T-Head C908: h264_biweight2_8_c: 58.0 h264_biweight2_8_rvv_i32: 11.2 h264_biweight4_8_c: 106.0 h264_biweight4_8_rvv_i32: 22.7 h264_biweight8_8_c: 205.7 h264_biweight8_8_rvv_i32: 50.0 h264_biweight16_8_c: 403.5 h264_biweight16_8_rvv_i32: 83.2 SpacemiT X60: h264_weight2_8_c: 48.2 h264_weight2_8_rvv_i32: 8.2 h264_weight4_8_c: 90.5 h264_weight4_8_rvv_i32: 16.5 h264_weight8_8_c: 175.2 h264_weight8_8_rvv_i32: 38.0 h264_weight16_8_c: 342.2 h264_weight16_8_rvv_i32: 66.0
Unified · Split
| @@ -28,7 +28,10 @@ · 27 unmodified lines | ||
| 28 | 28 | #include "libavutil/riscv/cpu.h" |
| 29 | 29 | #include "libavcodec/h264dsp.h" |
| 30 | 30 | |
| 31 | −extern const h264_weight_func ff_h264_weight_funcs_8_rvv[]; | |
| 31 | +extern const struct { | |
| 32 | + const h264_weight_func weight; | |
| 33 | + const h264_biweight_func biweight; | |
| 34 | +} ff_h264_weight_funcs_8_rvv[]; | |
| 32 | 35 | |
| 33 | 36 | void ff_h264_v_loop_filter_luma_8_rvv(uint8_t *pix, ptrdiff_t stride, |
| 34 | 37 | int alpha, int beta, int8_t *tc0); |
| @@ -63,9 +66,12 @@ av_cold void ff_h264dsp_init_riscv(H264DSPContext *dsp, const int bit_depth, · 28 unmodified lines | ||
| 63 | 66 | # if HAVE_RVV |
| 64 | 67 | if (flags & AV_CPU_FLAG_RVV_I32) { |
| 65 | 68 | if (bit_depth == 8 && ff_rv_vlen_least(128)) { |
| 66 | − memcpy(dsp->weight_h264_pixels_tab, | |
| 67 | − ff_h264_weight_funcs_8_rvv, | |
| 68 | − sizeof (dsp->weight_h264_pixels_tab)); | |
| 69 | + for (int i = 0; i < 4; i++) { | |
| 70 | + dsp->weight_h264_pixels_tab[i] = | |
| 71 | + ff_h264_weight_funcs_8_rvv[i].weight; | |
| 72 | + dsp->biweight_h264_pixels_tab[i] = | |
| 73 | + ff_h264_weight_funcs_8_rvv[i].biweight; | |
| 74 | + } | |
| 69 | 75 | |
| 70 | 76 | dsp->h264_v_loop_filter_luma = ff_h264_v_loop_filter_luma_8_rvv; |
| 71 | 77 | dsp->h264_h_loop_filter_luma = ff_h264_h_loop_filter_luma_8_rvv; |
| @@ -48,6 +48,35 @@ func ff_h264_weight_pixels_simple_8_rvv, zve32x · 47 unmodified lines | ||
| 48 | 48 | ret |
| 49 | 49 | endfunc |
| 50 | 50 | |
| 51 | + .variant_cc ff_h264_biweight_pixels_simple_8_rvv | |
| 52 | +func ff_h264_biweight_pixels_simple_8_rvv, zve32x | |
| 53 | + csrwi vxrm, 2 | |
| 54 | + addi a7, a7, 1 | |
| 55 | + ori a7, a7, 1 | |
| 56 | + sll a7, a7, a4 | |
| 57 | +1: | |
| 58 | + vsetvli zero, t6, e32, m4, ta, ma | |
| 59 | + vle8.v v8, (a0) | |
| 60 | + addi a3, a3, -1 | |
| 61 | + vle8.v v12, (a1) | |
| 62 | + add a1, a1, a2 | |
| 63 | + vmv.v.x v16, a7 | |
| 64 | + vsetvli zero, zero, e16, m2, ta, ma | |
| 65 | + vzext.vf2 v24, v8 | |
| 66 | + vzext.vf2 v28, v12 | |
| 67 | + vwmaccsu.vx v16, a5, v24 | |
| 68 | + vwmaccsu.vx v16, a6, v28 | |
| 69 | + vnclip.wx v16, v16, a4 | |
| 70 | + vmax.vx v16, v16, zero | |
| 71 | + vsetvli zero, zero, e8, m1, ta, ma | |
| 72 | + vnclipu.wi v8, v16, 1 | |
| 73 | + vse8.v v8, (a0) | |
| 74 | + add a0, a0, a2 | |
| 75 | + bnez a3, 1b | |
| 76 | + | |
| 77 | + ret | |
| 78 | +endfunc | |
| 79 | + | |
| 51 | 80 | func ff_h264_weight_pixels_8_rvv, zve32x |
