1; 2; jfdctint.asm - accurate integer FDCT (MMX) 3; 4; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB 5; 6; Based on the x86 SIMD extension for IJG JPEG library 7; Copyright (C) 1999-2006, MIYASAKA Masaru. 8; For conditions of distribution and use, see copyright notice in jsimdext.inc 9; 10; This file should be assembled with NASM (Netwide Assembler), 11; can *not* be assembled with Microsoft's MASM or any compatible 12; assembler (including Borland's Turbo Assembler). 13; NASM is available from http://nasm.sourceforge.net/ or 14; http://sourceforge.net/project/showfiles.php?group_id=6208 15; 16; This file contains a slow-but-accurate integer implementation of the 17; forward DCT (Discrete Cosine Transform). The following code is based 18; directly on the IJG's original jfdctint.c; see the jfdctint.c for 19; more details. 20; 21; [TAB8] 22 23%include "jsimdext.inc" 24%include "jdct.inc" 25 26; -------------------------------------------------------------------------- 27 28%define CONST_BITS 13 29%define PASS1_BITS 2 30 31%define DESCALE_P1 (CONST_BITS-PASS1_BITS) 32%define DESCALE_P2 (CONST_BITS+PASS1_BITS) 33 34%if CONST_BITS == 13 35F_0_298 equ 2446 ; FIX(0.298631336) 36F_0_390 equ 3196 ; FIX(0.390180644) 37F_0_541 equ 4433 ; FIX(0.541196100) 38F_0_765 equ 6270 ; FIX(0.765366865) 39F_0_899 equ 7373 ; FIX(0.899976223) 40F_1_175 equ 9633 ; FIX(1.175875602) 41F_1_501 equ 12299 ; FIX(1.501321110) 42F_1_847 equ 15137 ; FIX(1.847759065) 43F_1_961 equ 16069 ; FIX(1.961570560) 44F_2_053 equ 16819 ; FIX(2.053119869) 45F_2_562 equ 20995 ; FIX(2.562915447) 46F_3_072 equ 25172 ; FIX(3.072711026) 47%else 48; NASM cannot do compile-time arithmetic on floating-point constants. 49%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n)) 50F_0_298 equ DESCALE( 320652955,30-CONST_BITS) ; FIX(0.298631336) 51F_0_390 equ DESCALE( 418953276,30-CONST_BITS) ; FIX(0.390180644) 52F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100) 53F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865) 54F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223) 55F_1_175 equ DESCALE(1262586813,30-CONST_BITS) ; FIX(1.175875602) 56F_1_501 equ DESCALE(1612031267,30-CONST_BITS) ; FIX(1.501321110) 57F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065) 58F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560) 59F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869) 60F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447) 61F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026) 62%endif 63 64; -------------------------------------------------------------------------- 65 SECTION SEG_CONST 66 67 alignz 16 68 global EXTN(jconst_fdct_islow_mmx) 69 70EXTN(jconst_fdct_islow_mmx): 71 72PW_F130_F054 times 2 dw (F_0_541+F_0_765), F_0_541 73PW_F054_MF130 times 2 dw F_0_541, (F_0_541-F_1_847) 74PW_MF078_F117 times 2 dw (F_1_175-F_1_961), F_1_175 75PW_F117_F078 times 2 dw F_1_175, (F_1_175-F_0_390) 76PW_MF060_MF089 