mpeg2enc/predcomp_mmx.s

   1 ;;;
   2 ;;;  predcomp_00_mmx.s:
   3 ;;;
   4 ;;;               Extended MMX prediction composition
   5 ;;;  routines handling the four different interpolation cases...
   6 ;;;
   7 ;;;  Copyright (C) 2000 Andrew Stevens <as@comlab.ox.ac.uk>
   8
   9 ;;;
  10 ;;;  This program is free software; you can reaxstribute it and/or
  11 ;;;  modify it under the terms of the GNU General Public License
  12 ;;;  as published by the Free Software Foundation; either version 2
  13 ;;;  of the License, or (at your option) any later version.
  14 ;;;
  15 ;;;  This program is distributed in the hope that it will be useful,
  16 ;;;  but WITHOUT ANY WARRANTY; without even the implied warranty of
  17 ;;;  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18 ;;;  GNU General Public License for more details.
  19 ;;;
  20 ;;;  You should have received a copy of the GNU General Public License
  21 ;;;  along with this program; if not, write to the Free Software
  22 ;;;  Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
  23 ;;;  02111-1307, USA.
  24 ;;;
  25 ;;;
  26 ;;;
  27
  28 ;;; The no interpolation case...
  29
  30 global predcomp_00_mmx
  31
  32 ;;; void predcomp_<ix><iy>_mmx(char *src,char *dst,int lx, int w, int h, int addflag);
  33
  34 ;;; ix - Interpolation in x iy - Interpolation in y
  35
  36 ;;; eax = pdst
  37 ;;; ebx = psrc
  38 ;;; ecx = h left
  39 ;;; edx = lx;
  40 ;;; edi = w (8 or 16)
  41
  42
  43 ;;; mm1 = one's mask for src
  44 ;;; mm0 = zero mask for src...
  45
  46
  47
  48 align 32
  49 predcomp_00_mmx:
  50         push ebp                                        ; save frame pointer
  51         mov ebp, esp                            ; link
  52
  53         push eax
  54         push ebx
  55         push ecx
  56         push edx
  57         push edi
  58         push esi
  59
  60         mov     eax, 0x00010001
  61         movd  mm1, eax
  62         punpckldq mm1,mm1
  63
  64         mov ebx, [ebp+8]        ; get psrc
  65         mov eax, [ebp+12]       ; get pdst
  66         mov edx, [ebp+16]       ; get lx
  67         mov edi, [ebp+20]       ;  get w
  68         mov ecx, [ebp+24]       ; get h
  69         mov esi, [ebp+28]                       ; get addflag
  70                 ;; Extend addflag into bit-mask
  71         pxor mm0, mm0
  72         jmp predrow00m          ; align for speed
  73 align 32
  74 predrow00m:
  75         movq mm4, [ebx]         ; first 8 bytes of row
  76         cmp esi, 0
  77         jz      noadd00
  78
  79         movq mm5, mm4
  80         punpcklbw       mm4, mm0
  81         punpckhbw       mm5, mm0
  82
  83         movq mm2, [eax]
  84         movq mm3, mm2
  85         punpcklbw       mm2, mm0
  86         punpckhbw       mm3, mm0
  87         paddw           mm4, mm2
  88         paddw           mm5, mm3
  89         paddw           mm4, mm1
  90         paddw           mm5, mm1
  91         psrlw           mm4, 1
  92         psrlw           mm5, 1
  93         packuswb        mm4, mm5
  94 noadd00:
  95         movq            [eax], mm4
  96
  97         cmp     edi, 8
  98         jz  eightwide00
  99
 100         movq mm4, [ebx+8]               ; first 8 bytes of row
 101         cmp esi, 0
 102         jz      noadd00w
 103
 104         movq mm5, mm4
 105         punpcklbw       mm4, mm0
 106         punpckhbw       mm5, mm0
 107
 108         movq mm2, [eax+8]
 109         movq mm3, mm2
 110         punpcklbw       mm2, mm0
 111         punpckhbw       mm3, mm0
 112         paddw           mm4, mm2
 113         paddw           mm5, mm3
 114         paddw           mm4, mm1
 115         paddw           mm5, mm1
 116         psrlw           mm4, 1
 117         psrlw           mm5, 1
 118         packuswb        mm4, mm5
 119 noadd00w:
 120         movq    [eax+8], mm4
 121
 122 eightwide00:
 123         add eax, edx            ; update pointer to next row
 124         add ebx, edx            ; ditto
 125
 126         sub  ecx, 1                                     ; check h left
 127         jnz near predrow00m
 128
 129         pop esi
 130         pop edi
 131         pop edx
 132         pop ecx
 133         pop ebx
 134         pop eax
 135         pop ebp
 136         emms
 137         ret
 138
 139
 140 ;;; The x-axis interpolation case...
