quant-a.S 16.4 KB
Newer Older
1 2 3
/****************************************************************************
 * quant.S: arm quantization and level-run
 *****************************************************************************
Henrik Gramner's avatar
Henrik Gramner committed
4
 * Copyright (C) 2009-2017 x264 project
5 6
 *
 * Authors: David Conrad <lessen42@gmail.com>
7
 *          Janne Grunau <janne-x264@jannau.net>
8
 *          Martin Storsjo <martin@martin.st>
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
 *
 * This program is also available under a commercial proprietary license.
 * For more information, contact us at licensing@x264.com.
 *****************************************************************************/

#include "asm.S"

.macro QUANT_TWO bias0 bias1 mf0_1 mf2_3 mask
    add         v18.8h, v18.8h, \bias0
    add         v19.8h, v19.8h, \bias1
    umull       v20.4s, v18.4h, \mf0_1\().4h
    umull2      v21.4s, v18.8h, \mf0_1\().8h
    umull       v22.4s, v19.4h, \mf2_3\().4h
    umull2      v23.4s, v19.8h, \mf2_3\().8h
    sshr        v16.8h, v16.8h, #15
    sshr        v17.8h, v17.8h, #15
    shrn        v18.4h, v20.4s, #16
    shrn2       v18.8h, v21.4s, #16
    shrn        v19.4h, v22.4s, #16
    shrn2       v19.8h, v23.4s, #16
    eor         v18.16b, v18.16b, v16.16b
    eor         v19.16b, v19.16b, v17.16b
    sub         v18.8h, v18.8h, v16.8h
    sub         v19.8h, v19.8h, v17.8h
    orr         \mask,  v18.16b, v19.16b
    st1        {v18.8h,v19.8h}, [x0], #32
.endm

.macro QUANT_END d
    fmov        x2,  \d
    mov         w0,  #0
    tst         x2,  x2
    cinc        w0,  w0,  ne
    ret
.endm

// quant_2x2_dc( int16_t dct[4], int mf, int bias )
function x264_quant_2x2_dc_neon, export=1
    ld1        {v0.4h}, [x0]
    dup         v2.4h,  w2
    dup         v1.4h,  w1
    abs         v3.4h,  v0.4h
    add         v3.4h,  v3.4h,  v2.4h
    umull       v3.4s,  v3.4h,  v1.4h
    sshr        v0.4h,  v0.4h,  #15
    shrn        v3.4h,  v3.4s,  #16
    eor         v3.8b,  v3.8b,  v0.8b
    sub         v3.4h,  v3.4h,  v0.4h
    st1        {v3.4h}, [x0]
    QUANT_END   d3
endfunc

// quant_4x4_dc( int16_t dct[16], int mf, int bias )
function x264_quant_4x4_dc_neon, export=1
    ld1        {v16.8h,v17.8h}, [x0]
    abs         v18.8h,  v16.8h
    abs         v19.8h,  v17.8h
    dup         v0.8h,  w2
    dup         v2.8h,  w1
    QUANT_TWO   v0.8h,  v0.8h,  v2,  v2,  v0.16b
    uqxtn       v0.8b,  v0.8h
    QUANT_END   d0
endfunc

// quant_4x4( int16_t dct[16], uint16_t mf[16], uint16_t bias[16] )
function x264_quant_4x4_neon, export=1
    ld1        {v16.8h,v17.8h}, [x0]
    abs         v18.8h,  v16.8h
    abs         v19.8h,  v17.8h
    ld1        {v0.8h,v1.8h}, [x2]
    ld1        {v2.8h,v3.8h}, [x1]
    QUANT_TWO   v0.8h,  v1.8h,  v2,  v3,  v0.16b
    uqxtn       v0.8b,  v0.8h
    QUANT_END   d0
endfunc

