h264_qpel_10bit.asm 19.8 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
;*****************************************************************************
;* MMX/SSE2/AVX-optimized 10-bit H.264 qpel code
;*****************************************************************************
;* Copyright (C) 2011 x264 project
;*
;* Authors: Daniel Kang <daniel.d.kang@gmail.com>
;*
;* This file is part of Libav.
;*
;* Libav is free software; you can redistribute it and/or
;* 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.
;*
;* Libav 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
;* Lesser General Public License for more details.
;*
;* You should have received a copy of the GNU Lesser General Public
;* License along with Libav; if not, write to the Free Software
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************

25
%include "libavutil/x86/x86util.asm"
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

SECTION_RODATA 32

cextern pw_16
cextern pw_1
cextern pb_0

pw_pixel_max: times 8 dw ((1 << 10)-1)

pad10: times 8 dw 10*1023
pad20: times 8 dw 20*1023
pad30: times 8 dw 30*1023
depad: times 4 dd 32*20*1023 + 512
depad2: times 8 dw 20*1023 + 16*1022 + 16
unpad: times 8 dw 16*1022/32 ; needs to be mod 16

tap1: times 4 dw  1, -5
tap2: times 4 dw 20, 20
tap3: times 4 dw -5,  1
pd_0f: times 4 dd 0xffff

SECTION .text


%macro AVG_MOV 2
    pavgw %2, %1
    mova  %1, %2
%endmacro

%macro ADDW 3
%if mmsize == 8
    paddw %1, %2
%else
    movu  %3, %2
    paddw %1, %3
%endif
%endmacro

%macro FILT_H 4
    paddw  %1, %4
    psubw  %1, %2  ; a-b
    psraw  %1, 2   ; (a-b)/4
    psubw  %1, %2  ; (a-b)/4-b
    paddw  %1, %3  ; (a-b)/4-b+c
    psraw  %1, 2   ; ((a-b)/4-b+c)/4
    paddw  %1, %3  ; ((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16
%endmacro

%macro PRELOAD_V 0
    lea      r3, [r2*3]
    sub      r1, r3
    movu     m0, [r1+r2]
    movu     m1, [r1+r2*2]
    add      r1, r3
    movu     m2, [r1]
    movu     m3, [r1+r2]
    movu     m4, [r1+r2*2]
    add      r1, r3
%endmacro

%macro FILT_V 8
    movu     %6, [r1]
    paddw    %1, %6
    mova     %7, %2
    paddw    %7, %5
    mova     %8, %3
    paddw    %8, %4
    FILT_H   %1, %7, %8, [pw_16]
    psraw    %1, 1
    CLIPW    %1, [pb_0], [pw_pixel_max]
%endmacro

%macro MC 1
%define OP_MOV mova
100 101 102 103
INIT_MMX mmxext
%1 put, 4
INIT_XMM sse2
%1 put, 8
104 105

%define OP_MOV AVG_MOV
106 107 108 109
INIT_MMX mmxext
%1 avg, 4
INIT_XMM sse2
%1 avg, 8
110 111
%endmacro

