Difference between .equ and .word in ARM Assembly?

Question

I am curious - What is the difference between .equ and .word directives in ARM assembly, when defining constants?

Answer 1

.equ is like #define in C:

#define bob 10
.equ bob, 10

.word is like unsigned int in C:

unsigned int ted;
ted: 
.word 0

Or initialized with a value:

unsigned int alice = 42;
alice:
.word 42

Answer 2

.word is a directive that allocates a word-sized amount of storage space (memory) in that location. It can additionally have that location initialized with a given value.

.equ is more like a C preprocessor #define statement - it gets substituted in any subsequent code.

https://sourceware.org/binutils/docs-2.24/as/Equ.html#Equ

https://sourceware.org/binutils/docs-2.24/as/Word.html#Word

This is not actually ARM-specific, but applies to all gas targets.

Answer 3

As mentioned in the accepted answer (written by old_timer) label: .word value is like assigning a value to that label. I just want to add that you can assign multiple values to the same label just like an array, as the following:

g_pfnVectors:
  .word  _estack
  .word  Reset_Handler
  .word  NMI_Handler
  .word  HardFault_Handler
  .word  MemManage_Handler
  ...

The previous example has been taken from a STM32 MCU official startup file. This is exactly the machanism to initialize the NVIC.
So g_pfnVectors label got multiple values assigned (as an array, where the values are aligned next to each other in the memory so to say).

Answer 4

NASM 2.10.09 ELF output:

• .word is simple: it outputs 2 bytes to the object file no matter where we are.

  Consequences of this:

  • if .word is after a symbol x:, x will point to those bytes
  • if .word is in the text segment, those bytes might get executed

  It has absolutely no other side effect. In particular, it does not set the st_size field of the symbol table entry (e.g. int often == 4 bytes), which is something sensible compilers should do. You need the .size x, 2 directive for that.

• .equ does two things:

  1. updates a macro-like variable
  2. the last time you call it, it generates a symbol with st_shndx == SHN_ABS and the given value

  Sample code:

  .text
  
      .equ x, 123
      mov $x, %eax
      /* eax == 123 */
  
      .equ x, 456
      mov $x, %eax
      /* eax == 456 */
  

  Now:

  as --32 -o main.o main.S
  objdump -Sr main.o
  

  Gives:

  00000000 <.text>:
    0:   b8 7b 00 00 00          mov    $0x7b,%eax
    5:   b8 c8 01 00 00          mov    $0x1c8,%eax
  

  which confirms the macro-like effect, and:

  readelf -s main.o
  

  contains:

  Num:    Value  Size Type    Bind   Vis      Ndx Name
    4: 000001c8     0 NOTYPE  LOCAL  DEFAULT  ABS x
  

  which confirms the SHN_ABS effect: a symbol was created, and it could be used from another file by linking if it were global. I have explained this in more detail at https://stackoverflow.com/a/33148242/895245

  The situation is analogous for NASM's equ, except that the NASM version can only be used once per symbol.

  .set and the equals sign = (source) are the same as .equ.

  You should also look into .equiv https://sourceware.org/binutils/docs-2.25/as/Equiv.html , which prevents redefinition.