// quant_4x4x4( int16_t dct[4][16], uint16_t mf[16], uint16_t bias[16] )
function x264_quant_4x4x4_neon, export=1
    ld1        {v16.8h,v17.8h}, [x0]
    abs         v18.8h, v16.8h
    abs         v19.8h, v17.8h
    ld1        {v0.8h,v1.8h}, [x2]
    ld1        {v2.8h,v3.8h}, [x1]
    QUANT_TWO   v0.8h,  v1.8h,  v2,  v3,  v4.16b
    ld1        {v16.8h,v17.8h}, [x0]
    abs         v18.8h, v16.8h
    abs         v19.8h, v17.8h
    QUANT_TWO   v0.8h,  v1.8h,  v2,  v3,  v5.16b
    ld1        {v16.8h,v17.8h}, [x0]
    abs         v18.8h, v16.8h
    abs         v19.8h, v17.8h
    QUANT_TWO   v0.8h,  v1.8h,  v2,  v3,  v6.16b
    ld1        {v16.8h,v17.8h}, [x0]
    abs         v18.8h, v16.8h
    abs         v19.8h, v17.8h
    QUANT_TWO   v0.8h,  v1.8h,  v2,  v3,  v7.16b
    uqxtn       v4.8b,  v4.8h
    uqxtn       v7.8b,  v7.8h
    uqxtn       v6.8b,  v6.8h
    uqxtn       v5.8b,  v5.8h
    fmov        x7,  d7
    fmov        x6,  d6
    fmov        x5,  d5
    fmov        x4,  d4
    mov         w0,  #0
    tst         x7,  x7
    cinc        w0,  w0,  ne
    lsl         w0,  w0,  #1
    tst         x6,  x6
    cinc        w0,  w0,  ne
    lsl         w0,  w0,  #1
    tst         x5,  x5
    cinc        w0,  w0,  ne
    lsl         w0,  w0,  #1
    tst         x4,  x4
    cinc        w0,  w0,  ne
    ret
endfunc

// quant_8x8( int16_t dct[64], uint16_t mf[64], uint16_t bias[64] )
function x264_quant_8x8_neon, export=1
    ld1        {v16.8h,v17.8h}, [x0]
    abs         v18.8h, v16.8h
    abs         v19.8h, v17.8h
    ld1        {v0.8h,v1.8h}, [x2], #32
    ld1        {v2.8h,v3.8h}, [x1], #32
    QUANT_TWO   v0.8h,  v1.8h,  v2,  v3,  v4.16b
.rept 3
    ld1        {v16.8h,v17.8h}, [x0]
    abs         v18.8h, v16.8h
    abs         v19.8h, v17.8h
    ld1        {v0.8h,v1.8h}, [x2], #32
    ld1        {v2.8h,v3.8h}, [x1], #32
    QUANT_TWO   v0.8h,  v1.8h,  v2,  v3,  v5.16b
    orr         v4.16b, v4.16b, v5.16b
.endr
    uqxtn       v0.8b,  v4.8h
    QUANT_END   d0
endfunc

.macro DEQUANT_START mf_size offset dc=no
    mov         w3,  #0x2b
    mul         w3,  w3,  w2
    lsr         w3,  w3,  #8            // i_qbits = i_qp / 6
    add         w5,  w3,  w3,  lsl #1
    sub         w2,  w2,  w5,  lsl #1   // i_mf = i_qp % 6
    lsl         w2,  w2,  #\mf_size
.ifc \dc,no
    add         x1,  x1,  w2, sxtw      // dequant_mf[i_mf]
.else
    ldr         x1, [x1,  w2, sxtw]     // dequant_mf[i_mf][0][0]
.endif
    subs        w3,  w3,  #\offset      // 6 for 8x8
.endm