112
%macro MCAxA_OP 7
113
%if ARCH_X86_32
114 115
cglobal %1_h264_qpel%4_%2_10, %5,%6,%7
    call stub_%1_h264_qpel%3_%2_10 %+ SUFFIX
116 117
    mov  r0, r0m
    mov  r1, r1m
118 119 120
    add  r0, %3*2
    add  r1, %3*2
    call stub_%1_h264_qpel%3_%2_10 %+ SUFFIX
121 122
    mov  r0, r0m
    mov  r1, r1m
123 124 125
    lea  r0, [r0+r2*%3]
    lea  r1, [r1+r2*%3]
    call stub_%1_h264_qpel%3_%2_10 %+ SUFFIX
126 127
    mov  r0, r0m
    mov  r1, r1m
128 129 130
    lea  r0, [r0+r2*%3+%3*2]
    lea  r1, [r1+r2*%3+%3*2]
    call stub_%1_h264_qpel%3_%2_10 %+ SUFFIX
131 132
    RET
%else ; ARCH_X86_64
133 134 135
cglobal %1_h264_qpel%4_%2_10, %5,%6 + 2,%7
    mov r%6, r0
%assign p1 %6+1
136
    mov r %+ p1, r1
137 138 139 140 141 142 143 144 145
    call stub_%1_h264_qpel%3_%2_10 %+ SUFFIX
    lea  r0, [r%6+%3*2]
    lea  r1, [r %+ p1+%3*2]
    call stub_%1_h264_qpel%3_%2_10 %+ SUFFIX
    lea  r0, [r%6+r2*%3]
    lea  r1, [r %+ p1+r2*%3]
    call stub_%1_h264_qpel%3_%2_10 %+ SUFFIX
    lea  r0, [r%6+r2*%3+%3*2]
    lea  r1, [r %+ p1+r2*%3+%3*2]
146
%if UNIX64 == 0 ; fall through to function
147
    call stub_%1_h264_qpel%3_%2_10 %+ SUFFIX
148 149 150 151 152 153
    RET
%endif
%endif
%endmacro

;cpu, put/avg, mc, 4/8, ...
154 155 156 157 158
%macro cglobal_mc 6
%assign i %3*2
%if ARCH_X86_32 || cpuflag(sse2)
MCAxA_OP %1, %2, %3, i, %4,%5,%6
%endif
159

160
cglobal %1_h264_qpel%3_%2_10, %4,%5,%6
161
%if UNIX64 == 0 ; no prologue or epilogue for UNIX64
162
    call stub_%1_h264_qpel%3_%2_10 %+ SUFFIX
163 164 165
    RET
%endif

166
stub_%1_h264_qpel%3_%2_10 %+ SUFFIX:
167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183
%endmacro

;-----------------------------------------------------------------------------
; void h264_qpel_mc00(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
%macro COPY4 0
    movu          m0, [r1     ]
    OP_MOV [r0     ], m0
    movu          m0, [r1+r2  ]
    OP_MOV [r0+r2  ], m0
    movu          m0, [r1+r2*2]
    OP_MOV [r0+r2*2], m0
    movu          m0, [r1+r3  ]
    OP_MOV [r0+r3  ], m0
%endmacro

%macro MC00 1
184 185
INIT_MMX mmxext
cglobal_mc %1, mc00, 4, 3,4,0
186 187 188 189
    lea           r3, [r2*3]
    COPY4
    ret

190 191
INIT_XMM sse2
cglobal %1_h264_qpel8_mc00_10, 3,4
192 193 194 195 196 197 198
    lea  r3, [r2*3]
    COPY4
    lea  r0, [r0+r2*4]
    lea  r1, [r1+r2*4]
    COPY4
    RET

199
cglobal %1_h264_qpel16_mc00_10, 3,4
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
    mov r3d, 8
.loop:
    movu           m0, [r1      ]
    movu           m1, [r1   +16]
    OP_MOV [r0      ], m0
    OP_MOV [r0   +16], m1
    movu           m0, [r1+r2   ]
    movu           m1, [r1+r2+16]
    OP_MOV [r0+r2   ], m0
    OP_MOV [r0+r2+16], m1
    lea            r0, [r0+r2*2]
    lea            r1, [r1+r2*2]
    dec r3d
    jg .loop
    REP_RET
%endmacro

%define OP_MOV mova
MC00 put

%define OP_MOV AVG_MOV
MC00 avg

;-----------------------------------------------------------------------------
; void h264_qpel_mc20(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
%macro MC_CACHE 1
%define OP_MOV mova
228 229 230 231 232 233 234
INIT_MMX mmxext
%1 put, 4
INIT_XMM sse2, cache64
%1 put, 8
INIT_XMM ssse3, cache64
%1 put, 8
INIT_XMM sse2
235
%1 put, 8
236 237

%define OP_MOV AVG_MOV
238 239 240 241 242 243 244
INIT_MMX mmxext
%1 avg, 4
INIT_XMM sse2, cache64
%1 avg, 8
INIT_XMM ssse3, cache64
%1 avg, 8
INIT_XMM sse2
245
%1 avg, 8
246 247
%endmacro