times 2 dw (F_0_298-F_0_899),-F_0_899 77PW_MF089_F060 times 2 dw -F_0_899, (F_1_501-F_0_899) 78PW_MF050_MF256 times 2 dw (F_2_053-F_2_562),-F_2_562 79PW_MF256_F050 times 2 dw -F_2_562, (F_3_072-F_2_562) 80PD_DESCALE_P1 times 2 dd 1 << (DESCALE_P1-1) 81PD_DESCALE_P2 times 2 dd 1 << (DESCALE_P2-1) 82PW_DESCALE_P2X times 4 dw 1 << (PASS1_BITS-1) 83 84 alignz 16 85 86; -------------------------------------------------------------------------- 87 SECTION SEG_TEXT 88 BITS 32 89; 90; Perform the forward DCT on one block of samples. 91; 92; GLOBAL(void) 93; jsimd_fdct_islow_mmx (DCTELEM *data) 94; 95 96%define data(b) (b)+8 ; DCTELEM *data 97 98%define original_ebp ebp+0 99%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_MMWORD ; mmword wk[WK_NUM] 100%define WK_NUM 2 101 102 align 16 103 global EXTN(jsimd_fdct_islow_mmx) 104 105EXTN(jsimd_fdct_islow_mmx): 106 push ebp 107 mov eax,esp ; eax = original ebp 108 sub esp, byte 4 109 and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits 110 mov [esp],eax 111 mov ebp,esp ; ebp = aligned ebp 112 lea esp, [wk(0)] 113 pushpic ebx 114; push ecx ; need not be preserved 115; push edx ; need not be preserved 116; push esi ; unused 117; push edi ; unused 118 119 get_GOT ebx ; get GOT address 120 121 ; ---- Pass 1: process rows. 122 123 mov edx, POINTER [data(eax)] ; (DCTELEM *) 124 mov ecx, DCTSIZE/4 125 alignx 16,7 126.rowloop: 127 128 movq mm0, MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)] 129 movq mm1, MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)] 130 movq mm2, MMWORD [MMBLOCK(2,1,edx,SIZEOF_DCTELEM)] 131 movq mm3, MMWORD [MMBLOCK(3,1,edx,SIZEOF_DCTELEM)] 132 133 ; mm0=(20 21 22 23), mm2=(24 25 26 27) 134 ; mm1=(30 31 32 33), mm3=(34 35 36 37) 135 136 movq mm4,mm0 ; transpose coefficients(phase 1) 137 punpcklwd mm0,mm1 ; mm0=(20 30 21 31) 138 punpckhwd mm4,mm1 ; mm4=(22 32 23 33) 139 movq mm5,mm2 ; transpose coefficients(phase 1) 140 punpcklwd mm2,mm3 ; mm2=(24 34 25 35) 141 punpckhwd mm5,mm3 ; mm5=(26 36 27 37) 142 143 movq mm6, MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)] 144 movq mm7, MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)] 145 movq mm1, MMWORD [MMBLOCK(0,1,edx,SIZEOF_DCTELEM)] 146 movq mm3, MMWORD [MMBLOCK(1,1,edx,SIZEOF_DCTELEM)] 147 148 ; mm6=(00 01 02 03), mm1=(04 05 06 07) 149 ; mm7=(10 11 12 13), mm3=(14 15 16 17) 150 151 movq MMWORD [wk(0)], mm4 ; wk(0)=(22 32 23 33) 152 movq MMWORD [wk(1)], mm2 ; wk(1)=(24 34 25 35) 153 154 movq mm4,mm6 ; transpose coefficients(phase 1) 155 punpcklwd mm6,mm7 ; mm6=(00 10 01 11) 156 punpckhwd mm4,mm7 ; mm4=(02 12 03 13) 157 movq mm2,mm1 ; transpose coefficients(phase 1) 158 punpcklwd mm1,mm3 ; mm1=(04 14 05 15) 159 punpckhwd mm2,mm3 ; mm2=(06 16 07 17) 160 161 movq mm7,mm6 ; transpose coefficients(phase 2) 162 punpckldq mm6,mm0 ; mm6=(00 10 20 30)=data0 163 punpckhdq mm7,mm0 ; mm7=(01 11 21 31)=data1 164 movq mm3,mm2 ; transpose coefficients(phase 2) 