qpeldsp.asm 13.8 KB
Newer Older
1 2
;******************************************************************************
;* mpeg4 qpel
3
;* Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4
;* Copyright (c) 2008 Loren Merritt
5
;* Copyright (c) 2013 Daniel Kang
6
;*
7
;* This file is part of FFmpeg.
8
;*
9
;* FFmpeg is free software; you can redistribute it and/or
10 11 12 13
;* modify it under the terms of the GNU Lesser General Public
;* License as published by the Free Software Foundation; either
;* version 2.1 of the License, or (at your option) any later version.
;*
14
;* FFmpeg is distributed in the hope that it will be useful,
15 16 17 18 19
;* but WITHOUT ANY WARRANTY; without even the implied warranty of
;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
;* Lesser General Public License for more details.
;*
;* You should have received a copy of the GNU Lesser General Public
20
;* License along with FFmpeg; if not, write to the Free Software
21 22 23 24 25 26 27 28 29 30 31 32 33
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************

%include "libavutil/x86/x86util.asm"

SECTION_RODATA
cextern pb_1
cextern pw_3
cextern pw_15
cextern pw_16
cextern pw_20


34
SECTION .text
35

36
; void ff_put_no_rnd_pixels8_l2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
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
%macro PUT_NO_RND_PIXELS8_L2 0
cglobal put_no_rnd_pixels8_l2, 6,6
    movsxdifnidn r4, r4d
    movsxdifnidn r3, r3d
    pcmpeqb      m6, m6
    test        r5d, 1
    je .loop
    mova         m0, [r1]
    mova         m1, [r2]
    add          r1, r4
    add          r2, 8
    pxor         m0, m6
    pxor         m1, m6
    PAVGB        m0, m1
    pxor         m0, m6
    mova       [r0], m0
    add          r0, r3
    dec r5d
.loop:
    mova         m0, [r1]
    add          r1, r4
    mova         m1, [r1]
    add          r1, r4
    mova         m2, [r2]
    mova         m3, [r2+8]
    pxor         m0, m6
    pxor         m1, m6
    pxor         m2, m6
    pxor         m3, m6
    PAVGB        m0, m2
    PAVGB        m1, m3
    pxor         m0, m6
    pxor         m1, m6
    mova       [r0], m0
    add          r0, r3
    mova       [r0], m1
    add          r0, r3
    mova         m0, [r1]
    add          r1, r4
    mova         m1, [r1]
    add          r1, r4
    mova         m2, [r2+16]
    mova         m3, [r2+24]
    pxor         m0, m6
    pxor         m1, m6
    pxor         m2, m6
    pxor         m3, m6
    PAVGB        m0, m2
    PAVGB        m1, m3
    pxor         m0, m6
    pxor         m1, m6
    mova       [r0], m0
    add          r0, r3
    mova       [r0], m1
    add          r0, r3
    add          r2, 32
    sub         r5d, 4
    jne .loop
    REP_RET
%endmacro

