| 1 |
/* |
| 2 |
* Copyright © 2018, VideoLAN and dav1d authors |
| 3 |
* Copyright © 2020, Martin Storsjo |
| 4 |
* Copyright © 2019, B Krishnan Iyer |
| 5 |
* All rights reserved. |
| 6 |
* |
| 7 |
* Redistribution and use in source and binary forms, with or without |
| 8 |
* modification, are permitted provided that the following conditions are met: |
| 9 |
* |
| 10 |
* 1. Redistributions of source code must retain the above copyright notice, this |
| 11 |
* list of conditions and the following disclaimer. |
| 12 |
* |
| 13 |
* 2. Redistributions in binary form must reproduce the above copyright notice, |
| 14 |
* this list of conditions and the following disclaimer in the documentation |
| 15 |
* and/or other materials provided with the distribution. |
| 16 |
* |
| 17 |
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND |
| 18 |
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED |
| 19 |
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
| 20 |
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR |
| 21 |
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES |
| 22 |
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 23 |
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND |
| 24 |
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 25 |
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
| 26 |
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 27 |
*/ |
| 28 |
|
| 29 |
#include "src/arm/asm.S" |
| 30 |
#include "util.S" |
| 31 |
|
| 32 |
// void ipred_dc_128_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 33 |
// const pixel *const topleft, |
| 34 |
// const int width, const int height, const int a, |
| 35 |
// const int max_width, const int max_height); |
| 36 |
function ipred_dc_128_8bpc_neon, export=1 |
| 37 |
push {r4, lr} |
| 38 |
ldr r4, [sp, #8] |
| 39 |
clz r3, r3 |
| 40 |
adr r2, L(ipred_dc_128_tbl) |
| 41 |
sub r3, r3, #25 |
| 42 |
ldr r3, [r2, r3, lsl #2] |
| 43 |
vmov.i8 q0, #128 |
| 44 |
add r2, r2, r3 |
| 45 |
add r12, r0, r1 |
| 46 |
lsl r1, r1, #1 |
| 47 |
bx r2 |
| 48 |
|
| 49 |
.align 2 |
| 50 |
L(ipred_dc_128_tbl): |
| 51 |
.word 640f - L(ipred_dc_128_tbl) + CONFIG_THUMB |
| 52 |
.word 320f - L(ipred_dc_128_tbl) + CONFIG_THUMB |
| 53 |
.word 16f - L(ipred_dc_128_tbl) + CONFIG_THUMB |
| 54 |
.word 8f - L(ipred_dc_128_tbl) + CONFIG_THUMB |
| 55 |
.word 4f - L(ipred_dc_128_tbl) + CONFIG_THUMB |
| 56 |
4: |
| 57 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 58 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 59 |
subs r4, r4, #4 |
| 60 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 61 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 62 |
bgt 4b |
| 63 |
pop {r4, pc} |
| 64 |
8: |
| 65 |
vst1.8 {d0}, [r0, :64], r1 |
| 66 |
vst1.8 {d0}, [r12, :64], r1 |
| 67 |
subs r4, r4, #4 |
| 68 |
vst1.8 {d0}, [r0, :64], r1 |
| 69 |
vst1.8 {d0}, [r12, :64], r1 |
| 70 |
bgt 8b |
| 71 |
pop {r4, pc} |
| 72 |
16: |
| 73 |
vst1.8 {d0, d1}, [r0, :128], r1 |
| 74 |
vst1.8 {d0, d1}, [r12, :128], r1 |
| 75 |
subs r4, r4, #4 |
| 76 |
vst1.8 {d0, d1}, [r0, :128], r1 |
| 77 |
vst1.8 {d0, d1}, [r12, :128], r1 |
| 78 |
bgt 16b |
| 79 |
pop {r4, pc} |
| 80 |
320: |
| 81 |
vmov.i8 q1, #128 |
| 82 |
32: |
| 83 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 84 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 85 |
subs r4, r4, #4 |
| 86 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 87 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 88 |
bgt 32b |
| 89 |
pop {r4, pc} |
| 90 |
640: |
| 91 |
vmov.i8 q1, #128 |
| 92 |
sub r1, r1, #32 |
| 93 |
64: |
| 94 |
vst1.8 {d0, d1, d2, d3}, [r0, :128]! |
| 95 |
vst1.8 {d0, d1, d2, d3}, [r12, :128]! |
| 96 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 97 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 98 |
subs r4, r4, #4 |
| 99 |
vst1.8 {d0, d1, d2, d3}, [r0, :128]! |
| 100 |
vst1.8 {d0, d1, d2, d3}, [r12, :128]! |
| 101 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 102 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 103 |
bgt 64b |
| 104 |
pop {r4, pc} |
| 105 |
endfunc |
| 106 |
|
| 107 |
// void ipred_v_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 108 |
// const pixel *const topleft, |
| 109 |
// const int width, const int height, const int a, |
| 110 |
// const int max_width, const int max_height); |
| 111 |
function ipred_v_8bpc_neon, export=1 |
| 112 |
push {r4, lr} |
| 113 |
ldr lr, [sp, #8] |
| 114 |
clz r3, r3 |
| 115 |
adr r4, L(ipred_v_tbl) |
| 116 |
sub r3, r3, #25 |
| 117 |
ldr r3, [r4, r3, lsl #2] |
| 118 |
add r2, r2, #1 |
| 119 |
add r4, r4, r3 |
| 120 |
add r12, r0, r1 |
| 121 |
lsl r1, r1, #1 |
| 122 |
bx r4 |
| 123 |
|
| 124 |
.align 2 |
| 125 |
L(ipred_v_tbl): |
| 126 |
.word 640f - L(ipred_v_tbl) + CONFIG_THUMB |
| 127 |
.word 320f - L(ipred_v_tbl) + CONFIG_THUMB |
| 128 |
.word 160f - L(ipred_v_tbl) + CONFIG_THUMB |
| 129 |
.word 80f - L(ipred_v_tbl) + CONFIG_THUMB |
| 130 |
.word 40f - L(ipred_v_tbl) + CONFIG_THUMB |
| 131 |
40: |
| 132 |
vld1.32 {d0[]}, [r2] |
| 133 |
4: |
| 134 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 135 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 136 |
subs lr, lr, #4 |
| 137 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 138 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 139 |
bgt 4b |
| 140 |
pop {r4, pc} |
| 141 |
80: |
| 142 |
vld1.8 {d0}, [r2] |
| 143 |
8: |
| 144 |
vst1.8 {d0}, [r0, :64], r1 |
| 145 |
vst1.8 {d0}, [r12, :64], r1 |
| 146 |
subs lr, lr, #4 |
| 147 |
vst1.8 {d0}, [r0, :64], r1 |
| 148 |
vst1.8 {d0}, [r12, :64], r1 |
| 149 |
bgt 8b |
| 150 |
pop {r4, pc} |
| 151 |
160: |
| 152 |
vld1.8 {q0}, [r2] |
| 153 |
16: |
| 154 |
vst1.8 {d0, d1}, [r0, :128], r1 |
| 155 |
vst1.8 {d0, d1}, [r12, :128], r1 |
| 156 |
subs lr, lr, #4 |
| 157 |
vst1.8 {d0, d1}, [r0, :128], r1 |
| 158 |
vst1.8 {d0, d1}, [r12, :128], r1 |
| 159 |
bgt 16b |
| 160 |
pop {r4, pc} |
| 161 |
320: |
| 162 |
vld1.8 {q0, q1}, [r2] |
| 163 |
32: |
| 164 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 165 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 166 |
subs lr, lr, #4 |
| 167 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 168 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 169 |
bgt 32b |
| 170 |
pop {r4, pc} |
| 171 |
640: |
| 172 |
vld1.8 {q0, q1}, [r2]! |
| 173 |
sub r1, r1, #32 |
| 174 |
vld1.8 {q2, q3}, [r2] |
| 175 |
64: |
| 176 |
vst1.8 {d0, d1, d2, d3}, [r0, :128]! |
| 177 |
vst1.8 {d0, d1, d2, d3}, [r12, :128]! |
| 178 |
vst1.8 {d4, d5, d6, d7}, [r0, :128], r1 |
| 179 |
vst1.8 {d4, d5, d6, d7}, [r12, :128], r1 |
| 180 |
subs lr, lr, #4 |
| 181 |
vst1.8 {d0, d1, d2, d3}, [r0, :128]! |
| 182 |
vst1.8 {d0, d1, d2, d3}, [r12, :128]! |
| 183 |
vst1.8 {d4, d5, d6, d7}, [r0, :128], r1 |
| 184 |
vst1.8 {d4, d5, d6, d7}, [r12, :128], r1 |
| 185 |
bgt 64b |
| 186 |
pop {r4, pc} |
| 187 |
endfunc |
| 188 |
|
| 189 |
// void ipred_h_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 190 |
// const pixel *const topleft, |
| 191 |
// const int width, const int height, const int a, |
| 192 |
// const int max_width, const int max_height); |
| 193 |
function ipred_h_8bpc_neon, export=1 |
| 194 |
push {r4-r5, lr} |
| 195 |
ldr r4, [sp, #12] |
| 196 |
clz r3, r3 |
| 197 |
adr r5, L(ipred_h_tbl) |
| 198 |
sub r3, r3, #25 |
| 199 |
ldr r3, [r5, r3, lsl #2] |
| 200 |
sub r2, r2, #4 |
| 201 |
mov lr, #-4 |
| 202 |
add r5, r5, r3 |
| 203 |
add r12, r0, r1 |
| 204 |
lsl r1, r1, #1 |
| 205 |
bx r5 |
| 206 |
|
| 207 |
.align 2 |
| 208 |
L(ipred_h_tbl): |
| 209 |
.word 640f - L(ipred_h_tbl) + CONFIG_THUMB |
| 210 |
.word 320f - L(ipred_h_tbl) + CONFIG_THUMB |
| 211 |
.word 160f - L(ipred_h_tbl) + CONFIG_THUMB |
| 212 |
.word 8f - L(ipred_h_tbl) + CONFIG_THUMB |
| 213 |
.word 4f - L(ipred_h_tbl) + CONFIG_THUMB |
| 214 |
4: |
| 215 |
vld4.8 {d0[], d1[], d2[], d3[]}, [r2, :32], lr |
| 216 |
vst1.32 {d3[0]}, [r0, :32], r1 |
| 217 |
vst1.32 {d2[0]}, [r12, :32], r1 |
| 218 |
subs r4, r4, #4 |
| 219 |
vst1.32 {d1[0]}, [r0, :32], r1 |
| 220 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 221 |
bgt 4b |
| 222 |
pop {r4-r5, pc} |
| 223 |
8: |
| 224 |
vld4.8 {d0[], d1[], d2[], d3[]}, [r2, :32], lr |
| 225 |
vst1.8 {d3}, [r0, :64], r1 |
| 226 |
vst1.8 {d2}, [r12, :64], r1 |
| 227 |
subs r4, r4, #4 |
| 228 |
vst1.8 {d1}, [r0, :64], r1 |
| 229 |
vst1.8 {d0}, [r12, :64], r1 |
| 230 |
bgt 8b |
| 231 |
pop {r4-r5, pc} |
| 232 |
160: |
| 233 |
add r2, r2, #3 |
| 234 |
mov lr, #-1 |
| 235 |
16: |
| 236 |
vld1.8 {d0[], d1[]}, [r2], lr |
| 237 |
subs r4, r4, #4 |
| 238 |
vld1.8 {d2[], d3[]}, [r2], lr |
| 239 |
vst1.8 {q0}, [r0, :128], r1 |
| 240 |
vld1.8 {d4[], d5[]}, [r2], lr |
| 241 |
vst1.8 {q1}, [r12, :128], r1 |
| 242 |
vld1.8 {d6[], d7[]}, [r2], lr |
| 243 |
vst1.8 {q2}, [r0, :128], r1 |
| 244 |
vst1.8 {q3}, [r12, :128], r1 |
| 245 |
bgt 16b |
| 246 |
pop {r4-r5, pc} |
| 247 |
320: |
| 248 |
add r2, r2, #3 |
| 249 |
mov lr, #-1 |
| 250 |
sub r1, r1, #16 |
| 251 |
32: |
| 252 |
vld1.8 {d0[], d1[]}, [r2], lr |
| 253 |
subs r4, r4, #4 |
| 254 |
vld1.8 {d2[], d3[]}, [r2], lr |
| 255 |
vst1.8 {q0}, [r0, :128]! |
| 256 |
vld1.8 {d4[], d5[]}, [r2], lr |
| 257 |
vst1.8 {q1}, [r12, :128]! |
| 258 |
vld1.8 {d6[], d7[]}, [r2], lr |
| 259 |
vst1.8 {q0}, [r0, :128], r1 |
| 260 |
vst1.8 {q1}, [r12, :128], r1 |
| 261 |
vst1.8 {q2}, [r0, :128]! |
| 262 |
vst1.8 {q3}, [r12, :128]! |
| 263 |
vst1.8 {q2}, [r0, :128], r1 |
| 264 |
vst1.8 {q3}, [r12, :128], r1 |
| 265 |
bgt 32b |
| 266 |
pop {r4-r5, pc} |
| 267 |
640: |
| 268 |
add r2, r2, #3 |
| 269 |
mov lr, #-1 |
| 270 |
sub r1, r1, #48 |
| 271 |
64: |
| 272 |
vld1.8 {d0[], d1[]}, [r2], lr |
| 273 |
subs r4, r4, #4 |
| 274 |
vld1.8 {d2[], d3[]}, [r2], lr |
| 275 |
vst1.8 {q0}, [r0, :128]! |
| 276 |
vld1.8 {d4[], d5[]}, [r2], lr |
| 277 |
vst1.8 {q1}, [r12, :128]! |
| 278 |
vld1.8 {d6[], d7[]}, [r2], lr |
| 279 |
vst1.8 {q0}, [r0, :128]! |
| 280 |
vst1.8 {q1}, [r12, :128]! |
| 281 |
vst1.8 {q0}, [r0, :128]! |
| 282 |
vst1.8 {q1}, [r12, :128]! |
| 283 |
vst1.8 {q0}, [r0, :128], r1 |
| 284 |
vst1.8 {q1}, [r12, :128], r1 |
| 285 |
vst1.8 {q2}, [r0, :128]! |
| 286 |
vst1.8 {q3}, [r12, :128]! |
| 287 |
vst1.8 {q2}, [r0, :128]! |
| 288 |
vst1.8 {q3}, [r12, :128]! |
| 289 |
vst1.8 {q2}, [r0, :128]! |
| 290 |
vst1.8 {q3}, [r12, :128]! |
| 291 |
vst1.8 {q2}, [r0, :128], r1 |
| 292 |
vst1.8 {q3}, [r12, :128], r1 |
| 293 |
bgt 64b |
| 294 |
pop {r4-r5, pc} |
| 295 |
endfunc |
| 296 |
|
| 297 |
// void ipred_dc_top_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 298 |
// const pixel *const topleft, |
| 299 |
// const int width, const int height, const int a, |
| 300 |
// const int max_width, const int max_height); |
| 301 |
function ipred_dc_top_8bpc_neon, export=1 |
| 302 |
push {r4-r5, lr} |
| 303 |
ldr r4, [sp, #12] |
| 304 |
clz r3, r3 |
| 305 |
adr r5, L(ipred_dc_top_tbl) |
| 306 |
sub r3, r3, #25 |
| 307 |
ldr r3, [r5, r3, lsl #2] |
| 308 |
add r2, r2, #1 |
| 309 |
add r5, r5, r3 |
| 310 |
add r12, r0, r1 |
| 311 |
lsl r1, r1, #1 |
| 312 |
bx r5 |
| 313 |
|
| 314 |
.align 2 |
| 315 |
L(ipred_dc_top_tbl): |
| 316 |
.word 640f - L(ipred_dc_top_tbl) + CONFIG_THUMB |
| 317 |
.word 320f - L(ipred_dc_top_tbl) + CONFIG_THUMB |
| 318 |
.word 160f - L(ipred_dc_top_tbl) + CONFIG_THUMB |
| 319 |
.word 80f - L(ipred_dc_top_tbl) + CONFIG_THUMB |
| 320 |
.word 40f - L(ipred_dc_top_tbl) + CONFIG_THUMB |
| 321 |
40: |
| 322 |
vld1.32 {d0[]}, [r2] |
| 323 |
vpaddl.u8 d0, d0 |
| 324 |
vpadd.u16 d0, d0 |
| 325 |
vrshrn.u16 d0, q0, #2 |
| 326 |
vdup.8 d0, d0[0] |
| 327 |
4: |
| 328 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 329 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 330 |
subs r4, r4, #4 |
| 331 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 332 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 333 |
bgt 4b |
| 334 |
pop {r4-r5, pc} |
| 335 |
80: |
| 336 |
vld1.8 {d0}, [r2] |
| 337 |
vpaddl.u8 d0, d0 |
| 338 |
vpadd.u16 d0, d0 |
| 339 |
vpadd.u16 d0, d0 |
| 340 |
vrshrn.u16 d0, q0, #3 |
| 341 |
vdup.8 d0, d0[0] |
| 342 |
8: |
| 343 |
vst1.8 {d0}, [r0, :64], r1 |
| 344 |
vst1.8 {d0}, [r12, :64], r1 |
| 345 |
subs r4, r4, #4 |
| 346 |
vst1.8 {d0}, [r0, :64], r1 |
| 347 |
vst1.8 {d0}, [r12, :64], r1 |
| 348 |
bgt 8b |
| 349 |
pop {r4-r5, pc} |
| 350 |
160: |
| 351 |
vld1.8 {d0, d1}, [r2] |
| 352 |
vaddl.u8 q0, d0, d1 |
| 353 |
vadd.u16 d0, d0, d1 |
| 354 |
vpadd.u16 d0, d0 |
| 355 |
vpadd.u16 d0, d0 |
| 356 |
vrshrn.u16 d0, q0, #4 |
| 357 |
vdup.8 q0, d0[0] |
| 358 |
16: |
| 359 |
vst1.8 {d0, d1}, [r0, :128], r1 |
| 360 |
vst1.8 {d0, d1}, [r12, :128], r1 |
| 361 |
subs r4, r4, #4 |
| 362 |
vst1.8 {d0, d1}, [r0, :128], r1 |
| 363 |
vst1.8 {d0, d1}, [r12, :128], r1 |
| 364 |
bgt 16b |
| 365 |
