kc3-lang/libjpeg-turbo/jisseflt.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
-
jisseflt.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
; ; jisseflt.asm - floating-point IDCT (SSE & MMX) ; ; 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 a floating-point implementation of the inverse DCT ; (Discrete Cosine Transform). The following code is based directly on ; the IJG's original jidctflt.c; see the jidctflt.c for more details. ; ; Last Modified : February 4, 2006 ; ; [TAB8] %include "jsimdext.inc" %include "jdct.inc" %ifdef DCT_FLOAT_SUPPORTED %ifdef JIDCT_FLT_SSE_MMX_SUPPORTED ; This module is specialized to the case DCTSIZE = 8. ; %if DCTSIZE != 8 %error "Sorry, this code only copes with 8x8 DCTs." %endif ; -------------------------------------------------------------------------- %macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5) shufps %1,%2,0x44 %endmacro %macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7) shufps %1,%2,0xEE %endmacro ; -------------------------------------------------------------------------- SECTION SEG_CONST alignz 16 global EXTN(jconst_idct_float_sse) EXTN(jconst_idct_float_sse): PD_1_414 times 4 dd 1.414213562373095048801689 PD_1_847 times 4 dd 1.847759065022573512256366 PD_1_082 times 4 dd 1.082392200292393968799446 PD_M2_613 times 4 dd -2.613125929752753055713286 PD_0_125 times 4 dd 0.125 ; 1/8 PB_CENTERJSAMP times 8 db CENTERJSAMPLE alignz 16 ; -------------------------------------------------------------------------- SECTION SEG_TEXT BITS 32 ; ; Perform dequantization and inverse DCT on one block of coefficients. ; ; GLOBAL(void) ; jpeg_idct_float_sse (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 %define workspace wk(0)-DCTSIZE2*SIZEOF_FAST_FLOAT ; FAST_FLOAT workspace[DCTSIZE2] align 16 global EXTN(jpeg_idct_float_sse) EXTN(jpeg_idct_float_sse): 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, [workspace] 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, store into work array. ; mov eax, [original_ebp] mov edx, POINTER [compptr(eax)] mov edx, POINTER [jcompinfo_dct_table(edx)] ; quantptr mov esi, JCOEFPTR [coef_block(eax)] ; inptr lea edi, [workspace] ; FAST_FLOAT * wsptr mov ecx, DCTSIZE/4 ; ctr alignx 16,7 .columnloop: %ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] jnz near .columnDCT movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] por mm0, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] por mm1, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] por mm0, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] por mm1, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] por mm0, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] por mm1,mm0 packsswb mm1,mm1 movd eax,mm1 test eax,eax jnz short .columnDCT ; -- AC terms all zero movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] punpckhwd mm1,mm0 ; mm1=(** 