mc.S 113 KB
Newer Older
1 2 3
/*
 * Copyright © 2018, VideoLAN and dav1d authors
 * Copyright © 2018, Janne Grunau
4
 * Copyright © 2018, Martin Storsjo
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *
 * 1. Redistributions of source code must retain the above copyright notice, this
 *    list of conditions and the following disclaimer.
 *
 * 2. Redistributions in binary form must reproduce the above copyright notice,
 *    this list of conditions and the following disclaimer in the documentation
 *    and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#include "src/arm/asm.S"
30
#include "util.S"
31

32
.macro avg dst, t0, t1, t2, t3
33 34 35 36 37 38 39 40
        ld1             {\t0\().8h,\t1\().8h},   [x2],  32
        ld1             {\t2\().8h,\t3\().8h},   [x3],  32
        add             \t0\().8h,   \t0\().8h,   \t2\().8h
        add             \t1\().8h,   \t1\().8h,   \t3\().8h
        sqrshrun        \dst\().8b,  \t0\().8h,   #5
        sqrshrun2       \dst\().16b, \t1\().8h,   #5
.endm

41
.macro w_avg dst, t0, t1, t2, t3
42 43 44 45 46 47 48 49 50 51 52 53
        ld1             {\t0\().8h,\t1\().8h},   [x2],  32
        ld1             {\t2\().8h,\t3\().8h},   [x3],  32
        sub             \t0\().8h,   \t2\().8h,   \t0\().8h
        sub             \t1\().8h,   \t3\().8h,   \t1\().8h
        sqdmulh         \t0\().8h,   \t0\().8h,   v30.8h
        sqdmulh         \t1\().8h,   \t1\().8h,   v30.8h
        add             \t0\().8h,   \t2\().8h,   \t0\().8h
        add             \t1\().8h,   \t3\().8h,   \t1\().8h
        sqrshrun        \dst\().8b,  \t0\().8h,   #4
        sqrshrun2       \dst\().16b, \t1\().8h,   #4
.endm

54
.macro mask dst, t0, t1, t2, t3
55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72
        ld1             {v30.16b}, [x6],  16
        ld1             {\t0\().8h,\t1\().8h},   [x2],  32
        mul             v30.16b, v30.16b, v31.16b
        ld1             {\t2\().8h,\t3\().8h},   [x3],  32
        shll            v28.8h, v30.8b,  #8
        shll2           v29.8h, v30.16b, #8
        sub             \t0\().8h,   \t2\().8h,   \t0\().8h
        sub             \t1\().8h,   \t3\().8h,   \t1\().8h
        sqdmulh         \t0\().8h,   \t0\().8h,   v28.8h
        sqdmulh         \t1\().8h,   \t1\().8h,   v29.8h
        add             \t0\().8h,   \t2\().8h,   \t0\().8h
        add             \t1\().8h,   \t3\().8h,   \t1\().8h
        sqrshrun        \dst\().8b,  \t0\().8h,   #4
        sqrshrun2       \dst\().16b, \t1\().8h,   #4
.endm

.macro bidir_fn type
function \type\()_8bpc_neon, export=1
Janne Grunau's avatar
Janne Grunau committed
73
        clz             w4,  w4
74 75 76 77 78 79 80 81
.ifc \type, w_avg
        dup             v30.8h, w6
        neg             v30.8h, v30.8h
        shl             v30.8h, v30.8h, #11
.endif
.ifc \type, mask
        movi            v31.16b, #256-2
.endif
82
        adr             x7,  L(\type\()_tbl)
Janne Grunau's avatar
Janne Grunau committed
83
        sub             w4,  w4,  #24
84
        ldrh            w4,  [x7, x4, lsl #1]
85
        \type           v4,  v0,  v1,  v2,  v3
86
        sub             x7,  x7,  w4, uxtw
87
        br              x7
88 89 90
40:
        add             x7,  x0,  x1
        lsl             x1,  x1,  #1
91 92 93
4:
        cmp             w5,  #4
        st1             {v4.s}[0],  [x0], x1
94
        st1             {v4.s}[1],  [x7], x1
95
        st1             {v4.s}[2],  [x0], x1
96
        st1             {v4.s}[3],  [x7], x1
97
        b.eq            0f
98
        \type           v5,  v0,  v1,  v2,  v3
99
        cmp             w5,  #8
100
        st1             {v5.s}[0],  [x0], x1
101
        st1             {v5.s}[1],  [x7], x1
102
        st1             {v5.s}[2],  [x0], x1
103
        st1             {v5.s}[3],  [x7], x1
104
        b.eq            0f
105
        \type           v4,  v0,  v1,  v2,  v3
106
        st1             {v4.s}[0],  [x0], x1
107
        st1             {v4.s}[1],  [x7], x1
108 109
        \type           v5,  v0,  v1,  v2,  v3
        st1             {v4.s}[2],  [x0], x1
110
        st1             {v4.s}[3],  [x7], x1
111
        st1             {v5.s}[0],  [x0], x1
112
        st1             {v5.s}[1],  [x7], x1
113
        st1             {v5.s}[2],  [x0], x1
114
        st1             {v5.s}[3],  [x7], x1
115
        ret
116 117 118
80:
        add             x7,  x0,  x1
        lsl             x1,  x1,  #1
119
8:
120 121
        st1             {v4.d}[0],  [x0], x1
        \type           v5,  v0,  v1,  v2,  v3
122
        st1             {v4.d}[1],  [x7], x1
123
        st1             {v5.d}[0],  [x0], x1
124
        subs            w5,  w5,  #4
125
        st1             {v5.d}[1],  [x7], x1
126
        b.le            0f
127
        \type           v4,  v0,  v1,  v2,  v3
128 129
        b               8b
16:
130
        \type           v5,  v0,  v1,  v2,  v3
131
        st1             {v4.16b}, [x0], x1
132
        \type           v6,  v0,  v1,  v2,  v3
133
        st1             {v5.16b}, [x0], x1
134
        \type           v7,  v0,  v1,  v2,  v3
135 136 137 138
        st1             {v6.16b}, [x0], x1
        subs            w5,  w5,  #4
        st1             {v7.16b}, [x0], x1
        b.le            0f
139
        \type           v4,  v0,  v1,  v2,  v3
140 141 142 143 144
        b               16b
320:
        add             x7,  x0,  x1
        lsl             x1,  x1,  #1
32:
145 146
        \type           v5,  v0,  v1,  v2,  v3
        \type           v6,  v0,  v1,  v2,  v3
147
        st1             {v4.16b,v5.16b}, [x0], x1
148
        \type           v7,  v0,  v1,  v2,  v3
149 150 151
        subs            w5,  w5,  #2
        st1             {v6.16b,v7.16b}, [x7], x1
        b.le            0f
152
        \type           v4,  v0,  v1,  v2,  v3
153 154 155 156 157
        b               32b
640:
        add             x7,  x0,  x1
        lsl             x1,  x1,  #1
64:
158 159 160 161 162
        \type           v5,  v0,  v1,  v2,  v3
        \type           v6,  v0,  v1,  v2,  v3
        \type           v7,  v0,  v1,  v2,  v3
        \type           v16, v0,  v1,  v2,  v3
        \type           v17, v0,  v1,  v2,  v3
163
        st1             {v4.16b,v5.16b,v6.16b,v7.16b}, [x0], x1
164 165
        \type           v18, v0,  v1,  v2,  v3
        \type           v19, v0,  v1,  v2,  v3
166 167 168
        subs            w5,  w5,  #2
        st1             {v16.16b,v17.16b,v18.16b,v19.16b}, [x7], x1
        b.le            0f
169
        \type           v4, v0,  v1,  v2,  v3
170 171 172 173
        b               64b
1280:
        add             x7,  x0,  #64
128:
174 175 176 177 178
        \type           v5,  v0,  v1,  v2,  v3
        \type           v6,  v0,  v1,  v2,  v3
        \type           v7,  v0,  v1,  v2,  v3
        \type           v16, v0,  v1,  v2,  v3
        \type           v17, v0,  v1,  v2,  v3
179
        st1             {v4.16b,v5.16b,v6.16b,v7.16b}, [x0], x1
180 181
        \type           v18, v0,  v1,  v2,  v3
        \type           v19, v0,  v1,  v2,  v3
182 183 184
        subs            w5,  w5,  #1
        st1             {v16.16b,v17.16b,v18.16b,v19.16b}, [x7], x1
        b.le            0f
185
        \type           v4, v0,  v1,  v2,  v3
186 187 188
        b               128b
0:
        ret
189 190
L(\type\()_tbl):
        .hword L(\type\()_tbl) - 1280b
Janne Grunau's avatar
Janne Grunau committed
191 192
        .hword L(\type\()_tbl) -  640b
        .hword L(\type\()_tbl) -  320b
193
        .hword L(\type\()_tbl) -   16b
194 195
        .hword L(\type\()_tbl) -   80b
        .hword L(\type\()_tbl) -   40b
196 197 198 199 200 201
endfunc
.endm

