Table 7 Impact of memory layout and loop-fusion transform
From: Covariance tracking: architecture optimizations for embedded systems
Penryn-M | Haswell-M | Cortex-A9 | Cortex-A15 | |
|---|---|---|---|---|
cpp | ||||
SoA | 447 | 300 | 830 | 646 |
AoS | 207 | 178 | 836 | 520 |
AoS + T | 126 | 66 | 503 | 238 |
AoS + SIMD | 165 | 69 | 476 | 325 |
AoS + T SIMD | 92 | 38 | 201 | 169 |
speedups | ||||
SoA/AoS | ×2.2 | ×1.7 | ×1.0 | ×1.2 |
AoS/AoS + T | ×1.6 | ×2.7 | ×1.7 | ×2.2 |
SIMD/SIMD + T | ×1.8 | ×1.8 | ×2.4 | ×1.9 |
SoA/AoS+SIMD + T | ×4 . 9 | ×7 . 9 | ×4 . 1 | ×3 . 8 |
Execution time (ms) for 512 × 512 images | ||||
AoS+T SIMD | 20.1 | 6.1 | 43.9 | 26.1 |
Estimated energy consumption (mJ) for 512 × 512 images | ||||
AoS+T SIMD | 201 | 91.4 | 52.7 | 44.3 |