mpeg2enc/dist2_mmx.s

   1 ;
   2 ;  dist2_mmx.s:  mmX optimized squared distance sum
   3 ;
   4 ;  Original believed to be Copyright (C) 2000 Brent Byeler
   5 ;
   6 ;  This program is free software; you can reaxstribute it and/or
   7 ;  modify it under the terms of the GNU General Public License
   8 ;  as published by the Free Software Foundation; either version 2
   9 ;  of the License, or (at your option) any later version.
  10 ;
  11 ;  This program is distributed in the hope that it will be useful,
  12 ;  but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 ;  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14 ;  GNU General Public License for more details.
  15 ;
  16 ;  You should have received a copy of the GNU General Public License
  17 ;  along with this program; if not, write to the Free Software
  18 ;  Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
  19 ;
  20
  21 ; total squared difference between two (16*h) blocks
  22 ; including optional half pel interpolation of [ebp+8] ; blk1 (hx,hy)
  23 ; blk1,blk2: addresses of top left pels of both blocks
  24 ; lx:        distance (in bytes) of vertically adjacent pels
  25 ; hx,hy:     flags for horizontal and/or vertical interpolation
  26 ; h:         height of block (usually 8 or 16)
  27 ; mmX version
  28
  29 global dist2_mmx
  30 ; int dist2_mmx(unsigned char *blk1, unsigned char *blk2,
  31 ;                 int lx, int hx, int hy, int h)
  32
  33 ; mm7 = 0
  34
  35 ; eax = pblk1
  36 ; ebx = pblk2
  37 ; ecx = temp
  38 ; edx = distance_sum
  39 ; edi = h
  40 ; esi = lx
  41
  42                 ;;
  43                 ;;  private constants needed
  44                 ;;
  45
  46 SECTION .data
  47 align 16
  48 twos:
  49                         dw      2
  50                         dw      2
  51                         dw      2
  52                         dw      2
  53
  54 align 32
  55 dist2_mmx:
  56         push ebp                        ; save frame pointer
  57         mov ebp, esp            ; link
  58         push ebx
  59         push ecx
  60         push edx
  61         push esi
  62         push edi
  63
  64         mov             esi, [ebp+16] ; lx
  65         mov     eax, [ebp+20] ; hx
  66         mov     edx, [ebp+24] ; hy
  67         mov     edi, [ebp+28] ; h
  68
  69     pxor      mm5, mm5      ; sum
  70         test      edi, edi     ; h = 0?
  71         jle       near d2exit
  72
  73         pxor      mm7, mm7     ; get zeros i mm7
  74
  75         test      eax, eax     ; hx != 0?
  76         jne       near d2is10
  77         test      edx, edx     ; hy != 0?
  78         jne       near d2is10
  79
  80         mov       eax, [ebp+8]
  81     mov       ebx, [ebp+12]
  82         jmp      d2top00
  83
  84 align 32
  85 d2top00:
  86         movq      mm0, [eax]
  87         movq      mm1, mm0
  88         punpcklbw mm0, mm7
  89         punpckhbw mm1, mm7
  90
  91         movq      mm2, [ebx]
  92         movq      mm3, mm2
  93         punpcklbw mm2, mm7
  94         punpckhbw mm3, mm7
  95
  96         psubw     mm0, mm2
  97         psubw     mm1, mm3
  98         pmaddwd   mm0, mm0
  99         pmaddwd   mm1, mm1
 100         paddd     mm0, mm1
 101
 102         movq      mm1, [eax+8]
 103         movq      mm2, mm1
 104         punpcklbw mm1, mm7
 105         punpckhbw mm2, mm7
 106
 107         movq      mm3, [ebx+8]
 108         movq      mm4, mm3
 109         punpcklbw mm3, mm7
 110         punpckhbw mm4, mm7
 111
 112         psubw     mm1, mm3
 113         psubw     mm2, mm4
 114         pmaddwd   mm1, mm1
 115         pmaddwd   mm2, mm2
 116         paddd     mm1, mm2
 117
 118         paddd     mm0, mm1
 119
 120                 ;; Accumulate sum in edx... we use mm5
 121                 ;movd     ecx, mm0
 122         ;add       edx, ecx
 123             ;psrlq        mm0, 32
 124             ;movd         ecx, mm0
 125         ;add       edx, ecx
 126                 paddd    mm5, mm0
 127
 128         add       eax, esi
 129         add       ebx, esi
 130         dec       edi
 131         jg        d2top00
 132         jmp       d2exit
 133
 134
 135 d2is10:
 136         test      eax, eax
 137         je        near d2is01
 138         test      edx, edx
 139         jne       near d2is01
 140
 141
 142         mov       eax, [ebp+8] ; blk1
 143         mov       ebx, [ebp+12] ; blk1
 144
 145         pxor      mm6, mm6    ; mm6 = 0 and isn't changed anyplace in the loop..
