An assembly (or assembler) language, often abbreviated asm, is a low-level programming language for a computer, or other programmable device, in which there is a very strong (but often not one-to-one) correspondence between the language and the architecture’s machine code instructions. Each assembly language is specific to a particular computer architecture. In contrast, most high-level programming languages are generally portable across multiple architectures but require interpreting or compiling. Assembly language may also be called symbolic machine code.
Assembly language is converted into executable machine code by a utility program referred to as an assembler. The conversion process is referred to as assembly, or assembling the source code. Assembly time is the computational step where an assembler is run.
Assembly language uses a mnemonic to represent each low-level machine instruction or opcode, typically also each architectural register, flag, etc. Many operations require one or more operands in order to form a complete instruction and most assemblers can take expressions of numbers and named constants as well as registers and labels as operands, freeing the programmer from tedious repetitive calculations. Depending on the architecture, these elements may also be combined for specific instructions or addressing modes using offsets or other data as well as fixed addresses. Many assemblers offer additional mechanisms to facilitate program development, to control the assembly process, and to aid debugging.
- A macro assembler includes a macroinstruction facility so that (parameterized) assembly language text can be represented by a name, and that name can be used to insert the expanded text into other code.
- A cross assembler (see also cross compiler) is an assembler that is run on a computer or operating system (the host system) of a different type from the system on which the resulting code is to run (the target system). Cross-assembling facilitates the development of programs for systems that do not have the resources to support software development, such as an embedded system. In such a case, the resulting object code must be transferred to the target system, either via read-only memory (ROM, EPROM, etc.) or a data link using an exact bit-by-bit copy of the object code or a text-based representation of that code, such as Motorola S-record or Intel HEX.
- A high-level assembler is a program that provides language abstractions more often associated with high-level languages, such as advanced control structures (IF/THEN/ELSE, DO CASE, etc.) and high-level abstract data types, including structures/records, unions, classes, and sets.
- A microassembler is a program that helps prepare a microprogram, called firmware, to control the low level operation of a computer.
- A meta-assembler is a term used in some circles for “a program that accepts the syntactic and semantic description of an assembly language, and generates an assembler for that language.
An assembler program creates object code by translating combinations of mnemonics and syntax for operations and addressing modes into their numerical equivalents. This representation typically includes an operation code (“opcode”) as well as other control bits and data. The assembler also calculates constant expressions and resolves symbolic names for memory locations and other entities. The use of symbolic references is a key feature of assemblers, saving tedious calculations and manual address updates after program modifications. Most assemblers also include macro facilities for performing textual substitution – e.g., to generate common short sequences of instructions as inline, instead of called subroutines.
Some assemblers may also be able to perform some simple types of instruction set-specific optimizations. One concrete example of this may be the ubiquitous x86 assemblers from various vendors. Most of them are able to perform jump-instruction replacements (long jumps replaced by short or relative jumps) in any number of passes, on request. Others may even do simple rearrangement or insertion of instructions, such as some assemblers for RISC architectures that can help optimize a sensible instruction scheduling to exploit the CPU pipeline as efficiently as possible.
Like early programming languages such as Fortran, Algol, Cobol and Lisp, assemblers have been available since the 1950s and the first generations of text based computer interfaces. However, assemblers came first as they are far simpler to write than compilers for high-level languages. This is because each mnemonic along with the addressing modes and operands of an instruction translates rather directly into the numeric representations of that particular instruction, without much context or analysis. There have also been several classes of translators and semi automatic code generators with properties similar to both assembly and high level languages, with speedcode as perhaps one of the better known examples.
There may be several assemblers with different syntax for a particular CPU or instruction set architecture. For instance, an instruction to add memory data to a register in a x86-family processor might be add eax,[ebx], in original Intel syntax, whereas this would be written addl (%ebx),%eax in the AT&T syntax used by the GNU Assembler. Despite different appearances, different syntactic forms generally generate the same numeric machine code, see further below. A single assembler may also have different modes in order to support variations in syntactic forms as well as their exact semantic interpretations (such as FASM-syntax, TASM-syntax, ideal mode etc., in the special case of x86 assembly programming).