165 punpckldq mm2,mm5 ; mm2=(06 16 26 36)=data6 166 punpckhdq mm3,mm5 ; mm3=(07 17 27 37)=data7 167 168 movq mm0,mm7 169 movq mm5,mm6 170 psubw mm7,mm2 ; mm7=data1-data6=tmp6 171 psubw mm6,mm3 ; mm6=data0-data7=tmp7 172 paddw mm0,mm2 ; mm0=data1+data6=tmp1 173 paddw mm5,mm3 ; mm5=data0+data7=tmp0 174 175 movq mm2, MMWORD [wk(0)] ; mm2=(22 32 23 33) 176 movq mm3, MMWORD [wk(1)] ; mm3=(24 34 25 35) 177 movq MMWORD [wk(0)], mm7 ; wk(0)=tmp6 178 movq MMWORD [wk(1)], mm6 ; wk(1)=tmp7 179 180 movq mm7,mm4 ; transpose coefficients(phase 2) 181 punpckldq mm4,mm2 ; mm4=(02 12 22 32)=data2 182 punpckhdq mm7,mm2 ; mm7=(03 13 23 33)=data3 183 movq mm6,mm1 ; transpose coefficients(phase 2) 184 punpckldq mm1,mm3 ; mm1=(04 14 24 34)=data4 185 punpckhdq mm6,mm3 ; mm6=(05 15 25 35)=data5 186 187 movq mm2,mm7 188 movq mm3,mm4 189 paddw mm7,mm1 ; mm7=data3+data4=tmp3 190 paddw mm4,mm6 ; mm4=data2+data5=tmp2 191 psubw mm2,mm1 ; mm2=data3-data4=tmp4 192 psubw mm3,mm6 ; mm3=data2-data5=tmp5 193 194 ; -- Even part 195 196 movq mm1,mm5 197 movq mm6,mm0 198 paddw mm5,mm7 ; mm5=tmp10 199 paddw mm0,mm4 ; mm0=tmp11 200 psubw mm1,mm7 ; mm1=tmp13 201 psubw mm6,mm4 ; mm6=tmp12 202 203 movq mm7,mm5 204 paddw mm5,mm0 ; mm5=tmp10+tmp11 205 psubw mm7,mm0 ; mm7=tmp10-tmp11 206 207 psllw mm5,PASS1_BITS ; mm5=data0 208 psllw mm7,PASS1_BITS ; mm7=data4 209 210 movq MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)], mm5 211 movq MMWORD [MMBLOCK(0,1,edx,SIZEOF_DCTELEM)], mm7 212 213 ; (Original) 214 ; z1 = (tmp12 + tmp13) * 0.541196100; 215 ; data2 = z1 + tmp13 * 0.765366865; 216 ; data6 = z1 + tmp12 * -1.847759065; 217 ; 218 ; (This implementation) 219 ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; 220 ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); 221 222 movq mm4,mm1 ; mm1=tmp13 223 movq mm0,mm1 224 punpcklwd mm4,mm6 ; mm6=tmp12 225 punpckhwd mm0,mm6 226 movq mm1,mm4 227 movq mm6,mm0 228 pmaddwd mm4,[GOTOFF(ebx,PW_F130_F054)] ; mm4=data2L 229 pmaddwd mm0,[GOTOFF(ebx,PW_F130_F054)] ; mm0=data2H 230 pmaddwd mm1,[GOTOFF(ebx,PW_F054_MF130)] ; mm1=data6L 231 pmaddwd mm6,[GOTOFF(ebx,PW_F054_MF130)] ; mm6=data6H 232 233 paddd mm4,[GOTOFF(ebx,PD_DESCALE_P1)] 234 paddd mm0,[GOTOFF(ebx,PD_DESCALE_P1)] 235 psrad mm4,DESCALE_P1 236 psrad mm0,DESCALE_P1 237 paddd mm1,[GOTOFF(ebx,PD_DESCALE_P1)] 238 paddd mm6,[GOTOFF(ebx,PD_DESCALE_P1)] 239 psrad mm1,DESCALE_P1 240 psrad mm6,DESCALE_P1 241 242 packssdw mm4,mm0 ; mm4=data2 243 packssdw mm1,mm6 ; mm1=data6 244 245 movq MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)], mm4 246 movq MMWORD [MMBLOCK(2,1,edx,SIZEOF_DCTELEM)], mm1 247 248 ; -- Odd part 249 250 movq mm5, MMWORD [wk(0)] ; mm5=tmp6 251 movq mm7, MMWORD [wk(1)] ; mm7=tmp7 252 253 movq mm0,mm2 ; mm2=tmp4 254 movq mm6,mm3 ; mm3=tmp5 255 paddw mm0,mm5 ; mm0=z3 256 paddw mm6,mm7 ; mm6=z4 257 258 ; (Original) 259 ; z5 = (z3 + z4) * 1.175875602; 260 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; 261 ; z3 += z5; z4 += z5; 262 ; 263 ; (This implementation) 264 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; 265 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); 266 267 movq mm4,mm0 268 movq mm1,mm0 269 punpcklwd mm4,mm6 270 punpckhwd mm1,mm6 271 movq mm0,mm4 272 movq mm6,mm1 273 pmaddwd mm4,[GOTOFF(ebx,PW_MF078_F117)] ; mm4=z3L 274 pmaddwd mm1,[GOTOFF(ebx,PW_MF078_F117)] ; mm1=z3H 275 pmaddwd mm0,[GOTOFF(ebx,PW_F117_F078)] ; mm0=z4L 276 pmaddwd mm6,[GOTOFF(ebx,PW_F117_F078)] ; mm6=z4H 277 278 movq MMWORD [wk(0)], mm4 ; wk(0)=z3L 279 movq MMWORD [wk(1)], mm1 ; wk(1)=z3H 280 281 ; (Original) 282 ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; 283 ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; 284 ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; 285 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; 286 ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; 287 ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; 288 ; 289 ; (This implementation) 290 ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; 291 ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; 292 ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); 293 ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); 294 ; data7 = tmp4 + z3; data5 = tmp5 + z4; 295 ; data3 = tmp6 + z3; data1 = tmp7 + z4; 296 297 movq mm4,mm2 298 movq mm1,mm2 299 punpcklwd mm4,mm7 300 punpckhwd mm1,mm7 301 movq mm2,mm4 302 movq mm7,mm1 303 pmaddwd mm4,[GOTOFF(ebx,PW_MF060_MF089)] ; mm4=tmp4L 304 pmaddwd mm1,[GOTOFF(ebx,PW_MF060_MF089)] ; mm1=tmp4H 305 pmaddwd mm2,[GOTOFF(ebx,PW_MF089_F060)] ; mm2=tmp7L 306 pmaddwd mm7,[GOTOFF(ebx,PW_MF089_F060)] ; mm7=tmp7H 307 308 paddd mm4, MMWORD [wk(0)] ; mm4=data7L 309 paddd mm1, MMWORD [wk(1)] ; mm1=data7H 310 paddd mm2,mm0 ; mm2=data1L 311 paddd mm7,mm6 ; mm7=data1H 312 313 paddd mm4,[GOTOFF(ebx,PD_DESCALE_P1)] 314 paddd mm1,[GOTOFF(ebx,PD_DESCALE_P1)] 315 psrad mm4,DESCALE_P1 316 psrad mm1,DESCALE_P1 317 paddd mm2,[GOTOFF(ebx,PD_DESCALE_P1)] 318 paddd mm7,[GOTOFF(ebx,PD_DESCALE_P1)] 319 psrad mm2,DESCALE_P1 320 psrad mm7,DESCALE_P1 321 322 packssdw mm4,mm1 ; mm4=data7 323 packssdw mm2,mm7 ; mm2=data1 324 325 movq MMWORD [MMBLOCK(3,1,edx,SIZEOF_DCTELEM)], mm4 326 movq MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)], mm2 327 328 movq mm1,mm3 329 movq mm7,mm3 330 punpcklwd mm1,mm5 331 punpckhwd mm7,mm5 332 movq mm3,mm1 333 movq mm5,mm7 334 pmaddwd mm1,[GOTOFF(ebx,PW_MF050_MF256)] ; mm1=tmp5L 335 pmaddwd mm7,[GOTOFF(ebx,PW_MF050_MF256)] ; mm7=tmp5H 336 pmaddwd mm3,[GOTOFF(ebx,PW_MF256_F050)] ; mm3=tmp6L 337 pmaddwd mm5,[GOTOFF(ebx,PW_MF256_F050)] ; mm5=tmp6H 338 339 paddd mm1,mm0 ; mm1=data5L 340 paddd mm7,mm6 ; mm7=data5H 341 paddd mm3, MMWORD [wk(0)] ; mm3=data3L 342 paddd mm5, MMWORD [wk(1)] ; mm5=data3H 343 344 paddd mm1,[GOTOFF(ebx,PD_DESCALE_P1)] 345 paddd mm7,[GOTOFF(ebx,PD_DESCALE_P1)] 346 psrad mm1,DESCALE_P1 347 psrad mm7,DESCALE_P1 348 paddd mm3,[GOTOFF(ebx,PD_DESCALE_P1)] 349 paddd mm5,[GOTOFF(ebx,PD_DESCALE_P1)] 350 psrad mm3,DESCALE_P1 351 psrad mm5,DESCALE_P1 352 353 packssdw mm1,mm7 ; mm1=data5 354 packssdw mm3,mm5 ; mm3=data3 355 356 movq MMWORD [MMBLOCK(1,1,edx,SIZEOF_DCTELEM)], mm1 357 movq MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)], mm3 358 359 add edx, byte 4*DCTSIZE*SIZEOF_DCTELEM 360 dec ecx 361 jnz near .rowloop 362 363 ; ---- Pass 2: process columns. 