Skip to main content

Table 5 Complexity, memory accesses, and arithmetic intensity of AoS versions with/without loop-fusion

From: Covariance tracking: architecture optimizations for embedded systems

Instructions

MUL

ADD

LOAD

STORE

AI

AoS version with 3 loops

Product of features

n P

0

2n P

n P

-

Integral of features

0

3n F

4n F

n F

-

Integral of products

0

3n P

4n P

n P

-

Total

n P

3(n P +n F )

6n P +4n F

2n P +n F

-

Total with n P = n F (n F + 1)/2

2 n F 2 +5 n F

4 n F 2 +9 n F

0.5

AoS version + loop fusion with 1 loop

Integral of features

0

2n F

2n F

n F

-

Integral product of features

n P

2n P

n P

n P

-

Total

n P

2(n P + n F )

n P +2n F

n P n F

-

Total with n P = n F (n F + 1)/2

1.5 n F 2 +3.5 n F

n F 2 +4 n F

1.5