AN2203 Freescale Semiconductor / Motorola, AN2203 Datasheet - Page 34

AN2203

Manufacturer Part Number

AN2203

Description

MPC7450 RISC Microprocessor Family Software Optimization Guide

Manufacturer

Freescale Semiconductor / Motorola

Datasheet

1.AN2203.pdf (76 pages)

Available stocks

Company

Part Number

Manufacturer

Quantity

Price

Company:

Meier Automation Equipment Co., Limited

Part Number:

AN22030A

Manufacturer:

PANASONIC/松下

Quantity:

20 000

Current page: 34 of 76
Download datasheet (650Kb)

FPU Considerations

The timing for this sequence in Table 3-17 assumes that the load misses in the data cache. Here, after the

ﬁrst four fadds, the MPC7450 runs out of FPSCR rename registers and the pipeline stalls. When the load

completes, the pipeline restarts after an additional 2-cycle lag.

Note that denormalized numbers can cause problems for the FPU pipeline, so the normal latencies in

Table A-6 may not apply. Output denormalization in the very unlikely worst case can add as many as three

cycles of latency. Input denormalization takes four to six additional cycles, depending on whether one, two,

or three input source operands are denormalized.

3.6.1

On the MPC7450, the four vector execution units are fully independent and fully pipelined. Table 3-18

shows the latencies.

VFPU latency is usually four cycles, but some instructions, particularly the vector ﬂoat compares and vector

ﬂoat min/max (see Table A-8 to Table A-11 for a list) have only a 2-cycle latency. This can create

competition for the VFPU register forwarding bus. This is solved by forcing a partial stall when a bypass is

needed. Consider the following code example:

xxxxxx20

xxxxxx24

xxxxxx28

xxxxxx2C

xxxxxx30

Table 3-19 shows the timing for this vector compare bypass/stall situation. In cycle 6 the vcmp bypasses

from E0 to E3, stalling the vsubfp and vlogefp for a cycle in stages E1 and E2. Note that an instruction in

E1 stalls in E1 under a bypass scenario even if no instruction is in E2.

lfdu

fadd

Instruction

Vector Units

MPC7450 RISC Microprocessor Family Software Optimization Guide

—

vaddfp v10,v11,v12

vsubfp v11,v14,v13

vaddfp v12,v13,v14

vcmpbfp. v13,v18,v19

vmaddfp v14,v20,v21,v14

—

Table 3-18. Vector Execution Latency Summary

Freescale Semiconductor, Inc.

—

Table 3-17. FPSCR Rename Timing Example

For More Information On This Product,

—

Go to: www.freescale.com

—

VFPU

VIU1

VIU2

VPU

Unit

—

Typical Latency

—

MOTOROLA

AN2203 Freescale Semiconductor / Motorola, AN2203 Datasheet - Page 34

AN2203

Available stocks

Related parts for AN2203