02 ** 03) punpcklwd mm0,mm0 ; mm0=(00 00 01 01) psrad mm1,(DWORD_BIT-WORD_BIT) ; mm1=in0H=(02 03) psrad mm0,(DWORD_BIT-WORD_BIT) ; mm0=in0L=(00 01) cvtpi2ps xmm3,mm1 ; xmm3=(02 03 ** **) cvtpi2ps xmm0,mm0 ; xmm0=(00 01 ** **) movlhps xmm0,xmm3 ; xmm0=in0=(00 01 02 03) mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)] movaps xmm1,xmm0 movaps xmm2,xmm0 movaps xmm3,xmm0 shufps xmm0,xmm0,0x00 ; xmm0=(00 00 00 00) shufps xmm1,xmm1,0x55 ; xmm1=(01 01 01 01) shufps xmm2,xmm2,0xAA ; xmm2=(02 02 02 02) shufps xmm3,xmm3,0xFF ; xmm3=(03 03 03 03) movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm0 movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm0 movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm1 movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm1 movaps XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_FAST_FLOAT)], xmm2 movaps XMMWORD [XMMBLOCK(2,1,edi,SIZEOF_FAST_FLOAT)], xmm2 movaps XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_FAST_FLOAT)], xmm3 movaps XMMWORD [XMMBLOCK(3,1,edi,SIZEOF_FAST_FLOAT)], xmm3 jmp near .nextcolumn alignx 16,7 %endif .columnDCT: ; -- Even part movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] punpckhwd mm4,mm0 ; mm4=(** 02 ** 03) punpcklwd mm0,mm0 ; mm0=(00 00 01 01) punpckhwd mm5,mm1 ; mm5=(** 22 ** 23) punpcklwd mm1,mm1 ; mm1=(20 20 21 21) psrad mm4,(DWORD_BIT-WORD_BIT) ; mm4=in0H=(02 03) psrad mm0,(DWORD_BIT-WORD_BIT) ; mm0=in0L=(00 01) cvtpi2ps xmm4,mm4 ; xmm4=(02 03 ** **) cvtpi2ps xmm0,mm0 ; xmm0=(00 01 ** **) psrad mm5,(DWORD_BIT-WORD_BIT) ; mm5=in2H=(22 23) psrad mm1,(DWORD_BIT-WORD_BIT) ; mm1=in2L=(20 21) cvtpi2ps xmm5,mm5 ; xmm5=(22 23 ** **) cvtpi2ps xmm1,mm1 ; xmm1=(20 21 ** **) punpckhwd mm6,mm2 ; mm6=(** 42 ** 43) punpcklwd mm2,mm2 ; mm2=(40 40 41 41) punpckhwd mm7,mm3 ; mm7=(** 62 ** 63) punpcklwd mm3,mm3 ; mm3=(60 60 61 61) psrad mm6,(DWORD_BIT-WORD_BIT) ; mm6=in4H=(42 43) psrad mm2,(DWORD_BIT-WORD_BIT) ; mm2=in4L=(40 41) cvtpi2ps xmm6,mm6 ; xmm6=(42 43 ** **) cvtpi2ps xmm2,mm2 ; xmm2=(40 41 ** **) psrad mm7,(DWORD_BIT-WORD_BIT) ; mm7=in6H=(62 63) psrad mm3,(DWORD_BIT-WORD_BIT) ; mm3=in6L=(60 61) cvtpi2ps xmm7,mm7 ; xmm7=(62 63 ** **) cvtpi2ps xmm3,mm3 ; xmm3=(60 61 ** **) movlhps xmm0,xmm4 ; xmm0=in0=(00 01 02 03) movlhps xmm1,xmm5 ; xmm1=in2=(20 21 22 23) mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)] mulps xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FLOAT_MULT_TYPE)] movlhps xmm2,xmm6 ; xmm2=in4=(40 41 42 43) movlhps xmm3,xmm7 ; xmm3=in6=(60 61 62 63) mulps xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FLOAT_MULT_TYPE)] mulps xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FLOAT_MULT_TYPE)] movaps xmm4,xmm0 movaps xmm5,xmm1 subps xmm0,xmm2 ; xmm0=tmp11 subps xmm1,xmm3 addps xmm4,xmm2 ; xmm4=tmp10 addps xmm5,xmm3 ; xmm5=tmp13 mulps xmm1,[GOTOFF(ebx,PD_1_414)] subps xmm1,xmm5 ; xmm1=tmp12 movaps