bidir_fn avg
bidir_fn w_avg
bidir_fn mask
202 203


204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425
.macro w_mask_fn type
function w_mask_\type\()_8bpc_neon, export=1
        clz             w8,  w4
        adr             x9,  L(w_mask_\type\()_tbl)
        sub             w8,  w8,  #24
        ldrh            w8,  [x9,  x8,  lsl #1]
        sub             x9,  x9,  w8,  uxtw
        mov             w10, #6903
        dup             v0.8h,   w10
.if \type == 444
        movi            v1.16b,  #64
.elseif \type == 422
        dup             v2.8b,   w7
        movi            v3.8b,   #129
        sub             v3.8b,   v3.8b,   v2.8b
.elseif \type == 420
        dup             v2.8h,   w7
        movi            v3.8h,   #1, lsl #8
        sub             v3.8h,   v3.8h,   v2.8h
.endif
        add             x12,  x0,  x1
        lsl             x1,   x1,  #1
        br              x9
4:
        ld1             {v4.8h,   v5.8h},   [x2],  #32  // tmp1 (four rows at once)
        ld1             {v6.8h,   v7.8h},   [x3],  #32  // tmp2 (four rows at once)
        subs            w5,  w5,  #4
        sub             v16.8h,  v6.8h,   v4.8h
        sub             v17.8h,  v7.8h,   v5.8h
        sabd            v18.8h,  v4.8h,   v6.8h
        sabd            v19.8h,  v5.8h,   v7.8h
        uqsub           v18.8h,  v0.8h,   v18.8h
        uqsub           v19.8h,  v0.8h,   v19.8h
        ushr            v18.8h,  v18.8h,  #8
        ushr            v19.8h,  v19.8h,  #8
        shl             v20.8h,  v18.8h,  #9
        shl             v21.8h,  v19.8h,  #9
        sqdmulh         v20.8h,  v20.8h,  v16.8h
        sqdmulh         v21.8h,  v21.8h,  v17.8h
        add             v20.8h,  v20.8h,  v4.8h
        add             v21.8h,  v21.8h,  v5.8h
        sqrshrun        v22.8b,  v20.8h,  #4
        sqrshrun        v23.8b,  v21.8h,  #4
.if \type == 444
        xtn             v18.8b,   v18.8h
        xtn2            v18.16b,  v19.8h
        sub             v18.16b,  v1.16b,  v18.16b
        st1             {v18.16b}, [x6],  #16
.elseif \type == 422
        addp            v18.8h,   v18.8h,  v19.8h
        xtn             v18.8b,   v18.8h
        uhsub           v18.8b,   v3.8b,   v18.8b
        st1             {v18.8b},  [x6],  #8
.elseif \type == 420
        trn1            v24.2d,   v18.2d,  v19.2d
        trn2            v25.2d,   v18.2d,  v19.2d
        add             v24.8h,   v24.8h,  v25.8h
        addp            v18.8h,   v24.8h,  v24.8h
        sub             v18.4h,   v3.4h,   v18.4h
        rshrn           v18.8b,   v18.8h,  #2
        st1             {v18.s}[0],  [x6],  #4
.endif
        st1             {v22.s}[0],  [x0],  x1
        st1             {v22.s}[1],  [x12], x1
        st1             {v23.s}[0],  [x0],  x1
        st1             {v23.s}[1],  [x12], x1
        b.gt            4b
        ret
8:
        ld1             {v4.8h,   v5.8h},   [x2],  #32
        ld1             {v6.8h,   v7.8h},   [x3],  #32
        subs            w5,  w5,  #2
        sub             v16.8h,  v6.8h,   v4.8h
        sub             v17.8h,  v7.8h,   v5.8h
        sabd            v18.8h,  v4.8h,   v6.8h
        sabd            v19.8h,  v5.8h,   v7.8h
        uqsub           v18.8h,  v0.8h,   v18.8h
        uqsub           v19.8h,  v0.8h,   v19.8h
        ushr            v18.8h,  v18.8h,  #8
        ushr            v19.8h,  v19.8h,  #8
        shl             v20.8h,  v18.8h,  #9
        shl             v21.8h,  v19.8h,  #9
        sqdmulh         v20.8h,  v20.8h,  v16.8h
        sqdmulh         v21.8h,  v21.8h,  v17.8h
        add             v20.8h,  v20.8h,  v4.8h
        add             v21.8h,  v21.8h,  v5.8h
        sqrshrun        v22.8b,  v20.8h,  #4
        sqrshrun        v23.8b,  v21.8h,  #4
.if \type == 444
        xtn             v18.8b,  v18.8h
        xtn2            v18.16b, v19.8h
        sub             v18.16b, v1.16b,  v18.16b
        st1             {v18.16b}, [x6],  #16
.elseif \type == 422
        addp            v18.8h,  v18.8h,  v19.8h
        xtn             v18.8b,  v18.8h
        uhsub           v18.8b,  v3.8b,   v18.8b
        st1             {v18.8b},  [x6],  #8
.elseif \type == 420
        add             v18.8h,  v18.8h,  v19.8h
        addp            v18.8h,  v18.8h,  v18.8h
        sub             v18.4h,  v3.4h,   v18.4h
        rshrn           v18.8b,  v18.8h,  #2
        st1             {v18.s}[0],  [x6],  #4
.endif
        st1             {v22.8b},  [x0],  x1
        st1             {v23.8b},  [x12], x1
        b.gt            8b
        ret
1280:
640:
320:
160:
        mov             w11, w4
        sub             x1,  x1,  w4,  uxtw
.if \type == 444
        add             x10, x6,  w4,  uxtw
.elseif \type == 422
        add             x10, x6,  x11, lsr #1
.endif
        add             x9,  x3,  w4,  uxtw #1
        add             x7,  x2,  w4,  uxtw #1
161:
        mov             w8,  w4
16:
        ld1             {v4.8h,   v5.8h},   [x2],  #32
        ld1             {v6.8h,   v7.8h},   [x3],  #32
        ld1             {v16.8h,  v17.8h},  [x7],  #32
        ld1             {v18.8h,  v19.8h},  [x9],  #32
        subs            w8,  w8,  #16
        sub             v6.8h,   v6.8h,   v4.8h
        sub             v7.8h,   v7.8h,   v5.8h
        sub             v18.8h,  v18.8h,  v16.8h
        sub             v19.8h,  v19.8h,  v17.8h
        abs             v20.8h,  v6.8h
        abs             v21.8h,  v7.8h
        abs             v22.8h,  v18.8h
        abs             v23.8h,  v19.8h
        uqsub           v20.8h,  v0.8h,   v20.8h
        uqsub           v21.8h,  v0.8h,   v21.8h
        uqsub           v22.8h,  v0.8h,   v22.8h
        uqsub           v23.8h,  v0.8h,   v23.8h
        ushr            v20.8h,  v20.8h,  #8
        ushr            v21.8h,  v21.8h,  #8
        ushr            v22.8h,  v22.8h,  #8
        ushr            v23.8h,  v23.8h,  #8
        shl             v24.8h,  v20.8h,  #9
        shl             v25.8h,  v21.8h,  #9
        shl             v26.8h,  v22.8h,  #9
        shl             v27.8h,  v23.8h,  #9
        sqdmulh         v24.8h,  v24.8h,  v6.8h
        sqdmulh         v25.8h,  v25.8h,  v7.8h
        sqdmulh         v26.8h,  v26.8h,  v18.8h
        sqdmulh         v27.8h,  v27.8h,  v19.8h
        add             v24.8h,  v24.8h,  v4.8h
        add             v25.8h,  v25.8h,  v5.8h
        add             v26.8h,  v26.8h,  v16.8h
        add             v27.8h,  v27.8h,  v17.8h
        sqrshrun        v24.8b,  v24.8h,  #4
        sqrshrun        v25.8b,  v25.8h,  #4
        sqrshrun        v26.8b,  v26.8h,  #4
        sqrshrun        v27.8b,  v27.8h,  #4
.if \type == 444
        xtn             v20.8b,  v20.8h
        xtn2            v20.16b, v21.8h
        xtn             v21.8b,  v22.8h
        xtn2            v21.16b, v23.8h
        sub             v20.16b, v1.16b,  v20.16b
        sub             v21.16b, v1.16b,  v21.16b
        st1             {v20.16b}, [x6],  #16
        st1             {v21.16b}, [x10], #16
.elseif \type == 422
        addp            v20.8h,  v20.8h,  v21.8h
        addp            v21.8h,  v22.8h,  v23.8h
        xtn             v20.8b,  v20.8h
        xtn             v21.8b,  v21.8h
        uhsub           v20.8b,  v3.8b,   v20.8b
        uhsub           v21.8b,  v3.8b,   v21.8b
        st1             {v20.8b},  [x6],  #8
        st1             {v21.8b},  [x10], #8
.elseif \type == 420
        add             v20.8h,  v20.8h,  v22.8h
        add             v21.8h,  v21.8h,  v23.8h
        addp            v20.8h,  v20.8h,  v21.8h
        sub             v20.8h,  v3.8h,   v20.8h
        rshrn           v20.8b,  v20.8h,  #2
        st1             {v20.8b},  [x6],  #8
.endif
        st1             {v24.8b,  v25.8b},  [x0],  #16
        st1             {v26.8b,  v27.8b},  [x12], #16
        b.gt            16b
        subs            w5,  w5,  #2
        add             x2,  x2,  w4,  uxtw #1
        add             x3,  x3,  w4,  uxtw #1
        add             x7,  x7,  w4,  uxtw #1
        add             x9,  x9,  w4,  uxtw #1
.if \type == 444
        add             x6,  x6,  w4,  uxtw
        add             x10, x10, w4,  uxtw
.elseif \type == 422
        add             x6,  x6,  x11, lsr #1
        add             x10, x10, x11, lsr #1
.endif
        add             x0,  x0,  x1
        add             x12, x12, x1
        b.gt            161b
        ret
L(w_mask_\type\()_tbl):
        .hword L(w_mask_\type\()_tbl) - 1280b
        .hword L(w_mask_\type\()_tbl) -  640b
        .hword L(w_mask_\type\()_tbl) -  320b
        .hword L(w_mask_\type\()_tbl) -  160b
        .hword L(w_mask_\type\()_tbl) -    8b
        .hword L(w_mask_\type\()_tbl) -    4b
endfunc
.endm

