kc3-lang/libjpeg-turbo/jiss2red.asm
-
Show log
Commit
-
Hash : a2e6a9dd
Author :
Date : 2006-02-04T00:00:00
IJG R6b with x86SIMD V1.02 Independent JPEG Group's JPEG software release 6b with x86 SIMD extension for IJG JPEG library version 1.02
-
jiss2red.asm
-
1 2 3 4 5 6 7 8 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607
; ; jiss2red.asm - reduced-size IDCT (SSE2) ; ; x86 SIMD extension for IJG JPEG library ; Copyright (C) 1999-2006, MIYASAKA Masaru. ; For conditions of distribution and use, see copyright notice in jsimdext.inc ; ; This file should be assembled with NASM (Netwide Assembler), ; can *not* be assembled with Microsoft's MASM or any compatible ; assembler (including Borland's Turbo Assembler). ; NASM is available from http://nasm.sourceforge.net/ or ; http://sourceforge.net/project/showfiles.php?group_id=6208 ; ; This file contains inverse-DCT routines that produce reduced-size ; output: either 4x4 or 2x2 pixels from an 8x8 DCT block. ; The following code is based directly on the IJG's original jidctred.c; ; see the jidctred.c for more details. ; ; Last Modified : February 4, 2006 ; ; [TAB8] %include "jsimdext.inc" %include "jdct.inc" %ifdef IDCT_SCALING_SUPPORTED %ifdef JIDCT_INT_SSE2_SUPPORTED ; This module is specialized to the case DCTSIZE = 8. ; %if DCTSIZE != 8 %error "Sorry, this code only copes with 8x8 DCTs." %endif ; -------------------------------------------------------------------------- %define CONST_BITS 13 %define PASS1_BITS 2 %define DESCALE_P1_4 (CONST_BITS-PASS1_BITS+1) %define DESCALE_P2_4 (CONST_BITS+PASS1_BITS+3+1) %define DESCALE_P1_2 (CONST_BITS-PASS1_BITS+2) %define DESCALE_P2_2 (CONST_BITS+PASS1_BITS+3+2) %if CONST_BITS == 13 F_0_211 equ 1730 ; FIX(0.211164243) F_0_509 equ 4176 ; FIX(0.509795579) F_0_601 equ 4926 ; FIX(0.601344887) F_0_720 equ 5906 ; FIX(0.720959822) F_0_765 equ 6270 ; FIX(0.765366865) F_0_850 equ 6967 ; FIX(0.850430095) F_0_899 equ 7373 ; FIX(0.899976223) F_1_061 equ 8697 ; FIX(1.061594337) F_1_272 equ 10426 ; FIX(1.272758580) F_1_451 equ 11893 ; FIX(1.451774981) F_1_847 equ 15137 ; FIX(1.847759065) F_2_172 equ 17799 ; FIX(2.172734803) F_2_562 equ 20995 ; FIX(2.562915447) F_3_624 equ 29692 ; FIX(3.624509785) %else ; NASM cannot do compile-time arithmetic on floating-point constants. %define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n)) F_0_211 equ DESCALE( 226735879,30-CONST_BITS) ; FIX(0.211164243) F_0_509 equ DESCALE( 547388834,30-CONST_BITS) ; FIX(0.509795579) F_0_601 equ DESCALE( 645689155,30-CONST_BITS) ; FIX(0.601344887) F_0_720 equ DESCALE( 774124714,30-CONST_BITS) ; FIX(0.720959822) F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865) F_0_850 equ DESCALE( 