pop {r4-r5, pc} |
| 366 |
320: |
| 367 |
vld1.8 {d0, d1, d2, d3}, [r2] |
| 368 |
vaddl.u8 q0, d0, d1 |
| 369 |
vaddl.u8 q1, d2, d3 |
| 370 |
vadd.u16 q0, q0, q1 |
| 371 |
vadd.u16 d0, d0, d1 |
| 372 |
vpadd.u16 d0, d0 |
| 373 |
vpadd.u16 d0, d0 |
| 374 |
vrshrn.u16 d4, q0, #5 |
| 375 |
vdup.8 q0, d4[0] |
| 376 |
vdup.8 q1, d4[0] |
| 377 |
32: |
| 378 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 379 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 380 |
subs r4, r4, #4 |
| 381 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 382 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 383 |
bgt 32b |
| 384 |
pop {r4-r5, pc} |
| 385 |
640: |
| 386 |
vld1.8 {d0, d1, d2, d3}, [r2]! |
| 387 |
vaddl.u8 q0, d0, d1 |
| 388 |
vld1.8 {d4, d5, d6, d7}, [r2] |
| 389 |
vaddl.u8 q1, d2, d3 |
| 390 |
vaddl.u8 q2, d4, d5 |
| 391 |
vaddl.u8 q3, d6, d7 |
| 392 |
vadd.u16 q0, q0, q1 |
| 393 |
vadd.u16 q1, q2, q3 |
| 394 |
vadd.u16 q0, q0, q1 |
| 395 |
vadd.u16 d0, d0, d1 |
| 396 |
vpadd.u16 d0, d0 |
| 397 |
vpadd.u16 d0, d0 |
| 398 |
vrshrn.u16 d18, q0, #6 |
| 399 |
vdup.8 q0, d18[0] |
| 400 |
vdup.8 q1, d18[0] |
| 401 |
sub r1, r1, #32 |
| 402 |
64: |
| 403 |
vst1.8 {d0, d1, d2, d3}, [r0, :128]! |
| 404 |
vst1.8 {d0, d1, d2, d3}, [r12, :128]! |
| 405 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 406 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 407 |
subs r4, r4, #4 |
| 408 |
vst1.8 {d0, d1, d2, d3}, [r0, :128]! |
| 409 |
vst1.8 {d0, d1, d2, d3}, [r12, :128]! |
| 410 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 411 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 412 |
bgt 64b |
| 413 |
pop {r4-r5, pc} |
| 414 |
endfunc |
| 415 |
|
| 416 |
// void ipred_dc_left_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 417 |
// const pixel *const topleft, |
| 418 |
// const int width, const int height, const int a, |
| 419 |
// const int max_width, const int max_height); |
| 420 |
function ipred_dc_left_8bpc_neon, export=1 |
| 421 |
push {r4-r5, lr} |
| 422 |
ldr r4, [sp, #12] |
| 423 |
sub r2, r2, r4 |
| 424 |
clz r3, r3 |
| 425 |
clz lr, r4 |
| 426 |
sub lr, lr, #25 |
| 427 |
adr r5, L(ipred_dc_left_tbl) |
| 428 |
sub r3, r3, #20 |
| 429 |
ldr r3, [r5, r3, lsl #2] |
| 430 |
ldr lr, [r5, lr, lsl #2] |
| 431 |
add r3, r5, r3 |
| 432 |
add r5, r5, lr |
| 433 |
add r12, r0, r1 |
| 434 |
lsl r1, r1, #1 |
| 435 |
bx r5 |
| 436 |
|
| 437 |
.align 2 |
| 438 |
L(ipred_dc_left_tbl): |
| 439 |
.word L(ipred_dc_left_h64) - L(ipred_dc_left_tbl) + CONFIG_THUMB |
| 440 |
.word L(ipred_dc_left_h32) - L(ipred_dc_left_tbl) + CONFIG_THUMB |
| 441 |
.word L(ipred_dc_left_h16) - L(ipred_dc_left_tbl) + CONFIG_THUMB |
| 442 |
.word L(ipred_dc_left_h8) - L(ipred_dc_left_tbl) + CONFIG_THUMB |
| 443 |
.word L(ipred_dc_left_h4) - L(ipred_dc_left_tbl) + CONFIG_THUMB |
| 444 |
.word L(ipred_dc_left_w64) - L(ipred_dc_left_tbl) + CONFIG_THUMB |
| 445 |
.word L(ipred_dc_left_w32) - L(ipred_dc_left_tbl) + CONFIG_THUMB |
| 446 |
.word L(ipred_dc_left_w16) - L(ipred_dc_left_tbl) + CONFIG_THUMB |
| 447 |
.word L(ipred_dc_left_w8) - L(ipred_dc_left_tbl) + CONFIG_THUMB |
| 448 |
.word L(ipred_dc_left_w4) - L(ipred_dc_left_tbl) + CONFIG_THUMB |
| 449 |
|
| 450 |
L(ipred_dc_left_h4): |
| 451 |
vld1.32 {d0[]}, [r2, :32] |
| 452 |
vpaddl.u8 d0, d0 |
| 453 |
vpadd.u16 d0, d0 |
| 454 |
vrshrn.u16 d0, q0, #2 |
| 455 |
vdup.8 q0, d0[0] |
| 456 |
bx r3 |
| 457 |
L(ipred_dc_left_w4): |
| 458 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 459 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 460 |
subs r4, r4, #4 |
| 461 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 462 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 463 |
bgt L(ipred_dc_left_w4) |
| 464 |
pop {r4-r5, pc} |
| 465 |
L(ipred_dc_left_h8): |
| 466 |
vld1.8 {d0}, [r2, :64] |
| 467 |
vpaddl.u8 d0, d0 |
| 468 |
vpadd.u16 d0, d0 |
| 469 |
vpadd.u16 d0, d0 |
| 470 |
vrshrn.u16 d0, q0, #3 |
| 471 |
vdup.8 q0, d0[0] |
| 472 |
bx r3 |
| 473 |
L(ipred_dc_left_w8): |
| 474 |
vst1.8 {d0}, [r0, :64], r1 |
| 475 |
vst1.8 {d0}, [r12, :64], r1 |
| 476 |
subs r4, r4, #4 |
| 477 |
vst1.8 {d0}, [r0, :64], r1 |
| 478 |
vst1.8 {d0}, [r12, :64], r1 |
| 479 |
bgt L(ipred_dc_left_w8) |
| 480 |
pop {r4-r5, pc} |
| 481 |
L(ipred_dc_left_h16): |
| 482 |
vld1.8 {d0, d1}, [r2, :128] |
| 483 |
vaddl.u8 q0, d0, d1 |
| 484 |
vadd.u16 d0, d0, d1 |
| 485 |
vpadd.u16 d0, d0 |
| 486 |
vpadd.u16 d0, d0 |
| 487 |
vrshrn.u16 d0, q0, #4 |
| 488 |
vdup.8 q0, d0[0] |
| 489 |
bx r3 |
| 490 |
L(ipred_dc_left_w16): |
| 491 |
vst1.8 {d0, d1}, [r0, :128], r1 |
| 492 |
vst1.8 {d0, d1}, [r12, :128], r1 |
| 493 |
subs r4, r4, #4 |
| 494 |
vst1.8 {d0, d1}, [r0, :128], r1 |
| 495 |
vst1.8 {d0, d1}, [r12, :128], r1 |
| 496 |
bgt L(ipred_dc_left_w16) |
| 497 |
pop {r4-r5, pc} |
| 498 |
L(ipred_dc_left_h32): |
| 499 |
vld1.8 {d0, d1, d2, d3}, [r2, :128] |
| 500 |
vaddl.u8 q0, d0, d1 |
| 501 |
vaddl.u8 q1, d2, d3 |
| 502 |
vadd.u16 q0, q0, q1 |
| 503 |
vadd.u16 d0, d0, d1 |
| 504 |
vpadd.u16 d0, d0 |
| 505 |
vpadd.u16 d0, d0 |
| 506 |
vrshrn.u16 d0, q0, #5 |
| 507 |
vdup.8 q0, d0[0] |
| 508 |
bx r3 |
| 509 |
L(ipred_dc_left_w32): |
| 510 |
vmov.8 q1, q0 |
| 511 |
1: |
| 512 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 513 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 514 |
subs r4, r4, #4 |
| 515 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 516 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 517 |
bgt 1b |
| 518 |
pop {r4-r5, pc} |
| 519 |
L(ipred_dc_left_h64): |
| 520 |
vld1.8 {d0, d1, d2, d3}, [r2, :128]! |
| 521 |
vld1.8 {d4, d5, d6, d7}, [r2, :128] |
| 522 |
vaddl.u8 q0, d0, d1 |
| 523 |
vaddl.u8 q1, d2, d3 |
| 524 |
vaddl.u8 q2, d4, d5 |
| 525 |
vaddl.u8 q3, d6, d7 |
| 526 |
vadd.u16 q0, q0, q1 |
| 527 |
vadd.u16 q1, q2, q3 |
| 528 |
vadd.u16 q0, q0, q1 |
| 529 |
vadd.u16 d0, d0, d1 |
| 530 |
vpadd.u16 d0, d0 |
| 531 |
vpadd.u16 d0, d0 |
| 532 |
vrshrn.u16 d0, q0, #6 |
| 533 |
vdup.8 q0, d0[0] |
| 534 |
bx r3 |
| 535 |
L(ipred_dc_left_w64): |
| 536 |
vmov.8 q1, q0 |
| 537 |
sub r1, r1, #32 |
| 538 |
1: |
| 539 |
vst1.8 {d0, d1, d2, d3}, [r0, :128]! |
| 540 |
vst1.8 {d0, d1, d2, d3}, [r12, :128]! |
| 541 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 542 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 543 |
subs r4, r4, #4 |
| 544 |
vst1.8 {d0, d1, d2, d3}, [r0, :128]! |
| 545 |
vst1.8 {d0, d1, d2, d3}, [r12, :128]! |
| 546 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 547 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 548 |
bgt 1b |
| 549 |
pop {r4-r5, pc} |
| 550 |
endfunc |
| 551 |
|
| 552 |
// void ipred_dc_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 553 |
// const pixel *const topleft, |
| 554 |
// const int width, const int height, const int a, |
| 555 |
// const int max_width, const int max_height); |
| 556 |
function ipred_dc_8bpc_neon, export=1 |
| 557 |
push {r4-r6, lr} |
| 558 |
ldr r4, [sp, #16] |
| 559 |
sub r2, r2, r4 |
| 560 |
add lr, r3, r4 // width + height |
| 561 |
clz r3, r3 |
| 562 |
clz r12, r4 |
| 563 |
vdup.16 q15, lr // width + height |
| 564 |
adr r5, L(ipred_dc_tbl) |
| 565 |
rbit lr, lr // rbit(width + height) |
| 566 |
sub r3, r3, #20 // 25 leading bits, minus table offset 5 |
| 567 |
sub r12, r12, #25 |
| 568 |
clz lr, lr // ctz(width + height) |
| 569 |
ldr r3, [r5, r3, lsl #2] |
| 570 |
ldr r12, [r5, r12, lsl #2] |
| 571 |
neg lr, lr // -ctz(width + height) |
| 572 |
add r3, r5, r3 |
| 573 |
add r5, r5, r12 |
| 574 |
vshr.u16 q15, q15, #1 // (width + height) >> 1 |
| 575 |
vdup.16 q14, lr // -ctz(width + height) |
| 576 |
add r12, r0, r1 |
| 577 |
lsl r1, r1, #1 |
| 578 |
bx r5 |
| 579 |
|
| 580 |
.align 2 |
| 581 |
L(ipred_dc_tbl): |
| 582 |
.word L(ipred_dc_h64) - L(ipred_dc_tbl) + CONFIG_THUMB |
| 583 |
.word L(ipred_dc_h32) - L(ipred_dc_tbl) + CONFIG_THUMB |
| 584 |
.word L(ipred_dc_h16) - L(ipred_dc_tbl) + CONFIG_THUMB |
| 585 |
.word L(ipred_dc_h8) - L(ipred_dc_tbl) + CONFIG_THUMB |
| 586 |
.word L(ipred_dc_h4) - L(ipred_dc_tbl) + CONFIG_THUMB |
| 587 |
.word L(ipred_dc_w64) - L(ipred_dc_tbl) + CONFIG_THUMB |
| 588 |
.word L(ipred_dc_w32) - L(ipred_dc_tbl) + CONFIG_THUMB |
| 589 |
.word L(ipred_dc_w16) - L(ipred_dc_tbl) + CONFIG_THUMB |
| 590 |
.word L(ipred_dc_w8) - L(ipred_dc_tbl) + CONFIG_THUMB |
| 591 |
.word L(ipred_dc_w4) - L(ipred_dc_tbl) + CONFIG_THUMB |
| 592 |
|
| 593 |
L(ipred_dc_h4): |
| 594 |
vld1.32 {d0[]}, [r2, :32]! |
| 595 |
vpaddl.u8 d0, d0 |
| 596 |
add r2, r2, #1 |
| 597 |
vpadd.u16 d0, d0 |
| 598 |
bx r3 |
| 599 |
L(ipred_dc_w4): |
| 600 |
vld1.32 {d1[]}, [r2] |
| 601 |
vadd.s16 d0, d0, d30 |
| 602 |
vpaddl.u8 d1, d1 |
| 603 |
vpadd.u16 d1, d1 |
| 604 |
cmp r4, #4 |
| 605 |
vadd.s16 d0, d0, d1 |
| 606 |
vshl.u16 d0, d0, d28 |
| 607 |
beq 1f |
| 608 |
// h = 8/16 |
| 609 |
movw lr, #(0x3334/2) |
| 610 |
movw r5, #(0x5556/2) |
| 611 |
cmp r4, #16 |
| 612 |
it ne |
| 613 |
movne lr, r5 |
| 614 |
vdup.16 d30, lr |
| 615 |
vqdmulh.s16 d0, d0, d30 |
| 616 |
1: |
| 617 |
vdup.8 d0, d0[0] |
| 618 |
2: |
| 619 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 620 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 621 |
subs r4, r4, #4 |
| 622 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 623 |
vst1.32 {d0[0]}, [r12, :32], r1 |
| 624 |
bgt 2b |
| 625 |
pop {r4-r6, pc} |
| 626 |
|
| 627 |
L(ipred_dc_h8): |
| 628 |
vld1.8 {d0}, [r2, :64]! |
| 629 |
vpaddl.u8 d0, d0 |
| 630 |
vpadd.u16 d0, d0 |
| 631 |
add r2, r2, #1 |
| 632 |
vpadd.u16 d0, d0 |
| 633 |
bx r3 |
| 634 |
L(ipred_dc_w8): |
| 635 |
vld1.8 {d2}, [r2] |
| 636 |
vadd.s16 d0, d0, d30 |
| 637 |
vpaddl.u8 d2, d2 |
| 638 |
vpadd.u16 d2, d2 |
| 639 |
vpadd.u16 d2, d2 |
| 640 |
cmp r4, #8 |
| 641 |
vadd.s16 d0, d0, d2 |
| 642 |
vshl.u16 d0, d0, d28 |
| 643 |
beq 1f |
| 644 |
// h = 4/16/32 |
| 645 |
cmp r4, #32 |
| 646 |
movw lr, #(0x3334/2) |
| 647 |
movw r5, #(0x5556/2) |
| 648 |
it ne |
| 649 |
movne lr, r5 |
| 650 |
vdup.16 d24, lr |
| 651 |
vqdmulh.s16 d0, d0, d24 |
| 652 |
1: |
| 653 |
vdup.8 d0, d0[0] |
| 654 |
2: |
| 655 |
vst1.8 {d0}, [r0, :64], r1 |
| 656 |
vst1.8 {d0}, [r12, :64], r1 |
| 657 |
subs r4, r4, #4 |
| 658 |
vst1.8 {d0}, [r0, :64], r1 |
| 659 |
vst1.8 {d0}, [r12, :64], r1 |
| 660 |
bgt 2b |
| 661 |
pop {r4-r6, pc} |
| 662 |
|
| 663 |
L(ipred_dc_h16): |
| 664 |
vld1.8 {d0, d1}, [r2, :128]! |
| 665 |
vaddl.u8 q0, d0, d1 |
| 666 |
vadd.u16 d0, d0, d1 |
| 667 |
vpadd.u16 d0, d0 |
| 668 |
add r2, r2, #1 |
| 669 |
vpadd.u16 d0, d0 |
| 670 |
bx r3 |
| 671 |
L(ipred_dc_w16): |
| 672 |
vld1.8 {d2, d3}, [r2] |
| 673 |
vadd.s16 d0, d0, d30 |
| 674 |
vaddl.u8 q1, d2, d3 |
| 675 |
vadd.u16 d2, d2, d3 |
| 676 |
vpadd.u16 d2, d2 |
| 677 |
vpadd.u16 d2, d2 |
| 678 |
cmp r4, #16 |
| 679 |
vadd.s16 d0, d0, d2 |
| 680 |
vshl.u16 d0, d0, d28 |
| 681 |
beq 1f |
| 682 |
// h = 4/8/32/64 |
| 683 |
tst r4, #(32+16+8) // 16 added to make a consecutive bitmask |
| 684 |
movw lr, #(0x3334/2) |
| 685 |
movw r5, #(0x5556/2) |
| 686 |
it ne |
| 687 |
movne lr, r5 |
| 688 |
vdup.16 d24, lr |
| 689 |
vqdmulh.s16 d0, d0, d24 |
| 690 |
1: |
| 691 |
vdup.8 q0, d0[0] |
| 692 |
2: |
| 693 |
vst1.8 {d0, d1}, [r0, :128], r1 |
| 694 |
vst1.8 {d0, d1}, [r12, :128], r1 |
| 695 |
subs r4, r4, #4 |
| 696 |
vst1.8 {d0, d1}, [r0, :128], r1 |
| 697 |
vst1.8 {d0, d1}, [r12, :128], r1 |
| 698 |
bgt 2b |
| 699 |
pop {r4-r6, pc} |
| 700 |
|
| 701 |
L(ipred_dc_h32): |
| 702 |
vld1.8 {d0, d1, d2, d3}, [r2, :128]! |
| 703 |
vaddl.u8 q0, d0, d1 |
| 704 |
vaddl.u8 q1, d2, d3 |
| 705 |
vadd.u16 q0, q0, q1 |
| 706 |
vadd.u16 d0, d0, d1 |
| 707 |
vpadd.u16 d0, d0 |
| 708 |
add r2, r2, #1 |
| 709 |
vpadd.u16 d0, d0 |
| 710 |
bx r3 |
| 711 |
L(ipred_dc_w32): |
| 712 |
vld1.8 {d2, d3, d4, d5}, [r2] |
| 713 |
vadd.s16 d0, d0, d30 |
| 714 |
vaddl.u8 q1, d2, d3 |
| 715 |
vaddl.u8 q2, d4, d5 |
| 716 |
vadd.u16 q1, q1, q2 |
| 717 |
vadd.u16 d2, d2, d3 |
| 718 |
vpadd.u16 d2, d2 |
| 719 |
vpadd.u16 d2, d2 |
| 720 |
cmp r4, #32 |
| 721 |
vadd.s16 d0, d0, d2 |
| 722 |
vshl.u16 d4, d0, d28 |
| 723 |
beq 1f |
| 724 |
// h = 8/16/64 |
| 725 |
cmp r4, #8 |
| 726 |
movw lr, #(0x3334/2) |
| 727 |
movw r5, #(0x5556/2) |
| 728 |
it ne |
| 729 |
movne lr, r5 |
| 730 |
vdup.16 d24, lr |
| 731 |
vqdmulh.s16 d4, d4, d24 |
| 732 |
1: |
| 733 |
vdup.8 q0, d4[0] |
| 734 |
vdup.8 q1, d4[0] |
| 735 |
2: |
| 736 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 737 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 738 |
subs r4, r4, #4 |
| 739 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 740 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 741 |
bgt 2b |
| 742 |
pop {r4-r6, pc} |
| 743 |
|
| 744 |
L(ipred_dc_h64): |
| 745 |
vld1.8 {d0, d1, d2, d3}, [r2, :128]! |
| 746 |
vaddl.u8 q0, d0, d1 |