248
%macro MC20 2
249 250
cglobal_mc %1, mc20, %2, 3,4,9
    mov     r3d, %2
251 252 253 254 255 256 257
    mova     m1, [pw_pixel_max]
%if num_mmregs > 8
    mova     m8, [pw_16]
    %define p16 m8
%else
    %define p16 [pw_16]
%endif
258
.nextrow:
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
%if %0 == 4
    movu     m2, [r1-4]
    movu     m3, [r1-2]
    movu     m4, [r1+0]
    ADDW     m2, [r1+6], m5
    ADDW     m3, [r1+4], m5
    ADDW     m4, [r1+2], m5
%else ; movu is slow on these processors
%if mmsize==16
    movu     m2, [r1-4]
    movu     m0, [r1+6]
    mova     m6, m0
    psrldq   m0, 6

    paddw    m6, m2
    PALIGNR  m3, m0, m2, 2, m5
    PALIGNR  m7, m0, m2, 8, m5
    paddw    m3, m7
    PALIGNR  m4, m0, m2, 4, m5
    PALIGNR  m7, m0, m2, 6, m5
    paddw    m4, m7
    SWAP      2, 6
%else
    movu     m2, [r1-4]
    movu     m6, [r1+4]
    PALIGNR  m3, m6, m2, 2, m5
    paddw    m3, m6
    PALIGNR  m4, m6, m2, 4, m5
    PALIGNR  m7, m6, m2, 6, m5
    paddw    m4, m7
    paddw    m2, [r1+6]
%endif
%endif

    FILT_H   m2, m3, m4, p16
    psraw    m2, 1
    pxor     m0, m0
    CLIPW    m2, m0, m1
    OP_MOV [r0], m2
    add      r0, r2
    add      r1, r2
    dec     r3d
    jg .nextrow
    rep ret
%endmacro

MC_CACHE MC20

;-----------------------------------------------------------------------------
; void h264_qpel_mc30(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
310
%macro MC30 2
311
cglobal_mc %1, mc30, %2, 3,5,9
312
    lea r4, [r1+2]
313
    jmp stub_%1_h264_qpel%2_mc10_10 %+ SUFFIX %+ .body
314 315 316 317 318 319 320
%endmacro

MC_CACHE MC30

;-----------------------------------------------------------------------------
; void h264_qpel_mc10(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
321
%macro MC10 2
322
cglobal_mc %1, mc10, %2, 3,5,9
323
    mov      r4, r1
324
.body:
325
    mov     r3d, %2
326 327 328 329 330 331 332
    mova     m1, [pw_pixel_max]
%if num_mmregs > 8
    mova     m8, [pw_16]
    %define p16 m8
%else
    %define p16 [pw_16]
%endif
333
.nextrow:
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
%if %0 == 4
    movu     m2, [r1-4]
    movu     m3, [r1-2]
    movu     m4, [r1+0]
    ADDW     m2, [r1+6], m5
    ADDW     m3, [r1+4], m5
    ADDW     m4, [r1+2], m5
%else ; movu is slow on these processors
%if mmsize==16
    movu     m2, [r1-4]
    movu     m0, [r1+6]
    mova     m6, m0
    psrldq   m0, 6

    paddw    m6, m2
    PALIGNR  m3, m0, m2, 2, m5
    PALIGNR  m7, m0, m2, 8, m5
    paddw    m3, m7
    PALIGNR  m4, m0, m2, 4, m5
    PALIGNR  m7, m0, m2, 6, m5
    paddw    m4, m7
    SWAP      2, 6
%else
    movu     m2, [r1-4]
    movu     m6, [r1+4]
    PALIGNR  m3, m6, m2, 2, m5
    paddw    m3, m6
    PALIGNR  m4, m6, m2, 4, m5
    PALIGNR  m7, m6, m2, 6, m5
    paddw    m4, m7
    paddw    m2, [r1+6]
%endif
%endif