Number of passes
There are two types of assemblers based on how many passes through the source are needed (how many times the assembler reads the source) to produce the object file.
- One-pass assemblers go through the source code once. Any symbol used before it is defined will require “errata” at the end of the object code (or, at least, no earlier than the point where the symbol is defined) telling the linker or the loader to “go back” and overwrite a placeholder which had been left where the as yet undefined symbol was used.
- Multi-pass assemblers create a table with all symbols and their values in the first passes, then use the table in later passes to generate code.
In both cases, the assembler must be able to determine the size of each instruction on the initial passes in order to calculate the addresses of subsequent symbols. This means that if the size of an operation referring to an operand defined later depends on the type or distance of the operand, the assembler will make a pessimistic estimate when first encountering the operation, and if necessary pad it with one or more “no-operation” instructions in a later pass or the errata. In an assembler with peephole optimization, addresses may be recalculated between passes to allow replacing pessimistic code with code tailored to the exact distance from the target.
The original reason for the use of one-pass assemblers was speed of assembly – often a second pass would require rewinding and rereading the program source on tape or rereading a deck of cards or punched paper tape. Later computers with much larger memories (especially disc storage), had the space to perform all necessary processing without such re-reading. The advantage of the multi-pass assembler is that the absence of errata makes the linking process (or the program load if the assembler directly produces executable code) faster.
Example: in the following code snippet a one-pass assembler would be able to determine the address of the backward reference BKWD when assembling statement S2, but would not be able to determine the address of the forward reference FWD when assembling the branch statement S1; indeed FWD may be undefined. A two-pass assembler would determine both addresses in pass 1, so they would be known when generating code in pass 2,
S1 B FWD
FWD EQU *
BKWD EQU *
S2 B BKWD
More sophisticated high-level assemblers provide language abstractions such as:
- High-level procedure/function declarations and invocations
- Advanced control structures
- High-level abstract data types, including structures/records, unions, classes, and sets
- Sophisticated macro processing (although available on ordinary assemblers since the late 1950s for IBM 700 series and since the 1960s for IBM/360, amongst other machines)
- Object-oriented programming features such as classes, objects, abstraction, polymorphism, and inheritance
See Language design below for more details.
A program written in assembly language consists of a series of mnemonic processor instructions and meta-statements (known variously as directives, pseudo-instructions and pseudo-ops), comments and data. Assembly language instructions usually consist of an opcode mnemonic followed by a list of data, arguments or parameters. These are translated by an assembler into machine language instructions that can be loaded into memory and executed.
For example, the instruction below tells an x86/IA-32 processor to move an immediate 8-bit value into a register. The binary code for this instruction is 10110 followed by a 3-bit identifier for which register to use. The identifier for the AL register is 000, so the following machine code loads the AL register with the data 01100001.
This binary computer code can be made more human-readable by expressing it in hexadecimal as follows.
Here, B0 means ‘Move a copy of the following value into AL’, and 61 is a hexadecimal representation of the value 01100001, which is 97 in decimal. Assembly language for the 8086 family provides the mnemonic MOV (an abbreviation of move) for instructions such as this, so the machine code above can be written as follows in assembly language, complete with an explanatory comment if required, after the semicolon. This is much easier to read and to remember.
MOV AL, 61h ; Load AL with 97 decimal (61 hex)
In some assembly languages the same mnemonic such as MOV may be used for a family of related instructions for loading, copying and moving data, whether these are immediate values, values in registers, or memory locations pointed to by values in registers. Other assemblers may use separate opcode mnemonics such as L for “move memory to register”, ST for “move register to memory”, LR for “move register to register”, MVI for “move immediate operand to memory”, etc.