w_mask_fn 444
w_mask_fn 422
w_mask_fn 420


426 427 428 429 430 431 432 433
function blend_8bpc_neon, export=1
        adr             x6,  L(blend_tbl)
        clz             w3,  w3
        sub             w3,  w3,  #26
        ldrh            w3,  [x6,  x3,  lsl #1]
        sub             x6,  x6,  w3,  uxtw
        movi            v4.16b,  #64
        add             x8,  x0,  x1
434
        lsl             x1,  x1,  #1
435 436
        br              x6
4:
437
        ld1             {v2.8b},     [x5],  #8
438 439 440 441 442 443 444 445 446 447 448 449 450
        ld1             {v1.d}[0],   [x2],  #8
        ld1             {v0.s}[0],   [x0]
        subs            w4,  w4,  #2
        ld1             {v0.s}[1],   [x8]
        sub             v3.8b,   v4.8b,   v2.8b
        umull           v5.8h,   v1.8b,   v2.8b
        umlal           v5.8h,   v0.8b,   v3.8b
        rshrn           v6.8b,   v5.8h,   #6
        st1             {v6.s}[0],   [x0],  x1
        st1             {v6.s}[1],   [x8],  x1
        b.gt            4b
        ret
8:
451 452
        ld1             {v2.16b},  [x5],  #16
        ld1             {v1.16b},  [x2],  #16
453 454 455 456 457 458 459 460 461 462 463 464 465 466 467
        ld1             {v0.d}[0],   [x0]
        ld1             {v0.d}[1],   [x8]
        sub             v3.16b,  v4.16b,  v2.16b
        subs            w4,  w4,  #2
        umull           v5.8h,   v1.8b,   v2.8b
        umlal           v5.8h,   v0.8b,   v3.8b
        umull2          v6.8h,   v1.16b,  v2.16b
        umlal2          v6.8h,   v0.16b,  v3.16b
        rshrn           v7.8b,   v5.8h,   #6
        rshrn2          v7.16b,  v6.8h,   #6
        st1             {v7.d}[0],   [x0],  x1
        st1             {v7.d}[1],   [x8],  x1
        b.gt            8b
        ret
16:
468 469 470
        ld1             {v1.16b,  v2.16b},  [x5],  #32
        ld1             {v5.16b,  v6.16b},  [x2],  #32
        ld1             {v0.16b},  [x0]
471 472 473
        subs            w4,  w4,  #2
        sub             v7.16b,  v4.16b,  v1.16b
        sub             v20.16b, v4.16b,  v2.16b
474
        ld1             {v3.16b},  [x8]
475 476 477 478 479 480 481 482 483 484 485 486
        umull           v16.8h,  v5.8b,   v1.8b
        umlal           v16.8h,  v0.8b,   v7.8b
        umull2          v17.8h,  v5.16b,  v1.16b
        umlal2          v17.8h,  v0.16b,  v7.16b
        umull           v21.8h,  v6.8b,   v2.8b
        umlal           v21.8h,  v3.8b,   v20.8b
        umull2          v22.8h,  v6.16b,  v2.16b
        umlal2          v22.8h,  v3.16b,  v20.16b
        rshrn           v18.8b,  v16.8h,  #6
        rshrn2          v18.16b, v17.8h,  #6
        rshrn           v19.8b,  v21.8h,  #6
        rshrn2          v19.16b, v22.8h,  #6
487 488
        st1             {v18.16b}, [x0],  x1
        st1             {v19.16b}, [x8],  x1
489 490 491
        b.gt            16b
        ret
32:
492 493 494
        ld1             {v0.16b,  v1.16b,  v2.16b,  v3.16b},  [x5],  #64
        ld1             {v16.16b, v17.16b, v18.16b, v19.16b}, [x2],  #64
        ld1             {v20.16b, v21.16b}, [x0]
495
        subs            w4,  w4,  #2
496
        ld1             {v22.16b, v23.16b}, [x8]
497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524
        sub             v5.16b,  v4.16b,  v0.16b
        sub             v6.16b,  v4.16b,  v1.16b
        sub             v30.16b, v4.16b,  v2.16b
        sub             v31.16b, v4.16b,  v3.16b
        umull           v24.8h,  v16.8b,  v0.8b
        umlal           v24.8h,  v20.8b,  v5.8b
        umull2          v26.8h,  v16.16b, v0.16b
        umlal2          v26.8h,  v20.16b, v5.16b
        umull           v28.8h,  v17.8b,  v1.8b
        umlal           v28.8h,  v21.8b,  v6.8b
        umull2          v7.8h,   v17.16b, v1.16b
        umlal2          v7.8h,   v21.16b, v6.16b
        umull           v27.8h,  v18.8b,  v2.8b
        umlal           v27.8h,  v22.8b,  v30.8b
        umull2          v1.8h,   v18.16b, v2.16b
        umlal2          v1.8h,   v22.16b, v30.16b
        umull           v29.8h,  v19.8b,  v3.8b
        umlal           v29.8h,  v23.8b,  v31.8b
        umull2          v21.8h,  v19.16b, v3.16b
        umlal2          v21.8h,  v23.16b, v31.16b
        rshrn           v24.8b,  v24.8h,  #6
        rshrn2          v24.16b, v26.8h,  #6
        rshrn           v25.8b,  v28.8h,  #6
        rshrn2          v25.16b, v7.8h,   #6
        rshrn           v27.8b,  v27.8h,  #6
        rshrn2          v27.16b, v1.8h,   #6
        rshrn           v28.8b,  v29.8h,  #6
        rshrn2          v28.16b, v21.8h,  #6
525 526
        st1             {v24.16b, v25.16b}, [x0],  x1
        st1             {v27.16b, v28.16b}, [x8],  x1
527 528 529 530 531 532 533 534 535 536
        b.gt            32b
        ret
L(blend_tbl):
        .hword L(blend_tbl) - 32b
        .hword L(blend_tbl) - 16b
        .hword L(blend_tbl) -  8b
        .hword L(blend_tbl) -  4b
endfunc

function blend_h_8bpc_neon, export=1
Martin Storsjö's avatar
Martin Storsjö committed
537
        adr             x6,  L(blend_h_tbl)
538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565
        movrel          x5,  X(obmc_masks)
        add             x5,  x5,  w4,  uxtw
        sub             w4,  w4,  w4,  lsr #2
        clz             w7,  w3
        movi            v4.16b,  #64
        add             x8,  x0,  x1
        lsl             x1,  x1,  #1
        sub             w7,  w7,  #24
