ac3_sum_square_bufferfly_int32_c: 61.0 ac3_sum_square_bufferfly_int32_rvv_i64: 14.7
Unified · Split
| @@ -28,6 +28,8 @@ · 27 unmodified lines | ||
| 28 | 28 | |
| 29 | 29 | void ff_extract_exponents_rvb(uint8_t *exp, int32_t *coef, int nb_coefs); |
| 30 | 30 | void ff_float_to_fixed24_rvv(int32_t *dst, const float *src, size_t len); |
| 31 | +void ff_sum_square_butterfly_int32_rvv(int64_t *, const int32_t *, | |
| 32 | + const int32_t *, int); | |
| 31 | 33 | |
| 32 | 34 | av_cold void ff_ac3dsp_init_riscv(AC3DSPContext *c) |
| 33 | 35 | { |
| @@ -39,6 +41,10 @@ av_cold void ff_ac3dsp_init_riscv(AC3DSPContext *c) · 5 unmodified lines | ||
| 39 | 41 | c->extract_exponents = ff_extract_exponents_rvb; |
| 40 | 42 | if (flags & AV_CPU_FLAG_RVV_F32) |
| 41 | 43 | c->float_to_fixed24 = ff_float_to_fixed24_rvv; |
| 44 | +# if __riscv_xlen >= 64 | |
| 45 | + if (flags & AV_CPU_FLAG_RVV_I64) | |
| 46 | + c->sum_square_butterfly_int32 = ff_sum_square_butterfly_int32_rvv; | |
| 47 | +# endif | |
| 42 | 48 | } |
| 43 | 49 | #endif |
| 44 | 50 | } |
| @@ -37,3 +37,44 @@ func ff_float_to_fixed24_rvv, zve32f · 36 unmodified lines | ||
| 37 | 37 | |
| 38 | 38 | ret |
| 39 | 39 | endfunc |
| 40 | + | |
| 41 | +#if __riscv_xlen >= 64 | |
| 42 | +func ff_sum_square_butterfly_int32_rvv, zve64x | |
| 43 | + vsetvli t0, zero, e64, m8, ta, ma | |
| 44 | + vmv.v.x v0, zero | |
| 45 | + vmv.v.x v8, zero | |
| 46 | +1: | |
| 47 | + vsetvli t0, a3, e32, m2, tu, ma | |
| 48 | + vle32.v v16, (a1) | |
| 49 | + sub a3, a3, t0 | |
| 50 | + vle32.v v20, (a2) | |
| 51 | + sh2add a1, t0, a1 | |
| 52 | + vadd.vv v24, v16, v20 | |
| 53 | + sh2add a2, t0, a2 | |
| 54 | + vsub.vv v28, v16, v20 | |
| 55 | + vwmacc.vv v0, v16, v16 | |
| 56 | + vwmacc.vv v4, v20, v20 | |
| 57 | + vwmacc.vv v8, v24, v24 | |
| 58 | + vwmacc.vv v12, v28, v28 | |
| 59 | + bnez a3, 1b | |
| 60 | + | |
| 61 | + vsetvli t0, zero, e64, m4, ta, ma | |
| 62 | + vmv.s.x v16, zero | |
| 63 | + vmv.s.x v17, zero | |
| 64 | + vredsum.vs v16, v0, v16 | |
| 65 | + vmv.s.x v18, zero | |
| 66 | + vredsum.vs v17, v4, v17 | |
| 67 | + vmv.s.x v19, zero | |
| 68 | + vredsum.vs v18, v8, v18 | |
| 69 | + vmv.x.s t0, v16 | |
| 70 | + vredsum.vs v19, v12, v19 | |
| 71 | + vmv.x.s t1, v17 | |
| 72 | + sd t0, (a0) | |
| 73 | + vmv.x.s t2, v18 | |
| 74 | + sd t1, 8(a0) | |
| 75 | + vmv.x.s t3, v19 | |
| 76 | + sd t2, 16(a0) | |
| 77 | + sd t3, 24(a0) | |
| 78 | + ret | |
| 79 | +endfunc | |
| 80 | +#endif | |