hf_gen_c: 2922.7 hf_gen_rvv_f32: 731.5
Unified · Split
| @@ -27,6 +27,9 @@ void ff_sbr_sum64x5_rvv(float *z); · 26 unmodified lines | ||
| 27 | 27 | float ff_sbr_sum_square_rvv(float (*x)[2], int n); |
| 28 | 28 | void ff_sbr_neg_odd_64_rvv(float *x); |
| 29 | 29 | void ff_sbr_autocorrelate_rvv(const float x[40][2], float phi[3][2][2]); |
| 30 | +void ff_sbr_hf_gen_rvv(float (*X_high)[2], const float (*X_low)[2], | |
| 31 | + const float alpha0[2], const float alpha1[2], | |
| 32 | + float bw, int start, int end); | |
| 30 | 33 | void ff_sbr_hf_g_filt_rvv(float (*Y)[2], const float (*X_high)[40][2], |
| 31 | 34 | const float *g_filt, int m_max, intptr_t ixh); |
| 32 | 35 | |
| @@ -39,6 +42,7 @@ av_cold void ff_sbrdsp_init_riscv(SBRDSPContext *c) · 6 unmodified lines | ||
| 39 | 42 | if (flags & AV_CPU_FLAG_RVB_ADDR) { |
| 40 | 43 | c->sum64x5 = ff_sbr_sum64x5_rvv; |
| 41 | 44 | c->sum_square = ff_sbr_sum_square_rvv; |
| 45 | + c->hf_gen = ff_sbr_hf_gen_rvv; | |
| 42 | 46 | c->hf_g_filt = ff_sbr_hf_g_filt_rvv; |
| 43 | 47 | } |
| 44 | 48 | c->autocorrelate = ff_sbr_autocorrelate_rvv; |
| @@ -174,6 +174,56 @@ func ff_sbr_autocorrelate_rvv, zve32f · 173 unmodified lines | ||
| 174 | 174 | ret |
| 175 | 175 | endfunc |
| 176 | 176 | |
| 177 | +func ff_sbr_hf_gen_rvv, zve32f | |
| 178 | +NOHWF fmv.w.x fa0, a4 | |
| 179 | +NOHWF mv a4, a5 | |
| 180 | +NOHWF mv a5, a6 | |
| 181 | + flw ft2, 0(a2) | |
| 182 | + fmul.s fa1, fa0, fa0 // bw * bw | |
| 183 | + sh3add a1, a5, a1 | |
| 184 | + flw ft3, 4(a2) | |
| 185 | + fmul.s fa2, ft2, fa0 // alpha[2] | |
| 186 | + sh3add a0, a5, a0 | |
| 187 | + flw ft0, 0(a3) | |
| 188 | + fmul.s fa3, ft3, fa0 // alpha[3] | |
| 189 | + sub a5, a5, a4 | |
| 190 | + flw ft1, 4(a3) | |
| 191 | + fmul.s fa0, ft0, fa1 // alpha[0] | |
| 192 | + flw ft0, -16(a1) // X_low[end - 2][0] | |
| 193 | + fmul.s fa1, ft1, fa1 // alpha[1] | |
| 194 | + flw ft1, -12(a1) // X_low[end - 2][1] | |
| 195 | + flw ft2, -8(a1) // X_low[end - 1][0] | |
| 196 | + flw ft3, -4(a1) // X_low[end - 1][1] | |
| 197 | + addi a1, a1, -16 | |
| 198 | +1: | |
| 199 | + vsetvli t0, a5, e32, m4, ta, ma | |
| 200 | + slli t1, t0, 3 | |
| 201 | + sub a1, a1, t1 | |
| 202 | + vlseg2e32.v v0, (a1) // X_low[i - 2] | |
| 203 | + sub a0, a0, t1 | |
| 204 | + vfslide1down.vf v8, v0, ft0 // X_low[i - 1][0] | |
| 205 | + sub a5, a5, t0 | |
| 206 | + vfslide1down.vf v12, v4, ft1 // X_low[i - 1][1] | |
| 207 | + vfslide1down.vf v16, v8, ft2 // X_low[i ][0] | |
| 208 | + vfslide1down.vf v20, v12, ft3 // X_low[i ][1] | |
| 209 | + vfmacc.vf v16, fa0, v0 | |
| 210 | + vfmacc.vf v20, fa0, v4 | |
| 211 | + vfmv.f.s ft0, v0 | |
| 212 | + vfnmsac.vf v16, fa1, v4 | |
| 213 | + vfmacc.vf v20, fa1, v0 | |
| 214 | + vfmv.f.s ft1, v4 | |
| 215 | + vfmacc.vf v16, fa2, v8 | |
| 216 | + vfmacc.vf v20, fa2, v12 | |
| 217 | + vfmv.f.s ft2, v8 | |
| 218 | + vfnmsac.vf v16, fa3, v12 | |
| 219 | + vfmacc.vf v20, fa3, v8 | |
| 220 | + vfmv.f.s ft3, v12 | |
| 221 | + vsseg2e32.v v16, (a0) | |
| 222 | + bnez a5, 1b | |
| 223 | + | |
| 224 | + ret | |
| 225 | +endfunc | |
| 226 | + | |
| 177 | 227 | func ff_sbr_hf_g_filt_rvv, zve32f |
| 178 | 228 | li t1, 40 * 2 * 4 |
| 179 | 229 | sh3add a1, a4, a1 |