        ldrh            w7,  [x6,  x7,  lsl #1]
        sub             x6,  x6,  w7, uxtw
        br              x6
2:
        ld1             {v0.h}[0],   [x5],  #2
        ld1             {v1.s}[0],   [x2],  #4
        subs            w4,  w4,  #2
        ld1             {v2.h}[0],   [x0]
        zip1            v0.8b,   v0.8b,   v0.8b
        sub             v3.8b,   v4.8b,   v0.8b
        ld1             {v2.h}[1],   [x8]
        umull           v5.8h,   v1.8b,   v0.8b
        umlal           v5.8h,   v2.8b,   v3.8b
        rshrn           v5.8b,   v5.8h,   #6
        st1             {v5.h}[0],   [x0],  x1
        st1             {v5.h}[1],   [x8],  x1
        b.gt            2b
        ret
4:
        ld2r            {v0.8b,   v1.8b},   [x5],  #2
566
        ld1             {v2.8b},   [x2],  #8
567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832
        subs            w4,  w4,  #2
        ext             v0.8b,   v0.8b,   v1.8b,   #4
        ld1             {v3.s}[0],   [x0]
        sub             v5.8b,   v4.8b,   v0.8b
        ld1             {v3.s}[1],   [x8]
        umull           v6.8h,   v2.8b,   v0.8b
        umlal           v6.8h,   v3.8b,   v5.8b
        rshrn           v6.8b,   v6.8h,   #6
        st1             {v6.s}[0],   [x0],  x1
        st1             {v6.s}[1],   [x8],  x1
        b.gt            4b
        ret
8:
        ld2r            {v0.16b,  v1.16b},  [x5],  #2
        ld1             {v2.16b},  [x2],  #16
        ld1             {v3.d}[0],   [x0]
        ext             v0.16b,  v0.16b,  v1.16b,  #8
        sub             v5.16b,  v4.16b,  v0.16b
        ld1             {v3.d}[1],   [x8]
        subs            w4,  w4,  #2
        umull           v6.8h,   v0.8b,   v2.8b
        umlal           v6.8h,   v3.8b,   v5.8b
        umull2          v7.8h,   v0.16b,  v2.16b
        umlal2          v7.8h,   v3.16b,  v5.16b
        rshrn           v16.8b,  v6.8h,   #6
        rshrn2          v16.16b, v7.8h,   #6
        st1             {v16.d}[0],  [x0],  x1
        st1             {v16.d}[1],  [x8],  x1
        b.gt            8b
        ret
16:
        ld2r            {v0.16b,  v1.16b},  [x5],  #2
        ld1             {v2.16b,  v3.16b},  [x2],  #32
        ld1             {v5.16b},  [x0]
        sub             v7.16b,  v4.16b,  v0.16b
        sub             v16.16b, v4.16b,  v1.16b
        ld1             {v6.16b},  [x8]
        subs            w4,  w4,  #2
        umull           v17.8h,  v0.8b,   v2.8b
        umlal           v17.8h,  v5.8b,   v7.8b
        umull2          v18.8h,  v0.16b,  v2.16b
        umlal2          v18.8h,  v5.16b,  v7.16b
        umull           v19.8h,  v1.8b,   v3.8b
        umlal           v19.8h,  v6.8b,   v16.8b
        umull2          v20.8h,  v1.16b,  v3.16b
        umlal2          v20.8h,  v6.16b,  v16.16b
        rshrn           v21.8b,  v17.8h,  #6
        rshrn2          v21.16b, v18.8h,  #6
        rshrn           v22.8b,  v19.8h,  #6
        rshrn2          v22.16b, v20.8h,  #6
        st1             {v21.16b}, [x0],  x1
        st1             {v22.16b}, [x8],  x1
        b.gt            16b
        ret
1280:
640:
320:
        sub             x1,  x1,  w3,  uxtw
        add             x7,  x2,  w3,  uxtw
321:
        ld2r            {v0.16b,  v1.16b},  [x5],  #2
        mov             w6,  w3
        sub             v20.16b, v4.16b,  v0.16b
        sub             v21.16b, v4.16b,  v1.16b
32:
        ld1             {v16.16b, v17.16b}, [x2],  #32
        ld1             {v2.16b,  v3.16b},  [x0]
        subs            w6,  w6,  #32
        umull           v23.8h,  v0.8b,   v16.8b
        umlal           v23.8h,  v2.8b,   v20.8b
        ld1             {v18.16b, v19.16b}, [x7],  #32
        umull2          v27.8h,  v0.16b,  v16.16b
        umlal2          v27.8h,  v2.16b,  v20.16b
        ld1             {v6.16b,  v7.16b},  [x8]
        umull           v24.8h,  v0.8b,   v17.8b
        umlal           v24.8h,  v3.8b,   v20.8b
        umull2          v28.8h,  v0.16b,  v17.16b
        umlal2          v28.8h,  v3.16b,  v20.16b
        umull           v25.8h,  v1.8b,   v18.8b
        umlal           v25.8h,  v6.8b,   v21.8b
        umull2          v5.8h,   v1.16b,  v18.16b
        umlal2          v5.8h,   v6.16b,  v21.16b
        rshrn           v29.8b,  v23.8h,  #6
        rshrn2          v29.16b, v27.8h,  #6
        umull           v26.8h,  v1.8b,   v19.8b
        umlal           v26.8h,  v7.8b,   v21.8b
        umull2          v31.8h,  v1.16b,  v19.16b
        umlal2          v31.8h,  v7.16b,  v21.16b
        rshrn           v30.8b,  v24.8h,  #6
        rshrn2          v30.16b, v28.8h,  #6
        rshrn           v23.8b,  v25.8h,  #6
        rshrn2          v23.16b, v5.8h,   #6
        rshrn           v24.8b,  v26.8h,  #6
        st1             {v29.16b, v30.16b}, [x0],  #32
        rshrn2          v24.16b, v31.8h,  #6
        st1             {v23.16b, v24.16b}, [x8],  #32
        b.gt            32b
        subs            w4,  w4,  #2
        add             x0,  x0,  x1
        add             x8,  x8,  x1
        add             x2,  x2,  w3,  uxtw
        add             x7,  x7,  w3,  uxtw
        b.gt            321b
        ret
L(blend_h_tbl):
        .hword L(blend_h_tbl) - 1280b
        .hword L(blend_h_tbl) -  640b
        .hword L(blend_h_tbl) -  320b
        .hword L(blend_h_tbl) -   16b
        .hword L(blend_h_tbl) -    8b
        .hword L(blend_h_tbl) -    4b
        .hword L(blend_h_tbl) -    2b
endfunc