 141
 142 global predcomp_10_mmx
 143
 144
 145 align 32
 146 predcomp_10_mmx:
 147         push ebp                                        ; save frame pointer
 148         mov ebp, esp                            ; link
 149
 150         push eax
 151         push ebx
 152         push ecx
 153         push edx
 154         push edi
 155         push esi
 156
 157         mov     eax, 0x00010001
 158         movd  mm1, eax
 159         punpckldq mm1,mm1
 160
 161         mov ebx, [ebp+8]        ; get psrc
 162         mov eax, [ebp+12]       ; get pdst
 163         mov edx, [ebp+16]       ; get lx
 164         mov edi, [ebp+20]       ;  get w
 165         mov ecx, [ebp+24]       ; get h
 166         mov esi, [ebp+28]                       ; get addflag
 167                 ;; Extend addflag into bit-mask
 168         pxor mm0, mm0
 169         jmp predrow10m          ; align for speed
 170 align 32
 171 predrow10m:
 172         movq mm4, [ebx]         ; first 8 bytes of row
 173         movq mm5, mm4
 174         punpcklbw       mm4, mm0
 175         punpckhbw       mm5, mm0
 176         movq mm2, [ebx+1]
 177         movq mm3, mm2
 178         punpcklbw       mm2, mm0
 179         punpckhbw       mm3, mm0
 180
 181         paddw           mm4, mm2                ;  Average mm4/mm5 and mm2/mm3
 182         paddw           mm5, mm3
 183         paddw           mm4, mm1
 184         paddw           mm5, mm1
 185         psrlw           mm4, 1
 186         psrlw           mm5, 1
 187
 188         cmp esi, 0
 189         jz      noadd10
 190
 191         movq mm2, [eax]                         ;  Add
 192         movq mm3, mm2
 193         punpcklbw       mm2, mm0
 194         punpckhbw       mm3, mm0
 195         paddw           mm4, mm2                ;  Average mm4/mm5 and mm2/mm3
 196         paddw           mm5, mm3
 197         paddw           mm4, mm1
 198         paddw           mm5, mm1
 199         psrlw           mm4, 1
 200         psrlw           mm5, 1
 201 noadd10:
 202         packuswb        mm4, mm5
 203         movq            [eax], mm4
 204
 205         cmp     edi, 8
 206         jz  eightwide10
 207
 208         movq mm4, [ebx+8]               ; first 8 bytes of row
 209         movq mm5, mm4
 210         punpcklbw       mm4, mm0
 211         punpckhbw       mm5, mm0
 212         movq mm2, [ebx+9]
 213         movq mm3, mm2
 214         punpcklbw       mm2, mm0
 215         punpckhbw       mm3, mm0
 216
 217         paddw           mm4, mm2                ;  Average mm4/mm5 and mm2/mm3
 218         paddw           mm5, mm3
 219         paddw           mm4, mm1
 220         paddw           mm5, mm1
 221         psrlw           mm4, 1
 222         psrlw           mm5, 1
 223
 224         cmp esi, 0
 225         jz      noadd10w
 226
 227         movq mm2, [eax+8]                               ;  Add
 228         movq mm3, mm2
 229         punpcklbw       mm2, mm0
 230         punpckhbw       mm3, mm0
 231         paddw           mm4, mm2                ;  Average mm4/mm5 and mm2/mm3
 232         paddw           mm5, mm3
 233         paddw           mm4, mm1
 234         paddw           mm5, mm1
 235         psrlw           mm4, 1
 236         psrlw           mm5, 1
 237 noadd10w:
 238         packuswb        mm4, mm5
 239         movq            [eax+8], mm4
 240
 241
 242 eightwide10:
 243         add eax, edx            ; update pointer to next row
 244         add ebx, edx            ; ditto
 245
 246         sub  ecx, 1                                     ; check h left
 247         jnz near predrow10m
 248
 249         pop esi
 250         pop edi
 251         pop edx
 252         pop ecx
 253         pop ebx
 254         pop eax
 255         pop ebp
 256         emms
 257         ret
 258
 259 ;;; The y-axis interpolation case...