364 365 mov edx, POINTER [data(eax)] ; (DCTELEM *) 366 mov ecx, DCTSIZE/4 367 alignx 16,7 368.columnloop: 369 370 movq mm0, MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)] 371 movq mm1, MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)] 372 movq mm2, MMWORD [MMBLOCK(6,0,edx,SIZEOF_DCTELEM)] 373 movq mm3, MMWORD [MMBLOCK(7,0,edx,SIZEOF_DCTELEM)] 374 375 ; mm0=(02 12 22 32), mm2=(42 52 62 72) 376 ; mm1=(03 13 23 33), mm3=(43 53 63 73) 377 378 movq mm4,mm0 ; transpose coefficients(phase 1) 379 punpcklwd mm0,mm1 ; mm0=(02 03 12 13) 380 punpckhwd mm4,mm1 ; mm4=(22 23 32 33) 381 movq mm5,mm2 ; transpose coefficients(phase 1) 382 punpcklwd mm2,mm3 ; mm2=(42 43 52 53) 383 punpckhwd mm5,mm3 ; mm5=(62 63 72 73) 384 385 movq mm6, MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)] 386 movq mm7, MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)] 387 movq mm1, MMWORD [MMBLOCK(4,0,edx,SIZEOF_DCTELEM)] 388 movq mm3, MMWORD [MMBLOCK(5,0,edx,SIZEOF_DCTELEM)] 389 390 ; mm6=(00 10 20 30), mm1=(40 50 60 70) 391 ; mm7=(01 11 21 31), mm3=(41 51 61 71) 392 393 movq MMWORD [wk(0)], mm4 ; wk(0)=(22 23 32 33) 394 movq MMWORD [wk(1)], mm2 ; wk(1)=(42 43 52 53) 395 396 movq mm4,mm6 ; transpose coefficients(phase 1) 397 punpcklwd mm6,mm7 ; mm6=(00 01 10 11) 398 punpckhwd mm4,mm7 ; mm4=(20 21 30 31) 399 movq mm2,mm1 ; transpose coefficients(phase 1) 400 punpcklwd mm1,mm3 ; mm1=(40 41 50 51) 401 punpckhwd mm2,mm3 ; mm2=(60 61 70 71) 402 403 movq mm7,mm6 ; transpose coefficients(phase 2) 404 punpckldq mm6,mm0 ; mm6=(00 01 02 03)=data0 405 punpckhdq mm7,mm0 ; mm7=(10 11 12 13)=data1 406 movq mm3,mm2 ; transpose coefficients(phase 2) 407 punpckldq mm2,mm5 ; mm2=(60 61 62 63)=data6 408 punpckhdq mm3,mm5 ; mm3=(70 71 72 73)=data7 409 410 movq mm0,mm7 411 movq mm5,mm6 412 psubw mm7,mm2 ; mm7=data1-data6=tmp6 413 psubw mm6,mm3 ; mm6=data0-data7=tmp7 414 paddw mm0,mm2 ; mm0=data1+data6=tmp1 415 paddw mm5,mm3 ; mm5=data0+data7=tmp0 416 417 movq mm2, MMWORD [wk(0)] ; mm2=(22 23 32 33) 418 movq mm3, MMWORD [wk(1)] ; mm3=(42 43 52 53) 419 movq MMWORD [wk(0)], mm7 ; wk(0)=tmp6 420 movq MMWORD [wk(1)], mm6 ; wk(1)=tmp7 421 422 movq mm7,mm4 ; transpose coefficients(phase 2) 423 punpckldq mm4,mm2 ; mm4=(20 21 22 23)=data2 424 punpckhdq mm7,mm2 ; mm7=(30 31 32 33)=data3 425 movq mm6,mm1 ; transpose coefficients(phase 2) 426 punpckldq mm1,mm3 ; mm1=(40 41 42 43)=data4 427 punpckhdq mm6,mm3 ; mm6=(50 51 52 53)=data5 428 429 movq mm2,mm7 430 movq mm3,mm4 431 paddw mm7,mm1 ; mm7=data3+data4=tmp3 