The x86 opcode 10110000 (B0) copies an 8-bit value into the AL register, while 10110001 (B1) moves it into CL and 10110010 (B2) does so into DL. Assembly language examples for these follow.
MOV AL, 1h ; Load AL with immediate value 1
MOV CL, 2h ; Load CL with immediate value 2
MOV DL, 3h ; Load DL with immediate value 3
The syntax of MOV can also be more complex as the following examples show.
MOV EAX, [EBX] ; Move the 4 bytes in memory at the address contained in EBX into EAX
MOV [ESI+EAX], CL ; Move the contents of CL into the byte at address ESI+EAX
In each case, the MOV mnemonic is translated directly into an opcode in the ranges 88-8E, A0-A3, B0-B8, C6 or C7 by an assembler, and the programmer does not have to know or remember which.
Transforming assembly language into machine code is the job of an assembler, and the reverse can at least partially be achieved by a disassembler. Unlike high-level languages, there is a one-to-one correspondence between many simple assembly statements and machine language instructions. However, in some cases, an assembler may provide pseudoinstructions (essentially macros) which expand into several machine language instructions to provide commonly needed functionality. For example, for a machine that lacks a “branch if greater or equal” instruction, an assembler may provide a pseudoinstruction that expands to the machine’s “set if less than” and “branch if zero (on the result of the set instruction)”. Most full-featured assemblers also provide a rich macro language (discussed below) which is used by vendors and programmers to generate more complex code and data sequences.
Each computer architecture has its own machine language. Computers differ in the number and type of operations they support, in the different sizes and numbers of registers, and in the representations of data in storage. While most general-purpose computers are able to carry out essentially the same functionality, the ways they do so differ; the corresponding assembly languages reflect these differences.
Multiple sets of mnemonics or assembly-language syntax may exist for a single instruction set, typically instantiated in different assembler programs. In these cases, the most popular one is usually that supplied by the manufacturer and used in its documentation.
There is a large degree of diversity in the way the authors of assemblers categorize statements and in the nomenclature that they use. In particular, some describe anything other than a machine mnemonic or extended mnemonic as a pseudo-operation (pseudo-op). A typical assembly language consists of 3 types of instruction statements that are used to define program operations:
- Opcode mnemonics
- Data definitions
- Assembly directives
Opcode mnemonics and extended mnemonics
Instructions (statements) in assembly language are generally very simple, unlike those in high-level languages. Generally, a mnemonic is a symbolic name for a single executable machine language instruction (an opcode), and there is at least one opcode mnemonic defined for each machine language instruction. Each instruction typically consists of an operation or opcode plus zero or more operands. Most instructions refer to a single value, or a pair of values. Operands can be immediate (value coded in the instruction itself), registers specified in the instruction or implied, or the addresses of data located elsewhere in storage. This is determined by the underlying processor architecture: the assembler merely reflects how this architecture works. Extended mnemonics are often used to specify a combination of an opcode with a specific operand, e.g., the System/360 assemblers use B as an extended mnemonic for BC with a mask of 15 and NOP (“NO OPeration” – do nothing for one step) for BC with a mask of 0.
Extended mnemonics are often used to support specialized uses of instructions, often for purposes not obvious from the instruction name. For example, many CPU’s do not have an explicit NOP instruction, but do have instructions that can be used for the purpose. In 8086 CPUs the instruction xchg ax,ax is used for nop, with nop being a pseudo-opcode to encode the instruction xchg ax,ax. Some disassemblers recognize this and will decode the xchg ax,ax instruction as nop. Similarly, IBM assemblers for System/360 and System/370 use the extended mnemonics NOP and NOPR for BC and BCR with zero masks. For the SPARC architecture, these are known as synthetic instructions.
Some assemblers also support simple built-in macro-instructions that generate two or more machine instructions. For instance, with some Z80 assemblers the instruction ld hl,bc is recognized to generate ld l,c followed by ld h,b. These are sometimes known as pseudo-opcodes.