    FILT_H   m2, m3, m4, p16
    psraw    m2, 1
    pxor     m0, m0
    CLIPW    m2, m0, m1
    movu     m3, [r4]
    pavgw    m2, m3
    OP_MOV [r0], m2
    add      r0, r2
    add      r1, r2
    add      r4, r2
    dec     r3d
    jg .nextrow
    rep ret
%endmacro

MC_CACHE MC10

;-----------------------------------------------------------------------------
; void h264_qpel_mc02(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
388 389
%macro V_FILT 10
v_filt%9_%10_10
390 391 392 393 394 395 396 397
    add    r4, r2
.no_addr4:
    FILT_V m0, m1, m2, m3, m4, m5, m6, m7
    add    r1, r2
    add    r0, r2
    ret
%endmacro

398
INIT_MMX mmxext
399 400 401
RESET_MM_PERMUTATION
%assign i 0
%rep 4
402
V_FILT m0, m1, m2, m3, m4, m5, m6, m7, 4, i
403 404 405 406
SWAP 0,1,2,3,4,5
%assign i i+1
%endrep

407
INIT_XMM sse2
408 409 410
RESET_MM_PERMUTATION
%assign i 0
%rep 6
411
V_FILT m0, m1, m2, m3, m4, m5, m6, m7, 8, i
412 413 414 415
SWAP 0,1,2,3,4,5
%assign i i+1
%endrep

416 417
%macro MC02 2
cglobal_mc %1, mc02, %2, 3,4,8
418 419 420 421
    PRELOAD_V

    sub      r0, r2
%assign j 0
422
%rep %2
423
    %assign i (j % 6)
424
    call v_filt%2_ %+ i %+ _10.no_addr4
425 426 427 428 429 430 431 432 433 434 435 436
    OP_MOV [r0], m0
    SWAP 0,1,2,3,4,5
    %assign j j+1
%endrep
    ret
%endmacro

MC MC02

;-----------------------------------------------------------------------------
; void h264_qpel_mc01(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
437 438
%macro MC01 2
cglobal_mc %1, mc01, %2, 3,5,8
439
    mov      r4, r1
440
.body:
441 442 443 444 445
    PRELOAD_V

    sub      r4, r2
    sub      r0, r2
%assign j 0
446
%rep %2
447
    %assign i (j % 6)
448
    call v_filt%2_ %+ i %+ _10
449 450 451 452 453 454 455 456 457 458 459 460 461 462
    movu     m7, [r4]
    pavgw    m0, m7
    OP_MOV [r0], m0
    SWAP 0,1,2,3,4,5
    %assign j j+1
%endrep
    ret
%endmacro

MC MC01

;-----------------------------------------------------------------------------
; void h264_qpel_mc03(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
463 464
%macro MC03 2
cglobal_mc %1, mc03, %2, 3,5,8
465
    lea r4, [r1+r2]
466
    jmp stub_%1_h264_qpel%2_mc01_10 %+ SUFFIX %+ .body
467 468 469 470 471 472 473
%endmacro

MC MC03

;-----------------------------------------------------------------------------
; void h264_qpel_mc11(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
474 475
%macro H_FILT_AVG 2-3
h_filt%1_%2_10:
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
;FILT_H with fewer registers and averaged with the FILT_V result
;m6,m7 are tmp registers, m0 is the FILT_V result, the rest are to be used next in the next iteration
;unfortunately I need three registers, so m5 will have to be re-read from memory
    movu     m5, [r4-4]
    ADDW     m5, [r4+6], m7
    movu     m6, [r4-2]
    ADDW     m6, [r4+4], m7
    paddw    m5, [pw_16]
    psubw    m5, m6  ; a-b
    psraw    m5, 2   ; (a-b)/4
    psubw    m5, m6  ; (a-b)/4-b
    movu     m6, [r4+0]
    ADDW     m6, [r4+2], m7
    paddw    m5, m6  ; (a-b)/4-b+c
    psraw    m5, 2   ; ((a-b)/4-b+c)/4
    paddw    m5, m6  ; ((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16
    psraw    m5, 1
    CLIPW    m5, [pb_0], [pw_pixel_max]
;avg FILT_V, FILT_H
    pavgw    m0, m5
%if %0!=4
    movu     m5, [r1+r5]
%endif
    ret
%endmacro

