Instead of calculating a^2, b^2, (a+b)^2 and (a-b)^2, calculate only a^2, b^2 and 2*a*b in each iteration and derive the latter parts from these three at the end. Before and after: A78 ac3_sum_square_bufferfly_int32_neon: 484.8 ( 2.00x) ac3_sum_square_bufferfly_int32_neon: 468.2 ( 2.08x) A72 ac3_sum_square_bufferfly_int32_neon: 793.6 ( 1.26x) ac3_sum_square_bufferfly_int32_neon: 527.3 ( 1.92x)
Unified · Split
| @@ -69,21 +69,20 @@ function ff_ac3_sum_square_butterfly_int32_neon, export=1 · 68 unmodified lines | ||
| 69 | 69 | movi v0.2d, #0 |
| 70 | 70 | movi v1.2d, #0 |
| 71 | 71 | movi v2.2d, #0 |
| 72 | − movi v3.2d, #0 | |
| 73 | 72 | 1: ld1 {v4.2s}, [x1], #8 |
| 74 | 73 | ld1 {v5.2s}, [x2], #8 |
| 75 | − add v6.2s, v4.2s, v5.2s | |
| 76 | − sub v7.2s, v4.2s, v5.2s | |
| 77 | − smlal v0.2d, v4.2s, v4.2s | |
| 78 | − smlal v1.2d, v5.2s, v5.2s | |
| 79 | − smlal v2.2d, v6.2s, v6.2s | |
| 80 | − smlal v3.2d, v7.2s, v7.2s | |
| 81 | 74 | subs w3, w3, #2 |
| 75 | + smlal v0.2d, v4.2s, v4.2s // sum of a^2 | |
| 76 | + smlal v1.2d, v5.2s, v5.2s // sum of b^2 | |
| 77 | + sqdmlal v2.2d, v4.2s, v5.2s // sum of 2ab | |
| 82 | 78 | b.gt 1b |
| 83 | 79 | addp d0, v0.2d |
| 84 | 80 | addp d1, v1.2d |
| 85 | 81 | addp d2, v2.2d |
| 86 | − addp d3, v3.2d | |
| 82 | + sub d3, d0, d2 // a^2 + b^2 - 2ab | |
| 83 | + add d2, d0, d2 | |
| 84 | + add d3, d3, d1 // a^2 + b^2 + 2ab | |
| 85 | + add d2, d2, d1 | |
| 87 | 86 | st1 {v0.1d-v3.1d}, [x0] |
| 88 | 87 | ret |
| 89 | 88 | endfunc |