AN2367

The optimized routine

5/13

2.3

Code description

There are 5 blocks in this routine.

The code listing is given in appendix A.

2.4

Performance results

The following table shows the number of cycles for different number of bytes all divisible by

16.

From this table the following formula can be used:

Cycles = 84 + NBytes*1.5

84 cycles represent the total overhead of: Caller (C) transmit the 3 parameters, branch to

routine, routine context save, routine context restore, return to caller.

When the number of bytes is not divisible by 16, then the number of cycles is not linear. The

following formula must be used:

Cycles = 84 + NBytes*1.5 + Additional Cycles for remaining bytes

Table 2.

Overview of code blocks

Name

Function

Implementation

save_context

save all registers used by the

routine

d_main

main loop; transfer 16 bytes per

iteration

single loop

d_swit

second level switch: transfer 4,

8 or 12 bytes

linear routine

d_switch

third level switch: transfer 1, 2 or

3 bytes

switch case

restore_context

restore all registers

Table 3.

Performance results (no. of bytes divisible by 16)

Number of bytes

Number of cycles

32

126

128

276

512

852

1024

1620

Table 4.

Performance results (no. of bytes not divisible by 16)

Remaining bytes

Additional cycles

00

127

227