function blend_v_8bpc_neon, export=1
        adr             x6,  L(blend_v_tbl)
        movrel          x5,  X(obmc_masks)
        add             x5,  x5,  w3,  uxtw
        clz             w3,  w3
        movi            v4.16b,  #64
        add             x8,  x0,  x1
        lsl             x1,  x1,  #1
        sub             w3,  w3,  #26
        ldrh            w3,  [x6,  x3,  lsl #1]
        sub             x6,  x6,  w3,  uxtw
        br              x6
20:
        ld1r            {v0.8b},   [x5]
        sub             v1.8b,   v4.8b,   v0.8b
2:
        ld1             {v2.h}[0],   [x2],  #2
        ld1             {v3.b}[0],   [x0]
        subs            w4,  w4,  #2
        ld1             {v2.b}[1],   [x2]
        ld1             {v3.b}[1],   [x8]
        umull           v5.8h,   v2.8b,   v0.8b
        umlal           v5.8h,   v3.8b,   v1.8b
        rshrn           v5.8b,   v5.8h,   #6
        add             x2,  x2,  #2
        st1             {v5.b}[0],   [x0],  x1
        st1             {v5.b}[1],   [x8],  x1
        b.gt            2b
        ret
40:
        ld1r            {v0.2s},   [x5]
        sub             v1.8b,   v4.8b,   v0.8b
        sub             x1,  x1,  #3
4:
        ld1             {v2.8b},   [x2],  #8
        ld1             {v3.s}[0],   [x0]
        ld1             {v3.s}[1],   [x8]
        subs            w4,  w4,  #2
        umull           v5.8h,   v2.8b,   v0.8b
        umlal           v5.8h,   v3.8b,   v1.8b
        rshrn           v5.8b,   v5.8h,   #6
        st1             {v5.h}[0],   [x0],  #2
        st1             {v5.h}[2],   [x8],  #2
        st1             {v5.b}[2],   [x0],  #1
        st1             {v5.b}[6],   [x8],  #1
        add             x0,  x0,  x1
        add             x8,  x8,  x1
        b.gt            4b
        ret
80:
        ld1r            {v0.2d},   [x5]
        sub             v1.16b,  v4.16b,  v0.16b
        sub             x1,  x1,  #6
8:
        ld1             {v2.16b},  [x2],  #16
        ld1             {v3.d}[0],   [x0]
        ld1             {v3.d}[1],   [x8]
        subs            w4,  w4,  #2
        umull           v5.8h,  v0.8b,  v2.8b
        umlal           v5.8h,  v3.8b,  v1.8b
        umull2          v6.8h,  v0.16b, v2.16b
        umlal2          v6.8h,  v3.16b, v1.16b
        rshrn           v7.8b,  v5.8h,  #6
        rshrn2          v7.16b, v6.8h,  #6
        st1             {v7.s}[0],   [x0],  #4
        st1             {v7.s}[2],   [x8],  #4
        st1             {v7.h}[2],   [x0],  #2
        st1             {v7.h}[6],   [x8],  #2
        add             x0,  x0,  x1
        add             x8,  x8,  x1
        b.gt            8b
        ret
160:
        ld1             {v0.16b},  [x5]
        sub             v2.16b,  v4.16b,  v0.16b
        sub             x1,  x1,  #12
16:
        ld1             {v5.16b,  v6.16b},  [x2],  #32
        ld1             {v7.16b},  [x0]
        subs            w4,  w4,  #2
        ld1             {v16.16b}, [x8]
        umull           v17.8h,  v5.8b,   v0.8b
        umlal           v17.8h,  v7.8b,   v2.8b
        umull2          v18.8h,  v5.16b,  v0.16b
        umlal2          v18.8h,  v7.16b,  v2.16b
        umull           v20.8h,  v6.8b,   v0.8b
        umlal           v20.8h,  v16.8b,  v2.8b
        umull2          v21.8h,  v6.16b,  v0.16b
        umlal2          v21.8h,  v16.16b, v2.16b
        rshrn           v19.8b,  v17.8h,  #6
        rshrn2          v19.16b, v18.8h,  #6
        rshrn           v22.8b,  v20.8h,  #6
        rshrn2          v22.16b, v21.8h,  #6
        st1             {v19.8b},  [x0],  #8
        st1             {v22.8b},  [x8],  #8
        st1             {v19.s}[2],  [x0],  #4
        st1             {v22.s}[2],  [x8],  #4
        add             x0,  x0,  x1
        add             x8,  x8,  x1
        b.gt            16b
        ret
320:
        ld1             {v0.16b,  v1.16b},  [x5]
        sub             v2.16b,  v4.16b,  v0.16b
        sub             v3.16b,  v4.16b,  v1.16b
        sub             x1,  x1,  #24
32:
        ld1             {v16.16b, v17.16b, v18.16b, v19.16b}, [x2],  #64
        ld1             {v5.16b,  v6.16b},  [x0]
        subs            w4,  w4,  #2
        ld1             {v20.16b, v21.16b}, [x8]
        umull           v22.8h,  v16.8b,  v0.8b
        umlal           v22.8h,  v5.8b,   v2.8b
        umull2          v23.8h,  v16.16b, v0.16b
        umlal2          v23.8h,  v5.16b,  v2.16b
        umull           v28.8h,  v17.8b,  v1.8b
        umlal           v28.8h,  v6.8b,   v3.8b
        umull2          v29.8h,  v17.16b, v1.16b
        umlal2          v29.8h,  v6.16b,  v3.16b
        umull           v30.8h,  v18.8b,  v0.8b
        umlal           v30.8h,  v20.8b,  v2.8b
        umull2          v31.8h,  v18.16b, v0.16b
        umlal2          v31.8h,  v20.16b, v2.16b
        umull           v25.8h,  v19.8b,  v1.8b
        umlal           v25.8h,  v21.8b,  v3.8b
        umull2          v26.8h,  v19.16b, v1.16b
        umlal2          v26.8h,  v21.16b, v3.16b
        rshrn           v24.8b,  v22.8h,  #6
        rshrn2          v24.16b, v23.8h,  #6
        rshrn           v28.8b,  v28.8h,  #6
        rshrn2          v28.16b, v29.8h,  #6
        rshrn           v30.8b,  v30.8h,  #6
        rshrn2          v30.16b, v31.8h,  #6
        rshrn           v27.8b,  v25.8h,  #6
        rshrn2          v27.16b, v26.8h,  #6
        st1             {v24.16b}, [x0],  #16
        st1             {v30.16b}, [x8],  #16
        st1             {v28.8b},  [x0],  #8
        st1             {v27.8b},  [x8],  #8
        add             x0,  x0,  x1
        add             x8,  x8,  x1
        b.gt            32b
        ret
L(blend_v_tbl):
        .hword L(blend_v_tbl) - 320b
        .hword L(blend_v_tbl) - 160b
        .hword L(blend_v_tbl) -  80b
        .hword L(blend_v_tbl) -  40b
        .hword L(blend_v_tbl) -  20b
endfunc