// dequant_4x4( int16_t dct[16], int dequant_mf[6][16], int i_qp )
.macro DEQUANT size bits
function x264_dequant_\size\()_neon, export=1
    DEQUANT_START \bits+2, \bits
.ifc \size, 8x8
    mov         w2,  #4
.endif
    b.lt        dequant_\size\()_rshift

    dup         v31.8h, w3
dequant_\size\()_lshift_loop:
.ifc \size, 8x8
    subs        w2,  w2,  #1
.endif
    ld1        {v16.4s}, [x1], #16
    ld1        {v17.4s}, [x1], #16
    sqxtn       v2.4h,  v16.4s
    ld1        {v18.4s}, [x1], #16
    sqxtn2      v2.8h,  v17.4s
    ld1        {v19.4s}, [x1], #16
    sqxtn       v3.4h,  v18.4s
    ld1        {v0.8h,v1.8h}, [x0]
    sqxtn2      v3.8h,  v19.4s
    mul         v0.8h,  v0.8h,  v2.8h
    mul         v1.8h,  v1.8h,  v3.8h
    sshl        v0.8h,  v0.8h,  v31.8h
    sshl        v1.8h,  v1.8h,  v31.8h
    st1        {v0.8h,v1.8h}, [x0], #32
.ifc \size, 8x8
    b.gt        dequant_\size\()_lshift_loop
.endif
    ret

dequant_\size\()_rshift:
    dup         v31.4s, w3
    neg         w3,  w3
    mov         w5,  #1
    sub         w3,  w3,  #1
    lsl         w5,  w5,  w3

.ifc \size, 8x8
dequant_\size\()_rshift_loop:
    subs        w2,  w2,  #1
.endif
    ld1        {v16.4s}, [x1], #16
    ld1        {v17.4s}, [x1], #16
    sqxtn       v2.4h,  v16.4s
    ld1        {v18.4s}, [x1], #16
    dup         v16.4s, w5
    sqxtn2      v2.8h,  v17.4s
    ld1        {v19.4s}, [x1], #16
    dup         v17.4s, w5
    sqxtn       v3.4h,  v18.4s
    ld1        {v0.8h,v1.8h}, [x0]
    dup         v18.4s, w5
    sqxtn2      v3.8h,  v19.4s
    dup         v19.4s, w5

    smlal       v16.4s, v0.4h,  v2.4h
    smlal2      v17.4s, v0.8h,  v2.8h
    smlal       v18.4s, v1.4h,  v3.4h
    smlal2      v19.4s, v1.8h,  v3.8h
    sshl        v16.4s, v16.4s, v31.4s
    sshl        v17.4s, v17.4s, v31.4s
    sshl        v18.4s, v18.4s, v31.4s
    sshl        v19.4s, v19.4s, v31.4s

    sqxtn       v0.4h,  v16.4s
    sqxtn2      v0.8h,  v17.4s
    sqxtn       v1.4h,  v18.4s
    sqxtn2      v1.8h,  v19.4s
    st1        {v0.8h,v1.8h}, [x0], #32
.ifc \size, 8x8
    b.gt        dequant_\size\()_rshift_loop
.endif
    ret
endfunc
.endm

DEQUANT 4x4, 4
DEQUANT 8x8, 6

// dequant_4x4_dc( int16_t dct[16], int dequant_mf[6][16], int i_qp )
function x264_dequant_4x4_dc_neon, export=1
    DEQUANT_START 6, 6, yes
    b.lt        dequant_4x4_dc_rshift

    lsl         w1,  w1,  w3
    dup         v2.8h,  w1
    ld1        {v0.8h,v1.8h},   [x0]

    mul         v0.8h,  v0.8h,  v2.8h
    mul         v1.8h,  v1.8h,  v2.8h
    st1        {v0.8h,v1.8h},   [x0]
    ret

dequant_4x4_dc_rshift:
    dup         v4.8h,  w1
    dup         v3.4s, w3
    neg         w3,  w3
    mov         w5,  #1
    sub         w3,  w3,  #1
    lsl         w5,  w5,  w3