INIT_MMX mmxext
PUT_NO_RND_PIXELS8_L2


102
; void ff_put_no_rnd_pixels16_l2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
103
%macro PUT_NO_RND_PIXELS16_l2 0
104
cglobal put_no_rnd_pixels16_l2, 6,6
105
    movsxdifnidn r3, r3d
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
    movsxdifnidn r4, r4d
    pcmpeqb      m6, m6
    test        r5d, 1
    je .loop
    mova         m0, [r1]
    mova         m1, [r1+8]
    mova         m2, [r2]
    mova         m3, [r2+8]
    pxor         m0, m6
    pxor         m1, m6
    pxor         m2, m6
    pxor         m3, m6
    PAVGB        m0, m2
    PAVGB        m1, m3
    pxor         m0, m6
    pxor         m1, m6
    add          r1, r4
    add          r2, 16
    mova       [r0], m0
    mova     [r0+8], m1
    add          r0, r3
    dec r5d
.loop:
    mova         m0, [r1]
    mova         m1, [r1+8]
    add          r1, r4
    mova         m2, [r2]
    mova         m3, [r2+8]
    pxor         m0, m6
    pxor         m1, m6
    pxor         m2, m6
    pxor         m3, m6
    PAVGB        m0, m2
    PAVGB        m1, m3
    pxor         m0, m6
    pxor         m1, m6
    mova       [r0], m0
    mova     [r0+8], m1
    add          r0, r3
    mova         m0, [r1]
    mova         m1, [r1+8]
    add          r1, r4
    mova         m2, [r2+16]
    mova         m3, [r2+24]
    pxor         m0, m6
    pxor         m1, m6
    pxor         m2, m6
    pxor         m3, m6
    PAVGB        m0, m2
    PAVGB        m1, m3
    pxor         m0, m6
    pxor         m1, m6
    mova       [r0], m0
    mova     [r0+8], m1
    add          r0, r3
    add          r2, 32
    sub         r5d, 2
    jne .loop
    REP_RET
%endmacro

INIT_MMX mmxext
PUT_NO_RND_PIXELS16_l2
INIT_MMX 3dnow
PUT_NO_RND_PIXELS16_l2

%macro MPEG4_QPEL16_H_LOWPASS 1
173
cglobal %1_mpeg4_qpel16_h_lowpass, 5, 5, 0, 16
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
    movsxdifnidn r2, r2d
    movsxdifnidn r3, r3d
    pxor         m7, m7
.loop:
    mova         m0, [r1]
    mova         m1, m0
    mova         m2, m0
    punpcklbw    m0, m7
    punpckhbw    m1, m7
    pshufw       m5, m0, 0x90
    pshufw       m6, m0, 0x41
    mova         m3, m2
    mova         m4, m2
    psllq        m2, 8
    psllq        m3, 16
    psllq        m4, 24
    punpckhbw    m2, m7
    punpckhbw    m3, m7
    punpckhbw    m4, m7
    paddw        m5, m3
    paddw        m6, m2
    paddw        m5, m5
    psubw        m6, m5
    pshufw       m5, m0, 6
    pmullw       m6, [pw_3]
    paddw        m0, m4
    paddw        m5, m1
    pmullw       m0, [pw_20]
    psubw        m0, m5
    paddw        m6, [PW_ROUND]
    paddw        m0, m6
    psraw        m0, 5
206
    mova    [rsp+8], m0
207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229
    mova         m0, [r1+5]
    mova         m5, m0
    mova         m6, m0
    psrlq        m0, 8
    psrlq        m5, 16
    punpcklbw    m0, m7
    punpcklbw    m5, m7
    paddw        m2, m0
    paddw        m3, m5
    paddw        m2, m2
    psubw        m3, m2
    mova         m2, m6
    psrlq        m6, 24
    punpcklbw    m2, m7
    punpcklbw    m6, m7
    pmullw       m3, [pw_3]
    paddw        m1, m2
    paddw        m4, m6
    pmullw       m1, [pw_20]
    psubw        m3, m4
    paddw        m1, [PW_ROUND]
    paddw        m3, m1
    psraw        m3, 5
230
    mova         m1, [rsp+8]
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560
    packuswb     m1, m3
    OP_MOV     [r0], m1, m4
    mova         m1, [r1+9]
    mova         m4, m1
    mova         m3, m1
    psrlq        m1, 8
    psrlq        m4, 16
    punpcklbw    m1, m7
    punpcklbw    m4, m7
    paddw        m5, m1
    paddw        m0, m4
    paddw        m5, m5
    psubw        m0, m5
    mova         m5, m3
    psrlq        m3, 24
    pmullw       m0, [pw_3]
    punpcklbw    m3, m7
    paddw        m2, m3
    psubw        m0, m2
    mova         m2, m5
    punpcklbw    m2, m7
    punpckhbw    m5, m7
    paddw        m6, m2
    pmullw       m6, [pw_20]
    paddw        m0, [PW_ROUND]
    paddw        m0, m6
    psraw        m0, 5
    paddw        m3, m5
    pshufw       m6, m5, 0xf9
    paddw        m6, m4
    pshufw       m4, m5, 0xbe
    pshufw       m5, m5, 0x6f
    paddw        m4, m1
    paddw        m5, m2
    paddw        m6, m6
    psubw        m4, m6
    pmullw       m3, [pw_20]
    pmullw       m4, [pw_3]
    psubw        m3, m5
    paddw        m4, [PW_ROUND]
    paddw        m4, m3
    psraw        m4, 5
    packuswb     m0, m4
    OP_MOV   [r0+8], m0, m4
    add          r1, r3
    add          r0, r2
    dec r4d
    jne .loop
    REP_RET
%endmacro

