Table 6 Complexity, memory accesses, and arithmetic intensity of scalar/SIMD versions with/without loop-fusion: numerical results for n F = 7 and n F = 8
From: Covariance tracking: architecture optimizations for embedded systems
Version | Without loop-fusion | With loop-fusion | |||||
|---|---|---|---|---|---|---|---|
SIMD | n F | Arith | Mem | AI | Arith | Mem | AI |
Scalar | 7 | 133 | 259 | 0.51 | 98 | 105 | 0.93 |
Scalar | 8 | 168 | 328 | 0.51 | 124 | 132 | 0.94 |
SSE | 7 | 49 | 66 | 0.74 | 40 | 27 | 1.48 |
SSE | 8 | 59 | 82 | 0.72 | 48 | 33 | 1.45 |
Neon | 7 | 65 | 66 | 0.98 | 56 | 27 | 2.07 |
Neon | 8 | 77 | 82 | 0.94 | 66 | 33 | 2.00 |
AVX | 7 | 27 | 37 | 0.73 | 22 | 15 | 1.47 |
AVX | 8 | 33 | 45 | 0.73 | 27 | 18 | 1.50 |