xmm6,xmm4 movaps xmm7,xmm0 subps xmm4,xmm5 ; xmm4=tmp3 subps xmm0,xmm1 ; xmm0=tmp2 addps xmm6,xmm5 ; xmm6=tmp0 addps xmm7,xmm1 ; xmm7=tmp1 movaps XMMWORD [wk(1)], xmm4 ; tmp3 movaps XMMWORD [wk(0)], xmm0 ; tmp2 ; -- Odd part movq mm4, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] movq mm0, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] movq mm5, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] movq mm1, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] punpckhwd mm6,mm4 ; mm6=(** 12 ** 13) punpcklwd mm4,mm4 ; mm4=(10 10 11 11) punpckhwd mm2,mm0 ; mm2=(** 32 ** 33) punpcklwd mm0,mm0 ; mm0=(30 30 31 31) psrad mm6,(DWORD_BIT-WORD_BIT) ; mm6=in1H=(12 13) psrad mm4,(DWORD_BIT-WORD_BIT) ; mm4=in1L=(10 11) cvtpi2ps xmm4,mm6 ; xmm4=(12 13 ** **) cvtpi2ps xmm2,mm4 ; xmm2=(10 11 ** **) psrad mm2,(DWORD_BIT-WORD_BIT) ; mm2=in3H=(32 33) psrad mm0,(DWORD_BIT-WORD_BIT) ; mm0=in3L=(30 31) cvtpi2ps xmm0,mm2 ; xmm0=(32 33 ** **) cvtpi2ps xmm3,mm0 ; xmm3=(30 31 ** **) punpckhwd mm7,mm5 ; mm7=(** 52 ** 53) punpcklwd mm5,mm5 ; mm5=(50 50 51 51) punpckhwd mm3,mm1 ; mm3=(** 72 ** 73) punpcklwd mm1,mm1 ; mm1=(70 70 71 71) movlhps xmm2,xmm4 ; xmm2=in1=(10 11 12 13) movlhps xmm3,xmm0 ; xmm3=in3=(30 31 32 33) psrad mm7,(DWORD_BIT-WORD_BIT) ; mm7=in5H=(52 53) psrad mm5,(DWORD_BIT-WORD_BIT) ; mm5=in5L=(50 51) cvtpi2ps xmm4,mm7 ; xmm4=(52 53 ** **) cvtpi2ps xmm5,mm5 ; xmm5=(50 51 ** **) psrad mm3,(DWORD_BIT-WORD_BIT) ; mm3=in7H=(72 73) psrad mm1,(DWORD_BIT-WORD_BIT) ; mm1=in7L=(70 71) cvtpi2ps xmm0,mm3 ; xmm0=(72 73 ** **) cvtpi2ps xmm1,mm1 ; xmm1=(70 71 ** **) mulps xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FLOAT_MULT_TYPE)] mulps xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FLOAT_MULT_TYPE)] movlhps xmm5,xmm4 ; xmm5=in5=(50 51 52 53) movlhps xmm1,xmm0 ; xmm1=in7=(70 71 72 73) mulps xmm5, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FLOAT_MULT_TYPE)] mulps xmm1, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FLOAT_MULT_TYPE)] movaps xmm4,xmm2 movaps xmm0,xmm5 addps xmm2,xmm1 ; xmm2=z11 addps xmm5,xmm3 ; xmm5=z13 subps xmm4,xmm1 ; xmm4=z12 subps xmm0,xmm3 ; xmm0=z10 movaps xmm1,xmm2 subps xmm2,xmm5 addps xmm1,xmm5 ; xmm1=tmp7 mulps xmm2,[GOTOFF(ebx,PD_1_414)] ; xmm2=tmp11 movaps xmm3,xmm0 addps xmm0,xmm4 mulps xmm0,[GOTOFF(ebx,PD_1_847)] ; xmm0=z5 mulps xmm3,[GOTOFF(ebx,PD_M2_613)] ; xmm3=(z10 * -2.613125930) mulps xmm4,[GOTOFF(ebx,PD_1_082)] ; xmm4=(z12 * 1.082392200) addps xmm3,xmm0 ; xmm3=tmp12 subps xmm4,xmm0 ; xmm4=tmp10 ; -- Final output stage subps xmm3,xmm1 ; xmm3=tmp6 movaps xmm5,xmm6 movaps xmm0,xmm7 addps xmm6,xmm1 ; xmm6=data0=(00 01 02 03) addps xmm7,xmm3 ; xmm7=data1=(10 11 12 13) subps xmm5,xmm1 ; xmm5=data7=(70 71 72 73) subps xmm0,xmm3 ; xmm0=data6=(60 61 62 63) subps xmm2,xmm3 ; xmm2=tmp5 movaps xmm1,xmm6 ; transpose coefficients(phase 1) unpcklps xmm6,xmm7 ; xmm6=(00 10 01 