913142361,30-CONST_BITS) ; FIX(0.850430095) F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223) F_1_061 equ DESCALE(1139878239,30-CONST_BITS) ; FIX(1.061594337) F_1_272 equ DESCALE(1366614119,30-CONST_BITS) ; FIX(1.272758580) F_1_451 equ DESCALE(1558831516,30-CONST_BITS) ; FIX(1.451774981) F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065) F_2_172 equ DESCALE(2332956230,30-CONST_BITS) ; FIX(2.172734803) F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447) F_3_624 equ DESCALE(3891787747,30-CONST_BITS) ; FIX(3.624509785) %endif ; -------------------------------------------------------------------------- SECTION SEG_CONST alignz 16 global EXTN(jconst_idct_red_sse2) EXTN(jconst_idct_red_sse2): PW_F184_MF076 times 4 dw F_1_847,-F_0_765 PW_F256_F089 times 4 dw F_2_562, F_0_899 PW_F106_MF217 times 4 dw F_1_061,-F_2_172 PW_MF060_MF050 times 4 dw -F_0_601,-F_0_509 PW_F145_MF021 times 4 dw F_1_451,-F_0_211 PW_F362_MF127 times 4 dw F_3_624,-F_1_272 PW_F085_MF072 times 4 dw F_0_850,-F_0_720 PD_DESCALE_P1_4 times 4 dd 1 << (DESCALE_P1_4-1) PD_DESCALE_P2_4 times 4 dd 1 << (DESCALE_P2_4-1) PD_DESCALE_P1_2 times 4 dd 1 << (DESCALE_P1_2-1) PD_DESCALE_P2_2 times 4 dd 1 << (DESCALE_P2_2-1) PB_CENTERJSAMP times 16 db CENTERJSAMPLE alignz 16 ; -------------------------------------------------------------------------- SECTION SEG_TEXT BITS 32 ; ; Perform dequantization and inverse DCT on one block of coefficients, ; producing a reduced-size 4x4 output block. ; ; GLOBAL(void) ; jpeg_idct_4x4_sse2 (j_decompress_ptr cinfo, jpeg_component_info * compptr, ; JCOEFPTR coef_block, ; JSAMPARRAY output_buf, JDIMENSION output_col) ; %define cinfo(b) (b)+8 ; j_decompress_ptr cinfo %define compptr(b) (b)+12 ; jpeg_component_info * compptr %define coef_block(b) (b)+16 ; JCOEFPTR coef_block %define output_buf(b) (b)+20 ; JSAMPARRAY output_buf %define output_col(b) (b)+24 ; JDIMENSION output_col %define original_ebp ebp+0 %define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM] %define WK_NUM 2 align 16 global EXTN(jpeg_idct_4x4_sse2) EXTN(jpeg_idct_4x4_sse2): push ebp mov eax,esp ; eax = original ebp sub esp, byte 4 and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits mov [esp],eax mov ebp,esp ; ebp = aligned ebp lea esp, [wk(0)] pushpic ebx ; push ecx ; unused ; push edx ; need not be preserved push esi push edi get_GOT ebx ; get GOT address ; ---- Pass 1: process columns from input. ; mov eax, [original_ebp] mov edx, POINTER [compptr(eax)] mov edx, POINTER [jcompinfo_dct_table(edx)] ; quantptr mov esi, JCOEFPTR [coef_block(eax)] ; inptr %ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2 mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] jnz short .columnDCT movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] por xmm1, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] por xmm0, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] por xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] por xmm0,xmm1 packsswb xmm0,xmm0 packsswb xmm0,xmm0 movd eax,xmm0 test eax,eax jnz short .columnDCT ; -- AC terms all zero movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] psllw xmm0,PASS1_BITS movdqa xmm3,xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07) punpcklwd xmm0,xmm0 ; xmm0=(00 00 01 01 02 02 03 03) punpckhwd xmm3,xmm3 ; xmm3=(04 04 05 05 06 06 07 07) pshufd xmm1,xmm0,0x50 ; xmm1=[col0 col1]=(00 00 00 00 01 01 01 01) pshufd xmm0,xmm0,0xFA ; xmm0=[col2 col3]=(02 02 02 02 03 03 03 03) pshufd xmm6,xmm3,0x50 ; xmm6=[col4 col5]=(04 04 04 04 05 05 05 05) pshufd xmm3,xmm3,0xFA ; xmm3=[col6 col7]=(06 06 06 06 07 07 07 07) jmp near .column_end alignx 16,7 %endif .columnDCT: ; -- Odd part movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] movdqa xmm1, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] pmullw xmm0, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)] pmullw xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)] movdqa xmm2, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] pmullw xmm2, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)] pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)] movdqa xmm4,xmm0 movdqa xmm5,xmm0 punpcklwd xmm4,xmm1 punpckhwd xmm5,xmm1 movdqa xmm0,xmm4 movdqa xmm1,xmm5 pmaddwd xmm4,[GOTOFF(ebx,PW_F256_F089)] ; xmm4=(tmp2L) pmaddwd xmm5,[GOTOFF(ebx,PW_F256_F089)] ; xmm5=(tmp2H) pmaddwd xmm0,[GOTOFF(ebx,PW_F106_MF217)] ; xmm0=(tmp0L) pmaddwd xmm1,[GOTOFF(ebx,PW_F106_MF217)] ; xmm1=(tmp0H) movdqa xmm6,xmm2 movdqa xmm7,xmm2 punpcklwd xmm6,xmm3 punpckhwd xmm7,xmm3 movdqa xmm2,xmm6 movdqa xmm3,xmm7 pmaddwd xmm6,[GOTOFF(ebx,PW_MF060_MF050)] ; xmm6=(tmp2L) pmaddwd xmm7,[GOTOFF(ebx,PW_MF060_MF050)] ; xmm7=(tmp2H) pmaddwd xmm2,[GOTOFF(ebx,PW_F145_MF021)] ; xmm2=(tmp0L) pmaddwd xmm3,[GOTOFF(ebx,PW_F145_MF021)] ; xmm3=(tmp0H) paddd xmm6,xmm4 ; xmm6=tmp2L paddd xmm7,xmm5 ; xmm7=tmp2H paddd xmm2,xmm0 ; xmm2=tmp0L paddd xmm3,xmm1 ; xmm3=tmp0H movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp0L movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=tmp0H ; -- Even part movdqa xmm4, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] movdqa xmm5, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] movdqa xmm0, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] pmullw xmm4, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] pmullw xmm5, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)] pmullw xmm0, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)] pxor xmm1,xmm1 pxor xmm2,xmm2 punpcklwd xmm1,xmm4 ; xmm1=tmp0L punpckhwd xmm2,xmm4 ; xmm2=tmp0H psrad xmm1,(16-CONST_BITS-1) ; psrad xmm1,16 & pslld xmm1,CONST_BITS+1 psrad xmm2,(16-CONST_BITS-1) ; psrad xmm2,16 & pslld xmm2,CONST_BITS+1 movdqa xmm3,xmm5 ; xmm5=in2=z2 punpcklwd xmm5,xmm0 ; xmm0=in6=z3 punpckhwd xmm3,xmm0 pmaddwd xmm5,[GOTOFF(ebx,PW_F184_MF076)] ; xmm5=tmp2L pmaddwd xmm3,[GOTOFF(ebx,PW_F184_MF076)] ; xmm3=tmp2H movdqa xmm4,xmm1 movdqa xmm0,xmm2 paddd xmm1,xmm5 ; xmm1=tmp10L paddd xmm2,xmm3 ; xmm2=tmp10H psubd xmm4,xmm5 ; xmm4=tmp12L psubd xmm0,xmm3 ; xmm0=tmp12H ; -- Final output stage movdqa xmm5,xmm1 movdqa xmm3,xmm2 paddd xmm1,xmm6 ; xmm1=data0L paddd xmm2,xmm7 ; xmm2=data0H psubd xmm5,xmm6 ; xmm5=data3L psubd xmm3,xmm7 ; xmm3=data3H movdqa xmm6,[GOTOFF(ebx,PD_DESCALE_P1_4)] ; xmm6=[PD_DESCALE_P1_4] paddd xmm1,xmm6 paddd xmm2,xmm6 psrad xmm1,DESCALE_P1_4 psrad xmm2,DESCALE_P1_4 paddd xmm5,xmm6 paddd xmm3,xmm6 psrad xmm5,DESCALE_P1_4 psrad xmm3,DESCALE_P1_4 packssdw xmm1,xmm2 ; xmm1=data0=(00 01 02 03 04 05 06 07) packssdw xmm5,xmm3 ; xmm5=data3=(30 31 32 33 34 35 36 37) movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp0L movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp0H movdqa xmm2,xmm4 movdqa xmm3,xmm0 paddd xmm4,xmm7 ; xmm4=data1L paddd xmm0,xmm6 ; xmm0=data1H psubd xmm2,xmm7 ; xmm2=data2L psubd xmm3,xmm6 ; xmm3=data2H movdqa xmm7,[GOTOFF(ebx,PD_DESCALE_P1_4)] ; xmm7=[PD_DESCALE_P1_4] paddd xmm4,xmm7 paddd xmm0,xmm7 psrad xmm4,DESCALE_P1_4 psrad xmm0,DESCALE_P1_4 paddd xmm2,xmm7 paddd xmm3,xmm7 psrad xmm2,DESCALE_P1_4 psrad xmm3,DESCALE_P1_4 packssdw xmm4,xmm0 ; xmm4=data1=(10 11 12 13 14 15 16 17) packssdw xmm2,xmm3 ; xmm2=data2=(20 21 22 23 24 25 26 27) movdqa xmm6,xmm1 ; transpose coefficients(phase 1) punpcklwd xmm1,xmm4 ; xmm1=(00 10 01 11 02 12 03 13) punpckhwd xmm6,xmm4 ; xmm6=(04 14 05 15 06 16 07 17) movdqa xmm7,xmm2 ; transpose coefficients(phase 1) punpcklwd xmm2,xmm5 ; xmm2=(20 30 21 31 22 32 23 33) punpckhwd xmm7,xmm5 ; xmm7=(24 34 25 35 26 36 27 37) movdqa xmm0,xmm1 ; transpose coefficients(phase 2) punpckldq xmm1,xmm2 ; xmm1=[col0 col1]=(00 10 20 30 01 11 21 31) punpckhdq xmm0,xmm2 ; xmm0=[col2 col3]=(02 12 22 32 03 13 23 33) movdqa xmm3,xmm6 ; transpose coefficients(phase 2) punpckldq xmm6,xmm7 ; xmm6=[col4 col5]=(04 14 24 34 05 15 25 35) punpckhdq xmm3,xmm7 ; xmm3=[col6 col7]=(06 16 26 36 07 17 27 37) .column_end: ; -- Prefetch the next coefficient block prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32] prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32] prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32] prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32] ; ---- Pass 2: process rows, store into output array. mov eax, [original_ebp] mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) mov eax, JDIMENSION [output_col(eax)] ; -- Even part pxor xmm4,xmm4 punpcklwd xmm4,xmm1 ; xmm4=tmp0 psrad xmm4,(16-CONST_BITS-1) ; psrad xmm4,16 & pslld xmm4,CONST_BITS+1 ; -- Odd part punpckhwd xmm1,xmm0 punpckhwd xmm6,xmm3 movdqa xmm5,xmm1 movdqa xmm2,xmm6 pmaddwd xmm1,[GOTOFF(ebx,PW_F256_F089)] ; xmm1=(tmp2) pmaddwd xmm6,[GOTOFF(ebx,PW_MF060_MF050)] ; xmm6=(tmp2) pmaddwd xmm5,[GOTOFF(ebx,PW_F106_MF217)] ; xmm5=(tmp0) pmaddwd xmm2,[GOTOFF(ebx,PW_F145_MF021)] ; xmm2=(tmp0) paddd xmm6,xmm1 ; xmm6=tmp2 paddd xmm2,xmm5 ; xmm2=tmp0 ; -- Even part punpcklwd xmm0,xmm3 pmaddwd xmm0,[GOTOFF(ebx,PW_F184_MF076)] ; xmm0=tmp2 movdqa xmm7,xmm4 paddd xmm4,xmm0 ; xmm4=tmp10 psubd xmm7,xmm0 ; xmm7=tmp12 ; -- Final output stage movdqa xmm1,[GOTOFF(ebx,PD_DESCALE_P2_4)] ; xmm1=[PD_DESCALE_P2_4] movdqa xmm5,xmm4 movdqa xmm3,xmm7 paddd xmm4,xmm6 ; xmm4=data0=(00 10 20 30) paddd xmm7,xmm2 ; xmm7=data1=(01 11 21 31) psubd xmm5,xmm6 ; xmm5=data3=(03 13 23 33) psubd xmm3,xmm2 ; xmm3=data2=(02 12 22 32) paddd xmm4,xmm1 paddd xmm7,xmm1 psrad xmm4,DESCALE_P2_4 psrad xmm7,DESCALE_P2_4 paddd xmm5,xmm1 paddd xmm3,xmm1 psrad xmm5,DESCALE_P2_4 psrad xmm3,DESCALE_P2_4 packssdw xmm4,xmm3 ; xmm4=(00 10 20 30 02 12 22 32) packssdw xmm7,xmm5 ; xmm7=(01 11 21 31 03 13 23 33) movdqa xmm0,xmm4 ; transpose coefficients(phase 1) punpcklwd xmm4,xmm7 ; xmm4=(00 01 10 11 20 21 30 31) punpckhwd xmm0,xmm7 ; xmm0=(02 03 12 13 22 23 32 33) movdqa xmm6,xmm4 ; transpose coefficients(phase 2) punpckldq xmm4,xmm0 ; xmm4=(00 01 02 03 10 11 12 13) punpckhdq xmm6,xmm0 ; xmm6=(20 21 22 23 30 31 32 33) packsswb xmm4,xmm6 ; xmm4=(00 01 02 03 10 11 12 13 20 ..) paddb xmm4,[GOTOFF(ebx,PB_CENTERJSAMP)] pshufd xmm2,xmm4,0x39 ; xmm2=(10 11 12 13 20 21 22 23 30 ..) pshufd xmm1,xmm4,0x4E ; xmm1=(20 21 22 23 30 31 32 33 00 ..) pshufd xmm3,xmm4,0x93 ; xmm3=(30 31 32 33 00 01 02 03 10 ..) mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] movd _DWORD [edx+eax*SIZEOF_JSAMPLE], xmm4 movd _DWORD [esi+eax*SIZEOF_JSAMPLE], xmm2 mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW] mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW] movd _DWORD [edx+eax*SIZEOF_JSAMPLE], xmm1 movd _DWORD [esi+eax*SIZEOF_JSAMPLE], xmm3 pop edi pop esi ; pop edx ; need not be preserved ; pop ecx ; unused poppic ebx mov esp,ebp ; esp <- aligned ebp pop esp ; esp <- original ebp pop ebp ret ; -------------------------------------------------------------------------- ; ; Perform dequantization and inverse DCT on one block of coefficients, ; producing a