![](http://datasheet.mmic.net.cn/230000/V4ECFUM_datasheet_15625205/V4ECFUM_118.png)
6-2
ColdFire CF4e Core User’s Manual
For More Information On This Product,
Go to: www.freescale.com
Basic V4 Pipeline Strategy
Unrolling the OEP has two major ramifications on the processor complex:
To support both pipeline structures, significantly more bandwidth is needed than
earlier ColdFire versions. Both V2 and V3 instruction fetch and operand requests
share the K-Bus. For V4, a unified structure does not provide enough bandwidth for
instruction fetches, so a split bus, or Harvard architecture, is used so that separate
instruction and data memories can be accessed concurrently. The basic two-stage
pipeline V3 K-Bus structure is retained, with one K-Bus connecting the IFP to
instruction memory and the other connecting the OEP to data memory.
By increasing the number of OEP stages, coupled with a V3-style IFP, the combined
pipeline depth requires architectural enhancements to handle branch instructions.
Specifically, the V3 branch acceleration scheme alone does not achieve desired
performance levels. V4 processors implement a two-level adaptive prediction
scheme with a small, direct-mapped branch cache using V3-style branch
acceleration to minimize mispredicted branches.
The resulting pipeline structure is shown in Figure 6-1.
Table 6-1. CFxCore Processor Execution Latency
Instruction
Operation
Number of Processor Cycles
V2
V3
V4
<op> Ry,Rx
Register-to-register
1
1
1
mov.l <mem>y,Rx
32-bit load
2
3
1
mov.b <mem>y,Rx
8-bit load
3
4
1
mov.w <mem>y,Rx
16-bit load
3
4
1
mov.* Ry,<mem>x
store
1
1
1
mov.l <mem>y,<mem>x
Memory-to-memory
2
3
2
<op> <mem>y,Rx
Embedded load
3
4
1
<op> Ry,<mem>x
Read-modify-write
3
4
1
bsr, jsr label
Subroutine call
3
1
1
rts
Subroutine return
5
8
2
bra label
Branch always
2
1
1
bcc label
(forward, not taken)
(forward, taken)
(backward, not taken)
(backward, taken)
(predicted correctly)
(predicted incorrectly)
Conditional branch
1
3
3
2
1
5
0
8
F
Freescale Semiconductor, Inc.
n
.