    dup         v16.4s, w5
    dup         v17.4s, w5
    ld1        {v0.8h,v1.8h}, [x0]
    dup         v18.4s, w5
    dup         v19.4s, w5

    smlal       v16.4s, v0.4h,  v4.4h
    smlal2      v17.4s, v0.8h,  v4.8h
    smlal       v18.4s, v1.4h,  v4.4h
    smlal2      v19.4s, v1.8h,  v4.8h
    sshl        v16.4s, v16.4s, v3.4s
    sshl        v17.4s, v17.4s, v3.4s
    sshl        v18.4s, v18.4s, v3.4s
    sshl        v19.4s, v19.4s, v3.4s

    sqxtn       v0.4h,  v16.4s
    sqxtn2      v0.8h,  v17.4s
    sqxtn       v1.4h,  v18.4s
    sqxtn2      v1.8h,  v19.4s
    st1        {v0.8h,v1.8h}, [x0]
    ret
endfunc

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
.macro decimate_score_1x size
function x264_decimate_score\size\()_neon, export=1
    ld1        {v0.8h,v1.8h}, [x0]
    movrel      x5,  X(x264_decimate_table4)
    movi        v3.16b, #0x01
    sqxtn       v0.8b,  v0.8h
    sqxtn2      v0.16b, v1.8h
    abs         v2.16b, v0.16b
    cmeq        v1.16b, v0.16b, #0
    cmhi        v2.16b, v2.16b, v3.16b
    shrn        v1.8b,  v1.8h,  #4
    shrn        v2.8b,  v2.8h,  #4
    fmov        x2,  d2
    fmov        x1,  d1
    cbnz        x2,  9f
    mvn         x1,  x1
    mov         w0,  #0
    cbz         x1,  0f
.ifc \size, 15
    lsr         x1,  x1,  #1
.endif
    rbit        x1,  x1
1:
    clz         x3,  x1
    lsr         x6,  x3,  #2
    lsl         x1,  x1,  x3
    ldrb        w7,  [x5, x6]
    lsl         x1,  x1,  #4
    add         w0,  w0,  w7
    cbnz        x1,  1b
    ret
9:
    mov         w0,  #9
338
0:
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
    ret
endfunc
.endm

decimate_score_1x 15
decimate_score_1x 16

const mask64, align=6
    .byte  0x80, 0x40, 0x20, 0x10, 0x08, 0x04, 0x02, 0x01
    .byte  0x80, 0x40, 0x20, 0x10, 0x08, 0x04, 0x02, 0x01
endconst

function x264_decimate_score64_neon, export=1
    ld1        {v0.8h,v1.8h}, [x0], #32
    ld1        {v2.8h,v3.8h}, [x0], #32
    ld1        {v4.8h,v5.8h}, [x0], #32
    ld1        {v6.8h,v7.8h}, [x0]
    movrel      x6,  mask64
    movi        v31.16b, #0x01
    sqxtn       v16.8b,  v1.8h
    sqxtn2      v16.16b, v0.8h
    sqxtn       v17.8b,  v3.8h
    sqxtn2      v17.16b, v2.8h
    sqxtn       v18.8b,  v5.8h
    sqxtn2      v18.16b, v4.8h
    sqxtn       v19.8b,  v7.8h
    sqxtn2      v19.16b, v6.8h
    abs         v4.16b, v16.16b
    abs         v5.16b, v17.16b
    abs         v6.16b, v18.16b
    abs         v7.16b, v19.16b
    ld1        {v30.16b}, [x6]
    cmeq        v0.16b, v16.16b, #0
    cmeq        v1.16b, v17.16b, #0
    cmeq        v2.16b, v18.16b, #0
    cmeq        v3.16b, v19.16b, #0
    umax        v4.16b, v4.16b, v5.16b
    umax        v6.16b, v6.16b, v7.16b
    and         v0.16b, v0.16b, v30.16b
    and         v1.16b, v1.16b, v30.16b
    and         v2.16b, v2.16b, v30.16b
    and         v3.16b, v3.16b, v30.16b
    umax        v4.16b, v4.16b, v6.16b
    addp        v0.16b, v1.16b, v0.16b
    addp        v2.16b, v3.16b, v2.16b
    cmhi        v4.16b, v4.16b, v31.16b
    addp        v0.16b, v2.16b, v0.16b
    shrn        v4.8b,  v4.8h,  #4
    addp        v0.16b, v0.16b, v0.16b
    fmov        x2,  d4
    fmov        x1,  d0
    cbnz        x2,  9f
    mvn         x1,  x1
    mov         w0,  #0
    cbz         x1,  0f
    movrel      x5,  X(x264_decimate_table8)
1:
    clz         x3,  x1
    lsl         x1,  x1,  x3
    ldrb        w7,  [x5, x3]
    lsl         x1,  x1,  #1
    add         w0,  w0,  w7
    cbnz        x1,  1b
    ret
9:
    mov         w0,  #9
405
0:
406 407 408
    ret
endfunc

