as they were. But if you modify the low 32-bit part of the register in 64-bit mode, the

high 32 bits of the register will be cleared.

Supposedly, it was done to simplify porting code to x86-64.

MULunsigned multiply.IMULoften used instead ofMUL, read more about it:2.2.1.

NEGnegation:op=−opSame asNOT op / ADD op, 1.

NOPNOP. Its opcode is 0x90, it is in fact theXCHG EAX,EAXidle instruction. This implies that x86 does
not have a dedicatedNOPinstruction (as in manyRISC). This book has at least one listing where
GDB shows NOP as 16-bit XCHG instruction:1.8.1 on page 48.

More examples of such operations: (.1.7 on page 1038).

NOP may be generated by the compiler for aligning labels on a 16-byte boundary. Another very

popular usage ofNOPis to replace manually (patch) some instruction like a conditional jump toNOP

in order to disable its execution.

NOTop1:op1 =¬op 1. logical inversion Important feature—the instruction doesn’t change flags.

ORlogical “or”

POPget a value from the stack:value=SS:[ESP]; ESP=ESP+4 (or 8)

PUSHpush a value into the stack:ESP=ESP-4 (or 8); SS:[ESP]=value

RETreturn from subroutine:POP tmp; JMP tmp.

In fact, RET is an assembly language macro, in Windows and *NIX environment it is translated into

RETN (“return near”) or, in MS-DOS times, where the memory was addressed differently (11.6 on

page 1003), into RETF (“return far”).

RETcan have an operand. Then it works like this:

POP tmp; ADD ESP op1; JMP tmp.RETwith an operand usually ends functions in thestdcallcalling

convention, see also:6.1.2 on page 734.

SAHFcopy bits from AH to CPU flags: