1; 2; jfdctint.asm - accurate integer FDCT (SSE2) 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_sse2) 69 70EXTN(jconst_fdct_islow_sse2): 71 72PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541 73PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847) 74PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175 75PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390) 76PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899 77PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899) 78PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562 79PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562) 80PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1) 81PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1) 82PW_DESCALE_P2X times 8 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_sse2 (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_XMMWORD ; xmmword wk[WK_NUM] 100%define WK_NUM 6 101 102 align 16 103 global EXTN(jsimd_fdct_islow_sse2) 104 105EXTN(jsimd_fdct_islow_sse2): 106 push ebp 107 mov eax,esp ; eax = original ebp 108 sub esp, byte 4 109 and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits 110 mov [esp],eax 111 mov ebp,esp ; ebp = aligned ebp 112 lea esp, [wk(0)] 113 pushpic ebx 114; push ecx ; unused 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 *) 124 125 movdqa xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)] 126 movdqa xmm1, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)] 127 movdqa xmm2, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)] 128 movdqa xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)] 129 130 ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27) 131 ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37) 132 133 movdqa xmm4,xmm0 ; transpose coefficients(phase 1) 134 punpcklwd xmm0,xmm1 ; xmm0=(00 10 01 11 02 12 03 13) 135 punpckhwd xmm4,xmm1 ; xmm4=(04 14 05 15 06 16 07 17) 136 movdqa xmm5,xmm2 ; transpose coefficients(phase 1) 137 punpcklwd xmm2,xmm3 ; xmm2=(20 30 21 31 22 32 23 33) 138 punpckhwd xmm5,xmm3 ; xmm5=(24 34 25 35 26 36 27 37) 139 140 movdqa xmm6, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)] 141 movdqa xmm7, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)] 142 movdqa xmm1, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)] 143 movdqa xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)] 144 145 ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62) 146 ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63) 147 148 movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33) 149 movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37) 150 151 movdqa xmm2,xmm6 ; transpose coefficients(phase 1) 152 punpcklwd xmm6,xmm7 ; xmm6=(40 50 41 51 42 52 43 53) 153 punpckhwd xmm2,xmm7 ; xmm2=(44 54 45 55 46 56 47 57) 154 movdqa xmm5,xmm1 ; transpose coefficients(phase 1) 155 punpcklwd xmm1,xmm3 ; xmm1=(60 70 61 71 62 72 63 73) 156 punpckhwd xmm5,xmm3 ; xmm5=(64 74 65 75 66 76 67 77) 157 158 movdqa xmm7,xmm6 ; transpose coefficients(phase 2) 159 punpckldq xmm6,xmm1 ; xmm6=(40 50 60 70 41 51 61 71) 160 punpckhdq xmm7,xmm1 ; xmm7=(42 52 62 72 43 53 63 73) 161 movdqa xmm3,xmm2 ; transpose coefficients(phase 2) 162 punpckldq xmm2,xmm5 ; xmm2=(44 54 64 74 45 55 65 75) 163 punpckhdq xmm3,xmm5 ; xmm3=(46 56 66 76 47 57 67 77) 164 165 movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33) 166 movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37) 167 movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=(42 52 62 72 43 53 63 73) 168 movdqa XMMWORD [wk(3)], xmm2 ; wk(3)=(44 54 64 74 45 55 65 75) 169 170 movdqa xmm7,xmm0 ; transpose coefficients(phase 2) 171 punpckldq xmm0,xmm1 ; xmm0=(00 