409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492
// int coeff_last( int16_t *l )
function x264_coeff_last4_aarch64, export=1
    ldr         x2,  [x0]
    mov         w4,  #3
    clz         x0,  x2
    sub         w0,  w4,  w0, lsr #4
    ret
endfunc

function x264_coeff_last8_aarch64, export=1
    ldr         x3,  [x0, #8]
    mov         w4,  #7
    clz         x2,  x3
    cmp         w2,  #64
    b.ne        1f
    ldr         x3,  [x0]
    sub         w4,  w4,  #4
    clz         x2,  x3
1:
    sub         w0,  w4,  w2, lsr #4
    ret
endfunc

.macro COEFF_LAST_1x size
function x264_coeff_last\size\()_neon, export=1
.if \size == 15
    sub         x0,  x0,  #2
.endif
    ld1        {v0.8h,v1.8h}, [x0]
    uqxtn       v0.8b,  v0.8h
    uqxtn2      v0.16b, v1.8h
    cmtst       v0.16b, v0.16b, v0.16b
    shrn        v0.8b,  v0.8h,  #4
    fmov        x1,  d0
    mov         w3,  #\size - 1
    clz         x2,  x1
    sub         w0,  w3,  w2, lsr #2
    ret
endfunc
.endm

COEFF_LAST_1x 15
COEFF_LAST_1x 16

function x264_coeff_last64_neon, export=1
    ld1        {v0.8h,v1.8h,v2.8h,v3.8h}, [x0], 64
    movi        v31.8h,  #8
    movi        v30.8h,  #1
    uqxtn       v0.8b,  v0.8h
    uqxtn2      v0.16b, v1.8h
    ld1        {v4.8h,v5.8h,v6.8h,v7.8h}, [x0], 64
    uqxtn       v1.8b,  v2.8h
    uqxtn2      v1.16b, v3.8h
    uqxtn       v2.8b,  v4.8h
    uqxtn2      v2.16b, v5.8h
    uqxtn       v3.8b,  v6.8h
    uqxtn2      v3.16b, v7.8h

    cmtst       v0.16b, v0.16b, v0.16b
    cmtst       v1.16b, v1.16b, v1.16b
    cmtst       v2.16b, v2.16b, v2.16b
    cmtst       v3.16b, v3.16b, v3.16b

    shrn        v0.8b,  v0.8h,  #4
    shrn2       v0.16b, v1.8h,  #4
    shrn        v1.8b,  v2.8h,  #4
    shrn2       v1.16b, v3.8h,  #4

    clz         v0.4s,  v0.4s
    clz         v1.4s,  v1.4s

    shrn        v0.4h,  v0.4s,  #2
    shrn2       v0.8h,  v1.4s,  #2

    sub         v0.8h,  v31.8h,  v0.8h
    sshl        v0.8h,  v30.8h,  v0.8h
    shrn        v0.8b,  v0.8h,  #1

    fmov        x2,  d0
    mov         w3,  #63
    clz         x2,  x2
    sub         w0,  w3,  w2
    ret
endfunc
493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558