%macro PUT_OP 2-3
    mova %1, %2
%endmacro

%macro AVG_OP 2-3
    mova  %3, %1
    pavgb %2, %3
    mova  %1, %2
%endmacro

INIT_MMX mmxext
%define PW_ROUND pw_16
%define OP_MOV PUT_OP
MPEG4_QPEL16_H_LOWPASS put
%define PW_ROUND pw_16
%define OP_MOV AVG_OP
MPEG4_QPEL16_H_LOWPASS avg
%define PW_ROUND pw_15
%define OP_MOV PUT_OP
MPEG4_QPEL16_H_LOWPASS put_no_rnd



%macro MPEG4_QPEL8_H_LOWPASS 1
cglobal %1_mpeg4_qpel8_h_lowpass, 5, 5, 0, 8
    movsxdifnidn r2, r2d
    movsxdifnidn r3, r3d
    pxor         m7, m7
.loop:
    mova         m0, [r1]
    mova         m1, m0
    mova         m2, m0
    punpcklbw    m0, m7
    punpckhbw    m1, m7
    pshufw       m5, m0, 0x90
    pshufw       m6, m0, 0x41
    mova         m3, m2
    mova         m4, m2
    psllq        m2, 8
    psllq        m3, 16
    psllq        m4, 24
    punpckhbw    m2, m7
    punpckhbw    m3, m7
    punpckhbw    m4, m7
    paddw        m5, m3
    paddw        m6, m2
    paddw        m5, m5
    psubw        m6, m5
    pshufw       m5, m0, 0x6
    pmullw       m6, [pw_3]
    paddw        m0, m4
    paddw        m5, m1
    pmullw       m0, [pw_20]
    psubw        m0, m5
    paddw        m6, [PW_ROUND]
    paddw        m0, m6
    psraw        m0, 5
    movh         m5, [r1+5]
    punpcklbw    m5, m7
    pshufw       m6, m5, 0xf9
    paddw        m1, m5
    paddw        m2, m6
    pshufw       m6, m5, 0xbe
    pshufw       m5, m5, 0x6f
    paddw        m3, m6
    paddw        m4, m5
    paddw        m2, m2
    psubw        m3, m2
    pmullw       m1, [pw_20]
    pmullw       m3, [pw_3]
    psubw        m3, m4
    paddw        m1, [PW_ROUND]
    paddw        m3, m1
    psraw        m3, 5
    packuswb     m0, m3
    OP_MOV     [r0], m0, m4
    add          r1, r3
    add          r0, r2
    dec r4d
    jne .loop
    REP_RET
%endmacro

INIT_MMX mmxext
%define PW_ROUND pw_16
%define OP_MOV PUT_OP
MPEG4_QPEL8_H_LOWPASS put
%define PW_ROUND pw_16
%define OP_MOV AVG_OP
MPEG4_QPEL8_H_LOWPASS avg
%define PW_ROUND pw_15
%define OP_MOV PUT_OP
MPEG4_QPEL8_H_LOWPASS put_no_rnd