502
INIT_MMX mmxext
503 504 505
RESET_MM_PERMUTATION
%assign i 0
%rep 3
506
H_FILT_AVG 4, i
507 508 509
SWAP 0,1,2,3,4,5
%assign i i+1
%endrep
510
H_FILT_AVG 4, i, 0
511

512
INIT_XMM sse2
513 514 515 516
RESET_MM_PERMUTATION
%assign i 0
%rep 6
%if i==1
517
H_FILT_AVG 8, i, 0
518
%else
519
H_FILT_AVG 8, i
520 521 522 523 524
%endif
SWAP 0,1,2,3,4,5
%assign i i+1
%endrep

525
%macro MC11 2
526
; this REALLY needs x86_64
527
cglobal_mc %1, mc11, %2, 3,6,8
528
    mov      r4, r1
529
.body:
530 531 532 533 534 535 536
    PRELOAD_V

    sub      r0, r2
    sub      r4, r2
    mov      r5, r2
    neg      r5
%assign j 0
537
%rep %2
538
    %assign i (j % 6)
539 540 541
    call v_filt%2_ %+ i %+ _10
    call h_filt%2_ %+ i %+ _10
%if %2==8 && i==1
542 543 544 545 546 547 548 549 550 551 552 553 554 555
    movu     m5, [r1+r5]
%endif
    OP_MOV [r0], m0
    SWAP 0,1,2,3,4,5
    %assign j j+1
%endrep
    ret
%endmacro

MC MC11

;-----------------------------------------------------------------------------
; void h264_qpel_mc31(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
556 557
%macro MC31 2
cglobal_mc %1, mc31, %2, 3,6,8
558 559
    mov r4, r1
    add r1, 2
560
    jmp stub_%1_h264_qpel%2_mc11_10 %+ SUFFIX %+ .body
561 562 563 564 565 566 567
%endmacro

MC MC31

;-----------------------------------------------------------------------------
; void h264_qpel_mc13(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
568 569
%macro MC13 2
cglobal_mc %1, mc13, %2, 3,7,12
570
    lea r4, [r1+r2]
571
    jmp stub_%1_h264_qpel%2_mc11_10 %+ SUFFIX %+ .body
572 573 574 575 576 577 578
%endmacro

MC MC13

;-----------------------------------------------------------------------------
; void h264_qpel_mc33(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
579 580
%macro MC33 2
cglobal_mc %1, mc33, %2, 3,6,8
581 582
    lea r4, [r1+r2]
    add r1, 2
583
    jmp stub_%1_h264_qpel%2_mc11_10 %+ SUFFIX %+ .body
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
%endmacro

MC MC33

;-----------------------------------------------------------------------------
; void h264_qpel_mc22(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
%macro FILT_H2 3
    psubw  %1, %2  ; a-b
    psubw  %2, %3  ; b-c
    psllw  %2, 2
    psubw  %1, %2  ; a-5*b+4*c
    psllw  %3, 4
    paddw  %1, %3  ; a-5*b+20*c
%endmacro

%macro FILT_VNRD 8
    movu     %6, [r1]
    paddw    %1, %6
    mova     %7, %2
    paddw    %7, %5
    mova     %8, %3
    paddw    %8, %4
    FILT_H2  %1, %7, %8
%endmacro

