Wikipedia: "Compilers"
Read the paragraph on one-pass and multi-pass compilers. This distinction was more relevant in the early days of computing, when computer memory was limited and processing time was much slower compared to current computers.
A "pass" refers to reading and processing the input, i.e., source or source representation. A one-pass compiler is simpler to implement. However, more sophisticated processing, such as optimizations, may require multiple passes over the source representation.
The compiler is composed of components, which perform the steps of the compilation process. Some of the components may make a separate pass over the source representation--thus, the name of multi-pass. Multi-pass compilers are used for translating more complex languages (for example, high-level language to high-level language), and for performing more complete optimizations. Moreover, multi-pass compilers are easier to verify and validate, because testing and proof of correctness can be accomplished for the smaller components. Multi-pass compilers are also used for translation to an assembler or machine language for a theoretical machine, and for translation to an intermediate representation such as bytecode. Bytecode, or pcode (portable code), is a compact encoding that uses a one-byte instruction, and parameters for data or addresses, to represent the results of syntax and semantic analysis. The bytecode or pcode is independent of a particular machine. It is executed by a virtual machine residing on the target machine.