833
// This has got the same signature as the put_8tap functions,
Martin Storsjö's avatar
Martin Storsjö committed
834
// and assumes that x8 is set to (clz(w)-24).
835
function put_neon
836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928
        adr             x9,  L(put_tbl)
        ldrh            w8,  [x9, x8, lsl #1]
        sub             x9,  x9,  w8, uxtw
        br              x9

2:
        ld1             {v0.h}[0], [x2], x3
        ld1             {v1.h}[0], [x2], x3
        subs            w5,  w5,  #2
        st1             {v0.h}[0], [x0], x1
        st1             {v1.h}[0], [x0], x1
        b.gt            2b
        ret
4:
        ld1             {v0.s}[0], [x2], x3
        ld1             {v1.s}[0], [x2], x3
        subs            w5,  w5,  #2
        st1             {v0.s}[0], [x0], x1
        st1             {v1.s}[0], [x0], x1
        b.gt            4b
        ret
8:
        ld1             {v0.8b}, [x2], x3
        ld1             {v1.8b}, [x2], x3
        subs            w5,  w5,  #2
        st1             {v0.8b}, [x0], x1
        st1             {v1.8b}, [x0], x1
        b.gt            8b
        ret
160:
        add             x8,  x0,  x1
        lsl             x1,  x1,  #1
        add             x9,  x2,  x3
        lsl             x3,  x3,  #1
16:
        ld1             {v0.16b}, [x2], x3
        ld1             {v1.16b}, [x9], x3
        subs            w5,  w5,  #2
        st1             {v0.16b}, [x0], x1
        st1             {v1.16b}, [x8], x1
        b.gt            16b
        ret
32:
        ldp             x6,  x7,  [x2]
        ldp             x8,  x9,  [x2, #16]
        stp             x6,  x7,  [x0]
        subs            w5,  w5,  #1
        stp             x8,  x9,  [x0, #16]
        add             x2,  x2,  x3
        add             x0,  x0,  x1
        b.gt            32b
        ret
64:
        ldp             x6,  x7,  [x2]
        ldp             x8,  x9,  [x2, #16]
        stp             x6,  x7,  [x0]
        ldp             x10, x11, [x2, #32]
        stp             x8,  x9,  [x0, #16]
        subs            w5,  w5,  #1
        ldp             x12, x13, [x2, #48]
        stp             x10, x11, [x0, #32]
        stp             x12, x13, [x0, #48]
        add             x2,  x2,  x3
        add             x0,  x0,  x1
        b.gt            64b
        ret
128:
        ldp             q0,  q1,  [x2]
        ldp             q2,  q3,  [x2, #32]
        stp             q0,  q1,  [x0]
        ldp             q4,  q5,  [x2, #64]
        stp             q2,  q3,  [x0, #32]
        ldp             q6,  q7,  [x2, #96]
        subs            w5,  w5,  #1
        stp             q4,  q5,  [x0, #64]
        stp             q6,  q7,  [x0, #96]
        add             x2,  x2,  x3
        add             x0,  x0,  x1
        b.gt            128b
        ret

L(put_tbl):
        .hword L(put_tbl) - 128b
        .hword L(put_tbl) -  64b
        .hword L(put_tbl) -  32b
        .hword L(put_tbl) - 160b
        .hword L(put_tbl) -   8b
        .hword L(put_tbl) -   4b
        .hword L(put_tbl) -   2b
endfunc


// This has got the same signature as the prep_8tap functions,
Martin Storsjö's avatar
Martin Storsjö committed
929
// and assumes that x8 is set to (clz(w)-24), and x7 to w*2.
930
function prep_neon
931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144
        adr             x9,  L(prep_tbl)
        ldrh            w8,  [x9, x8, lsl #1]
        sub             x9,  x9,  w8, uxtw
        br              x9

4:
        ld1             {v0.s}[0], [x1], x2
        ld1             {v1.s}[0], [x1], x2
        subs            w4,  w4,  #2
        ushll           v0.8h, v0.8b, #4
        ushll           v1.8h, v1.8b, #4
        st1             {v0.4h, v1.4h}, [x0], #16
        b.gt            4b
        ret
8:
        ld1             {v0.8b}, [x1], x2
        ld1             {v1.8b}, [x1], x2
        subs            w4,  w4,  #2
        ushll           v0.8h, v0.8b, #4
        ushll           v1.8h, v1.8b, #4
        st1             {v0.8h, v1.8h}, [x0], #32
        b.gt            8b
        ret
160:
        add             x9,  x1,  x2
        lsl             x2,  x2,  #1
16:
        ld1             {v0.16b}, [x1], x2
        ld1             {v1.16b}, [x9], x2
        subs            w4,  w4,  #2
        ushll           v4.8h, v0.8b,  #4
        ushll2          v5.8h, v0.16b, #4
        ushll           v6.8h, v1.8b,  #4
        ushll2          v7.8h, v1.16b, #4
        st1             {v4.8h, v5.8h, v6.8h, v7.8h}, [x0], #64
        b.gt            16b
        ret
320:
        add             x8,  x0,  w3, uxtw
32:
        ld1             {v0.16b, v1.16b},  [x1], x2
        subs            w4,  w4,  #2
        ushll           v4.8h,  v0.8b,  #4
        ushll2          v5.8h,  v0.16b, #4
        ld1             {v2.16b, v3.16b},  [x1], x2
        ushll           v6.8h,  v1.8b,  #4
        ushll2          v7.8h,  v1.16b, #4
        ushll           v16.8h, v2.8b,  #4
        st1             {v4.8h,  v5.8h},  [x0], x7
        ushll2          v17.8h, v2.16b, #4
        st1             {v6.8h,  v7.8h},  [x8], x7
        ushll           v18.8h, v3.8b,  #4
        st1             {v16.8h, v17.8h}, [x0], x7
        ushll2          v19.8h, v3.16b, #4
        st1             {v18.8h, v19.8h}, [x8], x7
        b.gt            32b
        ret
640:
        add             x8,  x0,  #32
        mov             x6,  #64
64:
        ldp             q0,  q1,  [x1]
        subs            w4,  w4,  #1
        ushll           v4.8h,  v0.8b,  #4
        ushll2          v5.8h,  v0.16b, #4
        ldp             q2,  q3,  [x1, #32]
        ushll           v6.8h,  v1.8b,  #4
        ushll2          v7.8h,  v1.16b, #4
        add             x1,  x1,  x2
        ushll           v16.8h, v2.8b,  #4
        st1             {v4.8h,  v5.8h},  [x0], x6
        ushll2          v17.8h, v2.16b, #4
        ushll           v18.8h, v3.8b,  #4
        st1             {v6.8h,  v7.8h},  [x8], x6
        ushll2          v19.8h, v3.16b, #4
        st1             {v16.8h, v17.8h}, [x0], x6
        st1             {v18.8h, v19.8h}, [x8], x6
        b.gt            64b
        ret
1280:
        add             x8,  x0,  #64
        mov             x6,  #128
128:
        ldp             q0,  q1,  [x1]
        ldp             q2,  q3,  [x1, #32]
        ushll           v16.8h,  v0.8b,  #4
        ushll2          v17.8h,  v0.16b, #4
        ushll           v18.8h,  v1.8b,  #4
        ushll2          v19.8h,  v1.16b, #4
        ushll           v20.8h,  v2.8b,  #4
        ushll2          v21.8h,  v2.16b, #4
        ldp             q4,  q5,  [x1, #64]
        st1             {v16.8h, v17.8h, v18.8h, v19.8h}, [x0], x6
        ushll           v22.8h,  v3.8b,  #4
        ushll2          v23.8h,  v3.16b, #4
        ushll           v24.8h,  v4.8b,  #4
        ushll2          v25.8h,  v4.16b, #4
        ushll           v26.8h,  v5.8b,  #4
        ushll2          v27.8h,  v5.16b, #4
        ldp             q6,  q7,  [x1, #96]
        st1             {v20.8h, v21.8h, v22.8h, v23.8h}, [x8], x6
        ushll           v28.8h,  v6.8b,  #4
        ushll2          v29.8h,  v6.16b, #4
        ushll           v30.8h,  v7.8b,  #4
        ushll2          v31.8h,  v7.16b, #4
        subs            w4,  w4,  #1
        add             x1,  x1,  x2
        st1             {v24.8h, v25.8h, v26.8h, v27.8h}, [x0], x6
        st1             {v28.8h, v29.8h, v30.8h, v31.8h}, [x8], x6
        b.gt            128b
        ret

L(prep_tbl):
        .hword L(prep_tbl) - 1280b
        .hword L(prep_tbl) -  640b
        .hword L(prep_tbl) -  320b
        .hword L(prep_tbl) -  160b
        .hword L(prep_tbl) -    8b
        .hword L(prep_tbl) -    4b
endfunc