610 611
%macro HV 1
%if mmsize==16
612 613 614
%define PAD 12
%define COUNT 2
%else
615
%define PAD 4
616 617
%define COUNT 3
%endif
618
put_hv%1_10:
619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634
    neg      r2           ; This actually saves instructions
    lea      r1, [r1+r2*2-mmsize+PAD]
    lea      r4, [rsp+PAD+gprsize]
    mov     r3d, COUNT
.v_loop:
    movu     m0, [r1]
    sub      r1, r2
    movu     m1, [r1]
    sub      r1, r2
    movu     m2, [r1]
    sub      r1, r2
    movu     m3, [r1]
    sub      r1, r2
    movu     m4, [r1]
    sub      r1, r2
%assign i 0
635
%rep %1-1
636 637 638 639 640 641 642 643 644 645 646 647
    FILT_VNRD m0, m1, m2, m3, m4, m5, m6, m7
    psubw    m0, [pad20]
    movu     [r4+i*mmsize*3], m0
    sub      r1, r2
    SWAP 0,1,2,3,4,5
%assign i i+1
%endrep
    FILT_VNRD m0, m1, m2, m3, m4, m5, m6, m7
    psubw    m0, [pad20]
    movu     [r4+i*mmsize*3], m0
    add      r4, mmsize
    lea      r1, [r1+r2*8+mmsize]
648
%if %1==8
649 650 651 652 653 654 655 656
    lea      r1, [r1+r2*4]
%endif
    dec      r3d
    jg .v_loop
    neg      r2
    ret
%endmacro

657 658 659 660
INIT_MMX mmxext
HV 4
INIT_XMM sse2
HV 8
661

662
%macro H_LOOP 1
663 664 665 666 667 668 669 670 671 672 673
%if num_mmregs > 8
    %define s1 m8
    %define s2 m9
    %define s3 m10
    %define d1 m11
%else
    %define s1 [tap1]
    %define s2 [tap2]
    %define s3 [tap3]
    %define d1 [depad]
%endif
674
h%1_loop_op:
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
    movu       m1, [r1+mmsize-4]
    movu       m2, [r1+mmsize-2]
    mova       m3, [r1+mmsize+0]
    movu       m4, [r1+mmsize+2]
    movu       m5, [r1+mmsize+4]
    movu       m6, [r1+mmsize+6]
%if num_mmregs > 8
    pmaddwd    m1, s1
    pmaddwd    m2, s1
    pmaddwd    m3, s2
    pmaddwd    m4, s2
    pmaddwd    m5, s3
    pmaddwd    m6, s3
    paddd      m1, d1
    paddd      m2, d1
%else
    mova       m0, s1
    pmaddwd    m1, m0
    pmaddwd    m2, m0
    mova       m0, s2
    pmaddwd    m3, m0
    pmaddwd    m4, m0
    mova       m0, s3
    pmaddwd    m5, m0
    pmaddwd    m6, m0
    mova       m0, d1
    paddd      m1, m0
    paddd      m2, m0
%endif
    paddd      m3, m5
    paddd      m4, m6
    paddd      m1, m3
    paddd      m2, m4
    psrad      m1, 10
    psrad      m2, 10
    pslld      m2, 16
    pand       m1, [pd_0f]
    por        m1, m2
%if num_mmregs <= 8
    pxor       m0, m0
%endif
    CLIPW      m1, m0, m7
    add        r1, mmsize*3
    ret
%endmacro

721 722 723 724
INIT_MMX mmxext
H_LOOP 4
INIT_XMM sse2
H_LOOP 8
725

726 727
%macro MC22 2
cglobal_mc %1, mc22, %2, 3,7,12
728 729 730 731 732
%define PAD mmsize*8*4*2      ; SIZE*16*4*sizeof(pixel)
    mov      r6, rsp          ; backup stack pointer
    and     rsp, ~(mmsize-1)  ; align stack
    sub     rsp, PAD

733
    call put_hv%2_10
734

735
    mov       r3d, %2
736 737 738 739 740 741 742 743 744 745
    mova       m7, [pw_pixel_max]
%if num_mmregs > 8
    pxor       m0, m0
    mova       m8, [tap1]
    mova       m9, [tap2]
    mova      m10, [tap3]
    mova      m11, [depad]
%endif
    mov        r1, rsp
.h_loop:
746
    call h%2_loop_op
747 748 749 750 751 752 753 754 755 756 757 758 759 760 761

    OP_MOV   [r0], m1
    add        r0, r2
    dec       r3d
    jg .h_loop

    mov     rsp, r6          ; restore stack pointer
    ret
%endmacro

MC MC22

;-----------------------------------------------------------------------------
; void h264_qpel_mc12(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
762 763
%macro MC12 2
cglobal_mc %1, mc12, %2, 3,7,12
764 765 766 767 768
%define PAD mmsize*8*4*2        ; SIZE*16*4*sizeof(pixel)
    mov        r6, rsp          ; backup stack pointer
    and       rsp, ~(mmsize-1)  ; align stack
    sub       rsp, PAD