 260
 261 global predcomp_01_mmx
 262
 263
 264 align 32
 265 predcomp_01_mmx:
 266         push ebp                                        ; save frame pointer
 267         mov ebp, esp                            ; link
 268
 269         push eax
 270         push ebx
 271         push ecx
 272         push edx
 273         push edi
 274         push esi
 275
 276         mov     eax, 0x00010001
 277         movd  mm1, eax
 278         punpckldq mm1,mm1
 279
 280         mov ebx, [ebp+8]        ; get psrc
 281         mov eax, [ebp+12]       ; get pdst
 282         mov edx, [ebp+16]       ; get lx
 283         mov edi, [ebp+20]       ;  get w
 284         mov ecx, [ebp+24]       ; get h
 285         mov esi, [ebp+28]                       ; get addflag
 286         pxor mm0, mm0
 287         jmp predrow01m          ; align for speed
 288
 289 align 32
 290 predrow01m:
 291         movq mm4, [ebx]         ; first 8 bytes of row
 292         movq mm5, mm4
 293         add                     ebx, edx                                ; Next row
 294         punpcklbw       mm4, mm0
 295         punpckhbw       mm5, mm0
 296
 297         movq mm2, [ebx]
 298         movq mm3, mm2
 299         punpcklbw       mm2, mm0
 300         punpckhbw       mm3, mm0
 301
 302         paddw           mm4, mm2                ;  Average mm4/mm5 and mm2/mm3
 303         paddw           mm5, mm3
 304         paddw           mm4, mm1
 305         paddw           mm5, mm1
 306         psrlw           mm4, 1
 307         psrlw           mm5, 1
 308
 309         cmp esi, 0
 310         jz      noadd01
 311
 312         movq mm2, [eax]                         ;  Add
 313         movq mm3, mm2
 314         punpcklbw       mm2, mm0
 315         punpckhbw       mm3, mm0
 316         paddw           mm4, mm2                ;  Average mm4/mm5 and mm2/mm3
 317         paddw           mm5, mm3
 318         paddw           mm4, mm1
 319         paddw           mm5, mm1
 320         psrlw           mm4, 1
 321         psrlw           mm5, 1
 322 noadd01:
 323         packuswb        mm4, mm5
 324         movq            [eax], mm4
 325
 326         cmp     edi, 8
 327         jz  eightwide01
 328
 329         sub                     ebx, edx                ;  Back to first row...
 330         movq mm4, [ebx+8]               ; first 8 bytes of row
 331         movq mm5, mm4
 332         add                     ebx, edx                                ; Next row
 333         punpcklbw       mm4, mm0
 334         punpckhbw       mm5, mm0
 335         movq mm2, [ebx+8]
 336         movq mm3, mm2
 337         punpcklbw       mm2, mm0
 338         punpckhbw       mm3, mm0
 339
 340         paddw           mm4, mm2                ;  Average mm4/mm5 and mm2/mm3
 341         paddw           mm5, mm3
 342         paddw           mm4, mm1
 343         paddw           mm5, mm1
 344         psrlw           mm4, 1
 345         psrlw           mm5, 1
 346
 347         cmp esi, 0
 348         jz      noadd01w
 349
 350         movq mm2, [eax+8]                               ;  Add
 351         movq mm3, mm2
 352         punpcklbw       mm2, mm0
 353         punpckhbw       mm3, mm0
 354         paddw           mm4, mm2                ;  Average mm4/mm5 and mm2/mm3
 355         paddw           mm5, mm3
 356         paddw           mm4, mm1
 357         paddw           mm5, mm1
 358         psrlw           mm4, 1
 359         psrlw           mm5, 1
 360 noadd01w:
 361         packuswb        mm4, mm5
 362         movq            [eax+8], mm4
 363
 364
 365 eightwide01:
 366         add eax, edx            ; ditto
 367
 368         sub  ecx, 1                                     ; check h left
 369         jnz near predrow01m
 370
 371         pop esi
 372         pop edi
 373         pop edx
 374         pop ecx
 375         pop ebx
 376         pop eax
 377         pop ebp
 378         emms
 379         ret
 380
 381
 382 ;;; The x-axis and y-axis interpolation case...
 383
 384 global predcomp_11_mmx
 385
 386 ;;; mm0 = [0,0,0,0]W
 387 ;;; mm1 = [1,1,1,1]W
 388 ;;; mm2 = [2,2,2,2]W
 389 align 32
 390 predcomp_11_mmx:
 391         push ebp                                        ; save frame pointer
 392         mov ebp, esp                            ; link
 393
 394         push eax
 395         push ebx
 396         push ecx
 397         push edx
 398         push edi
 399         push esi
 400
 401         mov     eax, 0x00020002
 402         movd  mm2, eax
 403         punpckldq mm2,mm2
 404         mov     eax, 0x00010001
 405         movd  mm1, eax
 406         punpckldq mm1,mm1
 407         pxor mm0, mm0
 408
 409         mov ebx, [ebp+8]        ; get psrc
 410         mov eax, [ebp+12]       ; get pdst
 411         mov edx, [ebp+16]       ; get lx
 412         mov edi, [ebp+20]       ;  get w
 413         mov ecx, [ebp+24]       ; get h
 414         mov esi, [ebp+28]                       ;  Addflags
 415                 ;; Extend addflag into bit-mask
 416
 417
 418         jmp predrow11           ; align for speed
 419 align 32
 420 predrow11:
 421         movq mm4, [ebx]         ; mm4 and mm6 accumulate partial sums for interp.