| 52 | 81 | csrwi vxrm, 0 |
| 53 | 82 | sll a5, a5, a3 |
| @@ -84,6 +113,53 @@ func ff_h264_weight_pixels_8_rvv, zve32x · 30 unmodified lines | ||
| 84 | 113 | ret |
| 85 | 114 | endfunc |
| 86 | 115 | |
| 116 | + .variant_cc ff_h264_biweight_pixels_8_rvv | |
| 117 | +func ff_h264_biweight_pixels_8_rvv, zve32x | |
| 118 | + csrwi vxrm, 2 | |
| 119 | + addi a7, a7, 1 | |
| 120 | + ori a7, a7, 1 | |
| 121 | + sll a7, a7, a4 | |
| 122 | +1: | |
| 123 | + mv t0, a0 | |
| 124 | + mv t1, a1 | |
| 125 | + mv t5, t6 | |
| 126 | +2: | |
| 127 | + vsetvli t2, a3, e32, m8, ta, ma | |
| 128 | + vlsseg2e8.v v0, (t0), a2 | |
| 129 | + vlsseg2e8.v v4, (t1), a2 | |
| 130 | + addi t5, t5, -2 | |
| 131 | + vmv.v.x v16, a7 | |
| 132 | + vmv.v.x v24, a7 | |
| 133 | + vsetvli zero, zero, e16, m4, ta, ma | |
| 134 | + vzext.vf2 v8, v0 | |
| 135 | + vzext.vf2 v12, v2 | |
| 136 | + vwmaccsu.vx v16, a5, v8 | |
| 137 | + vwmaccsu.vx v24, a5, v12 | |
| 138 | + vzext.vf2 v8, v4 | |
| 139 | + vzext.vf2 v12, v6 | |
| 140 | + vwmaccsu.vx v16, a6, v8 | |
| 141 | + vwmaccsu.vx v24, a6, v12 | |
| 142 | + vnclip.wx v8, v16, a4 | |
| 143 | + vnclip.wx v12, v24, a4 | |
| 144 | + vmax.vx v8, v8, zero | |
| 145 | + vmax.vx v12, v12, zero | |
| 146 | + vsetvli zero, zero, e8, m2, ta, ma | |
| 147 | + vnclipu.wi v0, v8, 1 | |
| 148 | + vnclipu.wi v2, v12, 1 | |
| 149 | + vssseg2e8.v v0, (t0), a2 | |
| 150 | + addi t0, t0, 2 | |
| 151 | + addi t1, t1, 2 | |
| 152 | + bnez t5, 2b | |
| 153 | + | |
| 154 | + mul t3, a2, t2 | |
| 155 | + sub a3, a3, t2 | |
| 156 | + add a0, a0, t3 | |
| 157 | + add a1, a1, t3 | |
| 158 | + bnez a3, 1b | |
| 159 | + | |
| 160 | + ret | |
| 161 | +endfunc | |
| 162 | + | |
| 87 | 163 | .irp w, 16, 8, 4, 2 |
| 88 | 164 | func ff_h264_weight_pixels\w\()_8_rvv, zve32x |
| 89 | 165 | li a6, \w |
| @@ -93,6 +169,15 @@ func ff_h264_weight_pixels\w\()_8_rvv, zve32x · 3 unmodified lines | ||
| 93 | 169 | j ff_h264_weight_pixels_8_rvv |
| 94 | 170 | .endif |
| 95 | 171 | endfunc |
| 172 | + | |
| 173 | +func ff_h264_biweight_pixels\w\()_8_rvv, zve32x | |
| 174 | + li t6, \w | |
| 175 | + .if \w == 16 | |
| 176 | + j ff_h264_biweight_pixels_simple_8_rvv | |
| 177 | + .else | |
| 178 | + j ff_h264_biweight_pixels_8_rvv | |
| 179 | + .endif | |
| 180 | +endfunc | |
| 96 | 181 | .endr |
| 97 | 182 | |
| 98 | 183 | .global ff_h264_weight_funcs_8_rvv |
| @@ -101,10 +186,13 @@ const ff_h264_weight_funcs_8_rvv · 2 unmodified lines | ||
| 101 | 186 | .irp w, 16, 8, 4, 2 |
| 102 | 187 | #if __riscv_xlen == 32 |
| 103 | 188 | .word ff_h264_weight_pixels\w\()_8_rvv |
| 189 | + .word ff_h264_biweight_pixels\w\()_8_rvv | |
| 104 | 190 | #elif __riscv_xlen == 64 |
| 105 | 191 | .dword ff_h264_weight_pixels\w\()_8_rvv |
| 192 | + .dword ff_h264_biweight_pixels\w\()_8_rvv | |
| 106 | 193 | #else |
| 107 | 194 | .qword ff_h264_weight_pixels\w\()_8_rvv |
| 195 | + .qword ff_h264_biweight_pixels\w\()_8_rvv | |
| 108 | 196 | #endif |
| 109 | 197 | .endr |
| 110 | 198 | endconst |