432 paddw mm4,mm6 ; mm4=data2+data5=tmp2 433 psubw mm2,mm1 ; mm2=data3-data4=tmp4 434 psubw mm3,mm6 ; mm3=data2-data5=tmp5 435 436 ; -- Even part 437 438 movq mm1,mm5 439 movq mm6,mm0 440 paddw mm5,mm7 ; mm5=tmp10 441 paddw mm0,mm4 ; mm0=tmp11 442 psubw mm1,mm7 ; mm1=tmp13 443 psubw mm6,mm4 ; mm6=tmp12 444 445 movq mm7,mm5 446 paddw mm5,mm0 ; mm5=tmp10+tmp11 447 psubw mm7,mm0 ; mm7=tmp10-tmp11 448 449 paddw mm5,[GOTOFF(ebx,PW_DESCALE_P2X)] 450 paddw mm7,[GOTOFF(ebx,PW_DESCALE_P2X)] 451 psraw mm5,PASS1_BITS ; mm5=data0 452 psraw mm7,PASS1_BITS ; mm7=data4 453 454 movq MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)], mm5 455 movq MMWORD [MMBLOCK(4,0,edx,SIZEOF_DCTELEM)], mm7 456 457 ; (Original) 458 ; z1 = (tmp12 + tmp13) * 0.541196100; 459 ; data2 = z1 + tmp13 * 0.765366865; 460 ; data6 = z1 + tmp12 * -1.847759065; 461 ; 462 ; (This implementation) 463 ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; 464 ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); 465 466 movq mm4,mm1 ; mm1=tmp13 467 movq mm0,mm1 468 punpcklwd mm4,mm6 ; mm6=tmp12 469 punpckhwd mm0,mm6 470 movq mm1,mm4 471 movq mm6,mm0 472 pmaddwd mm4,[GOTOFF(ebx,PW_F130_F054)] ; mm4=data2L 473 pmaddwd mm0,[GOTOFF(ebx,PW_F130_F054)] ; mm0=data2H 474 pmaddwd mm1,[GOTOFF(ebx,PW_F054_MF130)] ; mm1=data6L 475 pmaddwd mm6,[GOTOFF(ebx,PW_F054_MF130)] ; mm6=data6H 476 477 paddd mm4,[GOTOFF(ebx,PD_DESCALE_P2)] 478 paddd mm0,[GOTOFF(ebx,PD_DESCALE_P2)] 479 psrad mm4,DESCALE_P2 480 psrad mm0,DESCALE_P2 481 paddd mm1,[GOTOFF(ebx,PD_DESCALE_P2)] 482 paddd mm6,[GOTOFF(ebx,PD_DESCALE_P2)] 483 psrad mm1,DESCALE_P2 484 psrad mm6,DESCALE_P2 485 486 packssdw mm4,mm0 ; mm4=data2 487 packssdw mm1,mm6 ; mm1=data6 488 489 movq MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)], mm4 490 movq MMWORD [MMBLOCK(6,0,edx,SIZEOF_DCTELEM)], mm1 491 492 ; -- Odd part 493 494 movq mm5, MMWORD [wk(0)] ; mm5=tmp6 495 movq mm7, MMWORD [wk(1)] ; mm7=tmp7 496 497 movq mm0,mm2 ; mm2=tmp4 498 movq mm6,mm3 ; mm3=tmp5 499 paddw mm0,mm5 ; mm0=z3 500 paddw mm6,mm7 ; mm6=z4 501 502 ; (Original) 503 ; z5 = (z3 + z4) * 1.175875602; 504 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; 505 ; z3 += z5; z4 += z5; 506 ; 507 ; (This implementation) 508 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; 509 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); 510 511 movq mm4,mm0 512 movq mm1,mm0 513 punpcklwd mm4,mm6 514 punpckhwd mm1,mm6 515 movq mm0,mm4 516 movq mm6,mm1 517 pmaddwd mm4,[GOTOFF(ebx,PW_MF078_F117)] ; mm4=z3L 518 pmaddwd mm1,[GOTOFF(ebx,PW_MF078_F117)] ; mm1=z3H 519 pmaddwd mm0,[GOTOFF(ebx,PW_F117_F078)] ; mm0=z4L 520 pmaddwd mm6,[GOTOFF(ebx,PW_F117_F078)] ; mm6=z4H 521 522 movq MMWORD [wk(0)], mm4 ; wk(0)=z3L 523 movq MMWORD [wk(1)], mm1 ; wk(1)=z3H 524 525 ; (Original) 526 ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; 527 ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; 528 ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; 