Registers have no leading `%' sign, unlike gas, and floating-point stack registers are referred to as st0, st1, and so on.
Floating-point instructions may use either the single-operand form or the double. A TO keyword is provided; thus, one could either write
or one could use the alternative single-operand forms
fadd to st1
Uninitialised storage is reserved using the RESB, RESW, RESD, RESQ and REST pseudo-opcodes, each taking one parameter which gives the number of bytes, words, doublewords, quadwords or ten-byte words to reserve.
Repetition of data items is not done by the DUP keyword as seen in DOS assemblers, but by the use of the TIMES prefix, like this:
message: times 3 db 'abc'
times 64-$+message db 0
which defines the string `abcabcabc', followed by the right number of zero bytes to make the total length up to 64 bytes.
Symbol references are always understood to be immediate (i.e. the address of the symbol), unless square brackets are used, in which case the contents of the memory location are used. Thus:
loads AX with the address of the variable `wordvar', whereas
all refer to the contents of memory locations. The syntaxes
es mov ax,wordvar
are not legal at all, although the use of a segment register name as an instruction prefix is valid, and can be used with instructions such as LODSB which can't be overridden any other way.
Constants may be expressed numerically in most formats: a trailing H, Q or B denotes hex, octal or binary respectively, and a leading `0x' or `$' denotes hex as well. Leading zeros are not treated specially at all. Character constants may be enclosed in single or double quotes; there is no escape character. The ordering is little-endian (reversed), so that the character constant 'abcd' denotes 0x64636261 and not 0x61626364.
Local labels begin with a period, and their `locality' is granted by the assembler prepending the name of the previous non-local symbol. Thus declaring a label `.loop' after a label `label' has actually defined a symbol called `label.loop'.
ABSOLUTE address causes nasm to position its notional assembly point at an absolute address: so no code or data may be generated, but you can use RESB, RESW and RESD to move the assembly point further on, and you can define labels. So this directive may be used to define data structures. When you have finished doing absolute assembly, you must issue another SECTION directive to return to normal assembly.
BITS 16 or BITS 32 switches the default processor mode for which nasm is generating code: it is equivalent to USE16 or USE32 in DOS assemblers.
EXTERN symbol and GLOBAL symbol import and export symbol definitions, respectively, from and to other modules. Note that the GLOBAL directive must appear before the definition of the symbol it refers to.
STRUC strucname and ENDSTRUC, when used to bracket a number of RESB, RESW or similar instructions, define a data structure. In addition to defining the offsets of the structure members, the construct also defines a symbol for the size of the structure, which is simply the structure name with _size tacked on to the end.
GROUP grpname seg1 seg2... is used by the obj (Microsoft 16-bit) output format, and defines segment groups. This format also uses UPPERCASE, which directs that all segment, group and symbol names output to the object file should be in uppercase. Note that the actual assembly is still case sensitive.
Multi-line macros are defined using %macro and %imacro (the distinction is the same as that between %define and %idefine), whose syntax is as follows:
%macro name minprm[-maxprm][+][.nolist] [defaults]
<some lines of macro expansion text>
Again, these macros may be overloaded. The trailing plus sign indicates that any parameters after the last one get subsumed, with their separating commas, into the last parameter. The defaults part can be used to specify defaults for unspecified macro parameters after minparam. %endm is a valid synonym for %endmacro.
To refer to the macro parameters within a macro expansion, you use %1, %2 and so on. You can also enforce that a macro parameter should contain a condition code by using %+1, and you can invert the condition code by using %-1. You can also define a label specific to a macro invocation by prefixing it with a double % sign.
Files can be included using the %include directive, which works like C.
The preprocessor has a `context stack', which may be used by one macro to store information that a later one will retrieve. You can push a context on the stack using %push, remove one using %pop, and change the name of the top context (without disturbing any associated definitions) using %repl. Labels and %define macros specific to the top context may be defined by prefixing their names with %$, and things specific to the next context down with %$$, and so on.
Conditional assembly is done by means of %ifdef, %ifndef, %else and %endif as in C. (Except that %ifdef can accept several putative macro names, and will evaluate TRUE if any of them is defined.) In addition, the directives %ifctx and %ifnctx can be used to condition on the name of the top context on the context stack. The obvious set of `else-if' directives, %elifdef, %elifndef, %elifctx and %elifnctx are also supported.