 146         pcmpeqw   mm1, mm1
 147         psubw     mm6, mm1
 148         jmp               d2top10
 149
 150 align 32
 151 d2top10:
 152         movq      mm0, [eax]
 153         movq      mm1, mm0
 154         punpcklbw mm0, mm7
 155         punpckhbw mm1, mm7
 156         movq      mm2, [eax+1]
 157         movq      mm3, mm2
 158         punpcklbw mm2, mm7
 159         punpckhbw mm3, mm7
 160         paddw     mm0, mm2
 161         paddw     mm1, mm3
 162         paddw     mm0, mm6   ; here we add mm6 = 0.... weird...
 163         paddw     mm1, mm6
 164         psrlw     mm0, 1
 165         psrlw     mm1, 1
 166
 167         movq      mm2, [ebx]
 168         movq      mm3, mm2
 169         punpcklbw mm2, mm7
 170         punpckhbw mm3, mm7
 171
 172         psubw     mm0, mm2
 173         psubw     mm1, mm3
 174         pmaddwd   mm0, mm0
 175         pmaddwd   mm1, mm1
 176         paddd     mm0, mm1
 177
 178         movq      mm1, [eax+8]
 179         movq      mm2, mm1
 180         punpcklbw mm1, mm7
 181         punpckhbw mm2, mm7
 182         movq      mm3, [eax+9]
 183         movq      mm4, mm3
 184         punpcklbw mm3, mm7
 185         punpckhbw mm4, mm7
 186         paddw     mm1, mm3
 187         paddw     mm2, mm4
 188         paddw     mm1, mm6
 189         paddw     mm2, mm6
 190         psrlw     mm1, 1
 191         psrlw     mm2, 1
 192
 193         movq      mm3, [ebx+8]
 194         movq      mm4, mm3
 195         punpcklbw mm3, mm7
 196         punpckhbw mm4, mm7
 197
 198         psubw     mm1, mm3
 199         psubw     mm2, mm4
 200         pmaddwd   mm1, mm1
 201         pmaddwd   mm2, mm2
 202         paddd     mm1, mm2
 203
 204
 205         paddd     mm0, mm1
 206                 ; Accumulate mm0 sum on edx... we'll use mm5 for this and add up at the end
 207                 ; movd    ecx, mm0
 208         ; add       edx, ecx
 209                 ; psrlq   mm0, 32
 210                 ; movd    ecx, mm0
 211         ; add       edx, ecx
 212         paddd     mm5, mm0
 213         add       eax, esi
 214         add       ebx, esi
 215         dec       edi
 216         jg        near d2top10
 217
 218
 219         jmp       d2exit
 220
 221 d2is01:
 222         test      eax, eax
 223         jne       near d2is11
 224         test      edx, edx
 225         je        near d2is11
 226
 227         mov       eax, [ebp+8] ; blk1
 228         mov       edx, [ebp+12] ; blk2
 229         mov       ebx, eax
 230         add       ebx, esi ;  blk1 + lx
 231
 232         pxor      mm6, mm6
 233         pcmpeqw   mm1, mm1
 234         psubw     mm6, mm1  ; mm6 = 1
 235         jmp               d2top01
 236
 237 align 32
 238 d2top01:
 239         movq      mm0, [eax]
 240         movq      mm1, mm0
 241         punpcklbw mm0, mm7
 242         punpckhbw mm1, mm7
 243         movq      mm2, [ebx]
 244         movq      mm3, mm2
 245         punpcklbw mm2, mm7
 246         punpckhbw mm3, mm7
 247         paddw     mm0, mm2
 248         paddw     mm1, mm3
 249         paddw     mm0, mm6
 250         paddw     mm1, mm6
 251         psrlw     mm0, 1
 252         psrlw     mm1, 1
 253
 254         movq      mm2, [edx]
 255         movq      mm3, mm2
 256     punpcklbw mm2, mm7
 257     punpckhbw mm3, mm7
 258
 259     psubw     mm0, mm2