| 747 |
vld1.8 {d4, d5, d6, d7}, [r2, :128]! |
| 748 |
vaddl.u8 q1, d2, d3 |
| 749 |
vaddl.u8 q2, d4, d5 |
| 750 |
vaddl.u8 q3, d6, d7 |
| 751 |
vadd.u16 q0, q0, q1 |
| 752 |
vadd.u16 q1, q2, q3 |
| 753 |
vadd.u16 q0, q0, q1 |
| 754 |
vadd.u16 d0, d0, d1 |
| 755 |
vpadd.u16 d0, d0 |
| 756 |
add r2, r2, #1 |
| 757 |
vpadd.u16 d0, d0 |
| 758 |
bx r3 |
| 759 |
L(ipred_dc_w64): |
| 760 |
vld1.8 {d2, d3, d4, d5}, [r2]! |
| 761 |
vadd.s16 d0, d0, d30 |
| 762 |
vaddl.u8 q2, d4, d5 |
| 763 |
vaddl.u8 q1, d2, d3 |
| 764 |
vadd.u16 d4, d4, d5 |
| 765 |
vadd.u16 d2, d2, d3 |
| 766 |
vld1.8 {d16, d17, d18, d19}, [r2] |
| 767 |
vpadd.u16 d4, d4 |
| 768 |
vpadd.u16 d2, d2 |
| 769 |
vpadd.u16 d4, d4 |
| 770 |
vpadd.u16 d2, d2 |
| 771 |
vaddl.u8 q8, d16, d17 |
| 772 |
vaddl.u8 q9, d18, d19 |
| 773 |
vadd.u16 d16, d16, d17 |
| 774 |
vadd.u16 d18, d18, d19 |
| 775 |
vpadd.u16 d16, d16 |
| 776 |
vpadd.u16 d18, d18 |
| 777 |
vpadd.u16 d16, d16 |
| 778 |
vpadd.u16 d18, d18 |
| 779 |
vadd.u16 d2, d2, d4 |
| 780 |
vadd.u16 d3, d16, d18 |
| 781 |
cmp r4, #64 |
| 782 |
vadd.s16 d0, d0, d2 |
| 783 |
vadd.s16 d0, d0, d3 |
| 784 |
vshl.u16 d18, d0, d28 |
| 785 |
beq 1f |
| 786 |
// h = 16/32 |
| 787 |
movw lr, #(0x5556/2) |
| 788 |
movt lr, #(0x3334/2) |
| 789 |
and r5, r4, #31 |
| 790 |
lsr lr, lr, r5 |
| 791 |
vdup.16 d30, lr |
| 792 |
vqdmulh.s16 d18, d18, d30 |
| 793 |
1: |
| 794 |
sub r1, r1, #32 |
| 795 |
vdup.8 q0, d18[0] |
| 796 |
vdup.8 q1, d18[0] |
| 797 |
2: |
| 798 |
vst1.8 {d0, d1, d2, d3}, [r0, :128]! |
| 799 |
vst1.8 {d0, d1, d2, d3}, [r12, :128]! |
| 800 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 801 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 802 |
subs r4, r4, #4 |
| 803 |
vst1.8 {d0, d1, d2, d3}, [r0, :128]! |
| 804 |
vst1.8 {d0, d1, d2, d3}, [r12, :128]! |
| 805 |
vst1.8 {d0, d1, d2, d3}, [r0, :128], r1 |
| 806 |
vst1.8 {d0, d1, d2, d3}, [r12, :128], r1 |
| 807 |
bgt 2b |
| 808 |
pop {r4-r6, pc} |
| 809 |
endfunc |
| 810 |
|
| 811 |
// void ipred_paeth_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 812 |
// const pixel *const topleft, |
| 813 |
// const int width, const int height, const int a, |
| 814 |
// const int max_width, const int max_height); |
| 815 |
function ipred_paeth_8bpc_neon, export=1 |
| 816 |
push {r4-r8, lr} |
| 817 |
ldr r4, [sp, #24] |
| 818 |
clz lr, r3 |
| 819 |
adr r5, L(ipred_paeth_tbl) |
| 820 |
sub lr, lr, #25 |
| 821 |
ldr lr, [r5, lr, lsl #2] |
| 822 |
vld1.8 {d4[], d5[]}, [r2] |
| 823 |
add r8, r2, #1 |
| 824 |
sub r2, r2, #4 |
| 825 |
add r5, r5, lr |
| 826 |
mov r7, #-4 |
| 827 |
add r6, r0, r1 |
| 828 |
lsl r1, r1, #1 |
| 829 |
bx r5 |
| 830 |
|
| 831 |
.align 2 |
| 832 |
L(ipred_paeth_tbl): |
| 833 |
.word 640f - L(ipred_paeth_tbl) + CONFIG_THUMB |
| 834 |
.word 320f - L(ipred_paeth_tbl) + CONFIG_THUMB |
| 835 |
.word 160f - L(ipred_paeth_tbl) + CONFIG_THUMB |
| 836 |
.word 80f - L(ipred_paeth_tbl) + CONFIG_THUMB |
| 837 |
.word 40f - L(ipred_paeth_tbl) + CONFIG_THUMB |
| 838 |
|
| 839 |
40: |
| 840 |
vld1.32 {d6[], d7[]}, [r8] |
| 841 |
vsubl.u8 q8, d6, d4 // top - topleft |
| 842 |
4: |
| 843 |
vld4.8 {d0[], d1[], d2[], d3[]}, [r2, :32], r7 |
| 844 |
vzip.32 d0, d1 |
| 845 |
vzip.32 d2, d3 |
| 846 |
vaddw.u8 q9, q8, d0 |
| 847 |
vaddw.u8 q10, q8, d2 |
| 848 |
vqmovun.s16 d18, q9 // base |
| 849 |
vqmovun.s16 d19, q10 |
| 850 |
vmov d1, d2 |
| 851 |
vabd.u8 q10, q3, q9 // tdiff |
| 852 |
vabd.u8 q11, q2, q9 // tldiff |
| 853 |
vabd.u8 q9, q0, q9 // ldiff |
| 854 |
vmin.u8 q12, q10, q11 // min(tdiff, tldiff) |
| 855 |
vcge.u8 q10, q11, q10 // tldiff >= tdiff |
| 856 |
vcge.u8 q9, q12, q9 // min(tdiff, tldiff) >= ldiff |
| 857 |
vbsl q10, q3, q2 // tdiff <= tldiff ? top : topleft |
| 858 |
vbit q10, q0, q9 // ldiff <= min ? left : ... |
| 859 |
vst1.32 {d21[1]}, [r0, :32], r1 |
| 860 |
vst1.32 {d21[0]}, [r6, :32], r1 |
| 861 |
subs r4, r4, #4 |
| 862 |
vst1.32 {d20[1]}, [r0, :32], r1 |
| 863 |
vst1.32 {d20[0]}, [r6, :32], r1 |
| 864 |
bgt 4b |
| 865 |
pop {r4-r8, pc} |
| 866 |
80: |
| 867 |
vld1.8 {d6}, [r8] |
| 868 |
vsubl.u8 q8, d6, d4 // top - topleft |
| 869 |
vmov d7, d6 |
| 870 |
8: |
| 871 |
vld4.8 {d0[], d1[], d2[], d3[]}, [r2, :32], r7 |
| 872 |
vaddw.u8 q9, q8, d0 |
| 873 |
vaddw.u8 q10, q8, d1 |
| 874 |
vaddw.u8 q11, q8, d2 |
| 875 |
vaddw.u8 q12, q8, d3 |
| 876 |
vqmovun.s16 d18, q9 // base |
| 877 |
vqmovun.s16 d19, q10 |
| 878 |
vqmovun.s16 d20, q11 |
| 879 |
vqmovun.s16 d21, q12 |
| 880 |
vabd.u8 q11, q3, q9 // tdiff |
| 881 |
vabd.u8 q12, q3, q10 |
| 882 |
vabd.u8 q13, q2, q9 // tldiff |
| 883 |
vabd.u8 q14, q2, q10 |
| 884 |
vabd.u8 q10, q1, q10 // ldiff |
| 885 |
vabd.u8 q9, q0, q9 |
| 886 |
vmin.u8 q15, q12, q14 // min(tdiff, tldiff) |
| 887 |
vcge.u8 q12, q14, q12 // tldiff >= tdiff |
| 888 |
vmin.u8 q14, q11, q13 // min(tdiff, tldiff) |
| 889 |
vcge.u8 q11, q13, q11 // tldiff >= tdiff |
| 890 |
vcge.u8 q10, q15, q10 // min(tdiff, tldiff) >= ldiff |
| 891 |
vcge.u8 q9, q14, q9 |
| 892 |
vbsl q12, q3, q2 // tdiff <= tldiff ? top : topleft |
| 893 |
vbsl q11, q3, q2 |
| 894 |
vbit q12, q1, q10 // ldiff <= min ? left : ... |
| 895 |
vbit q11, q0, q9 |
| 896 |
vst1.8 {d25}, [r0, :64], r1 |
| 897 |
vst1.8 {d24}, [r6, :64], r1 |
| 898 |
subs r4, r4, #4 |
| 899 |
vst1.8 {d23}, [r0, :64], r1 |
| 900 |
vst1.8 {d22}, [r6, :64], r1 |
| 901 |
bgt 8b |
| 902 |
pop {r4-r8, pc} |
| 903 |
160: |
| 904 |
320: |
| 905 |
640: |
| 906 |
vld1.8 {d6}, [r8]! |
| 907 |
mov r12, r3 |
| 908 |
// Set up pointers for four rows in parallel; r0, r6, r5, lr |
| 909 |
add r5, r0, r1 |
| 910 |
add lr, r6, r1 |
| 911 |
lsl r1, r1, #1 |
| 912 |
sub r1, r1, r3 |
| 913 |
1: |
| 914 |
vld4.8 {d0[], d1[], d2[], d3[]}, [r2, :32], r7 |
| 915 |
2: |
| 916 |
vsubl.u8 q8, d6, d4 // top - topleft |
| 917 |
vmov d7, d6 |
| 918 |
vaddw.u8 q9, q8, d0 |
| 919 |
vaddw.u8 q10, q8, d1 |
| 920 |
vaddw.u8 q11, q8, d2 |
| 921 |
vaddw.u8 q12, q8, d3 |
| 922 |
vqmovun.s16 d18, q9 // base |
| 923 |
vqmovun.s16 d19, q10 |
| 924 |
vqmovun.s16 d20, q11 |
| 925 |
vqmovun.s16 d21, q12 |
| 926 |
vabd.u8 q11, q3, q9 // tdiff |
| 927 |
vabd.u8 q12, q3, q10 |
| 928 |
vabd.u8 q13, q2, q9 // tldiff |
| 929 |
vabd.u8 q14, q2, q10 |
| 930 |
vabd.u8 q10, q1, q10 // ldiff |
| 931 |
vabd.u8 q9, q0, q9 |
| 932 |
vmin.u8 q15, q12, q14 // min(tdiff, tldiff) |
| 933 |
vcge.u8 q12, q14, q12 // tldiff >= tdiff |
| 934 |
vmin.u8 q14, q11, q13 // min(tdiff, tldiff) |
| 935 |
vcge.u8 q11, q13, q11 // tldiff >= tdiff |
| 936 |
vcge.u8 q10, q15, q10 // min(tdiff, tldiff) >= ldiff |
| 937 |
vcge.u8 q9, q14, q9 |
| 938 |
vbsl q12, q3, q2 // tdiff <= tldiff ? top : topleft |
| 939 |
vbsl q11, q3, q2 |
| 940 |
vbit q12, q1, q10 // ldiff <= min ? left : ... |
| 941 |
vbit q11, q0, q9 |
| 942 |
subs r3, r3, #8 |
| 943 |
vst1.8 {d25}, [r0, :64]! |
| 944 |
vst1.8 {d24}, [r6, :64]! |
| 945 |
vst1.8 {d23}, [r5, :64]! |
| 946 |
vst1.8 {d22}, [lr, :64]! |
| 947 |
ble 8f |
| 948 |
vld1.8 {d6}, [r8]! |
| 949 |
b 2b |
| 950 |
8: |
| 951 |
subs r4, r4, #4 |
| 952 |
ble 9f |
| 953 |
// End of horizontal loop, move pointers to next four rows |
| 954 |
sub r8, r8, r12 |
| 955 |
add r0, r0, r1 |
| 956 |
add r6, r6, r1 |
| 957 |
vld1.8 {d6}, [r8]! |
| 958 |
add r5, r5, r1 |
| 959 |
add lr, lr, r1 |
| 960 |
mov r3, r12 |
| 961 |
b 1b |
| 962 |
9: |
| 963 |
pop {r4-r8, pc} |
| 964 |
endfunc |
| 965 |
|
| 966 |
// void ipred_smooth_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 967 |
// const pixel *const topleft, |
| 968 |
// const int width, const int height, const int a, |
| 969 |
// const int max_width, const int max_height); |
| 970 |
function ipred_smooth_8bpc_neon, export=1 |
| 971 |
push {r4-r10, lr} |
| 972 |
ldr r4, [sp, #32] |
| 973 |
movrel r10, X(sm_weights) |
| 974 |
add r12, r10, r4 |
| 975 |
add r10, r10, r3 |
| 976 |
clz r9, r3 |
| 977 |
adr r5, L(ipred_smooth_tbl) |
| 978 |
sub lr, r2, r4 |
| 979 |
sub r9, r9, #25 |
| 980 |
ldr r9, [r5, r9, lsl #2] |
| 981 |
vld1.8 {d4[]}, [lr] // bottom |
| 982 |
add r8, r2, #1 |
| 983 |
add r5, r5, r9 |
| 984 |
add r6, r0, r1 |
| 985 |
lsl r1, r1, #1 |
| 986 |
bx r5 |
| 987 |
|
| 988 |
.align 2 |
| 989 |
L(ipred_smooth_tbl): |
| 990 |
.word 640f - L(ipred_smooth_tbl) + CONFIG_THUMB |
| 991 |
.word 320f - L(ipred_smooth_tbl) + CONFIG_THUMB |
| 992 |
.word 160f - L(ipred_smooth_tbl) + CONFIG_THUMB |
| 993 |
.word 80f - L(ipred_smooth_tbl) + CONFIG_THUMB |
| 994 |
.word 40f - L(ipred_smooth_tbl) + CONFIG_THUMB |
| 995 |
|
| 996 |
40: |
| 997 |
vld1.32 {d16[]}, [r8] // top |
| 998 |
vld1.32 {d18[]}, [r10, :32] // weights_hor |
| 999 |
sub r2, r2, #4 |
| 1000 |
mov r7, #-4 |
| 1001 |
vdup.8 q3, d16[3] // right |
| 1002 |
vsubl.u8 q8, d16, d4 // top-bottom |
| 1003 |
vmovl.u8 q9, d18 // weights_hor |
| 1004 |
4: |
| 1005 |
vld4.8 {d0[], d1[], d2[], d3[]}, [r2, :32], r7 // left |
| 1006 |
vld4.8 {d20[], d21[], d22[], d23[]}, [r12, :32]! // weights_ver |
| 1007 |
vshll.i8 q12, d6, #8 // right*256 |
| 1008 |
vshll.i8 q13, d6, #8 |
| 1009 |
vzip.32 d1, d0 // left, flipped |
| 1010 |
vzip.32 d3, d2 |
| 1011 |
vzip.32 d20, d21 // weights_ver |
| 1012 |
vzip.32 d22, d23 |
| 1013 |
vshll.i8 q14, d4, #8 // bottom*256 |
| 1014 |
vshll.i8 q15, d4, #8 |
| 1015 |
vsubl.u8 q0, d1, d6 // left-right |
| 1016 |
vsubl.u8 q1, d3, d6 |
| 1017 |
vmovl.u8 q10, d20 // weights_ver |
| 1018 |
vmovl.u8 q11, d22 |
| 1019 |
vmla.i16 q12, q1, q9 // right*256 + (left-right)*weights_hor |
| 1020 |
vmla.i16 q13, q0, q9 // (left flipped) |
| 1021 |
vmla.i16 q14, q8, q10 // bottom*256 + (top-bottom)*weights_ver |
| 1022 |
vmla.i16 q15, q8, q11 |
| 1023 |
vhadd.u16 q12, q12, q14 |
| 1024 |
vhadd.u16 q13, q13, q15 |
| 1025 |
vrshrn.i16 d24, q12, #8 |
| 1026 |
vrshrn.i16 d25, q13, #8 |
| 1027 |
vst1.32 {d24[0]}, [r0, :32], r1 |
| 1028 |
vst1.32 {d24[1]}, [r6, :32], r1 |
| 1029 |
subs r4, r4, #4 |
| 1030 |
vst1.32 {d25[0]}, [r0, :32], r1 |
| 1031 |
vst1.32 {d25[1]}, [r6, :32], r1 |
| 1032 |
bgt 4b |
| 1033 |
pop {r4-r10, pc} |
| 1034 |
80: |
| 1035 |
vld1.8 {d16}, [r8] // top |
| 1036 |
vld1.8 {d18}, [r10, :64] // weights_hor |
| 1037 |
sub r2, r2, #2 |
| 1038 |
mov r7, #-2 |
| 1039 |
vdup.8 q3, d16[7] // right |
| 1040 |
vsubl.u8 q8, d16, d4 // top-bottom |
| 1041 |
vmovl.u8 q9, d18 // weights_hor |
| 1042 |
8: |
| 1043 |
vld2.8 {d0[], d1[]}, [r2, :16], r7 // left |
| 1044 |
vld2.8 {d20[], d22[]}, [r12, :16]! // weights_ver |
| 1045 |
vshll.i8 q12, d6, #8 // right*256 |
| 1046 |
vshll.i8 q13, d6, #8 |
| 1047 |
vshll.i8 q14, d4, #8 // bottom*256 |
| 1048 |
vshll.i8 q15, d4, #8 |
| 1049 |
vsubl.u8 q1, d0, d6 // left-right (left flipped) |
| 1050 |
vsubl.u8 q0, d1, d6 |
| 1051 |
vmovl.u8 q10, d20 // weights_ver |
| 1052 |
vmovl.u8 q11, d22 |
| 1053 |
vmla.i16 q12, q0, q9 // right*256 + (left-right)*weights_hor |
| 1054 |
vmla.i16 q13, q1, q9 |
| 1055 |
vmla.i16 q14, q8, q10 // bottom*256 + (top-bottom)*weights_ver |
| 1056 |
vmla.i16 q15, q8, q11 |
| 1057 |
vhadd.u16 q12, q12, q14 |
| 1058 |
vhadd.u16 q13, q13, q15 |
| 1059 |
vrshrn.i16 d24, q12, #8 |
| 1060 |
vrshrn.i16 d25, q13, #8 |
| 1061 |
subs r4, r4, #2 |
| 1062 |
vst1.8 {d24}, [r0, :64], r1 |
| 1063 |
vst1.8 {d25}, [r6, :64], r1 |
| 1064 |
bgt 8b |
| 1065 |
pop {r4-r10, pc} |
| 1066 |
160: |
| 1067 |
320: |
| 1068 |
640: |
| 1069 |
add lr, r2, r3 |
| 1070 |
sub r2, r2, #2 |
| 1071 |
mov r7, #-2 |
| 1072 |
vld1.8 {d6[], d7[]}, [lr] // right |
| 1073 |
sub r1, r1, r3 |
| 1074 |
mov r9, r3 |
| 1075 |
|
| 1076 |
1: |
| 1077 |
vld2.8 {d0[], d1[]}, [r2, :16], r7 // left |
| 1078 |
vld2.8 {d20[], d22[]}, [r12, :16]! // weights_ver |
| 1079 |
vsubl.u8 q1, d0, d6 // left-right (left flipped) |
| 1080 |
vsubl.u8 q0, d1, d6 |
| 1081 |
vmovl.u8 q10, d20 // weights_ver |
| 1082 |
vmovl.u8 q11, d22 |
| 1083 |
2: |
| 1084 |
vld1.8 {d16}, [r8]! // top |
| 1085 |
vld1.8 {d18}, [r10, :64]! // weights_hor |
| 1086 |
vshll.i8 q12, d6, #8 // right*256 |
| 1087 |
vshll.i8 q13, d6, #8 |
| 1088 |
vmovl.u8 q9, d18 // weights_hor |
| 1089 |
vshll.i8 q14, d4, #8 // bottom*256 |
| 1090 |
vshll.i8 q15, d4, #8 |
| 1091 |
vsubl.u8 q8, d16, d4 // top-bottom |
| 1092 |
vmla.i16 q12, q0, q9 // right*256 + (left-right)*weights_hor |
| 1093 |
vmla.i16 q13, q1, q9 |
| 1094 |
vmla.i16 q14, q8, q10 // bottom*256 + (top-bottom)*weights_ver |
| 1095 |
vmla.i16 q15, q8, q11 |
| 1096 |
vhadd.u16 q12, q12, q14 |
| 1097 |
vhadd.u16 q13, q13, q15 |
| 1098 |
vrshrn.i16 d24, q12, #8 |
| 1099 |
vrshrn.i16 d25, q13, #8 |
| 1100 |
subs r3, r3, #8 |
| 1101 |
vst1.8 {d24}, [r0, :64]! |
| 1102 |
vst1.8 {d25}, [r6, :64]! |
| 1103 |
bgt 2b |
| 1104 |
subs r4, r4, #2 |
| 1105 |
ble 9f |
| 1106 |
sub r8, r8, r9 |
| 1107 |
sub r10, r10, r9 |
| 1108 |
add r0, r0, r1 |
| 1109 |
add r6, r6, r1 |
| 1110 |
mov r3, r9 |
| 1111 |
b 1b |
| 1112 |
9: |
| 1113 |