.macro load_slice s0, s1, strd, wd, d0, d1, d2, d3, d4, d5, d6
        ld1             {\d0\wd}[0], [\s0], \strd
        ld1             {\d1\wd}[0], [\s1], \strd
.ifnb \d2
        ld1             {\d2\wd}[0], [\s0], \strd
        ld1             {\d3\wd}[0], [\s1], \strd
.endif
.ifnb \d4
        ld1             {\d4\wd}[0], [\s0], \strd
.endif
.ifnb \d5
        ld1             {\d5\wd}[0], [\s1], \strd
.endif
.ifnb \d6
        ld1             {\d6\wd}[0], [\s0], \strd
.endif
.endm
.macro load_reg s0, s1, strd, wd, d0, d1, d2, d3, d4, d5, d6
        ld1             {\d0\wd}, [\s0], \strd
        ld1             {\d1\wd}, [\s1], \strd
.ifnb \d2
        ld1             {\d2\wd}, [\s0], \strd
        ld1             {\d3\wd}, [\s1], \strd
.endif
.ifnb \d4
        ld1             {\d4\wd}, [\s0], \strd
.endif
.ifnb \d5
        ld1             {\d5\wd}, [\s1], \strd
.endif
.ifnb \d6
        ld1             {\d6\wd}, [\s0], \strd
.endif
.endm
.macro load_h s0, s1, strd, d0, d1, d2, d3, d4, d5, d6
        load_slice      \s0, \s1, \strd, .h, \d0, \d1, \d2, \d3, \d4, \d5, \d6
.endm
.macro load_s s0, s1, strd, d0, d1, d2, d3, d4, d5, d6
        load_slice      \s0, \s1, \strd, .s, \d0, \d1, \d2, \d3, \d4, \d5, \d6
.endm
.macro load_8b s0, s1, strd, d0, d1, d2, d3, d4, d5, d6
        load_reg        \s0, \s1, \strd, .8b, \d0, \d1, \d2, \d3, \d4, \d5, \d6
.endm
.macro load_16b s0, s1, strd, d0, d1, d2, d3, d4, d5, d6
        load_reg        \s0, \s1, \strd, .16b, \d0, \d1, \d2, \d3, \d4, \d5, \d6
.endm
.macro interleave_1 wd, r0, r1, r2, r3, r4
        trn1            \r0\wd, \r0\wd, \r1\wd
        trn1            \r1\wd, \r1\wd, \r2\wd
.ifnb \r3
        trn1            \r2\wd, \r2\wd, \r3\wd
        trn1            \r3\wd, \r3\wd, \r4\wd
.endif
.endm
.macro interleave_1_h r0, r1, r2, r3, r4
        interleave_1    .4h, \r0, \r1, \r2, \r3, \r4
.endm
.macro interleave_1_s r0, r1, r2, r3, r4
        interleave_1    .2s, \r0, \r1, \r2, \r3, \r4
.endm
.macro interleave_2 wd, r0, r1, r2, r3, r4, r5
        trn1            \r0\wd,  \r0\wd, \r2\wd
        trn1            \r1\wd,  \r1\wd, \r3\wd
        trn1            \r2\wd,  \r2\wd, \r4\wd
        trn1            \r3\wd,  \r3\wd, \r5\wd
.endm
.macro interleave_2_s r0, r1, r2, r3, r4, r5
        interleave_2    .2s, \r0, \r1, \r2, \r3, \r4, \r5
.endm
.macro uxtl_b r0, r1, r2, r3, r4, r5, r6
        uxtl            \r0\().8h, \r0\().8b
        uxtl            \r1\().8h, \r1\().8b
.ifnb \r2
        uxtl            \r2\().8h, \r2\().8b
        uxtl            \r3\().8h, \r3\().8b
.endif
.ifnb \r4
        uxtl            \r4\().8h, \r4\().8b
.endif
.ifnb \r5
        uxtl            \r5\().8h, \r5\().8b
.endif
.ifnb \r6
        uxtl            \r6\().8h, \r6\().8b
.endif
.endm
.macro mul_mla_4 d, s0, s1, s2, s3, wd
        mul             \d\wd,  \s0\wd,  v0.h[0]
        mla             \d\wd,  \s1\wd,  v0.h[1]
        mla             \d\wd,  \s2\wd,  v0.h[2]
        mla             \d\wd,  \s3\wd,  v0.h[3]
.endm
1145 1146 1147
// Interleaving the mul/mla chains actually hurts performance
// significantly on Cortex A53, thus keeping mul/mla tightly
// chained like this.
1148 1149 1150 1151 1152 1153 1154 1155 1156
.macro mul_mla_8_1 d0, d1, s0, s1, s2, s3, s4, s5, s6, s7, s8
        mul             \d0\().8h, \s0\().8h, v0.h[0]
        mla             \d0\().8h, \s1\().8h, v0.h[1]
        mla             \d0\().8h, \s2\().8h, v0.h[2]
        mla             \d0\().8h, \s3\().8h, v0.h[3]
        mla             \d0\().8h, \s4\().8h, v0.h[4]
        mla             \d0\().8h, \s5\().8h, v0.h[5]
        mla             \d0\().8h, \s6\().8h, v0.h[6]
        mla             \d0\().8h, \s7\().8h, v0.h[7]
1157 1158 1159 1160 1161 1162 1163
        mul             \d1\().8h, \s1\().8h, v0.h[0]
        mla             \d1\().8h, \s2\().8h, v0.h[1]
        mla             \d1\().8h, \s3\().8h, v0.h[2]
        mla             \d1\().8h, \s4\().8h, v0.h[3]
        mla             \d1\().8h, \s5\().8h, v0.h[4]
        mla             \d1\().8h, \s6\().8h, v0.h[5]
        mla             \d1\().8h, \s7\().8h, v0.h[6]
1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174
        mla             \d1\().8h, \s8\().8h, v0.h[7]
.endm
.macro mul_mla_8_2 d0, d1, s0, s1, s2, s3, s4, s5, s6, s7, s8, s9
        mul             \d0\().8h, \s0\().8h, v0.h[0]
        mla             \d0\().8h, \s1\().8h, v0.h[1]
        mla             \d0\().8h, \s2\().8h, v0.h[2]
        mla             \d0\().8h, \s3\().8h, v0.h[3]
        mla             \d0\().8h, \s4\().8h, v0.h[4]
        mla             \d0\().8h, \s5\().8h, v0.h[5]
        mla             \d0\().8h, \s6\().8h, v0.h[6]
        mla             \d0\().8h, \s7\().8h, v0.h[7]
1175 1176 1177 1178 1179 1180 1181
        mul             \d1\().8h, \s2\().8h, v0.h[0]
        mla             \d1\().8h, \s3\().8h, v0.h[1]
        mla             \d1\().8h, \s4\().8h, v0.h[2]
        mla             \d1\().8h, \s5\().8h, v0.h[3]
        mla             \d1\().8h, \s6\().8h, v0.h[4]
        mla             \d1\().8h, \s7\().8h, v0.h[5]
        mla             \d1\().8h, \s8\().8h, v0.h[6]
1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192
        mla             \d1\().8h, \s9\().8h, v0.h[7]
.endm
.macro mul_mla_8_4 d0, d1, s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11
        mul             \d0\().8h, \s0\().8h,  v0.h[0]
        mla             \d0\().8h, \s1\().8h,  v0.h[1]
        mla             \d0\().8h, \s2\().8h,  v0.h[2]
        mla             \d0\().8h, \s3\().8h,  v0.h[3]
        mla             \d0\().8h, \s4\().8h,  v0.h[4]
        mla             \d0\().8h, \s5\().8h,  v0.h[5]
        mla             \d0\().8h, \s6\().8h,  v0.h[6]
        mla             \d0\().8h, \s7\().8h,  v0.h[7]
1193 1194 1195 1196 1197 1198 1199
        mul             \d1\().8h, \s4\().8h,  v0.h[0]
        mla             \d1\().8h, \s5\().8h,  v0.h[1]
        mla             \d1\().8h, \s6\().8h,  v0.h[2]
        mla             \d1\().8h, \s7\().8h,  v0.h[3]
        mla             \d1\().8h, \s8\().8h,  v0.h[4]
        mla             \d1\().8h, \s9\().8h,  v0.h[5]
        mla             \d1\().8h, \s10\().8h, v0.h[6]
1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229
        mla             \d1\().8h, \s11\().8h, v0.h[7]
.endm
.macro sqrshrun_b shift, r0, r1, r2, r3
        sqrshrun        \r0\().8b, \r0\().8h,  #\shift
.ifnb \r1
        sqrshrun        \r1\().8b, \r1\().8h,  #\shift
.endif
.ifnb \r2
        sqrshrun        \r2\().8b, \r2\().8h,  #\shift
        sqrshrun        \r3\().8b, \r3\().8h,  #\shift
.endif
.endm
.macro srshr_h shift, r0, r1, r2, r3
        srshr           \r0\().8h, \r0\().8h,  #\shift
.ifnb \r1
        srshr           \r1\().8h, \r1\().8h,  #\shift
.endif
.ifnb \r2
        srshr           \r2\().8h, \r2\().8h,  #\shift
        srshr           \r3\().8h, \r3\().8h,  #\shift
.endif
.endm
.macro st_h strd, reg, lanes
        st1             {\reg\().h}[0], [x0], \strd
        st1             {\reg\().h}[1], [x8], \strd
.if \lanes > 2
        st1             {\reg\().h}[2], [x0], \strd
        st1             {\reg\().h}[3], [x8], \strd
.endif
.endm
1230
.macro st_s strd, r0, r1
1231 1232 1233 1234 1235 1236 1237
        st1             {\r0\().s}[0], [x0], \strd
        st1             {\r0\().s}[1], [x8], \strd
.ifnb \r1
        st1             {\r1\().s}[0], [x0], \strd
        st1             {\r1\().s}[1], [x8], \strd
.endif
.endm
1238
.macro st_d strd, r0, r1
1239 1240 1241 1242 1243 1244 1245
        st1             {\r0\().d}[0], [x0], \strd
        st1             {\r0\().d}[1], [x8], \strd
.ifnb \r1
        st1             {\r1\().d}[0], [x0], \strd
        st1             {\r1\().d}[1], [x8], \strd
.endif
.endm
1246
.macro shift_store_4 type, strd, r0, r1
1247
.ifc \type, put
1248 1249
        sqrshrun_b      6,     \r0, \r1
        st_s            \strd, \r0, \r1
1250
.else
1251 1252
        srshr_h         2,     \r0, \r1
        st_d            \strd, \r0, \r1
1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284