 260     psubw     mm1, mm3
 261
 262         pmaddwd   mm0, mm0
 263     pmaddwd   mm1, mm1
 264     paddd     mm0, mm1
 265
 266         movq      mm1, [eax+8]
 267         movq      mm2, mm1
 268         punpcklbw mm1, mm7
 269         punpckhbw mm2, mm7
 270
 271         movq      mm3, [ebx+8]
 272         movq      mm4, mm3
 273         punpcklbw mm3, mm7
 274         punpckhbw mm4, mm7
 275
 276         paddw     mm1, mm3
 277         paddw     mm2, mm4
 278         paddw     mm1, mm6
 279         paddw     mm2, mm6
 280         psrlw     mm1, 1
 281         psrlw     mm2, 1
 282
 283         movq      mm3, [edx+8]
 284         movq      mm4, mm3
 285     punpcklbw mm3, mm7
 286     punpckhbw mm4, mm7
 287
 288     psubw     mm1, mm3
 289     psubw     mm2, mm4
 290
 291         pmaddwd   mm1, mm1
 292     pmaddwd   mm2, mm2
 293     paddd     mm0, mm1
 294         paddd     mm0, mm2
 295
 296         ;; Accumulate in "s" - we use mm5 for the purpose
 297         ;;
 298         ;movd     ecx, mm0
 299     ;add       s, ecx
 300         ;psrlq    mm0, 32
 301         ;movd     ecx, mm0
 302         ;add              s, ecx
 303         paddd     mm5, mm0
 304
 305         ;; Originally this moved
 306         mov       eax, ebx    ; eax = eax + lx
 307         add       edx, esi    ; edx = edx + lx
 308         add       ebx, esi    ; ebx = ebx + lx
 309         dec       edi
 310         jg        near d2top01
 311         jmp       d2exit
 312
 313 d2is11:
 314         mov       eax, [ebp+8] ; blk1
 315         mov       edx, [ebp+12] ; blk2
 316         mov       ebx, eax  ;  blk1
 317         add       ebx, esi  ; ebx = blk1 + lx
 318         jmp               d2top11
 319
 320 align 32
 321 d2top11:
 322         movq      mm0, [eax]
 323         movq      mm1, mm0
 324         punpcklbw mm0, mm7
 325         punpckhbw mm1, mm7
 326         movq      mm2, [eax+1]
 327         movq      mm3, mm2
 328         punpcklbw mm2, mm7
 329         punpckhbw mm3, mm7
 330         paddw     mm0, mm2
 331         paddw     mm1, mm3
 332         movq      mm2, [ebx]
 333         movq      mm3, mm2
 334         punpcklbw mm2, mm7
 335         punpckhbw mm3, mm7
 336         movq      mm4, [ebx+1]
 337         movq      mm6, mm4
 338         punpcklbw mm4, mm7
 339         punpckhbw mm6, mm7
 340         paddw     mm2, mm4
 341         paddw     mm3, mm6
 342         paddw     mm0, mm2
 343         paddw     mm1, mm3
 344         ;pxor         mm6, mm6    ; mm6 = 0
 345         ;pcmpeqw          mm5, mm5    ; mm5 = -1
 346         ;psubw        mm6, mm5    ; mm6 = 1
 347         ;paddw        mm6, mm6    ; mm6 = 2
 348         movq      mm6, [twos]
 349         paddw     mm0, mm6    ; round mm0
 350         paddw     mm1, mm6    ; round mm1
 351         psrlw     mm0, 2
 352         psrlw     mm1, 2
 353
 354         movq      mm2, [edx]
 355         movq      mm3, mm2
 356         punpcklbw mm2, mm7
 357         punpckhbw mm3, mm7
 358
 359         psubw     mm0, mm2
 360         psubw     mm1, mm3
 361         pmaddwd   mm0, mm0
 362         pmaddwd   mm1, mm1
 363         paddd     mm0, mm1
 364
 365         movq      mm1, [eax+8]
 366         movq      mm2, mm1
 367         punpcklbw mm1, mm7
 368         punpckhbw mm2, mm7
 369
 370         movq      mm3, [eax+9]