pop {r4-r10, pc} |
| 1114 |
endfunc |
| 1115 |
|
| 1116 |
// void ipred_smooth_v_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 1117 |
// const pixel *const topleft, |
| 1118 |
// const int width, const int height, const int a, |
| 1119 |
// const int max_width, const int max_height); |
| 1120 |
function ipred_smooth_v_8bpc_neon, export=1 |
| 1121 |
push {r4-r7, lr} |
| 1122 |
ldr r4, [sp, #20] |
| 1123 |
movrel r7, X(sm_weights) |
| 1124 |
add r7, r7, r4 |
| 1125 |
clz lr, r3 |
| 1126 |
adr r5, L(ipred_smooth_v_tbl) |
| 1127 |
sub r12, r2, r4 |
| 1128 |
sub lr, lr, #25 |
| 1129 |
ldr lr, [r5, lr, lsl #2] |
| 1130 |
vld1.8 {d4[]}, [r12] // bottom |
| 1131 |
add r2, r2, #1 |
| 1132 |
add r5, r5, lr |
| 1133 |
add r6, r0, r1 |
| 1134 |
lsl r1, r1, #1 |
| 1135 |
bx r5 |
| 1136 |
|
| 1137 |
.align 2 |
| 1138 |
L(ipred_smooth_v_tbl): |
| 1139 |
.word 640f - L(ipred_smooth_v_tbl) + CONFIG_THUMB |
| 1140 |
.word 320f - L(ipred_smooth_v_tbl) + CONFIG_THUMB |
| 1141 |
.word 160f - L(ipred_smooth_v_tbl) + CONFIG_THUMB |
| 1142 |
.word 80f - L(ipred_smooth_v_tbl) + CONFIG_THUMB |
| 1143 |
.word 40f - L(ipred_smooth_v_tbl) + CONFIG_THUMB |
| 1144 |
|
| 1145 |
40: |
| 1146 |
vld1.32 {d6[]}, [r2] // top |
| 1147 |
vsubl.u8 q3, d6, d4 // top-bottom |
| 1148 |
4: |
| 1149 |
vld4.8 {d16[], d17[], d18[], d19[]}, [r7, :32]! // weights_ver |
| 1150 |
vshll.i8 q10, d4, #8 // bottom*256 |
| 1151 |
vshll.i8 q11, d4, #8 |
| 1152 |
vzip.32 d16, d17 // weights_ver |
| 1153 |
vzip.32 d18, d19 |
| 1154 |
vmovl.u8 q8, d16 // weights_ver |
| 1155 |
vmovl.u8 q9, d18 |
| 1156 |
subs r4, r4, #4 |
| 1157 |
vmla.i16 q10, q3, q8 // bottom*256 + (top-bottom)*weights_ver |
| 1158 |
vmla.i16 q11, q3, q9 |
| 1159 |
vrshrn.i16 d20, q10, #8 |
| 1160 |
vrshrn.i16 d21, q11, #8 |
| 1161 |
vst1.32 {d20[0]}, [r0, :32], r1 |
| 1162 |
vst1.32 {d20[1]}, [r6, :32], r1 |
| 1163 |
vst1.32 {d21[0]}, [r0, :32], r1 |
| 1164 |
vst1.32 {d21[1]}, [r6, :32], r1 |
| 1165 |
bgt 4b |
| 1166 |
pop {r4-r7, pc} |
| 1167 |
80: |
| 1168 |
vld1.8 {d6}, [r2] // top |
| 1169 |
vsubl.u8 q3, d6, d4 // top-bottom |
| 1170 |
8: |
| 1171 |
vld4.8 {d16[], d18[], d20[], d22[]}, [r7, :32]! // weights_ver |
| 1172 |
vshll.i8 q12, d4, #8 // bottom*256 |
| 1173 |
vshll.i8 q13, d4, #8 |
| 1174 |
vshll.i8 q14, d4, #8 |
| 1175 |
vshll.i8 q15, d4, #8 |
| 1176 |
vmovl.u8 q8, d16 // weights_ver |
| 1177 |
vmovl.u8 q9, d18 |
| 1178 |
vmovl.u8 q10, d20 |
| 1179 |
vmovl.u8 q11, d22 |
| 1180 |
vmla.i16 q12, q3, q8 // bottom*256 + (top-bottom)*weights_ver |
| 1181 |
vmla.i16 q13, q3, q9 |
| 1182 |
vmla.i16 q14, q3, q10 |
| 1183 |
vmla.i16 q15, q3, q11 |
| 1184 |
vrshrn.i16 d24, q12, #8 |
| 1185 |
vrshrn.i16 d25, q13, #8 |
| 1186 |
vrshrn.i16 d26, q14, #8 |
| 1187 |
vrshrn.i16 d27, q15, #8 |
| 1188 |
vst1.8 {d24}, [r0, :64], r1 |
| 1189 |
vst1.8 {d25}, [r6, :64], r1 |
| 1190 |
subs r4, r4, #4 |
| 1191 |
vst1.8 {d26}, [r0, :64], r1 |
| 1192 |
vst1.8 {d27}, [r6, :64], r1 |
| 1193 |
bgt 8b |
| 1194 |
pop {r4-r7, pc} |
| 1195 |
160: |
| 1196 |
320: |
| 1197 |
640: |
| 1198 |
vpush {q4-q7} |
| 1199 |
// Set up pointers for four rows in parallel; r0, r6, r5, lr |
| 1200 |
add r5, r0, r1 |
| 1201 |
add lr, r6, r1 |
| 1202 |
lsl r1, r1, #1 |
| 1203 |
sub r1, r1, r3 |
| 1204 |
mov r12, r3 |
| 1205 |
|
| 1206 |
1: |
| 1207 |
vld4.8 {d8[], d10[], d12[], d14[]}, [r7, :32]! // weights_ver |
| 1208 |
vmovl.u8 q4, d8 // weights_ver |
| 1209 |
vmovl.u8 q5, d10 |
| 1210 |
vmovl.u8 q6, d12 |
| 1211 |
vmovl.u8 q7, d14 |
| 1212 |
2: |
| 1213 |
vld1.8 {q3}, [r2]! // top |
| 1214 |
vshll.i8 q8, d4, #8 // bottom*256 |
| 1215 |
vshll.i8 q9, d4, #8 |
| 1216 |
vshll.i8 q10, d4, #8 |
| 1217 |
vshll.i8 q11, d4, #8 |
| 1218 |
vsubl.u8 q0, d6, d4 // top-bottom |
| 1219 |
vsubl.u8 q1, d7, d4 |
| 1220 |
vshll.i8 q12, d4, #8 |
| 1221 |
vshll.i8 q13, d4, #8 |
| 1222 |
vshll.i8 q14, d4, #8 |
| 1223 |
vshll.i8 q15, d4, #8 |
| 1224 |
vmla.i16 q8, q0, q4 // bottom*256 + (top-bottom)*weights_ver |
| 1225 |
vmla.i16 q9, q1, q4 |
| 1226 |
vmla.i16 q10, q0, q5 |
| 1227 |
vmla.i16 q11, q1, q5 |
| 1228 |
vmla.i16 q12, q0, q6 // bottom*256 + (top-bottom)*weights_ver |
| 1229 |
vmla.i16 q13, q1, q6 |
| 1230 |
vmla.i16 q14, q0, q7 |
| 1231 |
vmla.i16 q15, q1, q7 |
| 1232 |
vrshrn.i16 d16, q8, #8 |
| 1233 |
vrshrn.i16 d17, q9, #8 |
| 1234 |
vrshrn.i16 d18, q10, #8 |
| 1235 |
vrshrn.i16 d19, q11, #8 |
| 1236 |
vrshrn.i16 d20, q12, #8 |
| 1237 |
vrshrn.i16 d21, q13, #8 |
| 1238 |
vrshrn.i16 d22, q14, #8 |
| 1239 |
vrshrn.i16 d23, q15, #8 |
| 1240 |
subs r3, r3, #16 |
| 1241 |
vst1.8 {q8}, [r0, :128]! |
| 1242 |
vst1.8 {q9}, [r6, :128]! |
| 1243 |
vst1.8 {q10}, [r5, :128]! |
| 1244 |
vst1.8 {q11}, [lr, :128]! |
| 1245 |
bgt 2b |
| 1246 |
subs r4, r4, #4 |
| 1247 |
ble 9f |
| 1248 |
sub r2, r2, r12 |
| 1249 |
add r0, r0, r1 |
| 1250 |
add r6, r6, r1 |
| 1251 |
add r5, r5, r1 |
| 1252 |
add lr, lr, r1 |
| 1253 |
mov r3, r12 |
| 1254 |
b 1b |
| 1255 |
9: |
| 1256 |
vpop {q4-q7} |
| 1257 |
pop {r4-r7, pc} |
| 1258 |
endfunc |
| 1259 |
|
| 1260 |
// void ipred_smooth_h_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 1261 |
// const pixel *const topleft, |
| 1262 |
// const int width, const int height, const int a, |
| 1263 |
// const int max_width, const int max_height); |
| 1264 |
function ipred_smooth_h_8bpc_neon, export=1 |
| 1265 |
push {r4-r8, lr} |
| 1266 |
ldr r4, [sp, #24] |
| 1267 |
movrel r8, X(sm_weights) |
| 1268 |
add r8, r8, r3 |
| 1269 |
clz lr, r3 |
| 1270 |
adr r5, L(ipred_smooth_h_tbl) |
| 1271 |
add r12, r2, r3 |
| 1272 |
sub lr, lr, #25 |
| 1273 |
ldr lr, [r5, lr, lsl #2] |
| 1274 |
vld1.8 {d4[]}, [r12] // right |
| 1275 |
add r5, r5, lr |
| 1276 |
add r6, r0, r1 |
| 1277 |
lsl r1, r1, #1 |
| 1278 |
bx r5 |
| 1279 |
|
| 1280 |
.align 2 |
| 1281 |
L(ipred_smooth_h_tbl): |
| 1282 |
.word 640f - L(ipred_smooth_h_tbl) + CONFIG_THUMB |
| 1283 |
.word 320f - L(ipred_smooth_h_tbl) + CONFIG_THUMB |
| 1284 |
.word 160f - L(ipred_smooth_h_tbl) + CONFIG_THUMB |
| 1285 |
.word 80f - L(ipred_smooth_h_tbl) + CONFIG_THUMB |
| 1286 |
.word 40f - L(ipred_smooth_h_tbl) + CONFIG_THUMB |
| 1287 |
|
| 1288 |
40: |
| 1289 |
vld1.32 {d6[]}, [r8, :32] // weights_hor |
| 1290 |
sub r2, r2, #4 |
| 1291 |
mov r7, #-4 |
| 1292 |
vmovl.u8 q3, d6 // weights_hor |
| 1293 |
4: |
| 1294 |
vld4.8 {d0[], d1[], d2[], d3[]}, [r2, :32], r7 // left |
| 1295 |
vshll.i8 q8, d4, #8 // right*256 |
| 1296 |
vshll.i8 q9, d4, #8 |
| 1297 |
vzip.32 d3, d2 // left, flipped |
| 1298 |
vzip.32 d1, d0 |
| 1299 |
vsubl.u8 q1, d3, d4 // left-right |
| 1300 |
vsubl.u8 q0, d1, d4 |
| 1301 |
subs r4, r4, #4 |
| 1302 |
vmla.i16 q8, q1, q3 // right*256 + (left-right)*weights_hor |
| 1303 |
vmla.i16 q9, q0, q3 |
| 1304 |
vrshrn.i16 d16, q8, #8 |
| 1305 |
vrshrn.i16 d17, q9, #8 |
| 1306 |
vst1.32 {d16[0]}, [r0, :32], r1 |
| 1307 |
vst1.32 {d16[1]}, [r6, :32], r1 |
| 1308 |
vst1.32 {d17[0]}, [r0, :32], r1 |
| 1309 |
vst1.32 {d17[1]}, [r6, :32], r1 |
| 1310 |
bgt 4b |
| 1311 |
pop {r4-r8, pc} |
| 1312 |
80: |
| 1313 |
vld1.8 {d6}, [r8, :64] // weights_hor |
| 1314 |
sub r2, r2, #4 |
| 1315 |
mov r7, #-4 |
| 1316 |
vmovl.u8 q3, d6 // weights_hor |
| 1317 |
8: |
| 1318 |
vld4.8 {d16[], d18[], d20[], d22[]}, [r2, :32], r7 // left |
| 1319 |
vshll.i8 q12, d4, #8 // right*256 |
| 1320 |
vshll.i8 q13, d4, #8 |
| 1321 |
vshll.i8 q14, d4, #8 |
| 1322 |
vshll.i8 q15, d4, #8 |
| 1323 |
vsubl.u8 q11, d22, d4 // left-right |
| 1324 |
vsubl.u8 q10, d20, d4 |
| 1325 |
vsubl.u8 q9, d18, d4 |
| 1326 |
vsubl.u8 q8, d16, d4 |
| 1327 |
vmla.i16 q12, q11, q3 // right*256 + (left-right)*weights_hor |
| 1328 |
vmla.i16 q13, q10, q3 // (left flipped) |
| 1329 |
vmla.i16 q14, q9, q3 |
| 1330 |
vmla.i16 q15, q8, q3 |
| 1331 |
vrshrn.i16 d24, q12, #8 |
| 1332 |
vrshrn.i16 d25, q13, #8 |
| 1333 |
vrshrn.i16 d26, q14, #8 |
| 1334 |
vrshrn.i16 d27, q15, #8 |
| 1335 |
vst1.8 {d24}, [r0, :64], r1 |
| 1336 |
vst1.8 {d25}, [r6, :64], r1 |
| 1337 |
subs r4, r4, #4 |
| 1338 |
vst1.8 {d26}, [r0, :64], r1 |
| 1339 |
vst1.8 {d27}, [r6, :64], r1 |
| 1340 |
bgt 8b |
| 1341 |
pop {r4-r8, pc} |
| 1342 |
160: |
| 1343 |
320: |
| 1344 |
640: |
| 1345 |
vpush {q4-q7} |
| 1346 |
sub r2, r2, #4 |
| 1347 |
mov r7, #-4 |
| 1348 |
// Set up pointers for four rows in parallel; r0, r6, r5, lr |
| 1349 |
add r5, r0, r1 |
| 1350 |
add lr, r6, r1 |
| 1351 |
lsl r1, r1, #1 |
| 1352 |
sub r1, r1, r3 |
| 1353 |
mov r12, r3 |
| 1354 |
|
| 1355 |
1: |
| 1356 |
vld4.8 {d8[], d10[], d12[], d14[]}, [r2, :32], r7 // left |
| 1357 |
vsubl.u8 q4, d8, d4 // left-right |
| 1358 |
vsubl.u8 q5, d10, d4 |
| 1359 |
vsubl.u8 q6, d12, d4 |
| 1360 |
vsubl.u8 q7, d14, d4 |
| 1361 |
2: |
| 1362 |
vld1.8 {q1}, [r8, :128]! // weights_hor |
| 1363 |
vshll.i8 q8, d4, #8 // right*256 |
| 1364 |
vshll.i8 q9, d4, #8 |
| 1365 |
vshll.i8 q10, d4, #8 |
| 1366 |
vshll.i8 q11, d4, #8 |
| 1367 |
vmovl.u8 q0, d2 // weights_hor |
| 1368 |
vmovl.u8 q1, d3 |
| 1369 |
vshll.i8 q12, d4, #8 |
| 1370 |
vshll.i8 q13, d4, #8 |
| 1371 |
vshll.i8 q14, d4, #8 |
| 1372 |
vshll.i8 q15, d4, #8 |
| 1373 |
vmla.i16 q8, q7, q0 // right*256 + (left-right)*weights_hor |
| 1374 |
vmla.i16 q9, q7, q1 // (left flipped) |
| 1375 |
vmla.i16 q10, q6, q0 |
| 1376 |
vmla.i16 q11, q6, q1 |
| 1377 |
vmla.i16 q12, q5, q0 |
| 1378 |
vmla.i16 q13, q5, q1 |
| 1379 |
vmla.i16 q14, q4, q0 |
| 1380 |
vmla.i16 q15, q4, q1 |
| 1381 |
vrshrn.i16 d16, q8, #8 |
| 1382 |
vrshrn.i16 d17, q9, #8 |
| 1383 |
vrshrn.i16 d18, q10, #8 |
| 1384 |
vrshrn.i16 d19, q11, #8 |
| 1385 |
vrshrn.i16 d20, q12, #8 |
| 1386 |
vrshrn.i16 d21, q13, #8 |
| 1387 |
vrshrn.i16 d22, q14, #8 |
| 1388 |
vrshrn.i16 d23, q15, #8 |
| 1389 |
subs r3, r3, #16 |
| 1390 |
vst1.8 {q8}, [r0, :128]! |
| 1391 |
vst1.8 {q9}, [r6, :128]! |
| 1392 |
vst1.8 {q10}, [r5, :128]! |
| 1393 |
vst1.8 {q11}, [lr, :128]! |
| 1394 |
bgt 2b |
| 1395 |
subs r4, r4, #4 |
| 1396 |
ble 9f |
| 1397 |
sub r8, r8, r12 |
| 1398 |
add r0, r0, r1 |
| 1399 |
add r6, r6, r1 |
| 1400 |
add r5, r5, r1 |
| 1401 |
add lr, lr, r1 |
| 1402 |
mov r3, r12 |
| 1403 |
b 1b |
| 1404 |
9: |
| 1405 |
vpop {q4-q7} |
| 1406 |
pop {r4-r8, pc} |
| 1407 |
endfunc |
| 1408 |
|
| 1409 |
// void ipred_filter_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 1410 |
// const pixel *const topleft, |
| 1411 |
// const int width, const int height, const int filt_idx, |
| 1412 |
// const int max_width, const int max_height); |
| 1413 |
function ipred_filter_8bpc_neon, export=1 |
| 1414 |
push {r4-r8, lr} |
| 1415 |
movw r12, #511 |
| 1416 |
ldrd r4, r5, [sp, #24] |
| 1417 |
and r5, r5, r12 // 511 |
| 1418 |
movrel r6, X(filter_intra_taps) |
| 1419 |
lsl r5, r5, #6 |
| 1420 |
add r6, r6, r5 |
| 1421 |
vld1.8 {d20, d21, d22, d23}, [r6, :128]! |
| 1422 |
clz lr, r3 |
| 1423 |
adr r5, L(ipred_filter_tbl) |
| 1424 |
vld1.8 {d27, d28, d29}, [r6, :64] |
| 1425 |
sub lr, lr, #26 |
| 1426 |
ldr lr, [r5, lr, lsl #2] |
| 1427 |
vmovl.s8 q8, d20 |
| 1428 |
vmovl.s8 q9, d21 |
| 1429 |
add r5, r5, lr |
| 1430 |
vmovl.s8 q10, d22 |
| 1431 |
vmovl.s8 q11, d23 |
| 1432 |
add r6, r0, r1 |
| 1433 |
lsl r1, r1, #1 |
| 1434 |
vmovl.s8 q12, d27 |
| 1435 |
vmovl.s8 q13, d28 |
| 1436 |
vmovl.s8 q14, d29 |
| 1437 |
add r8, r2, #1 |
| 1438 |
sub r2, r2, #2 |
| 1439 |
mov r7, #-2 |
| 1440 |
bx r5 |
| 1441 |
|
| 1442 |
.align 2 |
| 1443 |
L(ipred_filter_tbl): |
| 1444 |
.word 320f - L(ipred_filter_tbl) + CONFIG_THUMB |
| 1445 |
.word 160f - L(ipred_filter_tbl) + CONFIG_THUMB |
| 1446 |
.word 80f - L(ipred_filter_tbl) + CONFIG_THUMB |
| 1447 |
.word 40f - L(ipred_filter_tbl) + CONFIG_THUMB |
| 1448 |
|
| 1449 |
40: |
| 1450 |
vld1.32 {d0[]}, [r8] // top (0-3) |
| 1451 |
vmovl.u8 q0, d0 // top (0-3) |
| 1452 |
4: |
| 1453 |
vld1.32 {d2[]}, [r2], r7 // left (0-1) + topleft (2) |
| 1454 |
vmul.i16 q2, q9, d0[0] // p1(top[0]) * filter(1) |
| 1455 |
vmla.i16 q2, q10, d0[1] // p2(top[1]) * filter(2) |
| 1456 |
vmla.i16 q2, q11, d0[2] // p3(top[2]) * filter(3) |
| 1457 |
vmovl.u8 q1, d2 // left (0-1) + topleft (2) |
| 1458 |
vmla.i16 q2, q12, d0[3] // p4(top[3]) * filter(4) |
| 1459 |
vmla.i16 q2, q8, d2[2] // p0(topleft) * filter(0) |
| 1460 |
vmla.i16 q2, q13, d2[1] // p5(left[0]) * filter(5) |
| 1461 |
vmla.i16 q2, q14, d2[0] // p6(left[1]) * filter(6) |
| 1462 |
vqrshrun.s16 d4, q2, #4 |
| 1463 |
subs r4, r4, #2 |
| 1464 |
vst1.32 {d4[0]}, [r0, :32], r1 |
| 1465 |
vmovl.u8 q0, d4 |
| 1466 |
vst1.32 {d4[1]}, [r6, :32], r1 |
| 1467 |
vmov d0, d1 // move top from [4-7] to [0-3] |
| 1468 |
bgt 4b |
| 1469 |
pop {r4-r8, pc} |
| 1470 |
80: |
| 1471 |
vld1.8 {d0}, [r8] // top (0-7) |
| 1472 |
vmovl.u8 q0, d0 // top (0-7) |
| 1473 |
8: |
| 1474 |
vld1.32 {d2[]}, [r2], r7 // left (0-1) + topleft (2) |
| 1475 |
vmul.i16 q2, q9, d0[0] // p1(top[0]) * filter(1) |
| 1476 |
