Microops

Micro-ops, short for micro-operations, are the atomic steps that a CPU’s microarchitecture uses to realize machine instructions. They are produced by the instruction decoder from the architectural instruction stream and are the units that are scheduled and executed by the processor’s execution resources.

Most modern CPUs are out-of-order and superscalar, and they break down complex instructions into multiple micro-ops. An x86 instruction, for example, may expand into several micro-ops, while simpler instructions may decode into a single micro-op. Some processors use hardwired micro-ops, while others rely on microcode for certain instructions; many designs also employ a micro-op cache to avoid re-decoding frequently used sequences.

Once generated, micro-ops are placed into a micro-op queue or a reorder buffer, renamed to remove false dependencies, and dispatched to appropriate execution units such as integer ALUs, floating-point units, load/store units, or branch units. They are executed in parallel where possible, and their results are written back and retired in program order to preserve architectural correctness.

Origin and scope: The concept grew out of microprogramming in classic mainframe architectures, and in contemporary CPUs it is implemented in a variety of ways. In many designs, the term micro-op also encompasses a broader class of internal micro-operations used by the pipeline to support routing, memory addressing, and control flow.

See also: microcode, microinstruction, microarchitecture, x86, out-of-order execution.