Mnemonics are arbitrary symbols; in 1985 the IEEE published Standard 694 for a uniform set of mnemonics to be used by all assemblers. The standard has since been withdrawn.
There are instructions used to define data elements to hold data and variables. They define the type of data, the length and the alignment of data. These instructions can also define whether the data is available to outside programs (programs assembled separately) or only to the program in which the data section is defined. Some assemblers classify these as pseudo-ops.
Assembly directives, also called pseudo-opcodes, pseudo-operations or pseudo-ops, are commands given to an assembler “directing it to perform operations other than assembling instructions.” Directives affect how the assembler operates and “may affect the object code, the symbol table, the listing file, and the values of internal assembler parameters.” Sometimes the term pseudo-opcode is reserved for directives that generate object code, such as those that generate data.
The names of pseudo-ops often start with a dot to distinguish them from machine instructions. Pseudo-ops can make the assembly of the program dependent on parameters input by a programmer, so that one program can be assembled different ways, perhaps for different applications. Or, a pseudo-op can be used to manipulate presentation of a program to make it easier to read and maintain. Another common use of pseudo-ops is to reserve storage areas for run-time data and optionally initialize their contents to known values.
Symbolic assemblers let programmers associate arbitrary names (labels or symbols) with memory locations and various constants. Usually, every constant and variable is given a name so instructions can reference those locations by name, thus promoting self-documenting code. In executable code, the name of each subroutine is associated with its entry point, so any calls to a subroutine can use its name. Inside subroutines, GOTO destinations are given labels. Some assemblers support local symbols which are lexically distinct from normal symbols (e.g., the use of “10$” as a GOTO destination).
Some assemblers, such as NASM, provide flexible symbol management, letting programmers manage different namespaces, automatically calculate offsets within data structures, and assign labels that refer to literal values or the result of simple computations performed by the assembler. Labels can also be used to initialize constants and variables with relocatable addresses.
Assembly languages, like most other computer languages, allow comments to be added to program source code that will be ignored during assembly. Judicious commenting is essential in assembly language programs, as the meaning and purpose of a sequence of binary machine instructions can be difficult to determine. The “raw” (uncommented) assembly language generated by compilers or disassemblers is quite difficult to read when changes must be made.
Many assemblers support predefined macros, and others support programmer-defined (and repeatedly re-definable) macros involving sequences of text lines in which variables and constants are embedded. This sequence of text lines may include opcodes or directives. Once a macro has been defined its name may be used in place of a mnemonic. When the assembler processes such a statement, it replaces the statement with the text lines associated with that macro, then processes them as if they existed in the source code file (including, in some assemblers, expansion of any macros existing in the replacement text). Macros in this sense date to IBM autocoders of the 1950s.
In assembly language, the term “macro” represents a more comprehensive concept than it does in some other contexts, such as in the C programming language, where its #define directive typically is used to create short single line macros. Assembler macro instructions, like macros in PL/I and some other languages, can be lengthy “programs” by themselves, executed by interpretation by the assembler during assembly.
Since macros can have ‘short’ names but expand to several or indeed many lines of code, they can be used to make assembly language programs appear to be far shorter, requiring fewer lines of source code, as with higher level languages. They can also be used to add higher levels of structure to assembly programs, optionally introduce embedded debugging code via parameters and other similar features.
Macro assemblers often allow macros to take parameters. Some assemblers include quite sophisticated macro languages, incorporating such high-level language elements as optional parameters, symbolic variables, conditionals, string manipulation, and arithmetic operations, all usable during the execution of a given macro, and allowing macros to save context or exchange information. Thus a macro might generate numerous assembly language instructions or data definitions, based on the macro arguments. This could be used to generate record-style data structures or “unrolled” loops, for example, or could generate entire algorithms based on complex parameters. For instance, a “sort” macro could accept the specification of a complex sort key and generate code crafted for that specific key, not needing the run-time tests that would be required for a general procedure interpreting the specification. An organization using assembly language that has been heavily extended using such a macro suite can be considered to be working in a higher-level language, since such programmers are not working with a computer’s lowest-level conceptual elements. Underlining this point, macros were used to implement an early virtual machine in SNOBOL4 (1967), which was written in the SNOBOL Implementation Language (SIL), an assembly language for a virtual machine, which was then targeted to physical machines by transpiled to a native assembler via a macro assembler. This allowed a high degree of portability for the time.