reduced-size 2x2 output block. ; ; GLOBAL(void) ; jpeg_idct_2x2_sse2 (j_decompress_ptr cinfo, jpeg_component_info * compptr, ; JCOEFPTR coef_block, ; JSAMPARRAY output_buf, JDIMENSION output_col) ; %define cinfo(b) (b)+8 ; j_decompress_ptr cinfo %define compptr(b) (b)+12 ; jpeg_component_info * compptr %define coef_block(b) (b)+16 ; JCOEFPTR coef_block %define output_buf(b) (b)+20 ; JSAMPARRAY output_buf %define output_col(b) (b)+24 ; JDIMENSION output_col align 16 global EXTN(jpeg_idct_2x2_sse2) EXTN(jpeg_idct_2x2_sse2): push ebp mov ebp,esp push ebx ; push ecx ; need not be preserved ; push edx ; need not be preserved push esi push edi get_GOT ebx ; get GOT address ; ---- Pass 1: process columns from input. mov edx, POINTER [compptr(ebp)] mov edx, POINTER [jcompinfo_dct_table(edx)] ; quantptr mov esi, JCOEFPTR [coef_block(ebp)] ; inptr ; | input: | result: | ; | 00 01 ** 03 ** 05 ** 07 | | ; | 10 11 ** 13 ** 15 ** 17 | | ; | ** ** ** ** ** ** ** ** | | ; | 30 31 ** 33 ** 35 ** 37 | A0 A1 A3 A5 A7 | ; | ** ** ** ** ** ** ** ** | B0 B1 B3 B5 B7 | ; | 50 51 ** 53 ** 55 ** 57 | | ; | ** ** ** ** ** ** ** ** | | ; | 70 71 ** 73 ** 75 ** 77 | | ; -- Odd part movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] movdqa xmm1, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] pmullw xmm0, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)] pmullw xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)] movdqa xmm2, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] pmullw xmm2, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)] pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)] ; xmm0=(10 11 ** 13 ** 15 ** 17), xmm1=(30 31 ** 33 ** 35 ** 37) ; xmm2=(50 51 ** 53 ** 55 ** 57), xmm3=(70 71 ** 73 ** 75 ** 77) pcmpeqd xmm7,xmm7 pslld xmm7,WORD_BIT ; xmm7={0x0000 0xFFFF 0x0000 0xFFFF ..} movdqa xmm4,xmm0 ; xmm4=(10 11 ** 13 ** 15 ** 17) movdqa xmm5,xmm2 ; xmm5=(50 51 ** 53 ** 55 ** 57) punpcklwd xmm4,xmm1 ; xmm4=(10 30 11 31 ** ** 13 33) punpcklwd xmm5,xmm3 ; xmm5=(50 70 51 71 ** ** 53 73) pmaddwd xmm4,[GOTOFF(ebx,PW_F362_MF127)] pmaddwd xmm5,[GOTOFF(ebx,PW_F085_MF072)] psrld xmm0,WORD_BIT ; xmm0=(11 -- 13 -- 15 -- 17 --) pand xmm1,xmm7 ; xmm1=(-- 31 -- 33 -- 35 -- 37) psrld xmm2,WORD_BIT ; xmm2=(51 -- 53 -- 55 -- 57 --) pand xmm3,xmm7 ; xmm3=(-- 71 -- 73 -- 75 -- 77) por xmm0,xmm1 ; xmm0=(11 31 13 33 15 35 17 37) por xmm2,xmm3 ; xmm2=(51 71 53 73 55 75 57 77) pmaddwd xmm0,[GOTOFF(ebx,PW_F362_MF127)] pmaddwd xmm2,[GOTOFF(ebx,PW_F085_MF072)] paddd xmm4,xmm5 ; xmm4=tmp0[col0 col1 **** col3] paddd xmm0,xmm2 ; xmm0=tmp0[col1 col3 col5 col7] ; -- Even part movdqa xmm6, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] pmullw xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] ; xmm6=(00 01 ** 03 ** 05 ** 07) movdqa xmm1,xmm6 ; xmm1=(00 01 ** 03 ** 05 ** 07) pslld xmm6,WORD_BIT ; xmm6=(-- 00 -- ** -- ** -- **) pand xmm1,xmm7 ; xmm1=(-- 01 -- 03 -- 05 -- 07) psrad xmm6,(WORD_BIT-CONST_BITS-2) ; xmm6=tmp10[col0 **** **** ****] psrad xmm1,(WORD_BIT-CONST_BITS-2) ; xmm1=tmp10[col1 col3 col5 col7] ; -- Final output stage movdqa xmm3,xmm6 movdqa xmm5,xmm1 paddd xmm6,xmm4 ; xmm6=data0[col0 **** **** ****]=(A0 ** ** **) paddd xmm1,xmm0 ; xmm1=data0[col1 col3 col5 col7]=(A1 A3 A5 A7) psubd xmm3,xmm4 ; xmm3=data1[col0 **** **** ****]=(B0 ** ** **) psubd xmm5,xmm0 ; xmm5=data1[col1 col3 col5 col7]=(B1 B3 B5 B7) movdqa xmm2,[GOTOFF(ebx,PD_DESCALE_P1_2)] ; xmm2=[PD_DESCALE_P1_2] punpckldq xmm6,xmm3 ; xmm6=(A0 B0 ** **) movdqa xmm7,xmm1 punpcklqdq xmm1,xmm5 ; xmm1=(A1 A3 B1 B3) punpckhqdq xmm7,xmm5 ; xmm7=(A5 A7 B5 B7) paddd xmm6,xmm2 psrad xmm6,DESCALE_P1_2 paddd xmm1,xmm2 paddd xmm7,xmm2 psrad xmm1,DESCALE_P1_2 psrad xmm7,DESCALE_P1_2 ; -- Prefetch the next coefficient block prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32] prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32] prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32] prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32] ; ---- Pass 2: process rows, store into output array. mov edi, JSAMPARRAY [output_buf(ebp)] ; (JSAMPROW *) mov eax, JDIMENSION [output_col(ebp)] ; | input:| result:| ; | A0 B0 | | ; | A1 B1 | C0 C1 | ; | A3 B3 | D0 D1 | ; | A5 B5 | | ; | A7 B7 | | ; -- Odd part packssdw xmm1,xmm1 ; xmm1=(A1 A3 B1 B3 A1 A3 B1 B3) packssdw xmm7,xmm7 ; xmm7=(A5 A7 B5 B7 A5 A7 B5 B7) pmaddwd xmm1,[GOTOFF(ebx,PW_F362_MF127)] pmaddwd xmm7,[GOTOFF(ebx,PW_F085_MF072)] paddd xmm1,xmm7 ; xmm1=tmp0[row0 row1 row0 row1] ; -- Even part pslld xmm6,(CONST_BITS+2) ; xmm6=tmp10[row0 row1 **** ****] ; -- Final output stage movdqa xmm4,xmm6 paddd xmm6,xmm1 ; xmm6=data0[row0 row1 **** ****]=(C0 C1 ** **) psubd xmm4,xmm1 ; xmm4=data1[row0 row1 **** ****]=(D0 D1 ** **) punpckldq xmm6,xmm4 ; xmm6=(C0 D0 C1 D1) paddd xmm6,[GOTOFF(ebx,PD_DESCALE_P2_2)] psrad xmm6,DESCALE_P2_2 packssdw xmm6,xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1) packsswb xmm6,xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1 ..) paddb xmm6,[GOTOFF(ebx,PB_CENTERJSAMP)] pextrw ebx,xmm6,0x00 ; ebx=(C0 D0 -- --) pextrw ecx,xmm6,0x01 ; ecx=(C1 D1 -- --) mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] mov WORD [edx+eax*SIZEOF_JSAMPLE], bx mov WORD [esi+eax*SIZEOF_JSAMPLE], cx pop edi pop esi ; pop edx ; need not be preserved ; pop ecx ; need not be preserved pop ebx pop ebp ret %endif ; JIDCT_INT_SSE2_SUPPORTED %endif ; IDCT_SCALING_SUPPORTED