11) unpckhps xmm1,xmm7 ; xmm1=(02 12 03 13) movaps xmm3,xmm0 ; transpose coefficients(phase 1) unpcklps xmm0,xmm5 ; xmm0=(60 70 61 71) unpckhps xmm3,xmm5 ; xmm3=(62 72 63 73) movaps xmm7, XMMWORD [wk(0)] ; xmm7=tmp2 movaps xmm5, XMMWORD [wk(1)] ; xmm5=tmp3 movaps XMMWORD [wk(0)], xmm0 ; wk(0)=(60 70 61 71) movaps XMMWORD [wk(1)], xmm3 ; wk(1)=(62 72 63 73) addps xmm4,xmm2 ; xmm4=tmp4 movaps xmm0,xmm7 movaps xmm3,xmm5 addps xmm7,xmm2 ; xmm7=data2=(20 21 22 23) addps xmm5,xmm4 ; xmm5=data4=(40 41 42 43) subps xmm0,xmm2 ; xmm0=data5=(50 51 52 53) subps xmm3,xmm4 ; xmm3=data3=(30 31 32 33) movaps xmm2,xmm7 ; transpose coefficients(phase 1) unpcklps xmm7,xmm3 ; xmm7=(20 30 21 31) unpckhps xmm2,xmm3 ; xmm2=(22 32 23 33) movaps xmm4,xmm5 ; transpose coefficients(phase 1) unpcklps xmm5,xmm0 ; xmm5=(40 50 41 51) unpckhps xmm4,xmm0 ; xmm4=(42 52 43 53) movaps xmm3,xmm6 ; transpose coefficients(phase 2) unpcklps2 xmm6,xmm7 ; xmm6=(00 10 20 30) unpckhps2 xmm3,xmm7 ; xmm3=(01 11 21 31) movaps xmm0,xmm1 ; transpose coefficients(phase 2) unpcklps2 xmm1,xmm2 ; xmm1=(02 12 22 32) unpckhps2 xmm0,xmm2 ; xmm0=(03 13 23 33) movaps xmm7, XMMWORD [wk(0)] ; xmm7=(60 70 61 71) movaps xmm2, XMMWORD [wk(1)] ; xmm2=(62 72 63 73) movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm6 movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm3 movaps XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_FAST_FLOAT)], xmm1 movaps XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_FAST_FLOAT)], xmm0 movaps xmm6,xmm5 ; transpose coefficients(phase 2) unpcklps2 xmm5,xmm7 ; xmm5=(40 50 60 70) unpckhps2 xmm6,xmm7 ; xmm6=(41 51 61 71) movaps xmm3,xmm4 ; transpose coefficients(phase 2) unpcklps2 xmm4,xmm2 ; xmm4=(42 52 62 72) unpckhps2 xmm3,xmm2 ; xmm3=(43 53 63 73) movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm5 movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm6 movaps XMMWORD [XMMBLOCK(2,1,edi,SIZEOF_FAST_FLOAT)], xmm4 movaps XMMWORD [XMMBLOCK(3,1,edi,SIZEOF_FAST_FLOAT)], xmm3 .nextcolumn: add esi, byte 4*SIZEOF_JCOEF ; coef_block add edx, byte 4*SIZEOF_FLOAT_MULT_TYPE ; quantptr add edi, 4*DCTSIZE*SIZEOF_FAST_FLOAT ; wsptr dec ecx ; ctr jnz near .columnloop ; -- Prefetch the next coefficient block prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 0*32] prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 1*32] prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 2*32] prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 3*32] ; ---- Pass 2: process rows from work array, store into output array. mov eax, [original_ebp] lea esi, [workspace] ; FAST_FLOAT * wsptr mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) mov eax, JDIMENSION [output_col(eax)] mov ecx, DCTSIZE/4 ; ctr alignx 16,7 .rowloop: ; -- Even part movaps xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)] movaps xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_FAST_FLOAT)] movaps xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_FAST_FLOAT)] movaps xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_FAST_FLOAT)] movaps xmm4,xmm0 movaps xmm5,xmm1 subps xmm0,xmm2 ; xmm0=tmp11 subps xmm1,xmm3 addps xmm4,xmm2 ; xmm4=tmp10 addps xmm5,xmm3 ; xmm5=tmp13 mulps xmm1,[GOTOFF(ebx,PD_1_414)] subps xmm1,xmm5 ; xmm1=tmp12 movaps xmm6,xmm4 movaps xmm7,xmm0 subps xmm4,xmm5 ; xmm4=tmp3 subps xmm0,xmm1 ; xmm0=tmp2 addps xmm6,xmm5 ; xmm6=tmp0 addps xmm7,xmm1 ; xmm7=tmp1 movaps XMMWORD [wk(1)], xmm4 ; tmp3 movaps XMMWORD [wk(0)], xmm0 ; tmp2 ; -- Odd part movaps xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)] movaps xmm3, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_FAST_FLOAT)] movaps xmm5, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_FAST_FLOAT)] movaps xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_FAST_FLOAT)] movaps xmm4,xmm2 movaps xmm0,xmm5 addps xmm2,xmm1 ; xmm2=z11 addps xmm5,xmm3 ; xmm5=z13 subps xmm4,xmm1 ; xmm4=z12 subps xmm0,xmm3 ; xmm0=z10 movaps xmm1,xmm2 subps xmm2,xmm5 addps xmm1,xmm5 ; xmm1=tmp7 mulps xmm2,[GOTOFF(ebx,PD_1_414)] ; xmm2=tmp11 movaps xmm3,xmm0 addps xmm0,xmm4 mulps xmm0,[GOTOFF(ebx,PD_1_847)] ; xmm0=z5 mulps xmm3,[GOTOFF(ebx,PD_M2_613)] ; xmm3=(z10 * -2.613125930) mulps xmm4,[GOTOFF(ebx,PD_1_082)] ; xmm4=(z12 * 1.082392200) addps xmm3,xmm0 ; xmm3=tmp12 subps xmm4,xmm0 ; xmm4=tmp10 ; -- Final output stage subps xmm3,xmm1 ; xmm3=tmp6 movaps xmm5,xmm6 movaps xmm0,xmm7 addps xmm6,xmm1 ; xmm6=data0=(00 10 20 30) addps xmm7,xmm3 ; xmm7=data1=(01 11 21 31) subps xmm5,xmm1 ; xmm5=data7=(07 17 27 37) subps xmm0,xmm3 ; xmm0=data6=(06 16 26 36) subps xmm2,xmm3 ; xmm2=tmp5 movaps xmm1,[GOTOFF(ebx,PD_0_125)] ; xmm1=[PD_0_125] mulps xmm6,xmm1 ; descale(1/8) mulps xmm7,xmm1 ; descale(1/8) mulps xmm5,xmm1 ; descale(1/8) mulps xmm0,xmm1 ; descale(1/8) movhlps xmm3,xmm6 movhlps xmm1,xmm7 cvtps2pi mm0,xmm6 ; round to int32, mm0=data0L=(00 10) cvtps2pi mm1,xmm7 ; round to int32, mm1=data1L=(01 11) cvtps2pi mm2,xmm3 ; round to int32, mm2=data0H=(20 30) cvtps2pi mm3,xmm1 ; round to int32, mm3=data1H=(21 31) packssdw mm0,mm2 ; mm0=data0=(00 10 20 30) packssdw mm1,mm3 ; mm1=data1=(01 11 21 31) movhlps xmm6,xmm5 movhlps xmm7,xmm0 cvtps2pi mm4,xmm5 ; round to int32, mm4=data7L=(07 17) cvtps2pi mm5,xmm0 ; round to int32, mm5=data6L=(06 16) cvtps2pi mm6,xmm6 ; round to int32, mm6=data7H=(27 37) cvtps2pi mm7,xmm7 ; round to int32, mm7=data6H=(26 36) packssdw mm4,mm6 ; mm4=data7=(07 17 27 37) packssdw mm5,mm7 ; mm5=data6=(06 16 26 36) packsswb mm0,mm5 ; mm0=(00 10 20 30 06 16 26 36) packsswb mm1,mm4 ; mm1=(01 11 21 31 07 17 27 37) movaps xmm3, XMMWORD [wk(0)] ; xmm3=tmp2 movaps xmm1, XMMWORD [wk(1)] ; xmm1=tmp3 movaps xmm6,[GOTOFF(ebx,PD_0_125)] ; xmm6=[PD_0_125] addps xmm4,xmm2 ; xmm4=tmp4 movaps xmm5,xmm3 movaps xmm0,xmm1 addps xmm3,xmm2 ; xmm3=data2=(02 12 22 32) addps xmm1,xmm4 ; xmm1=data4=(04 14 24 34) subps xmm5,xmm2 ; xmm5=data5=(05 15 25 35) subps xmm0,xmm4 ; xmm0=data3=(03 13 23 33) mulps xmm3,xmm6 ; descale(1/8) mulps xmm1,xmm6 ; descale(1/8) mulps xmm5,xmm6 ; descale(1/8) mulps xmm0,xmm6 ; descale(1/8) movhlps xmm7,xmm3 movhlps xmm2,xmm1 cvtps2pi mm2,xmm3 ; round to int32, mm2=data2L=(02 12) cvtps2pi mm3,xmm1 ; round to int32, mm3=data4L=(04 14) cvtps2pi mm6,xmm7 ; round to int32, mm6=data2H=(22 32) cvtps2pi mm7,xmm2 ; round to int32, mm7=data4H=(24 34) packssdw mm2,mm6 ; mm2=data2=(02 12 22 32) packssdw mm3,mm7 ; mm3=data4=(04 14 24 34) movhlps xmm4,xmm5 movhlps xmm6,xmm0 cvtps2pi mm5,xmm5 ; round to int32, mm5=data5L=(05 15) cvtps2pi mm4,xmm0 ; round to int32, mm4=data3L=(03 13) cvtps2pi mm6,xmm4 ; round to int32, mm6=data5H=(25 35) cvtps2pi mm7,xmm6 ; round to int32, mm7=data3H=(23 33) packssdw mm5,mm6 ; mm5=data5=(05 15 25 35) packssdw mm4,mm7 ; mm4=data3=(03 13 23 33) movq mm6,[GOTOFF(ebx,PB_CENTERJSAMP)] ; mm6=[PB_CENTERJSAMP] packsswb mm2,mm3 ; mm2=(02 12 22 32 04 14 24 34) packsswb mm4,mm5 ; mm4=(03 13 23 33 05 15 25 35) paddb mm0,mm6 paddb mm1,mm6 paddb mm2,mm6 paddb mm4,mm6 movq mm7,mm0 ; transpose coefficients(phase 1) punpcklbw mm0,mm1 ; mm0=(00 01 10 11 20 21 30 31) punpckhbw mm7,mm1 ; mm7=(06 07 16 17 26 27 36 37) movq mm3,mm2 ; transpose coefficients(phase 1) punpcklbw mm2,mm4 ; mm2=(02 03 12 13 22 23 32 33) punpckhbw mm3,mm4 ; mm3=(04 05 14 15 24 25 34 35) movq mm5,mm0 ; transpose coefficients(phase 2) punpcklwd mm0,mm2 ; mm0=(00 01 02 03 10 11 12 13) punpckhwd mm5,mm2 ; mm5=(20 21 22 23 30 31 32 33) movq mm6,mm3 ; transpose coefficients(phase 2) punpcklwd mm3,mm7 ; mm3=(04 05 06 07 14 15 16 17) punpckhwd mm6,mm7 ; mm6=(24 25 26 27 34 35 36 37) movq mm1,mm0 ; transpose coefficients(phase 3) punpckldq mm0,mm3 ; mm0=(00 01 02 03 04 05 06 07) punpckhdq mm1,mm3 ; mm1=(10 11 12 13 14 15 16 17) movq mm4,mm5 ; transpose coefficients(phase 3) punpckldq mm5,mm6 ; mm5=(20 21 22 23 24 25 26 27) punpckhdq mm4,mm6 ; mm4=(30 31 32 33 34 35 36 37) pushpic ebx ; save GOT address mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] mov ebx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm0 movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm1 mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW] mov ebx, JSAMPROW [edi+3*SIZEOF_JSAMPROW] movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm5 movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm4 poppic ebx ; restore GOT address add esi, byte 4*SIZEOF_FAST_FLOAT ; wsptr add edi, byte 4*SIZEOF_JSAMPROW dec ecx ; ctr jnz near .rowloop emms ; empty MMX state pop edi pop esi ; pop edx ; need not be preserved ; pop ecx ; need not be preserved pop ebx mov esp,ebp ; esp <- aligned ebp pop esp ; esp <- original ebp pop ebp ret %endif ; JIDCT_FLT_SSE_MMX_SUPPORTED %endif ; DCT_FLOAT_SUPPORTED