Low-level programming language

In computer science, a low-level programming language is a programming language that provides little or no abstraction from a computer's instruction set architecture—commands or functions in the language map closely to processor instructions. Generally this refers to either machine code or assembly language. The word "low" refers to the small or nonexistent amount of abstraction between the language and machine language; because of this, low-level languages are sometimes described as being "close to the hardware." Because of the close relationship between the language and the hardware architecture programs written in low-level languages tend to be relatively non-portable.

Low-level languages can convert to machine code without a compiler or interpreter— second-generation programming languages use a simpler processor called an assembler— and the resulting code runs directly on the processor. A program written in a low-level language can be made to run very quickly, with a small memory footprint. An equivalent program in a high-level language can be less efficient and use more memory. Low-level languages are simple, but considered difficult to use, due to numerous technical details that the programmer must remember. By comparison, a high-level programming language isolates execution semantics of a computer architecture from the specification of the program, which simplifies development.

Low-level programming languages are sometimes divided into two categories: first generation, and second generation.

Machine code

Front panel of a PDP-8/E minicomputer. The row of switches at the bottom can be used to toggle in a machine language program

Machine code is the only language a computer can process directly without a previous transformation. Currently, programmers almost never write programs directly in machine code, because it requires attention to numerous details that a high-level language handles automatically, requires memorizing or looking up numerical codes for every instruction, and is extremely difficult to modify.

True machine code is a stream of raw, usually binary, data. A programmer coding in "machine code" normally codes instructions and data in a more readable form such as decimal, octal, or hexadecimal which is translated to internal format by a program called a loader or toggled into the computer's memory from a front panel.

Although few programs are written in machine language, programmers often become adept at reading it through working with core dumps or debugging from the front panel.

Example: A function in hexadecimal representation of 32-bit x86 machine code to calculate the nth Fibonacci number:

8B542408 83FA0077 06B80000 0000C383
FA027706 B8010000 00C353BB 01000000
C9010000 008D0419 83FA0376 078BD98B
B84AEBF1 5BC3

Assembly

Second-generation languages provide one abstraction level on top of the machine code. In the early days of coding on computers like the TX-0 and PDP-1, the first thing MIT hackers did was write assemblers.^[1] Assembly language has little semantics or formal specification, being only a mapping of human-readable symbols, including symbolic addresses, to opcodes, addresses, numeric constants, strings and so on. Typically, one machine instruction is represented as one line of assembly code. Assemblers produce object files that can link with other object files or be loaded on their own.

Most assemblers provide macros to generate common sequences of instructions.

Example: The same Fibonacci number calculator as above, but in x86 assembly language using MASM syntax:

fib:
    mov edx, [esp+8]
    cmp edx, 0
    ja @f
    mov eax, 0
    ret
    
    @@:
    cmp edx, 2
    ja @f
    mov eax, 1
    ret
    
    @@:
    push ebx
    mov ebx, 1
    mov ecx, 1
    
    @@:
        lea eax, [ebx+ecx]
        cmp edx, 3
        jbe @f
        mov ebx, ecx
        mov ecx, eax
        dec edx
    jmp @b
    
    @@:
    pop ebx
    ret

In this code example, hardware features of the x86 processor (its registers) are named and manipulated directly. The function loads its input from a precise location in the stack (8 bytes higher than the location stored in the ESP stack pointer) and performs its calculation by manipulating values in the EAX, EBX, ECX and EDX registers until it has finished and returns. Note that in this assembly language, there is no concept of returning a value. The result having been stored in the EAX register, the RET command simply moves code processing to the code location stored on the stack (usually the instruction immediately after the one that called this function) and it is up to the author of the calling code to know that this function stores its result in EAX and to retrieve it from there. x86 assembly language imposes no standard for returning values from a function (and so, in fact, has no concept of a function); it is up to the calling code to examine state after the procedure returns if it needs to extract a value.

Compare this with the same function in C:

unsigned int fib(unsigned int n)
{
    if (n <= 0)
        return 0;
    else if (n <= 2)
        return 1;
    else {
        unsigned int a,b,c;
        a = 1;
        b = 1;
        while (1) {
            c = a + b;
            if (n <= 3) return c;
            a = b;
            b = c;
            n--;
        }
    }
}

This code is very similar in structure to the assembly language example but there are significant differences in terms of abstraction:

• While the input (parameter n) is loaded from the stack, its precise position on the stack is not specified. The C compiler calculates this based on the calling conventions of the target architecture.
• The assembly language version loads the input parameter from the stack into a register and in each iteration of the loop decrements the value in the register, never altering the value in the memory location on the stack. The C compiler could do the same or could update the value in the stack. Which one it chooses is an implementation decision completely hidden from the code author (and one with no side effects, thanks to C language standards).
• The local variables a, b and c are abstractions that do not specify any specific storage location on the hardware. The C compiler decides how to actually store them for the target architecture.
• The return function specifies the value to return, but does not dictate how it is returned. The C compiler for any specific architecture implement a standard mechanism for returning the value. Compilers for the x86 architecture typically (but not always) use the EAX register to return a value, as in the assembly language example (the author of the assembly language example has chosen to copy the C convention but assembly language does not require this).

These abstractions make the C code compilable without modification on any architecture for which a C compiler has been written. The x86 assembly language code is specific to the x86 architecture.

Low-level programming in high-level languages

In the late 1960s, high-level languages such as PL/S, BLISS, BCPL, extended ALGOL (for Burroughs large systems) and C included some degree of access to low-level programming functions. One method for this is Inline assembly, in which assembly code is embedded in a high-level language that supports this feature. Some of these languages also allow architecture-dependent compiler optimization directives to adjust the way a compiler uses the target processor architecture.

References

1. Levy, Stephen (1994). Hackers: Heroes of the Computer Revolution, Penguin Books. p. 32. ISBN 0-14-100051-1