vmla.i16 q2, q10, d0[1] // p2(top[1]) * filter(2) |
| 1477 |
vmla.i16 q2, q11, d0[2] // p3(top[2]) * filter(3) |
| 1478 |
vmovl.u8 q1, d2 // left (0-1) + topleft (2) |
| 1479 |
vmla.i16 q2, q12, d0[3] // p4(top[3]) * filter(4) |
| 1480 |
vmla.i16 q2, q8, d2[2] // p0(topleft) * filter(0) |
| 1481 |
vmla.i16 q2, q13, d2[1] // p5(left[0]) * filter(5) |
| 1482 |
vmla.i16 q2, q14, d2[0] // p6(left[1]) * filter(6) |
| 1483 |
vmul.i16 q3, q9, d1[0] // p1(top[0]) * filter(1) |
| 1484 |
vmla.i16 q3, q10, d1[1] // p2(top[1]) * filter(2) |
| 1485 |
vmla.i16 q3, q11, d1[2] // p3(top[2]) * filter(3) |
| 1486 |
vqrshrun.s16 d4, q2, #4 |
| 1487 |
vmovl.u8 q1, d4 // first block, in 16 bit |
| 1488 |
vmla.i16 q3, q12, d1[3] // p4(top[3]) * filter(4) |
| 1489 |
vmla.i16 q3, q8, d0[3] // p0(topleft) * filter(0) |
| 1490 |
vmla.i16 q3, q13, d2[3] // p5(left[0]) * filter(5) |
| 1491 |
vmla.i16 q3, q14, d3[3] // p6(left[1]) * filter(6) |
| 1492 |
vqrshrun.s16 d5, q3, #4 |
| 1493 |
vzip.32 d4, d5 |
| 1494 |
subs r4, r4, #2 |
| 1495 |
vst1.8 {d4}, [r0, :64], r1 |
| 1496 |
vmovl.u8 q0, d5 |
| 1497 |
vst1.8 {d5}, [r6, :64], r1 |
| 1498 |
bgt 8b |
| 1499 |
pop {r4-r8, pc} |
| 1500 |
160: |
| 1501 |
320: |
| 1502 |
vpush {q4-q5} |
| 1503 |
sub r1, r1, r3 |
| 1504 |
mov lr, r3 |
| 1505 |
|
| 1506 |
1: |
| 1507 |
vld1.32 {d0[]}, [r2], r7 // left (0-1) + topleft (2) |
| 1508 |
vmovl.u8 q0, d0 // left (0-1) + topleft (2) |
| 1509 |
2: |
| 1510 |
vld1.8 {q2}, [r8]! // top(0-15) |
| 1511 |
vmul.i16 q3, q8, d0[2] // p0(topleft) * filter(0) |
| 1512 |
vmla.i16 q3, q13, d0[1] // p5(left[0]) * filter(5) |
| 1513 |
vmovl.u8 q1, d4 // top(0-7) |
| 1514 |
vmovl.u8 q2, d5 // top(8-15) |
| 1515 |
vmla.i16 q3, q14, d0[0] // p6(left[1]) * filter(6) |
| 1516 |
vmla.i16 q3, q9, d2[0] // p1(top[0]) * filter(1) |
| 1517 |
vmla.i16 q3, q10, d2[1] // p2(top[1]) * filter(2) |
| 1518 |
vmla.i16 q3, q11, d2[2] // p3(top[2]) * filter(3) |
| 1519 |
vmla.i16 q3, q12, d2[3] // p4(top[3]) * filter(4) |
| 1520 |
|
| 1521 |
vmul.i16 q4, q9, d3[0] // p1(top[0]) * filter(1) |
| 1522 |
vmla.i16 q4, q10, d3[1] // p2(top[1]) * filter(2) |
| 1523 |
vmla.i16 q4, q11, d3[2] // p3(top[2]) * filter(3) |
| 1524 |
vqrshrun.s16 d6, q3, #4 |
| 1525 |
vmovl.u8 q0, d6 // first block, in 16 bit |
| 1526 |
vmla.i16 q4, q12, d3[3] // p4(top[3]) * filter(4) |
| 1527 |
vmla.i16 q4, q8, d2[3] // p0(topleft) * filter(0) |
| 1528 |
vmla.i16 q4, q13, d0[3] // p5(left[0]) * filter(5) |
| 1529 |
vmla.i16 q4, q14, d1[3] // p6(left[1]) * filter(6) |
| 1530 |
|
| 1531 |
vmul.i16 q5, q9, d4[0] // p1(top[0]) * filter(1) |
| 1532 |
vmla.i16 q5, q10, d4[1] // p2(top[1]) * filter(2) |
| 1533 |
vmla.i16 q5, q11, d4[2] // p3(top[2]) * filter(3) |
| 1534 |
vqrshrun.s16 d7, q4, #4 |
| 1535 |
vmovl.u8 q0, d7 // second block, in 16 bit |
| 1536 |
vmla.i16 q5, q12, d4[3] // p4(top[3]) * filter(4) |
| 1537 |
vmla.i16 q5, q8, d3[3] // p0(topleft) * filter(0) |
| 1538 |
vmla.i16 q5, q13, d0[3] // p5(left[0]) * filter(5) |
| 1539 |
vmla.i16 q5, q14, d1[3] // p6(left[1]) * filter(6) |
| 1540 |
|
| 1541 |
vmul.i16 q15, q9, d5[0] // p1(top[0]) * filter(1) |
| 1542 |
vmla.i16 q15, q10, d5[1] // p2(top[1]) * filter(2) |
| 1543 |
vmla.i16 q15, q11, d5[2] // p3(top[2]) * filter(3) |
| 1544 |
vqrshrun.s16 d8, q5, #4 |
| 1545 |
vmovl.u8 q0, d8 // third block, in 16 bit |
| 1546 |
vmov.u8 r12, d5[6] |
| 1547 |
vmla.i16 q15, q12, d5[3] // p4(top[3]) * filter(4) |
| 1548 |
vmla.i16 q15, q8, d4[3] // p0(topleft) * filter(0) |
| 1549 |
vmla.i16 q15, q13, d0[3] // p5(left[0]) * filter(5) |
| 1550 |
vmla.i16 q15, q14, d1[3] // p6(left[1]) * filter(6) |
| 1551 |
vmov.8 d0[4], r12 |
| 1552 |
|
| 1553 |
subs r3, r3, #16 |
| 1554 |
vqrshrun.s16 d9, q15, #4 |
| 1555 |
|
| 1556 |
vst4.32 {d6[0], d7[0], d8[0], d9[0]}, [r0, :128]! |
| 1557 |
vst4.32 {d6[1], d7[1], d8[1], d9[1]}, [r6, :128]! |
| 1558 |
ble 8f |
| 1559 |
vmov.u8 r12, d9[7] |
| 1560 |
vmov.8 d0[0], r12 |
| 1561 |
vmov.u8 r12, d9[3] |
| 1562 |
vmov.8 d0[2], r12 |
| 1563 |
b 2b |
| 1564 |
8: |
| 1565 |
subs r4, r4, #2 |
| 1566 |
|
| 1567 |
ble 9f |
| 1568 |
sub r8, r6, lr |
| 1569 |
add r0, r0, r1 |
| 1570 |
add r6, r6, r1 |
| 1571 |
mov r3, lr |
| 1572 |
b 1b |
| 1573 |
9: |
| 1574 |
vpop {q4-q5} |
| 1575 |
pop {r4-r8, pc} |
| 1576 |
endfunc |
| 1577 |
|
| 1578 |
// void pal_pred_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 1579 |
// const uint16_t *const pal, const uint8_t *idx, |
| 1580 |
// const int w, const int h); |
| 1581 |
function pal_pred_8bpc_neon, export=1 |
| 1582 |
push {r4-r5, lr} |
| 1583 |
ldrd r4, r5, [sp, #12] |
| 1584 |
vld1.16 {q0}, [r2, :128] |
| 1585 |
clz lr, r4 |
| 1586 |
adr r12, L(pal_pred_tbl) |
| 1587 |
sub lr, lr, #25 |
| 1588 |
ldr lr, [r12, lr, lsl #2] |
| 1589 |
vmovn.i16 d0, q0 |
| 1590 |
add r12, r12, lr |
| 1591 |
add r2, r0, r1 |
| 1592 |
bx r12 |
| 1593 |
|
| 1594 |
.align 2 |
| 1595 |
L(pal_pred_tbl): |
| 1596 |
.word 640f - L(pal_pred_tbl) + CONFIG_THUMB |
| 1597 |
.word 320f - L(pal_pred_tbl) + CONFIG_THUMB |
| 1598 |
.word 160f - L(pal_pred_tbl) + CONFIG_THUMB |
| 1599 |
.word 80f - L(pal_pred_tbl) + CONFIG_THUMB |
| 1600 |
.word 40f - L(pal_pred_tbl) + CONFIG_THUMB |
| 1601 |
|
| 1602 |
40: |
| 1603 |
lsl r1, r1, #1 |
| 1604 |
4: |
| 1605 |
vld1.8 {q1}, [r3, :128]! |
| 1606 |
subs r5, r5, #4 |
| 1607 |
vtbl.8 d2, {d0}, d2 |
| 1608 |
vtbl.8 d3, {d0}, d3 |
| 1609 |
vst1.32 {d2[0]}, [r0, :32], r1 |
| 1610 |
vst1.32 {d2[1]}, [r2, :32], r1 |
| 1611 |
vst1.32 {d3[0]}, [r0, :32], r1 |
| 1612 |
vst1.32 {d3[1]}, [r2, :32], r1 |
| 1613 |
bgt 4b |
| 1614 |
pop {r4-r5, pc} |
| 1615 |
80: |
| 1616 |
lsl r1, r1, #1 |
| 1617 |
8: |
| 1618 |
vld1.8 {q1, q2}, [r3, :128]! |
| 1619 |
subs r5, r5, #4 |
| 1620 |
vtbl.8 d2, {d0}, d2 |
| 1621 |
vtbl.8 d3, {d0}, d3 |
| 1622 |
vst1.8 {d2}, [r0, :64], r1 |
| 1623 |
vtbl.8 d4, {d0}, d4 |
| 1624 |
vst1.8 {d3}, [r2, :64], r1 |
| 1625 |
vtbl.8 d5, {d0}, d5 |
| 1626 |
vst1.8 {d4}, [r0, :64], r1 |
| 1627 |
vst1.8 {d5}, [r2, :64], r1 |
| 1628 |
bgt 8b |
| 1629 |
pop {r4-r5, pc} |
| 1630 |
160: |
| 1631 |
lsl r1, r1, #1 |
| 1632 |
16: |
| 1633 |
vld1.8 {q8, q9}, [r3, :128]! |
| 1634 |
subs r5, r5, #4 |
| 1635 |
vld1.8 {q10, q11}, [r3, :128]! |
| 1636 |
vtbl.8 d16, {d0}, d16 |
| 1637 |
vtbl.8 d17, {d0}, d17 |
| 1638 |
vtbl.8 d18, {d0}, d18 |
| 1639 |
vtbl.8 d19, {d0}, d19 |
| 1640 |
vtbl.8 d20, {d0}, d20 |
| 1641 |
vtbl.8 d21, {d0}, d21 |
| 1642 |
vst1.8 {q8}, [r0, :128], r1 |
| 1643 |
vtbl.8 d22, {d0}, d22 |
| 1644 |
vst1.8 {q9}, [r2, :128], r1 |
| 1645 |
vtbl.8 d23, {d0}, d23 |
| 1646 |
vst1.8 {q10}, [r0, :128], r1 |
| 1647 |
vst1.8 {q11}, [r2, :128], r1 |
| 1648 |
bgt 16b |
| 1649 |
pop {r4-r5, pc} |
| 1650 |
320: |
| 1651 |
lsl r1, r1, #1 |
| 1652 |
32: |
| 1653 |
vld1.8 {q8, q9}, [r3, :128]! |
| 1654 |
subs r5, r5, #2 |
| 1655 |
vld1.8 {q10, q11}, [r3, :128]! |
| 1656 |
vtbl.8 d16, {d0}, d16 |
| 1657 |
vtbl.8 d17, {d0}, d17 |
| 1658 |
vtbl.8 d18, {d0}, d18 |
| 1659 |
vtbl.8 d19, {d0}, d19 |
| 1660 |
vtbl.8 d20, {d0}, d20 |
| 1661 |
vtbl.8 d21, {d0}, d21 |
| 1662 |
vst1.8 {q8, q9}, [r0, :128], r1 |
| 1663 |
vtbl.8 d22, {d0}, d22 |
| 1664 |
vtbl.8 d23, {d0}, d23 |
| 1665 |
vst1.8 {q10, q11}, [r2, :128], r1 |
| 1666 |
bgt 32b |
| 1667 |
pop {r4-r5, pc} |
| 1668 |
640: |
| 1669 |
sub r1, r1, #32 |
| 1670 |
64: |
| 1671 |
vld1.8 {q8, q9}, [r3, :128]! |
| 1672 |
subs r5, r5, #1 |
| 1673 |
vld1.8 {q10, q11}, [r3, :128]! |
| 1674 |
vtbl.8 d16, {d0}, d16 |
| 1675 |
vtbl.8 d17, {d0}, d17 |
| 1676 |
vtbl.8 d18, {d0}, d18 |
| 1677 |
vtbl.8 d19, {d0}, d19 |
| 1678 |
vtbl.8 d20, {d0}, d20 |
| 1679 |
vtbl.8 d21, {d0}, d21 |
| 1680 |
vst1.8 {q8, q9}, [r0, :128]! |
| 1681 |
vtbl.8 d22, {d0}, d22 |
| 1682 |
vtbl.8 d23, {d0}, d23 |
| 1683 |
vst1.8 {q10, q11}, [r0, :128], r1 |
| 1684 |
bgt 64b |
| 1685 |
pop {r4-r5, pc} |
| 1686 |
endfunc |
| 1687 |
|
| 1688 |
// void ipred_cfl_128_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 1689 |
// const pixel *const topleft, |
| 1690 |
// const int width, const int height, |
| 1691 |
// const int16_t *ac, const int alpha); |
| 1692 |
function ipred_cfl_128_8bpc_neon, export=1 |
| 1693 |
push {r4-r8, lr} |
| 1694 |
ldrd r4, r5, [sp, #24] |
| 1695 |
ldr r6, [sp, #32] |
| 1696 |
clz lr, r3 |
| 1697 |
adr r12, L(ipred_cfl_128_tbl) |
| 1698 |
sub lr, lr, #26 |
| 1699 |
ldr lr, [r12, lr, lsl #2] |
| 1700 |
vmov.i16 q0, #128 // dc |
| 1701 |
vdup.i16 q1, r6 // alpha |
| 1702 |
add r12, r12, lr |
| 1703 |
add r6, r0, r1 |
| 1704 |
lsl r1, r1, #1 |
| 1705 |
bx r12 |
| 1706 |
|
| 1707 |
.align 2 |
| 1708 |
L(ipred_cfl_128_tbl): |
| 1709 |
L(ipred_cfl_splat_tbl): |
| 1710 |
.word L(ipred_cfl_splat_w16) - L(ipred_cfl_128_tbl) + CONFIG_THUMB |
| 1711 |
.word L(ipred_cfl_splat_w16) - L(ipred_cfl_128_tbl) + CONFIG_THUMB |
| 1712 |
.word L(ipred_cfl_splat_w8) - L(ipred_cfl_128_tbl) + CONFIG_THUMB |
| 1713 |
.word L(ipred_cfl_splat_w4) - L(ipred_cfl_128_tbl) + CONFIG_THUMB |
| 1714 |
|
| 1715 |
L(ipred_cfl_splat_w4): |
| 1716 |
vld1.16 {q2, q3}, [r5, :128]! |
| 1717 |
vmul.i16 q2, q2, q1 // diff = ac * alpha |
| 1718 |
vmul.i16 q3, q3, q1 |
| 1719 |
vshr.s16 q8, q2, #15 // sign = diff >> 15 |
| 1720 |
vshr.s16 q9, q3, #15 |
| 1721 |
vadd.i16 q2, q2, q8 // diff + sign |
| 1722 |
vadd.i16 q3, q3, q9 |
| 1723 |
vrshr.s16 q2, q2, #6 // (diff + sign + 32) >> 6 = apply_sign() |
| 1724 |
vrshr.s16 q3, q3, #6 |
| 1725 |
vadd.i16 q2, q2, q0 // dc + apply_sign() |
| 1726 |
vadd.i16 q3, q3, q0 |
| 1727 |
vqmovun.s16 d4, q2 // iclip_pixel(dc + apply_sign()) |
| 1728 |
vqmovun.s16 d5, q3 |
| 1729 |
vst1.32 {d4[0]}, [r0, :32], r1 |
| 1730 |
vst1.32 {d4[1]}, [r6, :32], r1 |
| 1731 |
subs r4, r4, #4 |
| 1732 |
vst1.32 {d5[0]}, [r0, :32], r1 |
| 1733 |
vst1.32 {d5[1]}, [r6, :32], r1 |
| 1734 |
bgt L(ipred_cfl_splat_w4) |
| 1735 |
pop {r4-r8, pc} |
| 1736 |
L(ipred_cfl_splat_w8): |
| 1737 |
vld1.16 {q8, q9}, [r5, :128]! |
| 1738 |
vld1.16 {q10, q11}, [r5, :128]! |
| 1739 |
vmul.i16 q8, q8, q1 // diff = ac * alpha |
| 1740 |
vmul.i16 q9, q9, q1 |
| 1741 |
vmul.i16 q10, q10, q1 |
| 1742 |
vmul.i16 q11, q11, q1 |
| 1743 |
vshr.s16 q12, q8, #15 // sign = diff >> 15 |
| 1744 |
vshr.s16 q13, q9, #15 |
| 1745 |
vshr.s16 q14, q10, #15 |
| 1746 |
vshr.s16 q15, q11, #15 |
| 1747 |
vadd.i16 q8, q8, q12 // diff + sign |
| 1748 |
vadd.i16 q9, q9, q13 |
| 1749 |
vadd.i16 q10, q10, q14 |
| 1750 |
vadd.i16 q11, q11, q15 |
| 1751 |
vrshr.s16 q8, q8, #6 // (diff + sign + 32) >> 6 = apply_sign() |
| 1752 |
vrshr.s16 q9, q9, #6 |
| 1753 |
vrshr.s16 q10, q10, #6 |
| 1754 |
vrshr.s16 q11, q11, #6 |
| 1755 |
vadd.i16 q8, q8, q0 // dc + apply_sign() |
| 1756 |
vadd.i16 q9, q9, q0 |
| 1757 |
vadd.i16 q10, q10, q0 |
| 1758 |
vadd.i16 q11, q11, q0 |
| 1759 |
vqmovun.s16 d16, q8 // iclip_pixel(dc + apply_sign()) |
| 1760 |
vqmovun.s16 d17, q9 |
| 1761 |
vqmovun.s16 d18, q10 |
| 1762 |
vqmovun.s16 d19, q11 |
| 1763 |
vst1.8 {d16}, [r0, :64], r1 |
| 1764 |
vst1.8 {d17}, [r6, :64], r1 |
| 1765 |
subs r4, r4, #4 |
| 1766 |
vst1.8 {d18}, [r0, :64], r1 |
| 1767 |
vst1.8 {d19}, [r6, :64], r1 |
| 1768 |
bgt L(ipred_cfl_splat_w8) |
| 1769 |
pop {r4-r8, pc} |
| 1770 |
L(ipred_cfl_splat_w16): |
| 1771 |
add r12, r5, r3, lsl #1 |
| 1772 |
sub r1, r1, r3 |
| 1773 |
mov lr, r3 |
| 1774 |
1: |
| 1775 |
vld1.16 {q8, q9}, [r5, :128]! |
| 1776 |
vmul.i16 q8, q8, q1 // diff = ac * alpha |
| 1777 |
vld1.16 {q10, q11}, [r12, :128]! |
| 1778 |
vmul.i16 q9, q9, q1 |
| 1779 |
vmul.i16 q10, q10, q1 |
| 1780 |
vmul.i16 q11, q11, q1 |
| 1781 |
vshr.s16 q12, q8, #15 // sign = diff >> 15 |
| 1782 |
vshr.s16 q13, q9, #15 |
| 1783 |
vshr.s16 q14, q10, #15 |
| 1784 |
vshr.s16 q15, q11, #15 |
| 1785 |
vadd.i16 q8, q8, q12 // diff + sign |
| 1786 |
vadd.i16 q9, q9, q13 |
| 1787 |
vadd.i16 q10, q10, q14 |
| 1788 |
vadd.i16 q11, q11, q15 |
| 1789 |
vrshr.s16 q8, q8, #6 // (diff + sign + 32) >> 6 = apply_sign() |
| 1790 |
vrshr.s16 q9, q9, #6 |
| 1791 |
vrshr.s16 q10, q10, #6 |
| 1792 |
vrshr.s16 q11, q11, #6 |
| 1793 |
vadd.i16 q8, q8, q0 // dc + apply_sign() |
| 1794 |
vadd.i16 q9, q9, q0 |
| 1795 |
vadd.i16 q10, q10, q0 |
| 1796 |
vadd.i16 q11, q11, q0 |
| 1797 |
vqmovun.s16 d16, q8 // iclip_pixel(dc + apply_sign()) |
| 1798 |
vqmovun.s16 d17, q9 |
| 1799 |
vqmovun.s16 d18, q10 |
| 1800 |
vqmovun.s16 d19, q11 |
| 1801 |
subs r3, r3, #16 |
| 1802 |
vst1.16 {q8}, [r0, :128]! |
| 1803 |
vst1.16 {q9}, [r6, :128]! |
| 1804 |
bgt 1b |
| 1805 |
subs r4, r4, #2 |
| 1806 |
add r5, r5, lr, lsl #1 |
| 1807 |
add r12, r12, lr, lsl #1 |
| 1808 |
add r0, r0, r1 |
| 1809 |
add r6, r6, r1 |
| 1810 |
mov r3, lr |
| 1811 |
bgt 1b |
| 1812 |
pop {r4-r8, pc} |
| 1813 |
endfunc |
| 1814 |
|
| 1815 |
// void ipred_cfl_top_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 1816 |
// const pixel *const topleft, |
| 1817 |