Macros were used to customize large scale software systems for specific customers in the mainframe era and were also used by customer personnel to satisfy their employers’ needs by making specific versions of manufacturer operating systems. This was done, for example, by systems programmers working with IBM’s Conversational Monitor System / Virtual Machine (VM/CMS) and with IBM’s “real time transaction processing” add-ons, Customer Information Control System CICS, and ACP/TPF, the airline/financial system that began in the 1970s and still runs many large computer reservation systems (CRS) and credit card systems today.
It is also possible to use solely the macro processing abilities of an assembler to generate code written in completely different languages, for example, to generate a version of a program in COBOL using a pure macro assembler program containing lines of COBOL code inside assembly time operators instructing the assembler to generate arbitrary code. IBM OS/360 uses macros to perform system generation. The user specifies options by coding a series of assembler macros. Assembling these macros generates a job stream to build the system, including job control language and utility control statements.
This is because, as was realized in the 1960s, the concept of “macro processing” is independent of the concept of “assembly”, the former being in modern terms more word processing, text processing, than generating object code. The concept of macro processing appeared, and appears, in the C programming language, which supports “preprocessor instructions” to set variables, and make conditional tests on their values. Note that unlike certain previous macro processors inside assemblers, the C preprocessor is not Turing-complete because it lacks the ability to either loop or “go to”, the latter allowing programs to loop.
Despite the power of macro processing, it fell into disuse in many high level languages (major exceptions being C/C++ and PL/I) while remaining a perennial for assemblers.
Macro parameter substitution is strictly by name: at macro processing time, the value of a parameter is textually substituted for its name. The most famous class of bugs resulting was the use of a parameter that itself was an expression and not a simple name when the macro writer expected a name. In the macro:
foo: macro a
the intention was that the caller would provide the name of a variable, and the “global” variable or constant b would be used to multiply “a”. If foo is called with the parameter a-c, the macro expansion of load a-c*b occurs. To avoid any possible ambiguity, users of macro processors can parenthesize formal parameters inside macro definitions, or callers can parenthesize the input parameters.
Support for structured programming
Some assemblers have incorporated structured programming elements to encode execution flow. The earliest example of this approach was in the Concept-14 macro set, originally proposed by Dr. Harlan Mills (March 1970), and implemented by Marvin Kessler at IBM’s Federal Systems Division, which extended the S/360 macro assembler with IF/ELSE/ENDIF and similar control flow blocks. This was a way to reduce or eliminate the use of GOTO operations in assembly code, one of the main factors causing spaghetti code in assembly language. This approach was widely accepted in the early ’80s (the latter days of large-scale assembly language use).
A curious design was A-natural, a “stream-oriented” assembler for 8080/Z80 processors from Whitesmiths Ltd. (developers of the Unix-like Idris operating system, and what was reported to be the first commercial C compiler). The language was classified as an assembler, because it worked with raw machine elements such as opcodes, registers, and memory references; but it incorporated an expression syntax to indicate execution order. Parentheses and other special symbols, along with block-oriented structured programming constructs, controlled the sequence of the generated instructions. A-natural was built as the object language of a C compiler, rather than for hand-coding, but its logical syntax won some fans.
There has been little apparent demand for more sophisticated assemblers since the decline of large-scale assembly language development. In spite of that, they are still being developed and applied in cases where resource constraints or peculiarities in the target system’s architecture prevent the effective use of higher-level languages.
Assemblers with a strong macro engine allow structured programming via macros, such as the switch macro provided with the Masm32 package (note this code is a complete program):
For more details or to start new programming projects please Contact Us.