.macro coeff_level_run_start size
    add         x6,  x1,  #23            // runlevel->mask
    mov         w7,  #0
    mov         w8,  #0
    mov         w9,  #1
    and         x6,  x6,  #~15
    mov         w4,  #\size - 1
.endm

.macro coeff_level_run shift
    clz         x3,  x2
    subs        w4,  w4,  w3, lsr #\shift
    str         w4,  [x1], #4
1:
    ldrh        w5,  [x0, x4, lsl #1]
    strh        w5,  [x6], #2
    add         w7,  w7,  #1
    lsl         w10, w9, w4
    orr         w8,  w8,  w10
    b.le        2f
    add         w3,  w3,  #1 << \shift
    sub         w4,  w4,  #1
    and         x3,  x3,  #~((1 << \shift) - 1)
    lsl         x2,  x2,  x3
    clz         x3,  x2
    subs        w4,  w4,  w3, lsr #\shift
    b.ge        1b
2:
    str         w8,  [x1]
    mov         w0,  w7
.endm

function x264_coeff_level_run4_aarch64, export=1
    ldr         x2,  [x0]

    coeff_level_run_start 4

    coeff_level_run 4

    ret
endfunc

.macro X264_COEFF_LEVEL_RUN size
function x264_coeff_level_run\size\()_neon, export=1
.if \size == 15
    sub         x0,  x0,  #2
.endif
.if         \size < 15
    ld1         {v0.8h}, [x0]
    uqxtn       v0.8b,  v0.8h
    cmtst       v0.8b,  v0.8b,  v0.8b
.else
    ld1         {v0.8h,v1.8h}, [x0]
    uqxtn       v0.8b,  v0.8h
    uqxtn2      v0.16b, v1.8h
    cmtst       v0.16b, v0.16b, v0.16b
    shrn        v0.8b,  v0.8h,  #4
.endif
    fmov        x2,  d0
.if \size == 15
    add         x0,  x0,  #2
.endif

    coeff_level_run_start \size

559
    coeff_level_run (4 - (\size + 1) / 8)
560 561 562 563 564 565 566 567

    ret
endfunc
.endm

X264_COEFF_LEVEL_RUN 8
X264_COEFF_LEVEL_RUN 15
X264_COEFF_LEVEL_RUN 16
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

function x264_denoise_dct_neon, export=1
1:  subs        w3,  w3,  #16
    ld1         {v0.8h,v1.8h}, [x0]
    ld1         {v4.4s,v5.4s,v6.4s,v7.4s}, [x1]
    abs         v16.8h,  v0.8h
    abs         v17.8h,  v1.8h
    ld1         {v2.8h,v3.8h}, [x2], #32
    cmlt        v18.8h,  v0.8h,   #0
    cmlt        v19.8h,  v1.8h,   #0
    uaddw       v4.4s,   v4.4s,   v16.4h
    uaddw2      v5.4s,   v5.4s,   v16.8h
    uqsub       v20.8h,  v16.8h,  v2.8h
    uqsub       v21.8h,  v17.8h,  v3.8h
    uaddw       v6.4s,   v6.4s,   v17.4h
    uaddw2      v7.4s,   v7.4s,   v17.8h
    neg         v22.8h,  v20.8h
    neg         v23.8h,  v21.8h
    bsl         v18.16b, v22.16b, v20.16b
    bsl         v19.16b, v23.16b, v21.16b
    st1         {v4.4s,v5.4s,v6.4s,v7.4s}, [x1], #64
    st1         {v18.8h,v19.8h}, [x0], #32
    b.gt        1b
    ret
endfunc