// const int width, const int height, |
| 1818 |
// const int16_t *ac, const int alpha); |
| 1819 |
function ipred_cfl_top_8bpc_neon, export=1 |
| 1820 |
push {r4-r8, lr} |
| 1821 |
ldrd r4, r5, [sp, #24] |
| 1822 |
ldr r6, [sp, #32] |
| 1823 |
clz lr, r3 |
| 1824 |
adr r12, L(ipred_cfl_top_tbl) |
| 1825 |
sub lr, lr, #26 |
| 1826 |
ldr lr, [r12, lr, lsl #2] |
| 1827 |
vdup.16 q1, r6 // alpha |
| 1828 |
add r2, r2, #1 |
| 1829 |
add r12, r12, lr |
| 1830 |
add r6, r0, r1 |
| 1831 |
lsl r1, r1, #1 |
| 1832 |
bx r12 |
| 1833 |
|
| 1834 |
.align 2 |
| 1835 |
L(ipred_cfl_top_tbl): |
| 1836 |
.word 32f - L(ipred_cfl_top_tbl) + CONFIG_THUMB |
| 1837 |
.word 16f - L(ipred_cfl_top_tbl) + CONFIG_THUMB |
| 1838 |
.word 8f - L(ipred_cfl_top_tbl) + CONFIG_THUMB |
| 1839 |
.word 4f - L(ipred_cfl_top_tbl) + CONFIG_THUMB |
| 1840 |
|
| 1841 |
4: |
| 1842 |
vld1.32 {d0[]}, [r2] |
| 1843 |
vpaddl.u8 d0, d0 |
| 1844 |
vpadd.u16 d0, d0 |
| 1845 |
vrshr.u16 d0, d0, #2 |
| 1846 |
vdup.16 q0, d0[0] |
| 1847 |
b L(ipred_cfl_splat_w4) |
| 1848 |
8: |
| 1849 |
vld1.8 {d0}, [r2] |
| 1850 |
vpaddl.u8 d0, d0 |
| 1851 |
vpadd.u16 d0, d0 |
| 1852 |
vpadd.u16 d0, d0 |
| 1853 |
vrshr.u16 d0, d0, #3 |
| 1854 |
vdup.16 q0, d0[0] |
| 1855 |
b L(ipred_cfl_splat_w8) |
| 1856 |
16: |
| 1857 |
vld1.8 {q0}, [r2] |
| 1858 |
vaddl.u8 q0, d0, d1 |
| 1859 |
vadd.u16 d0, d0, d1 |
| 1860 |
vpadd.u16 d0, d0 |
| 1861 |
vpadd.u16 d0, d0 |
| 1862 |
vrshr.u16 d0, d0, #4 |
| 1863 |
vdup.16 q0, d0[0] |
| 1864 |
b L(ipred_cfl_splat_w16) |
| 1865 |
32: |
| 1866 |
vld1.8 {q2, q3}, [r2] |
| 1867 |
vaddl.u8 q2, d4, d5 |
| 1868 |
vaddl.u8 q3, d6, d7 |
| 1869 |
vadd.u16 q0, q2, q3 |
| 1870 |
vadd.u16 d0, d0, d1 |
| 1871 |
vpadd.u16 d0, d0 |
| 1872 |
vpadd.u16 d0, d0 |
| 1873 |
vrshr.u16 d0, d0, #5 |
| 1874 |
vdup.16 q0, d0[0] |
| 1875 |
b L(ipred_cfl_splat_w16) |
| 1876 |
endfunc |
| 1877 |
|
| 1878 |
// void ipred_cfl_left_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 1879 |
// const pixel *const topleft, |
| 1880 |
// const int width, const int height, |
| 1881 |
// const int16_t *ac, const int alpha); |
| 1882 |
function ipred_cfl_left_8bpc_neon, export=1 |
| 1883 |
push {r4-r8, lr} |
| 1884 |
ldrd r4, r5, [sp, #24] |
| 1885 |
ldr r6, [sp, #32] |
| 1886 |
sub r2, r2, r4 |
| 1887 |
clz lr, r3 |
| 1888 |
clz r8, r4 |
| 1889 |
adr r12, L(ipred_cfl_splat_tbl) |
| 1890 |
adr r7, L(ipred_cfl_left_tbl) |
| 1891 |
sub lr, lr, #26 |
| 1892 |
sub r8, r8, #26 |
| 1893 |
ldr lr, [r12, lr, lsl #2] |
| 1894 |
ldr r8, [r7, r8, lsl #2] |
| 1895 |
vdup.16 q1, r6 // alpha |
| 1896 |
add r12, r12, lr |
| 1897 |
add r7, r7, r8 |
| 1898 |
add r6, r0, r1 |
| 1899 |
lsl r1, r1, #1 |
| 1900 |
bx r7 |
| 1901 |
|
| 1902 |
.align 2 |
| 1903 |
L(ipred_cfl_left_tbl): |
| 1904 |
.word L(ipred_cfl_left_h32) - L(ipred_cfl_left_tbl) + CONFIG_THUMB |
| 1905 |
.word L(ipred_cfl_left_h16) - L(ipred_cfl_left_tbl) + CONFIG_THUMB |
| 1906 |
.word L(ipred_cfl_left_h8) - L(ipred_cfl_left_tbl) + CONFIG_THUMB |
| 1907 |
.word L(ipred_cfl_left_h4) - L(ipred_cfl_left_tbl) + CONFIG_THUMB |
| 1908 |
|
| 1909 |
L(ipred_cfl_left_h4): |
| 1910 |
vld1.32 {d0[]}, [r2, :32] |
| 1911 |
vpaddl.u8 d0, d0 |
| 1912 |
vpadd.u16 d0, d0 |
| 1913 |
vrshr.u16 d0, d0, #2 |
| 1914 |
vdup.16 q0, d0[0] |
| 1915 |
bx r12 |
| 1916 |
|
| 1917 |
L(ipred_cfl_left_h8): |
| 1918 |
vld1.8 {d0}, [r2, :64] |
| 1919 |
vpaddl.u8 d0, d0 |
| 1920 |
vpadd.u16 d0, d0 |
| 1921 |
vpadd.u16 d0, d0 |
| 1922 |
vrshr.u16 d0, d0, #3 |
| 1923 |
vdup.16 q0, d0[0] |
| 1924 |
bx r12 |
| 1925 |
|
| 1926 |
L(ipred_cfl_left_h16): |
| 1927 |
vld1.8 {q0}, [r2, :128] |
| 1928 |
vaddl.u8 q0, d0, d1 |
| 1929 |
vadd.u16 d0, d0, d1 |
| 1930 |
vpadd.u16 d0, d0 |
| 1931 |
vpadd.u16 d0, d0 |
| 1932 |
vrshr.u16 d0, d0, #4 |
| 1933 |
vdup.16 q0, d0[0] |
| 1934 |
bx r12 |
| 1935 |
|
| 1936 |
L(ipred_cfl_left_h32): |
| 1937 |
vld1.8 {q2, q3}, [r2, :128] |
| 1938 |
vaddl.u8 q2, d4, d5 |
| 1939 |
vaddl.u8 q3, d6, d7 |
| 1940 |
vadd.u16 q0, q2, q3 |
| 1941 |
vadd.u16 d0, d0, d1 |
| 1942 |
vpadd.u16 d0, d0 |
| 1943 |
vpadd.u16 d0, d0 |
| 1944 |
vrshr.u16 d0, d0, #5 |
| 1945 |
vdup.16 q0, d0[0] |
| 1946 |
bx r12 |
| 1947 |
endfunc |
| 1948 |
|
| 1949 |
// void ipred_cfl_8bpc_neon(pixel *dst, const ptrdiff_t stride, |
| 1950 |
// const pixel *const topleft, |
| 1951 |
// const int width, const int height, |
| 1952 |
// const int16_t *ac, const int alpha); |
| 1953 |
function ipred_cfl_8bpc_neon, export=1 |
| 1954 |
push {r4-r8, lr} |
| 1955 |
ldrd r4, r5, [sp, #24] |
| 1956 |
ldr r6, [sp, #32] |
| 1957 |
sub r2, r2, r4 |
| 1958 |
add r8, r3, r4 // width + height |
| 1959 |
vdup.16 q1, r6 // alpha |
| 1960 |
clz lr, r3 |
| 1961 |
clz r6, r4 |
| 1962 |
vdup.16 d16, r8 // width + height |
| 1963 |
adr r7, L(ipred_cfl_tbl) |
| 1964 |
rbit r8, r8 // rbit(width + height) |
| 1965 |
sub lr, lr, #22 // 26 leading bits, minus table offset 4 |
| 1966 |
sub r6, r6, #26 |
| 1967 |
clz r8, r8 // ctz(width + height) |
| 1968 |
ldr lr, [r7, lr, lsl #2] |
| 1969 |
ldr r6, [r7, r6, lsl #2] |
| 1970 |
neg r8, r8 // -ctz(width + height) |
| 1971 |
add r12, r7, lr |
| 1972 |
add r7, r7, r6 |
| 1973 |
vshr.u16 d16, d16, #1 // (width + height) >> 1 |
| 1974 |
vdup.16 d17, r8 // -ctz(width + height) |
| 1975 |
add r6, r0, r1 |
| 1976 |
lsl r1, r1, #1 |
| 1977 |
bx r7 |
| 1978 |
|
| 1979 |
.align 2 |
| 1980 |
L(ipred_cfl_tbl): |
| 1981 |
.word L(ipred_cfl_h32) - L(ipred_cfl_tbl) + CONFIG_THUMB |
| 1982 |
.word L(ipred_cfl_h16) - L(ipred_cfl_tbl) + CONFIG_THUMB |
| 1983 |
.word L(ipred_cfl_h8) - L(ipred_cfl_tbl) + CONFIG_THUMB |
| 1984 |
.word L(ipred_cfl_h4) - L(ipred_cfl_tbl) + CONFIG_THUMB |
| 1985 |
.word L(ipred_cfl_w32) - L(ipred_cfl_tbl) + CONFIG_THUMB |
| 1986 |
.word L(ipred_cfl_w16) - L(ipred_cfl_tbl) + CONFIG_THUMB |
| 1987 |
.word L(ipred_cfl_w8) - L(ipred_cfl_tbl) + CONFIG_THUMB |
| 1988 |
.word L(ipred_cfl_w4) - L(ipred_cfl_tbl) + CONFIG_THUMB |
| 1989 |
|
| 1990 |
L(ipred_cfl_h4): |
| 1991 |
vld1.32 {d0[]}, [r2, :32]! |
| 1992 |
vpaddl.u8 d0, d0 |
| 1993 |
add r2, r2, #1 |
| 1994 |
vpadd.i16 d0, d0 |
| 1995 |
bx r12 |
| 1996 |
L(ipred_cfl_w4): |
| 1997 |
vld1.32 {d1[]}, [r2] |
| 1998 |
vadd.i16 d0, d0, d16 |
| 1999 |
vpaddl.u8 d1, d1 |
| 2000 |
vpadd.u16 d1, d1 |
| 2001 |
cmp r4, #4 |
| 2002 |
vadd.i16 d0, d0, d1 |
| 2003 |
vshl.u16 d0, d0, d17 |
| 2004 |
beq 1f |
| 2005 |
// h = 8/16 |
| 2006 |
movw lr, #(0x3334/2) |
| 2007 |
movw r8, #(0x5556/2) |
| 2008 |
cmp r4, #16 |
| 2009 |
it ne |
| 2010 |
movne lr, r8 |
| 2011 |
vdup.16 d18, lr |
| 2012 |
vqdmulh.s16 d0, d0, d18 |
| 2013 |
1: |
| 2014 |
vdup.16 q0, d0[0] |
| 2015 |
b L(ipred_cfl_splat_w4) |
| 2016 |
|
| 2017 |
L(ipred_cfl_h8): |
| 2018 |
vld1.8 {d0}, [r2, :64]! |
| 2019 |
vpaddl.u8 d0, d0 |
| 2020 |
vpadd.i16 d0, d0 |
| 2021 |
add r2, r2, #1 |
| 2022 |
vpadd.i16 d0, d0 |
| 2023 |
bx r12 |
| 2024 |
L(ipred_cfl_w8): |
| 2025 |
vld1.8 {d1}, [r2] |
| 2026 |
vadd.i16 d0, d0, d16 |
| 2027 |
vpaddl.u8 d1, d1 |
| 2028 |
vpadd.i16 d1, d1 |
| 2029 |
vpadd.i16 d1, d1 |
| 2030 |
cmp r4, #8 |
| 2031 |
vadd.i16 d0, d0, d1 |
| 2032 |
vshl.u16 d0, d0, d17 |
| 2033 |
beq 1f |
| 2034 |
// h = 4/16/32 |
| 2035 |
cmp r4, #32 |
| 2036 |
movw lr, #(0x3334/2) |
| 2037 |
movw r8, #(0x5556/2) |
| 2038 |
it ne |
| 2039 |
movne lr, r8 |
| 2040 |
vdup.16 d18, lr |
| 2041 |
vqdmulh.s16 d0, d0, d18 |
| 2042 |
1: |
| 2043 |
vdup.16 q0, d0[0] |
| 2044 |
b L(ipred_cfl_splat_w8) |
| 2045 |
|
| 2046 |
L(ipred_cfl_h16): |
| 2047 |
vld1.8 {q0}, [r2, :128]! |
| 2048 |
vaddl.u8 q0, d0, d1 |
| 2049 |
vadd.i16 d0, d0, d1 |
| 2050 |
vpadd.i16 d0, d0 |
| 2051 |
add r2, r2, #1 |
| 2052 |
vpadd.i16 d0, d0 |
| 2053 |
bx r12 |
| 2054 |
L(ipred_cfl_w16): |
| 2055 |
vld1.8 {q2}, [r2] |
| 2056 |
vadd.i16 d0, d0, d16 |
| 2057 |
vaddl.u8 q2, d4, d5 |
| 2058 |
vadd.i16 d4, d4, d5 |
| 2059 |
vpadd.i16 d4, d4 |
| 2060 |
vpadd.i16 d4, d4 |
| 2061 |
cmp r4, #16 |
| 2062 |
vadd.i16 d0, d0, d4 |
| 2063 |
vshl.u16 d0, d0, d17 |
| 2064 |
beq 1f |
| 2065 |
// h = 4/8/32/64 |
| 2066 |
tst r4, #(32+16+8) // 16 added to make a consecutive bitmask |
| 2067 |
movw lr, #(0x3334/2) |
| 2068 |
movw r8, #(0x5556/2) |
| 2069 |
it ne |
| 2070 |
movne lr, r8 |
| 2071 |
vdup.16 d18, lr |
| 2072 |
vqdmulh.s16 d0, d0, d18 |
| 2073 |
1: |
| 2074 |
vdup.16 q0, d0[0] |
| 2075 |
b L(ipred_cfl_splat_w16) |
| 2076 |
|
| 2077 |
L(ipred_cfl_h32): |
| 2078 |
vld1.8 {q2, q3}, [r2, :128]! |
| 2079 |
vaddl.u8 q2, d4, d5 |
| 2080 |
vaddl.u8 q3, d6, d7 |
| 2081 |
vadd.i16 q0, q2, q3 |
| 2082 |
vadd.i16 d0, d0, d1 |
| 2083 |
vpadd.i16 d0, d0 |
| 2084 |
add r2, r2, #1 |
| 2085 |
vpadd.i16 d0, d0 |
| 2086 |
bx r12 |
| 2087 |
L(ipred_cfl_w32): |
| 2088 |
vld1.8 {q2, q3}, [r2] |
| 2089 |
vadd.i16 d0, d0, d16 |
| 2090 |
vaddl.u8 q2, d4, d5 |
| 2091 |
vaddl.u8 q3, d6, d7 |
| 2092 |
vadd.i16 q2, q2, q3 |
| 2093 |
vadd.i16 d4, d4, d5 |
| 2094 |
vpadd.i16 d4, d4 |
| 2095 |
vpadd.i16 d4, d4 |
| 2096 |
cmp r4, #32 |
| 2097 |
vadd.i16 d0, d0, d4 |
| 2098 |
vshl.u16 d0, d0, d17 |
| 2099 |
beq 1f |
| 2100 |
// h = 8/16/64 |
| 2101 |
cmp r4, #8 |
| 2102 |
movw lr, #(0x3334/2) |
| 2103 |
movw r8, #(0x5556/2) |
| 2104 |
it ne |
| 2105 |
movne lr, r8 |
| 2106 |
vdup.16 d18, lr |
| 2107 |
vqdmulh.s16 d0, d0, d18 |
| 2108 |
1: |
| 2109 |
vdup.16 q0, d0[0] |
| 2110 |
b L(ipred_cfl_splat_w16) |
| 2111 |
endfunc |
| 2112 |
|
| 2113 |
// void cfl_ac_420_8bpc_neon(int16_t *const ac, const pixel *const ypx, |
| 2114 |
// const ptrdiff_t stride, const int w_pad, |
| 2115 |
// const int h_pad, const int cw, const int ch); |
| 2116 |
function ipred_cfl_ac_420_8bpc_neon, export=1 |
| 2117 |
push {r4-r8,lr} |
| 2118 |
ldrd r4, r5, [sp, #24] |
| 2119 |
ldr r6, [sp, #32] |
| 2120 |
clz r8, r5 |
| 2121 |
lsl r4, r4, #2 |
| 2122 |
adr r7, L(ipred_cfl_ac_420_tbl) |
| 2123 |
sub r8, r8, #27 |
| 2124 |
ldr r8, [r7, r8, lsl #2] |
| 2125 |
vmov.i16 q8, #0 |
| 2126 |
vmov.i16 q9, #0 |
| 2127 |
vmov.i16 q10, #0 |
| 2128 |
vmov.i16 q11, #0 |
| 2129 |
add r7, r7, r8 |
| 2130 |
sub r8, r6, r4 // height - h_pad |
| 2131 |
rbit lr, r5 // rbit(width) |
| 2132 |
rbit r12, r6 // rbit(height) |
| 2133 |
clz lr, lr // ctz(width) |
| 2134 |
clz r12, r12 // ctz(height) |
| 2135 |
add lr, lr, r12 // log2sz |
| 2136 |
add r12, r1, r2 |
| 2137 |
vdup.32 d31, lr |
| 2138 |
lsl r2, r2, #1 |
| 2139 |
vneg.s32 d31, d31 // -log2sz |
| 2140 |
bx r7 |
| 2141 |
|
| 2142 |
.align 2 |
| 2143 |
L(ipred_cfl_ac_420_tbl): |
| 2144 |
.word L(ipred_cfl_ac_420_w16) - L(ipred_cfl_ac_420_tbl) + CONFIG_THUMB |
| 2145 |
.word L(ipred_cfl_ac_420_w8) - L(ipred_cfl_ac_420_tbl) + CONFIG_THUMB |
| 2146 |
.word L(ipred_cfl_ac_420_w4) - L(ipred_cfl_ac_420_tbl) + CONFIG_THUMB |
| 2147 |
|
| 2148 |
L(ipred_cfl_ac_420_w4): |
| 2149 |
1: // Copy and subsample input |
| 2150 |
vld1.8 {d0}, [r1, :64], r2 |
| 2151 |
vld1.8 {d2}, [r12, :64], r2 |
| 2152 |
vld1.8 {d1}, [r1, :64], r2 |
| 2153 |
vld1.8 {d3}, [r12, :64], r2 |
| 2154 |
vpaddl.u8 q0, q0 |
| 2155 |
vpaddl.u8 q1, q1 |
| 2156 |
vadd.i16 q0, q0, q1 |
| 2157 |
vshl.i16 q0, q0, #1 |
| 2158 |
subs r8, r8, #2 |
| 2159 |
vst1.16 {q0}, [r0, :128]! |
| 2160 |
vadd.i16 q8, q8, q0 |
| 2161 |
bgt 1b |
| 2162 |
cmp r4, #0 |
| 2163 |
vmov d0, d1 |
| 2164 |
vmov d2, d1 |
| 2165 |
vmov d3, d1 |
| 2166 |
L(ipred_cfl_ac_420_w4_hpad): |
| 2167 |
beq 3f // This assumes that all callers already did "cmp r4, #0" |
| 2168 |
2: // Vertical padding (h_pad > 0) |
| 2169 |
subs r4, r4, #4 |
| 2170 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2171 |
vadd.i16 q8, q8, q0 |
| 2172 |
vadd.i16 q8, q8, q1 |
| 2173 |
bgt 2b |
| 2174 |
3: |
| 2175 |
L(ipred_cfl_ac_420_w4_calc_subtract_dc): |
| 2176 |
// Aggregate the sums |
| 2177 |
vadd.i16 q0, q8, q9 |
| 2178 |
vadd.i16 q1, q10, q11 |
| 2179 |
vpaddl.u16 q0, q0 |
| 2180 |
vpaddl.u16 q1, q1 |
| 2181 |
vadd.i32 q0, q1 |
| 2182 |
vadd.i32 d0, d0, d1 |
| 2183 |
vpadd.i32 d0, d0, d0 // sum |
| 2184 |
sub r0, r0, r6, lsl #3 |
| 2185 |
vrshl.u32 d16, d0, d31 // (sum + (1 << (log2sz - 1))) >>= log2sz |
| 2186 |
vdup.16 q8, d16[0] |
| 2187 |
L(ipred_cfl_ac_420_w4_subtract_dc): |
| 2188 |
6: // Subtract dc from ac |
| 2189 |
vld1.16 {q0, q1}, [r0, :128] |
| 2190 |
subs r6, r6, #4 |
| 2191 |
vsub.i16 q0, q0, q8 |
| 2192 |
vsub.i16 q1, q1, q8 |
| 2193 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2194 |
bgt 6b |
| 2195 |
pop {r4-r8, pc} |
| 2196 |
|
| 2197 |
L(ipred_cfl_ac_420_w8): |
| 2198 |
cmp r3, #0 |
| 2199 |
bne L(ipred_cfl_ac_420_w8_wpad) |
| 2200 |
1: // Copy and subsample input, without padding |
| 2201 |
vld1.8 {q0}, [r1, :128], r2 |
| 2202 |
vld1.8 {q1}, [r12, :128], r2 |
| 2203 |
vld1.8 {q2}, [r1, :128], r2 |
| 2204 |
vpaddl.u8 q0, q0 |
| 2205 |
vld1.8 {q3}, [r12, :128], r2 |
| 2206 |
vpaddl.u8 q1, q1 |
| 2207 |
vpaddl.u8 q2, q2 |
| 2208 |
