This stores the constant coefficients deinterleaved, so that they can be loaded directly with NF=0. Unfortunately, we cannot optimise loading the input, due to insufficient memory alignment (not 32-bit). Before: g722_apply_qmf_c: 82.5 g722_apply_qmf_rvv_i32: 78.2 After: g722_apply_qmf_c: 82.5 g722_apply_qmf_rvv_i32: 65.2
Unified · Split
| @@ -24,7 +24,9 @@ func ff_g722_apply_qmf_rvv, zve32x · 23 unmodified lines | ||
| 24 | 24 | lla t0, qmf_coeffs |
| 25 | 25 | vsetivli zero, 12, e16, m2, ta, ma |
| 26 | 26 | vlseg2e16.v v28, (a0) |
| 27 | − vlseg2e16.v v24, (t0) | |
| 27 | + addi t1, t0, 12 * 2 | |
| 28 | + vle16.v v24, (t0) | |
| 29 | + vle16.v v26, (t1) | |
| 28 | 30 | vwmul.vv v16, v28, v24 |
| 29 | 31 | vwmul.vv v20, v30, v26 |
| 30 | 32 | vsetivli zero, 12, e32, m4, ta, ma |
| @@ -41,26 +43,26 @@ endfunc · 10 unmodified lines | ||
| 41 | 43 | const qmf_coeffs, align=2 |
| 42 | 44 | .short 3 |
| 43 | 45 | .short -11 |
| 44 | − .short -11 | |
| 45 | − .short 53 | |
| 46 | 46 | .short 12 |
| 47 | − .short -156 | |
| 48 | 47 | .short 32 |
| 49 | − .short 362 | |
| 50 | 48 | .short -210 |
| 51 | − .short -805 | |
| 52 | 49 | .short 951 |
| 53 | 50 | .short 3876 |
| 54 | − .short 3876 | |
| 55 | − .short 951 | |
| 56 | 51 | .short -805 |
| 57 | − .short -210 | |
| 58 | 52 | .short 362 |
| 59 | − .short 32 | |
| 60 | 53 | .short -156 |
| 61 | − .short 12 | |
| 62 | 54 | .short 53 |
| 63 | 55 | .short -11 |
| 64 | 56 | .short -11 |
| 57 | + .short 53 | |
| 58 | + .short -156 | |
| 59 | + .short 362 | |
| 60 | + .short -805 | |
| 61 | + .short 3876 | |
| 62 | + .short 951 | |
| 63 | + .short -210 | |
| 64 | + .short 32 | |
| 65 | + .short 12 | |
| 66 | + .short -11 | |
| 65 | 67 | .short 3 |
| 66 | 68 | endconst |