machine cycle
<processor> The four steps which the CPU carries out for each machine
language instruction: fetch, decode, execute, and store. These steps are
performed by the control unit, and may be fixed in the logic of the CPU or may
be programmed as microcode which is itself usually fixed (in ROM) but may be
(partially) modifiable (stored in RAM).
The fetch cycle places the current program counter contents (the address of the
next instruction to execute) on the address bus and reads in the word at that
location into the instruction register (IR). In RISC CPUs instructions are
usually a single word but in other architectures an instruction may be several
words long, necessitating several fetches.
The decode cycle uses the contents of the IR to determine which gates should be
opened between the CPU's various functional units and busses and what operation
the ALU(s) should perform (e.g. add, bitwise and). Each gate allows data to flow
from one unit to another (e.g. from register 0 to ALU input 1) or enables data
from one output onto a certain bus. In the simplest case ("horizontal encoding")
each bit of the instruction register controls a single gate or several bits may
control the ALU operation. This is rarely used because it requires long
instruction words (such an architecture is sometimes called a very long
instruction word architecture). Commonly, groups of bits from the IR are fed
through decoders to control higher level aspects of the CPU's operation, e.g.
source and destination registers, addressing mode and ALU operation. This is
known as vertical encoding. One way RISC processors gain their advantage in
speed is by having simple instruction decoding which can be performed quickly.
The execute cycle occurs when the decoding logic has settled and entails the
passing of values between the various function units and busses and the
operation of the ALU. A simple instruction will require only a single execute
cycle whereas a complex instruction (e.g. subroutine call or one using memory
indirect addressing) may require three or four. Instructions in a RISC typically
(but not invariably) take only a single cycle.
The store cycle is when the result of the instruction is written to its
destination, either a register or a memory location. This is really part of the
execute cycle because some instructions may write to multiple destinations as
part of their execution.
(1995-04-13)
Nearby terms:
machinable « machine « machine code « machine
cycle » machine language » machine learning »
Mach Interface Generator
|