Showing the changes to codec/processing/src/arm64 (1 of 9 files). Show all
Unified · Split
| @@ -84,7 +84,6 @@ comp_ds_bilinear_loop0: · 83 unmodified lines | ||
| 84 | 84 | |
| 85 | 85 | WELS_ASM_AARCH64_FUNC_END |
| 86 | 86 | |
| 87 | − | |
| 88 | 87 | WELS_ASM_AARCH64_FUNC_BEGIN DyadicBilinearDownsamplerWidthx32_AArch64_neon |
| 89 | 88 | sub w9, w3, w4 |
| 90 | 89 | sub w1, w1, w4, lsr #1 |
| @@ -123,6 +122,113 @@ comp_ds_bilinear_w_x32_loop1: · 32 unmodified lines | ||
| 123 | 122 | cbnz w5, comp_ds_bilinear_w_x32_loop0 |
| 124 | 123 | WELS_ASM_AARCH64_FUNC_END |
| 125 | 124 | |
| 125 | +WELS_ASM_AARCH64_FUNC_BEGIN DyadicBilinearOneThirdDownsampler_AArch64_neon | |
| 126 | + | |
| 127 | + //Initialize the register | |
| 128 | + mov x6, x2 | |
| 129 | + mov x8, x0 | |
| 130 | + mov w9, #0 | |
| 131 | + | |
| 132 | + //Save the tailer for the unasigned size | |
| 133 | + smaddl x7, w1, w5, x0 | |
| 134 | + ld1 {v16.16b}, [x7] | |
| 135 | + | |
| 136 | + add x7, x2, w3, sxtw | |
| 137 | + //processing a colume data | |
| 138 | +comp_ds_bilinear_onethird_loop0: | |
| 139 | + | |
| 140 | + ld3 {v0.16b, v1.16b, v2.16b}, [x2], #48 | |
| 141 | + ld3 {v4.16b, v5.16b, v6.16b}, [x7], #48 | |
| 142 | + | |
| 143 | + uaddl v2.8h, v0.8b, v1.8b | |
| 144 | + uaddl2 v3.8h, v0.16b, v1.16b | |
| 145 | + uaddl v6.8h, v4.8b, v5.8b | |
| 146 | + uaddl2 v7.8h, v4.16b, v5.16b | |
| 147 | + urshr v2.8h, v2.8h, #1 | |
| 148 | + urshr v3.8h, v3.8h, #1 | |
| 149 | + urshr v6.8h, v6.8h, #1 | |
| 150 | + urshr v7.8h, v7.8h, #1 | |
| 151 | + | |
| 152 | + urhadd v0.8h, v2.8h, v6.8h | |
| 153 | + urhadd v1.8h, v3.8h, v7.8h | |
| 154 | + xtn v0.8b, v0.8h | |
| 155 | + xtn v1.8b, v1.8h | |
| 156 | + st1 {v0.8b,v1.8b}, [x0], #16 | |
| 157 | + | |
| 158 | + add w9, w9, #48 | |
| 159 | + | |
| 160 | + cmp w9, w4 | |
| 161 | + b.cc comp_ds_bilinear_onethird_loop0 | |
| 162 | + | |
| 163 | + mov w9, #0 | |
| 164 | + add x6, x6, w3, sxtw #1 | |
| 165 | + add x6, x6, w3, sxtw | |
| 166 | + mov x2, x6 | |
| 167 | + add x7, x2, w3, sxtw | |
| 168 | + add x8, x8, w1, sxtw | |
| 169 | + mov x0, x8 | |
| 170 | + sub w5, w5, #1 | |
| 171 | + | |
| 172 | + cbnz w5, comp_ds_bilinear_onethird_loop0 | |
| 173 | + | |
| 174 | + //restore the tailer for the unasigned size | |
| 175 | + st1 {v16.16b}, [x0] | |
| 176 | +WELS_ASM_AARCH64_FUNC_END | |
| 177 | + | |
| 178 | +WELS_ASM_AARCH64_FUNC_BEGIN DyadicBilinearQuarterDownsampler_AArch64_neon | |
| 179 | + //Initialize the register | |
| 180 | + mov x6, x2 | |
| 181 | + mov x8, x0 | |
| 182 | + mov w9, #0 | |
| 183 | + lsr w5, w5, #2 | |
| 184 | + | |
| 185 | + //Save the tailer for the unasigned size | |
| 186 | + smaddl x7, w1, w5, x0 | |
| 187 | + ld1 {v16.16b}, [x7] | |
| 188 | + | |
| 189 | + add x7, x2, w3, sxtw | |
| 190 | + //processing a colume data | |
| 191 | +comp_ds_bilinear_quarter_loop0: | |
| 192 | + | |
| 193 | + ld2 {v0.8h, v1.8h}, [x2], #32 | |
| 194 | + ld2 {v2.8h, v3.8h}, [x2], #32 | |
| 195 | + ld2 {v4.8h, v5.8h}, [x7], #32 | |
| 196 | + ld2 {v6.8h, v7.8h}, [x7], #32 | |
| 197 | + | |
| 198 | + uaddlp v0.8h, v0.16b | |
| 199 | + uaddlp v1.8h, v2.16b | |
| 200 | + uaddlp v4.8h, v4.16b | |
| 201 | + uaddlp v5.8h, v6.16b | |
| 202 | + urshr v0.8h, v0.8h, #1 | |
| 203 | + urshr v1.8h, v1.8h, #1 | |
| 204 | + urshr v4.8h, v4.8h, #1 | |
| 205 | + urshr v5.8h, v5.8h, #1 | |
| 206 | + | |
| 207 | + urhadd v0.8h, v0.8h, v4.8h | |
| 208 | + urhadd v1.8h, v1.8h, v5.8h | |
| 209 | + xtn v0.8b, v0.8h | |
| 210 | + xtn v1.8b, v1.8h | |
| 211 | + st1 {v0.8b,v1.8b}, [x0], #16 | |
| 212 | + | |
| 213 | + add w9, w9, #64 | |
| 214 | + | |
| 215 | + cmp w9, w4 | |
| 216 | + b.cc comp_ds_bilinear_quarter_loop0 | |
| 217 | + | |
| 218 | + mov w9, #0 | |
| 219 | + add x6, x6, w3, sxtw #2 | |
| 220 | + mov x2, x6 | |
| 221 | + add x7, x2, w3, sxtw | |
| 222 | + add x8, x8, w1, sxtw | |
| 223 | + mov x0, x8 | |
| 224 | + sub w5, w5, #1 | |
| 225 | + | |
| 226 | + cbnz w5, comp_ds_bilinear_quarter_loop0 | |
| 227 | + | |
| 228 | + //restore the tailer for the unasigned size | |
| 229 | + st1 {v16.16b}, [x0] | |
| 230 | +WELS_ASM_AARCH64_FUNC_END | |
| 231 | + | |
| 126 | 232 | WELS_ASM_AARCH64_FUNC_BEGIN GeneralBilinearAccurateDownsampler_AArch64_neon |
| 127 | 233 | mov w10, #32767 |
| 128 | 234 | and w8, w6, w10 |