529 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; 530 ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; 531 ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; 532 ; 533 ; (This implementation) 534 ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; 535 ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; 536 ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); 537 ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); 538 ; data7 = tmp4 + z3; data5 = tmp5 + z4; 539 ; data3 = tmp6 + z3; data1 = tmp7 + z4; 540 541 movq mm4,mm2 542 movq mm1,mm2 543 punpcklwd mm4,mm7 544 punpckhwd mm1,mm7 545 movq mm2,mm4 546 movq mm7,mm1 547 pmaddwd mm4,[GOTOFF(ebx,PW_MF060_MF089)] ; mm4=tmp4L 548 pmaddwd mm1,[GOTOFF(ebx,PW_MF060_MF089)] ; mm1=tmp4H 549 pmaddwd mm2,[GOTOFF(ebx,PW_MF089_F060)] ; mm2=tmp7L 550 pmaddwd mm7,[GOTOFF(ebx,PW_MF089_F060)] ; mm7=tmp7H 551 552 paddd mm4, MMWORD [wk(0)] ; mm4=data7L 553 paddd mm1, MMWORD [wk(1)] ; mm1=data7H 554 paddd mm2,mm0 ; mm2=data1L 555 paddd mm7,mm6 ; mm7=data1H 556 557 paddd mm4,[GOTOFF(ebx,PD_DESCALE_P2)] 558 paddd mm1,[GOTOFF(ebx,PD_DESCALE_P2)] 559 psrad mm4,DESCALE_P2 560 psrad mm1,DESCALE_P2 561 paddd mm2,[GOTOFF(ebx,PD_DESCALE_P2)] 562 paddd mm7,[GOTOFF(ebx,PD_DESCALE_P2)] 563 psrad mm2,DESCALE_P2 564 psrad mm7,DESCALE_P2 565 566 packssdw mm4,mm1 ; mm4=data7 567 packssdw mm2,mm7 ; mm2=data1 568 569 movq MMWORD [MMBLOCK(7,0,edx,SIZEOF_DCTELEM)], mm4 570 movq MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)], mm2 571 572 movq mm1,mm3 573 movq mm7,mm3 574 punpcklwd mm1,mm5 575 punpckhwd mm7,mm5 576 movq mm3,mm1 577 movq mm5,mm7 578 pmaddwd mm1,[GOTOFF(ebx,PW_MF050_MF256)] ; mm1=tmp5L 579 pmaddwd mm7,[GOTOFF(ebx,PW_MF050_MF256)] ; mm7=tmp5H 580 pmaddwd mm3,[GOTOFF(ebx,PW_MF256_F050)] ; mm3=tmp6L 581 pmaddwd mm5,[GOTOFF(ebx,PW_MF256_F050)] ; mm5=tmp6H 582 583 paddd mm1,mm0 ; mm1=data5L 584 paddd mm7,mm6 ; mm7=data5H 585 paddd mm3, MMWORD [wk(0)] ; mm3=data3L 586 paddd mm5, MMWORD [wk(1)] ; mm5=data3H 587 588 paddd mm1,[GOTOFF(ebx,PD_DESCALE_P2)] 589 paddd mm7,[GOTOFF(ebx,PD_DESCALE_P2)] 590 psrad mm1,DESCALE_P2 591 psrad mm7,DESCALE_P2 592 paddd mm3,[GOTOFF(ebx,PD_DESCALE_P2)] 593 paddd mm5,[GOTOFF(ebx,PD_DESCALE_P2)] 594 psrad mm3,DESCALE_P2 595 psrad mm5,DESCALE_P2 596 597 packssdw mm1,mm7 ; mm1=data5 598 packssdw mm3,mm5 ; mm3=data3 599 600 movq MMWORD [MMBLOCK(5,0,edx,SIZEOF_DCTELEM)], mm1 601 movq MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)], mm3 602 603 add edx, byte 4*SIZEOF_DCTELEM 604 dec ecx 605 jnz near .columnloop 606 607 emms ; empty MMX state 608 609; pop edi ; unused 610; pop esi ; unused 611; pop edx ; need not be preserved 612; pop ecx ; need not be preserved 613 poppic ebx 614 mov esp,ebp ; esp <- aligned ebp 615 pop esp ; esp <- original ebp 616 pop ebp 617 ret 618 619; For some reason, the OS X linker does not honor the request to align the 620; segment unless we do this. 621 align 16 622