 422         movq mm6, mm4
 423         punpcklbw       mm4, mm0
 424         punpckhbw       mm6, mm0
 425
 426         movq mm5, [ebx+1]
 427         movq mm7, mm5
 428         punpcklbw       mm5, mm0
 429         paddw           mm4, mm5
 430         punpckhbw       mm7, mm0
 431         paddw           mm6, mm7
 432
 433         add ebx, edx            ; update pointer to next row
 434
 435         movq mm5, [ebx]         ; first 8 bytes 1st row:         avg src in x
 436         movq mm7, mm5
 437         punpcklbw       mm5, mm0                ;  Accumulate partial interpolation
 438         paddw           mm4, mm5
 439         punpckhbw   mm7, mm0
 440         paddw           mm6, mm7
 441
 442         movq mm5, [ebx+1]
 443         movq mm7, mm5
 444         punpcklbw       mm5, mm0
 445         paddw           mm4, mm5
 446         punpckhbw       mm7, mm0
 447         paddw           mm6, mm7
 448
 449                 ;; Now round
 450         paddw           mm4, mm2
 451         paddw           mm6, mm2
 452         psrlw           mm4, 2
 453         psrlw           mm6, 2
 454
 455         cmp esi, 0
 456         jz  noadd11
 457
 458         movq mm5, [eax]                         ;  Add
 459         movq mm7, mm5
 460         punpcklbw       mm5, mm0
 461         punpckhbw       mm7, mm0
 462         paddw           mm4, mm5                ;  Average mm4/mm6 and mm5/mm7
 463         paddw           mm6, mm7
 464         paddw           mm4, mm1
 465         paddw           mm6, mm1
 466         psrlw           mm4, 1
 467         psrlw           mm6, 1
 468
 469 noadd11:
 470         packuswb        mm4, mm6
 471         movq [eax], mm4
 472
 473         cmp   edi, 8
 474         jz near eightwide11
 475
 476         sub ebx, edx            ; Back to first row...
 477
 478         movq mm4, [ebx+8]               ; mm4 and mm6 accumulate partial sums for interp.
 479         movq mm6, mm4
 480         punpcklbw       mm4, mm0
 481         punpckhbw       mm6, mm0
 482
 483         movq mm5, [ebx+9]
 484         movq mm7, mm5
 485         punpcklbw       mm5, mm0
 486         paddw           mm4, mm5
 487         punpckhbw       mm7, mm0
 488         paddw           mm6, mm7
 489
 490         add ebx, edx            ; update pointer to next row
 491
 492         movq mm5, [ebx+8]               ; first 8 bytes 1st row:         avg src in x
 493         movq mm7, mm5
 494         punpcklbw       mm5, mm0                ;  Accumulate partial interpolation
 495         paddw           mm4, mm5
 496         punpckhbw   mm7, mm0
 497         paddw           mm6, mm7
 498
 499         movq mm5, [ebx+9]
 500         movq mm7, mm5
 501         punpcklbw       mm5, mm0
 502         paddw           mm4, mm5
 503         punpckhbw       mm7, mm0
 504         paddw           mm6, mm7
 505
 506                 ;; Now round
 507         paddw           mm4, mm2
 508         paddw           mm6, mm2
 509         psrlw           mm4, 2
 510         psrlw           mm6, 2
 511
 512         cmp esi, 0
 513         jz  noadd11w
 514
 515         movq mm5, [eax+8]                               ;  Add and average
 516         movq mm7, mm5
 517         punpcklbw       mm5, mm0
 518         punpckhbw       mm7, mm0
 519         paddw           mm4, mm5                ;  Average mm4/mm6 and mm5/mm7
 520         paddw           mm6, mm7
 521         paddw           mm4, mm1
 522         paddw           mm6, mm1
 523         psrlw           mm4, 1
 524         psrlw           mm6, 1
 525 noadd11w:
 526         packuswb        mm4, mm6
 527         movq [eax+8], mm4
 528
 529 eightwide11:
 530         add eax, edx            ; update pointer to next row
 531
 532
 533         sub  ecx, 1                     ; check h left
 534         jnz near predrow11
 535
 536         pop esi
 537         pop edi
 538         pop edx
 539         pop ecx
 540         pop ebx
 541         pop eax
 542         pop ebp
 543         emms
 544         ret
 545
 546
 547