%macro QPEL_V_LOW 5
    paddw      m0, m1
    mova       m4, [pw_20]
    pmullw     m4, m0
    mova       m0, %4
    mova       m5, %1
    paddw      m5, m0
    psubw      m4, m5
    mova       m5, %2
    mova       m6, %3
    paddw      m5, m3
    paddw      m6, m2
    paddw      m6, m6
    psubw      m5, m6
    pmullw     m5, [pw_3]
    paddw      m4, [PW_ROUND]
    paddw      m5, m4
    psraw      m5, 5
    packuswb   m5, m5
    OP_MOV     %5, m5, m7
    SWAP 0,1,2,3
%endmacro

%macro MPEG4_QPEL16_V_LOWPASS 1
cglobal %1_mpeg4_qpel16_v_lowpass, 4, 6, 0, 544
    movsxdifnidn r2, r2d
    movsxdifnidn r3, r3d

    mov         r4d, 17
    mov          r5, rsp
    pxor         m7, m7
.looph:
    mova         m0, [r1]
    mova         m1, [r1]
    mova         m2, [r1+8]
    mova         m3, [r1+8]
    punpcklbw    m0, m7
    punpckhbw    m1, m7
    punpcklbw    m2, m7
    punpckhbw    m3, m7
    mova       [r5], m0
    mova  [r5+0x88], m1
    mova [r5+0x110], m2
    mova [r5+0x198], m3
    add          r5, 8
    add          r1, r3
    dec r4d
    jne .looph


    ; NOTE: r1 CHANGES VALUES: r1 -> 4 - 14*dstStride
    mov         r4d, 4
    mov          r1, 4
    neg          r2
    lea          r1, [r1+r2*8]
    lea          r1, [r1+r2*4]
    lea          r1, [r1+r2*2]
    neg          r2
    mov          r5, rsp
.loopv:
    pxor         m7, m7
    mova         m0, [r5+ 0x0]
    mova         m1, [r5+ 0x8]
    mova         m2, [r5+0x10]
    mova         m3, [r5+0x18]
    QPEL_V_LOW [r5+0x10], [r5+ 0x8], [r5+ 0x0], [r5+0x20], [r0]
    QPEL_V_LOW [r5+ 0x8], [r5+ 0x0], [r5+ 0x0], [r5+0x28], [r0+r2]
    lea    r0, [r0+r2*2]
    QPEL_V_LOW [r5+ 0x0], [r5+ 0x0], [r5+ 0x8], [r5+0x30], [r0]
    QPEL_V_LOW [r5+ 0x0], [r5+ 0x8], [r5+0x10], [r5+0x38], [r0+r2]
    lea    r0, [r0+r2*2]
    QPEL_V_LOW [r5+ 0x8], [r5+0x10], [r5+0x18], [r5+0x40], [r0]
    QPEL_V_LOW [r5+0x10], [r5+0x18], [r5+0x20], [r5+0x48], [r0+r2]
    lea    r0, [r0+r2*2]
    QPEL_V_LOW [r5+0x18], [r5+0x20], [r5+0x28], [r5+0x50], [r0]
    QPEL_V_LOW [r5+0x20], [r5+0x28], [r5+0x30], [r5+0x58], [r0+r2]
    lea    r0, [r0+r2*2]
    QPEL_V_LOW [r5+0x28], [r5+0x30], [r5+0x38], [r5+0x60], [r0]
    QPEL_V_LOW [r5+0x30], [r5+0x38], [r5+0x40], [r5+0x68], [r0+r2]
    lea    r0, [r0+r2*2]
    QPEL_V_LOW [r5+0x38], [r5+0x40], [r5+0x48], [r5+0x70], [r0]
    QPEL_V_LOW [r5+0x40], [r5+0x48], [r5+0x50], [r5+0x78], [r0+r2]
    lea    r0, [r0+r2*2]
    QPEL_V_LOW [r5+0x48], [r5+0x50], [r5+0x58], [r5+0x80], [r0]
    QPEL_V_LOW [r5+0x50], [r5+0x58], [r5+0x60], [r5+0x80], [r0+r2]
    lea    r0, [r0+r2*2]
    QPEL_V_LOW [r5+0x58], [r5+0x60], [r5+0x68], [r5+0x78], [r0]
    QPEL_V_LOW [r5+0x60], [r5+0x68], [r5+0x70], [r5+0x70], [r0+r2]

    add    r5, 0x88
    add    r0, r1
    dec r4d
    jne .loopv
    REP_RET
%endmacro

