Fujitsu MB91F109, Hardware Manual

Page: 56 / 461

32
CHAPTER 2 CPU
Figure 2.2-2 Instruction Pipeline
Instructions are always executed in order. That is, instruction A that is put into the pipeline
before instruction B always reaches the write back stage before instruction B.
Instructions are normally executed at a rate of one instruction per cycle. However, a load/store
instruction involving memory wait, branch instruction without a delay slot, or multiple-cycle
instruction, requires multiple cycles to complete execution. The instruction execution speed is
also slowed down when instruction supply takes time.
❍
"32 bits <--> 16 bits" bus converter
The "32 bits <--> 16 bits" bus converter interfaces between the D-BUS that allows high-speed
32-bit wide access and R-BUS that allows 16-bit wide access. It thus enables the CPU to
access data in the internal peripheral circuits.
Upon receipt of a 32-bit wide access from the CPU, the bus converter converts it into two 16-bit
wide accesses to implement access to the R-BUS. Some internal peripheral circuits have
restrictions on access width.
❍
"Harvard <--> Princeton" bus converter
The "Harvard <--> Princeton" bus converter coordinates the instruction access and data access
of the CPU to implement smooth interfacing with the external bus.
The CPU has Harvard architecture in which the instruction bus and data bus are independent of
each other. The bus controller that controls the external bus has Princeton architecture
consisting of a single bus. The bus converter gives priority to the instruction and data accesses
of the CPU to control accesses to the bus controller. This control always optimizes the order of
access to the external bus.
The bus converter has a two-word write buffer to eliminate the CPU’s bus wait time and a one-
word prefetch buffer for instruction fetch.
Instruction 1
Instruction 2
Instruction 3
Instruction 4
Instruction 5
Instruction 6
CLK
WB
WB
WB
WB
WB
MA
MA
MA
MA
MA WB
EX
EX
EX
EX
ID
ID
ID
IF
IF