769
    call put_hv%2_10
770 771

    xor       r4d, r4d
772
.body:
773
    mov       r3d, %2
774 775 776 777 778 779 780 781 782 783
    pxor       m0, m0
    mova       m7, [pw_pixel_max]
%if num_mmregs > 8
    mova       m8, [tap1]
    mova       m9, [tap2]
    mova      m10, [tap3]
    mova      m11, [depad]
%endif
    mov        r1, rsp
.h_loop:
784
    call h%2_loop_op
785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806

    movu       m3, [r1+r4-2*mmsize] ; movu needed for mc32, etc
    paddw      m3, [depad2]
    psrlw      m3, 5
    psubw      m3, [unpad]
    CLIPW      m3, m0, m7
    pavgw      m1, m3

    OP_MOV   [r0], m1
    add        r0, r2
    dec       r3d
    jg .h_loop

    mov     rsp, r6          ; restore stack pointer
    ret
%endmacro

MC MC12

;-----------------------------------------------------------------------------
; void h264_qpel_mc32(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
807 808
%macro MC32 2
cglobal_mc %1, mc32, %2, 3,7,12
809 810 811 812 813
%define PAD mmsize*8*3*2  ; SIZE*16*4*sizeof(pixel)
    mov  r6, rsp          ; backup stack pointer
    and rsp, ~(mmsize-1)  ; align stack
    sub rsp, PAD

814
    call put_hv%2_10
815 816

    mov r4d, 2            ; sizeof(pixel)
817
    jmp stub_%1_h264_qpel%2_mc12_10 %+ SUFFIX %+ .body
818 819 820 821 822 823 824
%endmacro

MC MC32

;-----------------------------------------------------------------------------
; void h264_qpel_mc21(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
825 826
%macro H_NRD 1
put_h%1_10:
827
    add       rsp, gprsize
828
    mov       r3d, %1
829 830
    xor       r4d, r4d
    mova       m6, [pad20]
831
.nextrow:
832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849
    movu       m2, [r5-4]
    movu       m3, [r5-2]
    movu       m4, [r5+0]
    ADDW       m2, [r5+6], m5
    ADDW       m3, [r5+4], m5
    ADDW       m4, [r5+2], m5

    FILT_H2    m2, m3, m4
    psubw      m2, m6
    mova [rsp+r4], m2
    add       r4d, mmsize*3
    add        r5, r2
    dec       r3d
    jg .nextrow
    sub       rsp, gprsize
    ret
%endmacro

850 851 852 853
INIT_MMX mmxext
H_NRD 4
INIT_XMM sse2
H_NRD 8
854

855 856
%macro MC21 2
cglobal_mc %1, mc21, %2, 3,7,12
857
    mov   r5, r1
858
.body:
859 860 861 862 863
%define PAD mmsize*8*3*2   ; SIZE*16*4*sizeof(pixel)
    mov   r6, rsp          ; backup stack pointer
    and  rsp, ~(mmsize-1)  ; align stack

    sub  rsp, PAD
864
    call put_h%2_10
865 866

    sub  rsp, PAD
867
    call put_hv%2_10
868 869

    mov r4d, PAD-mmsize    ; H buffer
870
    jmp stub_%1_h264_qpel%2_mc12_10 %+ SUFFIX %+ .body
871 872 873 874 875 876 877
%endmacro

MC MC21

;-----------------------------------------------------------------------------
; void h264_qpel_mc23(uint8_t *dst, uint8_t *src, int stride)
;-----------------------------------------------------------------------------
878 879
%macro MC23 2
cglobal_mc %1, mc23, %2, 3,7,12
880
    lea   r5, [r1+r2]
881
    jmp stub_%1_h264_qpel%2_mc21_10 %+ SUFFIX %+ .body
882 883 884
%endmacro

MC MC23