vpaddl.u8 q3, q3 |
| 2209 |
vadd.i16 q0, q0, q1 |
| 2210 |
vadd.i16 q2, q2, q3 |
| 2211 |
vshl.i16 q0, q0, #1 |
| 2212 |
vshl.i16 q1, q2, #1 |
| 2213 |
subs r8, r8, #2 |
| 2214 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2215 |
vadd.i16 q8, q8, q0 |
| 2216 |
vadd.i16 q9, q9, q1 |
| 2217 |
bgt 1b |
| 2218 |
cmp r4, #0 |
| 2219 |
vmov q0, q1 |
| 2220 |
b L(ipred_cfl_ac_420_w8_hpad) |
| 2221 |
|
| 2222 |
L(ipred_cfl_ac_420_w8_wpad): |
| 2223 |
1: // Copy and subsample input, padding 4 |
| 2224 |
vld1.16 {d0}, [r1, :64], r2 |
| 2225 |
vld1.16 {d2}, [r12, :64], r2 |
| 2226 |
vld1.16 {d1}, [r1, :64], r2 |
| 2227 |
vld1.16 {d3}, [r12, :64], r2 |
| 2228 |
vpaddl.u8 q0, q0 |
| 2229 |
vpaddl.u8 q1, q1 |
| 2230 |
vadd.i16 q0, q0, q1 |
| 2231 |
vshl.i16 q0, q0, #1 |
| 2232 |
vdup.16 d3, d1[3] |
| 2233 |
vmov d2, d1 |
| 2234 |
vdup.16 d1, d0[3] |
| 2235 |
subs r8, r8, #2 |
| 2236 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2237 |
vadd.i16 q8, q8, q0 |
| 2238 |
vadd.i16 q9, q9, q1 |
| 2239 |
bgt 1b |
| 2240 |
cmp r4, #0 |
| 2241 |
vmov q0, q1 |
| 2242 |
|
| 2243 |
L(ipred_cfl_ac_420_w8_hpad): |
| 2244 |
beq 3f // This assumes that all callers already did "cmp r4, #0" |
| 2245 |
2: // Vertical padding (h_pad > 0) |
| 2246 |
subs r4, r4, #4 |
| 2247 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2248 |
vadd.i16 q8, q8, q0 |
| 2249 |
vadd.i16 q9, q9, q1 |
| 2250 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2251 |
vadd.i16 q10, q10, q0 |
| 2252 |
vadd.i16 q11, q11, q1 |
| 2253 |
bgt 2b |
| 2254 |
3: |
| 2255 |
|
| 2256 |
// Double the height and reuse the w4 summing/subtracting |
| 2257 |
lsl r6, r6, #1 |
| 2258 |
b L(ipred_cfl_ac_420_w4_calc_subtract_dc) |
| 2259 |
|
| 2260 |
L(ipred_cfl_ac_420_w16): |
| 2261 |
adr r7, L(ipred_cfl_ac_420_w16_tbl) |
| 2262 |
ldr r3, [r7, r3, lsl #2] |
| 2263 |
add r7, r7, r3 |
| 2264 |
bx r7 |
| 2265 |
|
| 2266 |
.align 2 |
| 2267 |
L(ipred_cfl_ac_420_w16_tbl): |
| 2268 |
.word L(ipred_cfl_ac_420_w16_wpad0) - L(ipred_cfl_ac_420_w16_tbl) + CONFIG_THUMB |
| 2269 |
.word L(ipred_cfl_ac_420_w16_wpad1) - L(ipred_cfl_ac_420_w16_tbl) + CONFIG_THUMB |
| 2270 |
.word L(ipred_cfl_ac_420_w16_wpad2) - L(ipred_cfl_ac_420_w16_tbl) + CONFIG_THUMB |
| 2271 |
.word L(ipred_cfl_ac_420_w16_wpad3) - L(ipred_cfl_ac_420_w16_tbl) + CONFIG_THUMB |
| 2272 |
|
| 2273 |
L(ipred_cfl_ac_420_w16_wpad0): |
| 2274 |
1: // Copy and subsample input, without padding |
| 2275 |
vld1.8 {q0, q1}, [r1, :128], r2 |
| 2276 |
vld1.8 {q2, q3}, [r12, :128], r2 |
| 2277 |
vpaddl.u8 q0, q0 |
| 2278 |
vld1.8 {q12, q13}, [r1, :128], r2 |
| 2279 |
vpaddl.u8 q1, q1 |
| 2280 |
vpaddl.u8 q2, q2 |
| 2281 |
vpaddl.u8 q3, q3 |
| 2282 |
vadd.i16 q0, q0, q2 |
| 2283 |
vadd.i16 q1, q1, q3 |
| 2284 |
vld1.8 {q2, q3}, [r12, :128], r2 |
| 2285 |
vpaddl.u8 q12, q12 |
| 2286 |
vpaddl.u8 q13, q13 |
| 2287 |
vpaddl.u8 q2, q2 |
| 2288 |
vpaddl.u8 q3, q3 |
| 2289 |
vadd.i16 q12, q12, q2 |
| 2290 |
vadd.i16 q13, q13, q3 |
| 2291 |
vshl.i16 q0, q0, #1 |
| 2292 |
vshl.i16 q1, q1, #1 |
| 2293 |
vshl.i16 q2, q12, #1 |
| 2294 |
vshl.i16 q3, q13, #1 |
| 2295 |
subs r8, r8, #2 |
| 2296 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2297 |
vadd.i16 q8, q8, q0 |
| 2298 |
vadd.i16 q9, q9, q1 |
| 2299 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2300 |
vadd.i16 q10, q10, q2 |
| 2301 |
vadd.i16 q11, q11, q3 |
| 2302 |
bgt 1b |
| 2303 |
cmp r4, #0 |
| 2304 |
vmov q0, q2 |
| 2305 |
vmov q1, q3 |
| 2306 |
b L(ipred_cfl_ac_420_w16_hpad) |
| 2307 |
|
| 2308 |
L(ipred_cfl_ac_420_w16_wpad1): |
| 2309 |
1: // Copy and subsample input, padding 4 |
| 2310 |
vldr d2, [r1, #16] |
| 2311 |
vld1.8 {q0}, [r1, :128], r2 |
| 2312 |
vldr d6, [r12, #16] |
| 2313 |
vld1.8 {q2}, [r12, :128], r2 |
| 2314 |
vpaddl.u8 d2, d2 |
| 2315 |
vldr d26, [r1, #16] |
| 2316 |
vpaddl.u8 q0, q0 |
| 2317 |
vld1.8 {q12}, [r1, :128], r2 |
| 2318 |
vpaddl.u8 d6, d6 |
| 2319 |
vldr d30, [r12, #16] |
| 2320 |
vpaddl.u8 q2, q2 |
| 2321 |
vld1.8 {q14}, [r12, :128], r2 |
| 2322 |
vpaddl.u8 d26, d26 |
| 2323 |
vpaddl.u8 q12, q12 |
| 2324 |
vpaddl.u8 d30, d30 |
| 2325 |
vpaddl.u8 q14, q14 |
| 2326 |
vadd.i16 d2, d2, d6 |
| 2327 |
vadd.i16 q0, q0, q2 |
| 2328 |
vadd.i16 d26, d26, d30 |
| 2329 |
vadd.i16 q12, q12, q14 |
| 2330 |
vshl.i16 d2, d2, #1 |
| 2331 |
vshl.i16 q0, q0, #1 |
| 2332 |
vshl.i16 d6, d26, #1 |
| 2333 |
vshl.i16 q2, q12, #1 |
| 2334 |
vdup.16 d3, d2[3] |
| 2335 |
vdup.16 d7, d6[3] |
| 2336 |
subs r8, r8, #2 |
| 2337 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2338 |
vadd.i16 q8, q8, q0 |
| 2339 |
vadd.i16 q9, q9, q1 |
| 2340 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2341 |
vadd.i16 q10, q10, q2 |
| 2342 |
vadd.i16 q11, q11, q3 |
| 2343 |
bgt 1b |
| 2344 |
cmp r4, #0 |
| 2345 |
vmov q0, q2 |
| 2346 |
vmov q1, q3 |
| 2347 |
b L(ipred_cfl_ac_420_w16_hpad) |
| 2348 |
|
| 2349 |
L(ipred_cfl_ac_420_w16_wpad2): |
| 2350 |
1: // Copy and subsample input, padding 8 |
| 2351 |
vld1.8 {q0}, [r1, :128], r2 |
| 2352 |
vld1.8 {q1}, [r12, :128], r2 |
| 2353 |
vld1.8 {q2}, [r1, :128], r2 |
| 2354 |
vpaddl.u8 q0, q0 |
| 2355 |
vld1.8 {q3}, [r12, :128], r2 |
| 2356 |
vpaddl.u8 q1, q1 |
| 2357 |
vpaddl.u8 q2, q2 |
| 2358 |
vpaddl.u8 q3, q3 |
| 2359 |
vadd.i16 q0, q0, q1 |
| 2360 |
vadd.i16 q2, q2, q3 |
| 2361 |
vshl.i16 q0, q0, #1 |
| 2362 |
vshl.i16 q2, q2, #1 |
| 2363 |
vdup.16 q1, d1[3] |
| 2364 |
vdup.16 q3, d5[3] |
| 2365 |
subs r8, r8, #2 |
| 2366 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2367 |
vadd.i16 q8, q8, q0 |
| 2368 |
vadd.i16 q9, q9, q1 |
| 2369 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2370 |
vadd.i16 q10, q10, q2 |
| 2371 |
vadd.i16 q11, q11, q3 |
| 2372 |
bgt 1b |
| 2373 |
cmp r4, #0 |
| 2374 |
vmov q0, q2 |
| 2375 |
vmov q1, q3 |
| 2376 |
b L(ipred_cfl_ac_420_w16_hpad) |
| 2377 |
|
| 2378 |
L(ipred_cfl_ac_420_w16_wpad3): |
| 2379 |
1: // Copy and subsample input, padding 12 |
| 2380 |
vld1.8 {d0}, [r1, :64], r2 |
| 2381 |
vld1.8 {d1}, [r12, :64], r2 |
| 2382 |
vld1.8 {d4}, [r1, :64], r2 |
| 2383 |
vpaddl.u8 q0, q0 |
| 2384 |
vld1.8 {d5}, [r12, :64], r2 |
| 2385 |
vpaddl.u8 q2, q2 |
| 2386 |
vadd.i16 d0, d0, d1 |
| 2387 |
vadd.i16 d4, d4, d5 |
| 2388 |
vshl.i16 d0, d0, #1 |
| 2389 |
vshl.i16 d4, d4, #1 |
| 2390 |
vdup.16 q1, d0[3] |
| 2391 |
vdup.16 q3, d4[3] |
| 2392 |
vdup.16 d1, d0[3] |
| 2393 |
vdup.16 d5, d4[3] |
| 2394 |
subs r8, r8, #2 |
| 2395 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2396 |
vadd.i16 q8, q8, q0 |
| 2397 |
vadd.i16 q9, q9, q1 |
| 2398 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2399 |
vadd.i16 q10, q10, q2 |
| 2400 |
vadd.i16 q11, q11, q3 |
| 2401 |
bgt 1b |
| 2402 |
cmp r4, #0 |
| 2403 |
vmov q0, q2 |
| 2404 |
vmov q1, q3 |
| 2405 |
b L(ipred_cfl_ac_420_w16_hpad) |
| 2406 |
|
| 2407 |
L(ipred_cfl_ac_420_w16_hpad): |
| 2408 |
beq 3f // This assumes that all callers already did "cmp r4, #0" |
| 2409 |
2: // Vertical padding (h_pad > 0) |
| 2410 |
subs r4, r4, #2 |
| 2411 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2412 |
vadd.i16 q8, q8, q0 |
| 2413 |
vadd.i16 q9, q9, q1 |
| 2414 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2415 |
vadd.i16 q10, q10, q2 |
| 2416 |
vadd.i16 q11, q11, q3 |
| 2417 |
bgt 2b |
| 2418 |
3: |
| 2419 |
|
| 2420 |
// Quadruple the height and reuse the w4 summing/subtracting |
| 2421 |
lsl r6, r6, #2 |
| 2422 |
b L(ipred_cfl_ac_420_w4_calc_subtract_dc) |
| 2423 |
endfunc |
| 2424 |
|
| 2425 |
// void cfl_ac_422_8bpc_neon(int16_t *const ac, const pixel *const ypx, |
| 2426 |
// const ptrdiff_t stride, const int w_pad, |
| 2427 |
// const int h_pad, const int cw, const int ch); |
| 2428 |
function ipred_cfl_ac_422_8bpc_neon, export=1 |
| 2429 |
push {r4-r8,lr} |
| 2430 |
ldrd r4, r5, [sp, #24] |
| 2431 |
ldr r6, [sp, #32] |
| 2432 |
clz r8, r5 |
| 2433 |
lsl r4, r4, #2 |
| 2434 |
adr r7, L(ipred_cfl_ac_422_tbl) |
| 2435 |
sub r8, r8, #27 |
| 2436 |
ldr r8, [r7, r8, lsl #2] |
| 2437 |
vmov.i16 q8, #0 |
| 2438 |
vmov.i16 q9, #0 |
| 2439 |
vmov.i16 q10, #0 |
| 2440 |
vmov.i16 q11, #0 |
| 2441 |
add r7, r7, r8 |
| 2442 |
sub r8, r6, r4 // height - h_pad |
| 2443 |
rbit lr, r5 // rbit(width) |
| 2444 |
rbit r12, r6 // rbit(height) |
| 2445 |
clz lr, lr // ctz(width) |
| 2446 |
clz r12, r12 // ctz(height) |
| 2447 |
add lr, lr, r12 // log2sz |
| 2448 |
add r12, r1, r2 |
| 2449 |
vdup.32 d31, lr |
| 2450 |
lsl r2, r2, #1 |
| 2451 |
vneg.s32 d31, d31 // -log2sz |
| 2452 |
bx r7 |
| 2453 |
|
| 2454 |
.align 2 |
| 2455 |
L(ipred_cfl_ac_422_tbl): |
| 2456 |
.word L(ipred_cfl_ac_422_w16) - L(ipred_cfl_ac_422_tbl) + CONFIG_THUMB |
| 2457 |
.word L(ipred_cfl_ac_422_w8) - L(ipred_cfl_ac_422_tbl) + CONFIG_THUMB |
| 2458 |
.word L(ipred_cfl_ac_422_w4) - L(ipred_cfl_ac_422_tbl) + CONFIG_THUMB |
| 2459 |
|
| 2460 |
L(ipred_cfl_ac_422_w4): |
| 2461 |
1: // Copy and subsample input |
| 2462 |
vld1.8 {d0}, [r1, :64], r2 |
| 2463 |
vld1.8 {d1}, [r12, :64], r2 |
| 2464 |
vld1.8 {d2}, [r1, :64], r2 |
| 2465 |
vld1.8 {d3}, [r12, :64], r2 |
| 2466 |
vpaddl.u8 q0, q0 |
| 2467 |
vpaddl.u8 q1, q1 |
| 2468 |
vshl.i16 q0, q0, #2 |
| 2469 |
vshl.i16 q1, q1, #2 |
| 2470 |
subs r8, r8, #4 |
| 2471 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2472 |
vadd.i16 q8, q8, q0 |
| 2473 |
vadd.i16 q9, q9, q1 |
| 2474 |
bgt 1b |
| 2475 |
cmp r4, #0 |
| 2476 |
vmov d0, d3 |
| 2477 |
vmov d1, d3 |
| 2478 |
vmov d2, d3 |
| 2479 |
b L(ipred_cfl_ac_420_w4_hpad) |
| 2480 |
|
| 2481 |
L(ipred_cfl_ac_422_w8): |
| 2482 |
cmp r3, #0 |
| 2483 |
bne L(ipred_cfl_ac_422_w8_wpad) |
| 2484 |
1: // Copy and subsample input, without padding |
| 2485 |
vld1.8 {q0}, [r1, :128], r2 |
| 2486 |
vld1.8 {q1}, [r12, :128], r2 |
| 2487 |
vld1.8 {q2}, [r1, :128], r2 |
| 2488 |
vpaddl.u8 q0, q0 |
| 2489 |
vld1.8 {q3}, [r12, :128], r2 |
| 2490 |
vpaddl.u8 q1, q1 |
| 2491 |
vpaddl.u8 q2, q2 |
| 2492 |
vpaddl.u8 q3, q3 |
| 2493 |
vshl.i16 q0, q0, #2 |
| 2494 |
vshl.i16 q1, q1, #2 |
| 2495 |
vshl.i16 q2, q2, #2 |
| 2496 |
vshl.i16 q3, q3, #2 |
| 2497 |
subs r8, r8, #4 |
| 2498 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2499 |
vadd.i16 q8, q8, q0 |
| 2500 |
vadd.i16 q9, q9, q1 |
| 2501 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2502 |
vadd.i16 q10, q10, q2 |
| 2503 |
vadd.i16 q11, q11, q3 |
| 2504 |
bgt 1b |
| 2505 |
cmp r4, #0 |
| 2506 |
vmov q0, q3 |
| 2507 |
vmov q1, q3 |
| 2508 |
b L(ipred_cfl_ac_420_w8_hpad) |
| 2509 |
|
| 2510 |
L(ipred_cfl_ac_422_w8_wpad): |
| 2511 |
1: // Copy and subsample input, padding 4 |
| 2512 |
vld1.8 {d0}, [r1, :64], r2 |
| 2513 |
vld1.8 {d1}, [r12, :64], r2 |
| 2514 |
vld1.8 {d2}, [r1, :64], r2 |
| 2515 |
vld1.8 {d3}, [r12, :64], r2 |
| 2516 |
vpaddl.u8 q0, q0 |
| 2517 |
vpaddl.u8 q1, q1 |
| 2518 |
vshl.i16 q0, q0, #2 |
| 2519 |
vshl.i16 q1, q1, #2 |
| 2520 |
vdup.16 d7, d3[3] |
| 2521 |
vmov d6, d3 |
| 2522 |
vdup.16 d5, d2[3] |
| 2523 |
vmov d4, d2 |
| 2524 |
vdup.16 d3, d1[3] |
| 2525 |
vmov d2, d1 |
| 2526 |
vdup.16 d1, d0[3] |
| 2527 |
subs r8, r8, #4 |
| 2528 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2529 |
vadd.i16 q8, q8, q0 |
| 2530 |
vadd.i16 q9, q9, q1 |
| 2531 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2532 |
vadd.i16 q10, q10, q2 |
| 2533 |
vadd.i16 q11, q11, q3 |
| 2534 |
bgt 1b |
| 2535 |
cmp r4, #0 |
| 2536 |
vmov q0, q3 |
| 2537 |
vmov q1, q3 |
| 2538 |
b L(ipred_cfl_ac_420_w8_hpad) |
| 2539 |
|
| 2540 |
L(ipred_cfl_ac_422_w16): |
| 2541 |
adr r7, L(ipred_cfl_ac_422_w16_tbl) |
| 2542 |
ldr r3, [r7, r3, lsl #2] |
| 2543 |
add r7, r7, r3 |
| 2544 |
bx r7 |
| 2545 |
|
| 2546 |
.align 2 |
| 2547 |
L(ipred_cfl_ac_422_w16_tbl): |
| 2548 |
.word L(ipred_cfl_ac_422_w16_wpad0) - L(ipred_cfl_ac_422_w16_tbl) + CONFIG_THUMB |
| 2549 |
.word L(ipred_cfl_ac_422_w16_wpad1) - L(ipred_cfl_ac_422_w16_tbl) + CONFIG_THUMB |
| 2550 |
.word L(ipred_cfl_ac_422_w16_wpad2) - L(ipred_cfl_ac_422_w16_tbl) + CONFIG_THUMB |
| 2551 |
.word L(ipred_cfl_ac_422_w16_wpad3) - L(ipred_cfl_ac_422_w16_tbl) + CONFIG_THUMB |
| 2552 |
|
| 2553 |
L(ipred_cfl_ac_422_w16_wpad0): |
| 2554 |
1: // Copy and subsample input, without padding |
| 2555 |
vld1.8 {q0, q1}, [r1, :128], r2 |
| 2556 |
vld1.8 {q2, q3}, [r12, :128], r2 |
| 2557 |
vpaddl.u8 q0, q0 |
| 2558 |
vpaddl.u8 q1, q1 |
| 2559 |
vpaddl.u8 q2, q2 |
| 2560 |
vpaddl.u8 q3, q3 |
| 2561 |
vshl.i16 q0, q0, #2 |
| 2562 |
vshl.i16 q1, q1, #2 |
| 2563 |
vshl.i16 q2, q2, #2 |
| 2564 |
vshl.i16 q3, q3, #2 |
| 2565 |
subs r8, r8, #2 |
| 2566 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2567 |
vadd.i16 q8, q8, q0 |
| 2568 |
vadd.i16 q9, q9, q1 |
| 2569 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2570 |
vadd.i16 q10, q10, q2 |
| 2571 |
vadd.i16 q11, q11, q3 |
| 2572 |
bgt 1b |
| 2573 |
cmp r4, #0 |
| 2574 |
vmov q0, q2 |
| 2575 |