 371         movq      mm4, mm3
 372         punpcklbw mm3, mm7
 373         punpckhbw mm4, mm7
 374
 375         paddw     mm1, mm3
 376         paddw     mm2, mm4
 377
 378         movq      mm3, [ebx+8]
 379         movq      mm4, mm3
 380         punpcklbw mm3, mm7
 381         punpckhbw mm4, mm7
 382         paddw     mm1, mm3
 383         paddw     mm2, mm4
 384
 385         movq      mm3, [ebx+9]
 386         movq      mm4, mm3
 387         punpcklbw mm3, mm7
 388         punpckhbw mm4, mm7
 389
 390         paddw     mm1, mm3
 391         paddw     mm2, mm4
 392
 393         ;pxor     mm6, mm6    ; Zero mm6
 394         ;pcmpeqw          mm5, mm5    ; mm5 = -1
 395         ;psubw    mm6, mm5    ; mm6 = 1
 396         ;paddw    mm6, mm6    ; mm6 = 2
 397         ;paddw    mm1, mm6    ; round mm1 and mm2
 398         ;paddw    mm2, mm6
 399         movq      mm6, [twos]
 400         paddw     mm1, mm6
 401         paddw     mm2, mm6
 402
 403         psrlw     mm1, 2
 404         psrlw     mm2, 2
 405
 406         movq      mm3, [edx+8]
 407         movq      mm4, mm3
 408         punpcklbw mm3, mm7
 409         punpckhbw mm4, mm7
 410
 411         psubw     mm1, mm3
 412         psubw     mm2, mm4
 413         pmaddwd   mm1, mm1
 414         pmaddwd   mm2, mm2
 415         paddd     mm1, mm2
 416
 417         paddd     mm0, mm1
 418
 419         ;;
 420         ;; Accumulate the result in "s" we use mm6 for the purpose...
 421         ;movd     ecx, mm0
 422     ;    add       s, ecx
 423         ;psrlq    mm0, 32
 424         ;movd     ecx, mm0
 425         ;add      s, ecx
 426         paddd     mm5, mm0
 427
 428         mov       eax, ebx    ; ahem ebx = eax at start of loop and wasn't changed...
 429         add       ebx, esi
 430         add       edx, esi
 431         dec       edi
 432         jg        near d2top11
 433
 434
 435 d2exit:
 436         ;; Put the final sum in eax for return...
 437         movd      eax, mm5
 438         psrlq     mm5, 32
 439         movd      ecx, mm5
 440     add       eax, ecx
 441
 442         pop edi
 443         pop esi
 444         pop edx
 445         pop ecx
 446         pop ebx
 447
 448         pop ebp                 ; restore stack pointer
 449
 450         emms                    ; clear mmx registers
 451         ret
 452
 453
 454 ; total squared difference between two (8*h) blocks
 455 ; blk1,blk2: addresses of top left pels of both blocks
 456 ; lx:        distance (in bytes) of vertically adjacent pels
 457 ; h:         height of block (usually 4, or 8)
 458 ; mmX version
 459
 460 global dist2_22_mmx
 461 ; int dist2_22_mmx(unsigned char *blk1, unsigned char *blk2,
 462 ;                 int lx, int h)
 463
 464 ; mm7 = 0
 465
 466 ; eax = pblk1
 467 ; ebx = pblk2
 468 ; ecx = temp
 469 ; edx = distance_sum
 470 ; edi = h
 471 ; esi = lx
 472
 473 align 32
 474 dist2_22_mmx:
 475         push ebp                        ; save frame pointer
 476         mov ebp, esp            ; link
 477         push ebx
 478         push ecx
 479         push edx
 480         push esi
 481         push edi
 482
 483         mov             esi, [ebp+16] ; lx
 484         mov     edi, [ebp+20] ; h
 485
 486     pxor      mm5, mm5      ; sum
 487         test      edi, edi     ; h = 0?