%macro PUT_OPH 2-3
    movh %1, %2
%endmacro

%macro AVG_OPH 2-3
    movh  %3, %1
    pavgb %2, %3
    movh  %1, %2
%endmacro

INIT_MMX mmxext
%define PW_ROUND pw_16
%define OP_MOV PUT_OPH
MPEG4_QPEL16_V_LOWPASS put
%define PW_ROUND pw_16
%define OP_MOV AVG_OPH
MPEG4_QPEL16_V_LOWPASS avg
%define PW_ROUND pw_15
%define OP_MOV PUT_OPH
MPEG4_QPEL16_V_LOWPASS put_no_rnd



%macro MPEG4_QPEL8_V_LOWPASS 1
cglobal %1_mpeg4_qpel8_v_lowpass, 4, 6, 0, 288
    movsxdifnidn r2, r2d
    movsxdifnidn r3, r3d

    mov         r4d, 9
    mov          r5, rsp
    pxor         m7, m7
.looph:
    mova         m0, [r1]
    mova         m1, [r1]
    punpcklbw    m0, m7
    punpckhbw    m1, m7
    mova       [r5], m0
    mova  [r5+0x48], m1
    add          r5, 8
    add          r1, r3
    dec r4d
    jne .looph


    ; NOTE: r1 CHANGES VALUES: r1 -> 4 - 6*dstStride
    mov         r4d, 2
    mov          r1, 4
    neg          r2
    lea          r1, [r1+r2*4]
    lea          r1, [r1+r2*2]
    neg          r2
    mov          r5, rsp
.loopv:
    pxor         m7, m7
    mova         m0, [r5+ 0x0]
    mova         m1, [r5+ 0x8]
    mova         m2, [r5+0x10]
    mova         m3, [r5+0x18]
    QPEL_V_LOW [r5+0x10], [r5+ 0x8], [r5+ 0x0], [r5+0x20], [r0]
    QPEL_V_LOW [r5+ 0x8], [r5+ 0x0], [r5+ 0x0], [r5+0x28], [r0+r2]
    lea    r0, [r0+r2*2]
    QPEL_V_LOW [r5+ 0x0], [r5+ 0x0], [r5+ 0x8], [r5+0x30], [r0]
    QPEL_V_LOW [r5+ 0x0], [r5+ 0x8], [r5+0x10], [r5+0x38], [r0+r2]
    lea    r0, [r0+r2*2]
    QPEL_V_LOW [r5+ 0x8], [r5+0x10], [r5+0x18], [r5+0x40], [r0]
    QPEL_V_LOW [r5+0x10], [r5+0x18], [r5+0x20], [r5+0x40], [r0+r2]
    lea    r0, [r0+r2*2]
    QPEL_V_LOW [r5+0x18], [r5+0x20], [r5+0x28], [r5+0x38], [r0]
    QPEL_V_LOW [r5+0x20], [r5+0x28], [r5+0x30], [r5+0x30], [r0+r2]

    add    r5, 0x48
    add    r0, r1
    dec r4d
    jne .loopv
    REP_RET
%endmacro

INIT_MMX mmxext
%define PW_ROUND pw_16
%define OP_MOV PUT_OPH
MPEG4_QPEL8_V_LOWPASS put
%define PW_ROUND pw_16
%define OP_MOV AVG_OPH
MPEG4_QPEL8_V_LOWPASS avg
%define PW_ROUND pw_15
%define OP_MOV PUT_OPH
MPEG4_QPEL8_V_LOWPASS put_no_rnd