vmov q1, q3 |
| 2576 |
b L(ipred_cfl_ac_420_w16_hpad) |
| 2577 |
|
| 2578 |
L(ipred_cfl_ac_422_w16_wpad1): |
| 2579 |
1: // Copy and subsample input, padding 4 |
| 2580 |
vldr d2, [r1, #16] |
| 2581 |
vld1.8 {q0}, [r1, :128], r2 |
| 2582 |
vldr d6, [r12, #16] |
| 2583 |
vld1.8 {q2}, [r12, :128], r2 |
| 2584 |
vpaddl.u8 d2, d2 |
| 2585 |
vpaddl.u8 q0, q0 |
| 2586 |
vpaddl.u8 d6, d6 |
| 2587 |
vpaddl.u8 q2, q2 |
| 2588 |
vshl.i16 d2, d2, #2 |
| 2589 |
vshl.i16 q0, q0, #2 |
| 2590 |
vshl.i16 d6, d6, #2 |
| 2591 |
vshl.i16 q2, q2, #2 |
| 2592 |
vdup.16 d3, d2[3] |
| 2593 |
vdup.16 d7, d6[3] |
| 2594 |
subs r8, r8, #2 |
| 2595 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2596 |
vadd.i16 q8, q8, q0 |
| 2597 |
vadd.i16 q9, q9, q1 |
| 2598 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2599 |
vadd.i16 q10, q10, q2 |
| 2600 |
vadd.i16 q11, q11, q3 |
| 2601 |
bgt 1b |
| 2602 |
cmp r4, #0 |
| 2603 |
vmov q0, q2 |
| 2604 |
vmov q1, q3 |
| 2605 |
b L(ipred_cfl_ac_420_w16_hpad) |
| 2606 |
|
| 2607 |
L(ipred_cfl_ac_422_w16_wpad2): |
| 2608 |
1: // Copy and subsample input, padding 8 |
| 2609 |
vld1.8 {q0}, [r1, :128], r2 |
| 2610 |
vld1.8 {q2}, [r12, :128], r2 |
| 2611 |
vpaddl.u8 q0, q0 |
| 2612 |
vpaddl.u8 q2, q2 |
| 2613 |
vshl.i16 q0, q0, #2 |
| 2614 |
vshl.i16 q2, q2, #2 |
| 2615 |
vdup.16 q1, d1[3] |
| 2616 |
vdup.16 q3, d5[3] |
| 2617 |
subs r8, r8, #2 |
| 2618 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2619 |
vadd.i16 q8, q8, q0 |
| 2620 |
vadd.i16 q9, q9, q1 |
| 2621 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2622 |
vadd.i16 q10, q10, q2 |
| 2623 |
vadd.i16 q11, q11, q3 |
| 2624 |
bgt 1b |
| 2625 |
cmp r4, #0 |
| 2626 |
vmov q0, q2 |
| 2627 |
vmov q1, q3 |
| 2628 |
b L(ipred_cfl_ac_420_w16_hpad) |
| 2629 |
|
| 2630 |
L(ipred_cfl_ac_422_w16_wpad3): |
| 2631 |
1: // Copy and subsample input, padding 12 |
| 2632 |
vld1.8 {d0}, [r1, :64], r2 |
| 2633 |
vld1.8 {d1}, [r12, :64], r2 |
| 2634 |
vpaddl.u8 q0, q0 |
| 2635 |
vshl.i16 q0, q0, #2 |
| 2636 |
vdup.16 q3, d1[3] |
| 2637 |
vdup.16 q1, d0[3] |
| 2638 |
vdup.16 d5, d1[3] |
| 2639 |
vmov d4, d1 |
| 2640 |
vdup.16 d1, d0[3] |
| 2641 |
subs r8, r8, #2 |
| 2642 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2643 |
vadd.i16 q8, q8, q0 |
| 2644 |
vadd.i16 q9, q9, q1 |
| 2645 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2646 |
vadd.i16 q10, q10, q2 |
| 2647 |
vadd.i16 q11, q11, q3 |
| 2648 |
bgt 1b |
| 2649 |
cmp r4, #0 |
| 2650 |
vmov q0, q2 |
| 2651 |
vmov q1, q3 |
| 2652 |
b L(ipred_cfl_ac_420_w16_hpad) |
| 2653 |
endfunc |
| 2654 |
|
| 2655 |
// void cfl_ac_444_8bpc_neon(int16_t *const ac, const pixel *const ypx, |
| 2656 |
// const ptrdiff_t stride, const int w_pad, |
| 2657 |
// const int h_pad, const int cw, const int ch); |
| 2658 |
function ipred_cfl_ac_444_8bpc_neon, export=1 |
| 2659 |
push {r4-r8,lr} |
| 2660 |
ldrd r4, r5, [sp, #24] |
| 2661 |
ldr r6, [sp, #32] |
| 2662 |
clz r8, r5 |
| 2663 |
lsl r4, r4, #2 |
| 2664 |
adr r7, L(ipred_cfl_ac_444_tbl) |
| 2665 |
sub r8, r8, #26 |
| 2666 |
ldr r8, [r7, r8, lsl #2] |
| 2667 |
vmov.i16 q8, #0 |
| 2668 |
vmov.i16 q9, #0 |
| 2669 |
vmov.i16 q10, #0 |
| 2670 |
vmov.i16 q11, #0 |
| 2671 |
add r7, r7, r8 |
| 2672 |
sub r8, r6, r4 // height - h_pad |
| 2673 |
rbit lr, r5 // rbit(width) |
| 2674 |
rbit r12, r6 // rbit(height) |
| 2675 |
clz lr, lr // ctz(width) |
| 2676 |
clz r12, r12 // ctz(height) |
| 2677 |
add lr, lr, r12 // log2sz |
| 2678 |
add r12, r1, r2 |
| 2679 |
vdup.32 d31, lr |
| 2680 |
lsl r2, r2, #1 |
| 2681 |
vneg.s32 d31, d31 // -log2sz |
| 2682 |
bx r7 |
| 2683 |
|
| 2684 |
.align 2 |
| 2685 |
L(ipred_cfl_ac_444_tbl): |
| 2686 |
.word L(ipred_cfl_ac_444_w32) - L(ipred_cfl_ac_444_tbl) + CONFIG_THUMB |
| 2687 |
.word L(ipred_cfl_ac_444_w16) - L(ipred_cfl_ac_444_tbl) + CONFIG_THUMB |
| 2688 |
.word L(ipred_cfl_ac_444_w8) - L(ipred_cfl_ac_444_tbl) + CONFIG_THUMB |
| 2689 |
.word L(ipred_cfl_ac_444_w4) - L(ipred_cfl_ac_444_tbl) + CONFIG_THUMB |
| 2690 |
|
| 2691 |
L(ipred_cfl_ac_444_w4): |
| 2692 |
1: // Copy and expand input |
| 2693 |
vld1.32 {d0[]}, [r1, :32], r2 |
| 2694 |
vld1.32 {d0[1]}, [r12, :32], r2 |
| 2695 |
vld1.32 {d2[]}, [r1, :32], r2 |
| 2696 |
vld1.32 {d2[1]}, [r12, :32], r2 |
| 2697 |
vshll.u8 q0, d0, #3 |
| 2698 |
vshll.u8 q1, d2, #3 |
| 2699 |
subs r8, r8, #4 |
| 2700 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2701 |
vadd.i16 q8, q8, q0 |
| 2702 |
vadd.i16 q9, q9, q1 |
| 2703 |
bgt 1b |
| 2704 |
cmp r4, #0 |
| 2705 |
vmov d0, d3 |
| 2706 |
vmov d1, d3 |
| 2707 |
vmov d2, d3 |
| 2708 |
b L(ipred_cfl_ac_420_w4_hpad) |
| 2709 |
|
| 2710 |
L(ipred_cfl_ac_444_w8): |
| 2711 |
1: // Copy and expand input |
| 2712 |
vld1.16 {d0}, [r1, :64], r2 |
| 2713 |
vld1.16 {d2}, [r12, :64], r2 |
| 2714 |
vld1.16 {d4}, [r1, :64], r2 |
| 2715 |
vshll.u8 q0, d0, #3 |
| 2716 |
vld1.16 {d6}, [r12, :64], r2 |
| 2717 |
vshll.u8 q1, d2, #3 |
| 2718 |
vshll.u8 q2, d4, #3 |
| 2719 |
vshll.u8 q3, d6, #3 |
| 2720 |
subs r8, r8, #4 |
| 2721 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2722 |
vadd.i16 q8, q8, q0 |
| 2723 |
vadd.i16 q9, q9, q1 |
| 2724 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2725 |
vadd.i16 q10, q10, q2 |
| 2726 |
vadd.i16 q11, q11, q3 |
| 2727 |
bgt 1b |
| 2728 |
cmp r4, #0 |
| 2729 |
vmov q0, q3 |
| 2730 |
vmov q1, q3 |
| 2731 |
b L(ipred_cfl_ac_420_w8_hpad) |
| 2732 |
|
| 2733 |
L(ipred_cfl_ac_444_w16): |
| 2734 |
cmp r3, #0 |
| 2735 |
bne L(ipred_cfl_ac_444_w16_wpad) |
| 2736 |
1: // Copy and expand input, without padding |
| 2737 |
vld1.8 {q1}, [r1, :128], r2 |
| 2738 |
vld1.8 {q3}, [r12, :128], r2 |
| 2739 |
vshll.u8 q0, d2, #3 |
| 2740 |
vshll.u8 q1, d3, #3 |
| 2741 |
vshll.u8 q2, d6, #3 |
| 2742 |
vshll.u8 q3, d7, #3 |
| 2743 |
subs r8, r8, #2 |
| 2744 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2745 |
vadd.i16 q8, q8, q0 |
| 2746 |
vadd.i16 q9, q9, q1 |
| 2747 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2748 |
vadd.i16 q10, q10, q2 |
| 2749 |
vadd.i16 q11, q11, q3 |
| 2750 |
bgt 1b |
| 2751 |
cmp r4, #0 |
| 2752 |
vmov q0, q2 |
| 2753 |
vmov q1, q3 |
| 2754 |
b L(ipred_cfl_ac_420_w16_hpad) |
| 2755 |
|
| 2756 |
L(ipred_cfl_ac_444_w16_wpad): |
| 2757 |
1: // Copy and expand input, padding 8 |
| 2758 |
vld1.8 {d0}, [r1, :64], r2 |
| 2759 |
vld1.8 {d4}, [r12, :64], r2 |
| 2760 |
vshll.u8 q0, d0, #3 |
| 2761 |
vshll.u8 q2, d4, #3 |
| 2762 |
vdup.16 q1, d1[3] |
| 2763 |
vdup.16 q3, d5[3] |
| 2764 |
subs r8, r8, #2 |
| 2765 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2766 |
vadd.i16 q8, q8, q0 |
| 2767 |
vadd.i16 q9, q9, q1 |
| 2768 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2769 |
vadd.i16 q10, q10, q2 |
| 2770 |
vadd.i16 q11, q11, q3 |
| 2771 |
bgt 1b |
| 2772 |
cmp r4, #0 |
| 2773 |
vmov q0, q2 |
| 2774 |
vmov q1, q3 |
| 2775 |
b L(ipred_cfl_ac_420_w16_hpad) |
| 2776 |
|
| 2777 |
L(ipred_cfl_ac_444_w32): |
| 2778 |
adr r7, L(ipred_cfl_ac_444_w32_tbl) |
| 2779 |
ldr r3, [r7, r3, lsl #1] // (w3>>1) << 2 |
| 2780 |
add r7, r7, r3 |
| 2781 |
bx r7 |
| 2782 |
|
| 2783 |
.align 2 |
| 2784 |
L(ipred_cfl_ac_444_w32_tbl): |
| 2785 |
.word L(ipred_cfl_ac_444_w32_wpad0) - L(ipred_cfl_ac_444_w32_tbl) + CONFIG_THUMB |
| 2786 |
.word L(ipred_cfl_ac_444_w32_wpad2) - L(ipred_cfl_ac_444_w32_tbl) + CONFIG_THUMB |
| 2787 |
.word L(ipred_cfl_ac_444_w32_wpad4) - L(ipred_cfl_ac_444_w32_tbl) + CONFIG_THUMB |
| 2788 |
.word L(ipred_cfl_ac_444_w32_wpad6) - L(ipred_cfl_ac_444_w32_tbl) + CONFIG_THUMB |
| 2789 |
|
| 2790 |
L(ipred_cfl_ac_444_w32_wpad0): |
| 2791 |
1: // Copy and expand input, without padding |
| 2792 |
vld1.8 {q2, q3}, [r1, :128], r2 |
| 2793 |
vld1.8 {q13, q14}, [r12, :128], r2 |
| 2794 |
vshll.u8 q0, d4, #3 |
| 2795 |
vshll.u8 q1, d5, #3 |
| 2796 |
vshll.u8 q2, d6, #3 |
| 2797 |
vshll.u8 q3, d7, #3 |
| 2798 |
vshll.u8 q12, d26, #3 |
| 2799 |
vshll.u8 q13, d27, #3 |
| 2800 |
subs r8, r8, #2 |
| 2801 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2802 |
vadd.i16 q8, q8, q0 |
| 2803 |
vadd.i16 q9, q9, q1 |
| 2804 |
vshll.u8 q0, d28, #3 |
| 2805 |
vshll.u8 q1, d29, #3 |
| 2806 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2807 |
vadd.i16 q10, q10, q2 |
| 2808 |
vadd.i16 q11, q11, q3 |
| 2809 |
vst1.16 {q12, q13}, [r0, :128]! |
| 2810 |
vadd.i16 q8, q8, q12 |
| 2811 |
vadd.i16 q9, q9, q13 |
| 2812 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2813 |
vadd.i16 q10, q10, q0 |
| 2814 |
vadd.i16 q11, q11, q1 |
| 2815 |
bgt 1b |
| 2816 |
cmp r4, #0 |
| 2817 |
b L(ipred_cfl_ac_444_w32_hpad) |
| 2818 |
|
| 2819 |
L(ipred_cfl_ac_444_w32_wpad2): |
| 2820 |
1: // Copy and expand input, padding 8 |
| 2821 |
vldr d4, [r1, #16] |
| 2822 |
vld1.8 {q1}, [r1, :128], r2 |
| 2823 |
vldr d28, [r12, #16] |
| 2824 |
vld1.8 {q13}, [r12, :128], r2 |
| 2825 |
vshll.u8 q2, d4, #3 |
| 2826 |
vshll.u8 q0, d2, #3 |
| 2827 |
vshll.u8 q1, d3, #3 |
| 2828 |
vshll.u8 q12, d26, #3 |
| 2829 |
vshll.u8 q13, d27, #3 |
| 2830 |
vdup.16 q3, d5[3] |
| 2831 |
subs r8, r8, #2 |
| 2832 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2833 |
vadd.i16 q8, q8, q0 |
| 2834 |
vadd.i16 q9, q9, q1 |
| 2835 |
vshll.u8 q0, d28, #3 |
| 2836 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2837 |
vadd.i16 q10, q10, q2 |
| 2838 |
vadd.i16 q11, q11, q3 |
| 2839 |
vdup.16 q1, d1[3] |
| 2840 |
vst1.16 {q12, q13}, [r0, :128]! |
| 2841 |
vadd.i16 q8, q8, q12 |
| 2842 |
vadd.i16 q9, q9, q13 |
| 2843 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2844 |
vadd.i16 q10, q10, q0 |
| 2845 |
vadd.i16 q11, q11, q1 |
| 2846 |
bgt 1b |
| 2847 |
cmp r4, #0 |
| 2848 |
b L(ipred_cfl_ac_444_w32_hpad) |
| 2849 |
|
| 2850 |
L(ipred_cfl_ac_444_w32_wpad4): |
| 2851 |
1: // Copy and expand input, padding 16 |
| 2852 |
vld1.8 {q1}, [r1, :128], r2 |
| 2853 |
vld1.8 {q13}, [r12, :128], r2 |
| 2854 |
vshll.u8 q0, d2, #3 |
| 2855 |
vshll.u8 q1, d3, #3 |
| 2856 |
vshll.u8 q12, d26, #3 |
| 2857 |
vshll.u8 q13, d27, #3 |
| 2858 |
vdup.16 q2, d3[3] |
| 2859 |
vdup.16 q3, d3[3] |
| 2860 |
subs r8, r8, #2 |
| 2861 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2862 |
vadd.i16 q8, q8, q0 |
| 2863 |
vadd.i16 q9, q9, q1 |
| 2864 |
vdup.16 q0, d27[3] |
| 2865 |
vdup.16 q1, d27[3] |
| 2866 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2867 |
vadd.i16 q10, q10, q2 |
| 2868 |
vadd.i16 q11, q11, q3 |
| 2869 |
vst1.16 {q12, q13}, [r0, :128]! |
| 2870 |
vadd.i16 q8, q8, q12 |
| 2871 |
vadd.i16 q9, q9, q13 |
| 2872 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2873 |
vadd.i16 q10, q10, q0 |
| 2874 |
vadd.i16 q11, q11, q1 |
| 2875 |
bgt 1b |
| 2876 |
cmp r4, #0 |
| 2877 |
b L(ipred_cfl_ac_444_w32_hpad) |
| 2878 |
|
| 2879 |
L(ipred_cfl_ac_444_w32_wpad6): |
| 2880 |
1: // Copy and expand input, padding 24 |
| 2881 |
vld1.8 {d0}, [r1, :64], r2 |
| 2882 |
vld1.8 {d24}, [r12, :64], r2 |
| 2883 |
vshll.u8 q0, d0, #3 |
| 2884 |
vshll.u8 q12, d24, #3 |
| 2885 |
subs r8, r8, #2 |
| 2886 |
vdup.16 q1, d1[3] |
| 2887 |
vdup.16 q2, d1[3] |
| 2888 |
vdup.16 q3, d1[3] |
| 2889 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2890 |
vadd.i16 q8, q8, q0 |
| 2891 |
vadd.i16 q9, q9, q1 |
| 2892 |
vdup.16 q13, d25[3] |
| 2893 |
vdup.16 q0, d25[3] |
| 2894 |
vdup.16 q1, d25[3] |
| 2895 |
vst1.16 {q2, q3}, [r0, :128]! |
| 2896 |
vadd.i16 q10, q10, q2 |
| 2897 |
vadd.i16 q11, q11, q3 |
| 2898 |
vst1.16 {q12, q13}, [r0, :128]! |
| 2899 |
vadd.i16 q8, q8, q12 |
| 2900 |
vadd.i16 q9, q9, q13 |
| 2901 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2902 |
vadd.i16 q10, q10, q0 |
| 2903 |
vadd.i16 q11, q11, q1 |
| 2904 |
bgt 1b |
| 2905 |
cmp r4, #0 |
| 2906 |
|
| 2907 |
L(ipred_cfl_ac_444_w32_hpad): |
| 2908 |
beq 3f // This assumes that all callers already did "cmp r4, #0" |
| 2909 |
2: // Vertical padding (h_pad > 0) |
| 2910 |
subs r4, r4, #1 |
| 2911 |
vst1.16 {q12, q13}, [r0, :128]! |
| 2912 |
vadd.i16 q8, q8, q12 |
| 2913 |
vadd.i16 q9, q9, q13 |
| 2914 |
vst1.16 {q0, q1}, [r0, :128]! |
| 2915 |
vadd.i16 q10, q10, q0 |
| 2916 |
vadd.i16 q11, q11, q1 |
| 2917 |
bgt 2b |
| 2918 |
3: |
| 2919 |
|
| 2920 |
// Multiply the height by eight and reuse the w4 subtracting |
| 2921 |
lsl r6, r6, #3 |
| 2922 |
// Aggregate the sums, with wider intermediates earlier than in |
| 2923 |
// ipred_cfl_ac_420_w8_calc_subtract_dc. |
| 2924 |
vpaddl.u16 q0, q8 |
| 2925 |
vpaddl.u16 q1, q9 |
| 2926 |
vpaddl.u16 q2, q10 |
| 2927 |
vpaddl.u16 q3, q11 |
| 2928 |
vadd.i32 q0, q0, q1 |
| 2929 |
vadd.i32 q2, q2, q3 |
| 2930 |
vadd.i32 q0, q0, q2 |
| 2931 |
vadd.i32 d0, d0, d1 |
| 2932 |
vpadd.i32 d0, d0, d0 // sum |
| 2933 |
sub r0, r0, r6, lsl #3 |
| 2934 |
vrshl.u32 d16, d0, d31 // (sum + (1 << (log2sz - 1))) >>= log2sz |
| 2935 |
vdup.16 q8, d16[0] |
| 2936 |
b L(ipred_cfl_ac_420_w4_subtract_dc) |
| 2937 |
endfunc |