 488         jle       near d2exit
 489
 490         pxor      mm7, mm7     ; get zeros i mm7
 491
 492         mov       eax, [ebp+8]          ; blk1
 493     mov       ebx, [ebp+12]             ; blk2
 494         jmp      d2top22
 495
 496 align 32
 497 d2top22:
 498         movq      mm0, [eax]
 499         movq      mm1, mm0
 500         punpcklbw mm0, mm7
 501         punpckhbw mm1, mm7
 502
 503         movq      mm2, [ebx]
 504         movq      mm3, mm2
 505         punpcklbw mm2, mm7
 506         punpckhbw mm3, mm7
 507
 508         psubw     mm0, mm2
 509         psubw     mm1, mm3
 510         pmaddwd   mm0, mm0
 511         pmaddwd   mm1, mm1
 512         paddd     mm5, mm0
 513                 paddd    mm5, mm1
 514
 515                         add       eax, esi
 516         add       ebx, esi
 517         dec       edi
 518         jg        d2top22
 519         jmp       d2exit
 520
 521
 522 ; total squared difference between interpolation of two (8*h) blocks and
 523 ; another 8*h block
 524 ; blk1,blk2: addresses of top left pels of both blocks
 525 ; lx:        distance (in bytes) of vertically adjacent pels
 526 ; h:         height of block (usually 4, or 8)
 527 ; mmX version
 528
 529 global bdist2_22_mmx
 530 ; int bdist2_22_mmx(unsigned char *blk1f, unsigned char*blk1b,
 531 ;                                  unsigned char *blk2,
 532 ;                 int lx, int h)
 533
 534 ; mm7 = 0
 535
 536 ; eax = pblk1f
 537 ; ebx = pblk2
 538 ; ecx = pblk1b
 539 ; edx = distance_sum
 540 ; edi = h
 541 ; esi = lx
 542
 543 align 32
 544 bdist2_22_mmx:
 545         push ebp                        ; save frame pointer
 546         mov ebp, esp            ; link
 547         push ebx
 548         push ecx
 549         push edx
 550         push esi
 551         push edi
 552
 553         mov             esi, [ebp+20] ; lx
 554         mov     edi, [ebp+24] ; h
 555
 556     pxor      mm5, mm5      ; sum
 557         test      edi, edi     ; h = 0?
 558         jle       near d2exit
 559
 560         pxor      mm7, mm7     ; get zeros i mm7
 561
 562         mov       eax, [ebp+8]          ; blk1f
 563     mov       ebx, [ebp+12]             ; blk1b
 564     mov       ecx, [ebp+16]             ; blk2
 565         jmp      bd2top22
 566
 567 align 32
 568 bd2top22:
 569         movq      mm0, [eax]
 570         movq      mm1, mm0
 571                 movq      mm4, [ebx]
 572                 movq      mm6, mm4
 573         punpcklbw mm0, mm7
 574         punpckhbw mm1, mm7
 575                 punpcklbw mm4, mm7
 576                 punpckhbw mm6, mm7
 577
 578         movq      mm2, [ecx]
 579         movq      mm3, mm2
 580         punpcklbw mm2, mm7
 581         punpckhbw mm3, mm7
 582
 583                 paddw     mm0, mm4
 584                 psrlw     mm0, 1
 585         psubw     mm0, mm2
 586         pmaddwd   mm0, mm0
 587                 paddw     mm1, mm6
 588                 psrlw     mm1, 1
 589         psubw     mm1, mm3
 590         pmaddwd   mm1, mm1
 591         paddd     mm5, mm0
 592                 paddd    mm5, mm1
 593
 594                 add       eax, esi
 595                 add       ebx, esi
 596                 add               ecx, esi
 597                 dec       edi
 598                 jg        bd2top22
 599                 jmp       d2exit
 600