10 20 30 01 11 21 31) 172 punpckhdq xmm7,xmm1 ; xmm7=(02 12 22 32 03 13 23 33) 173 movdqa xmm2,xmm4 ; transpose coefficients(phase 2) 174 punpckldq xmm4,xmm5 ; xmm4=(04 14 24 34 05 15 25 35) 175 punpckhdq xmm2,xmm5 ; xmm2=(06 16 26 36 07 17 27 37) 176 177 movdqa xmm1,xmm0 ; transpose coefficients(phase 3) 178 punpcklqdq xmm0,xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0 179 punpckhqdq xmm1,xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1 180 movdqa xmm5,xmm2 ; transpose coefficients(phase 3) 181 punpcklqdq xmm2,xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6 182 punpckhqdq xmm5,xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7 183 184 movdqa xmm6,xmm1 185 movdqa xmm3,xmm0 186 psubw xmm1,xmm2 ; xmm1=data1-data6=tmp6 187 psubw xmm0,xmm5 ; xmm0=data0-data7=tmp7 188 paddw xmm6,xmm2 ; xmm6=data1+data6=tmp1 189 paddw xmm3,xmm5 ; xmm3=data0+data7=tmp0 190 191 movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(42 52 62 72 43 53 63 73) 192 movdqa xmm5, XMMWORD [wk(3)] ; xmm5=(44 54 64 74 45 55 65 75) 193 movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6 194 movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7 195 196 movdqa xmm1,xmm7 ; transpose coefficients(phase 3) 197 punpcklqdq xmm7,xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2 198 punpckhqdq xmm1,xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3 199 movdqa xmm0,xmm4 ; transpose coefficients(phase 3) 200 punpcklqdq xmm4,xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4 201 punpckhqdq xmm0,xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5 202 203 movdqa xmm2,xmm1 204 movdqa xmm5,xmm7 205 paddw xmm1,xmm4 ; xmm1=data3+data4=tmp3 206 paddw xmm7,xmm0 ; xmm7=data2+data5=tmp2 207 psubw xmm2,xmm4 ; xmm2=data3-data4=tmp4 208 psubw xmm5,xmm0 ; xmm5=data2-data5=tmp5 209 210 ; -- Even part 211 212 movdqa xmm4,xmm3 213 movdqa xmm0,xmm6 214 paddw xmm3,xmm1 ; xmm3=tmp10 215 paddw xmm6,xmm7 ; xmm6=tmp11 216 psubw xmm4,xmm1 ; xmm4=tmp13 217 psubw xmm0,xmm7 ; xmm0=tmp12 218 219 movdqa xmm1,xmm3 220 paddw xmm3,xmm6 ; xmm3=tmp10+tmp11 221 psubw xmm1,xmm6 ; xmm1=tmp10-tmp11 222 223 psllw xmm3,PASS1_BITS ; xmm3=data0 224 psllw xmm1,PASS1_BITS ; xmm1=data4 225 226 movdqa XMMWORD [wk(2)], xmm3 ; wk(2)=data0 227 movdqa XMMWORD [wk(3)], xmm1 ; wk(3)=data4 228 229 ; (Original) 230 ; z1 = (tmp12 + tmp13) * 0.541196100; 231 ; data2 = z1 + tmp13 * 0.765366865; 232 ; data6 = z1 + tmp12 * -1.847759065; 233 ; 234 ; (This implementation) 235 ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; 236 ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); 237 238 movdqa xmm7,xmm4 ; xmm4=tmp13 239 movdqa xmm6,xmm4 240 punpcklwd xmm7,xmm0 ; xmm0=tmp12 241 punpckhwd xmm6,xmm0 242 movdqa xmm4,xmm7 243 movdqa xmm0,xmm6 244 pmaddwd xmm7,[GOTOFF(ebx,PW_F130_F054)] ; xmm7=data2L 245 pmaddwd xmm6,[GOTOFF(ebx,PW_F130_F054)] ; xmm6=data2H 246 pmaddwd xmm4,[GOTOFF(ebx,PW_F054_MF130)] ; xmm4=data6L 247 pmaddwd xmm0,[GOTOFF(ebx,PW_F054_MF130)] ; xmm0=data6H 248 249 paddd xmm7,[GOTOFF(ebx,PD_DESCALE_P1)] 250 paddd xmm6,[GOTOFF(ebx,PD_DESCALE_P1)] 251 psrad xmm7,DESCALE_P1 252 psrad xmm6,DESCALE_P1 253 paddd xmm4,[GOTOFF(ebx,PD_DESCALE_P1)] 254 paddd xmm0,[GOTOFF(ebx,PD_DESCALE_P1)] 255 psrad xmm4,DESCALE_P1 256 psrad xmm0,DESCALE_P1 257 258 packssdw xmm7,xmm6 ; xmm7=data2 259 packssdw xmm4,xmm0 ; xmm4=data6 260 261 movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=data2 262 movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=data6 263 264 ; -- Odd part 265 266 movdqa xmm3, XMMWORD [wk(0)] ; xmm3=tmp6 267 movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp7 268 269 movdqa xmm6,xmm2 ; xmm2=tmp4 270 movdqa xmm0,xmm5 ; xmm5=tmp5 271 paddw xmm6,xmm3 ; xmm6=z3 272 paddw xmm0,xmm1 ; xmm0=z4 273 274 ; (Original) 275 ; z5 = (z3 + z4) * 1.175875602; 276 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; 277 ; z3 += z5; z4 += z5; 278 ; 279 ; (This implementation) 280 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; 281 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); 282 283 movdqa xmm7,xmm6 284 movdqa xmm4,xmm6 285 punpcklwd xmm7,xmm0 286 punpckhwd xmm4,xmm0 287 movdqa xmm6,xmm7 288 movdqa xmm0,xmm4 289 pmaddwd xmm7,[GOTOFF(ebx,PW_MF078_F117)] ; xmm7=z3L 290 pmaddwd xmm4,[GOTOFF(ebx,PW_MF078_F117)] ; xmm4=z3H 291 pmaddwd xmm6,[GOTOFF(ebx,PW_F117_F078)] ; xmm6=z4L 292 pmaddwd xmm0,[GOTOFF(ebx,PW_F117_F078)] ; xmm0=z4H 293 294 movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=z3L 295 movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=z3H 296 297 ; (Original) 298 ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; 299 ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; 300 ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; 301 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; 302 ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; 303 ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; 304 ; 305 ; (This implementation) 306 ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; 307 ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; 308 ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); 309 ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); 310 ; data7 = tmp4 + z3; data5 = tmp5 + z4; 311 ; data3 = tmp6 + z3; data1 = tmp7 + z4; 312 313 movdqa xmm7,xmm2 314 movdqa xmm4,xmm2 315 punpcklwd xmm7,xmm1 316 punpckhwd xmm4,xmm1 317 movdqa xmm2,xmm7 318 movdqa xmm1,xmm4 319 pmaddwd xmm7,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm7=tmp4L 320 pmaddwd xmm4,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm4=tmp4H 321 pmaddwd xmm2,[GOTOFF(ebx,PW_MF089_F060)] ; xmm2=tmp7L 322 pmaddwd xmm1,[GOTOFF(ebx,PW_MF089_F060)] ; xmm1=tmp7H 323 324 paddd xmm7, XMMWORD [wk(0)] ; xmm7=data7L 325 paddd xmm4, XMMWORD [wk(1)] ; xmm4=data7H 326 paddd xmm2,xmm6 ; xmm2=data1L 327 paddd xmm1,xmm0 ; xmm1=data1H 328 329 paddd xmm7,[GOTOFF(ebx,PD_DESCALE_P1)] 330 paddd xmm4,[GOTOFF(ebx,PD_DESCALE_P1)] 331 psrad xmm7,DESCALE_P1 332 psrad xmm4,DESCALE_P1 333 paddd xmm2,[GOTOFF(ebx,PD_DESCALE_P1)] 334 paddd xmm1,[GOTOFF(ebx,PD_DESCALE_P1)] 335 psrad xmm2,DESCALE_P1 336 psrad xmm1,DESCALE_P1 337 338 packssdw xmm7,xmm4 ; xmm7=data7 339 packssdw xmm2,xmm1 ; xmm2=data1 340 341 movdqa xmm4,xmm5 342 movdqa xmm1,xmm5 343 punpcklwd xmm4,xmm3 344 punpckhwd xmm1,xmm3 345 movdqa xmm5,xmm4 346 movdqa xmm3,xmm1 347 pmaddwd xmm4,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm4=tmp5L 348 pmaddwd xmm1,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm1=tmp5H 349 pmaddwd xmm5,[GOTOFF(ebx,PW_MF256_F050)] ; xmm5=tmp6L 350 pmaddwd xmm3,[GOTOFF(ebx,PW_MF256_F050)] ; xmm3=tmp6H 351 352 paddd xmm4,xmm6 ; xmm4=data5L 353 paddd xmm1,xmm0 ; xmm1=data5H 354 paddd xmm5, XMMWORD [wk(0)] ; xmm5=data3L 355 paddd xmm3, XMMWORD [wk(1)] ; xmm3=data3H 356 357 paddd xmm4,[GOTOFF(ebx,PD_DESCALE_P1)] 358 paddd xmm1,[GOTOFF(ebx,PD_DESCALE_P1)] 359 psrad xmm4,DESCALE_P1 360 psrad xmm1,DESCALE_P1 361 paddd xmm5,[GOTOFF(ebx,PD_DESCALE_P1)] 362 paddd xmm3,[GOTOFF(ebx,PD_DESCALE_P1)] 363 psrad xmm5,DESCALE_P1 364 psrad xmm3,DESCALE_P1 365 366 packssdw xmm4,xmm1 ; xmm4=data5 367 packssdw xmm5,xmm3 ; xmm5=data3 368 369 ; ---- Pass 2: process columns. 370 371; mov edx, POINTER [data(eax)] ; (DCTELEM *) 372 373 movdqa xmm6, XMMWORD [wk(2)] ; xmm6=col0 374 movdqa xmm0, XMMWORD [wk(4)] ; xmm0=col2 375 376 ; xmm6=(00 10 20 30 40 50 60 70), xmm0=(02 12 22 32 42 52 62 72) 377 ; xmm2=(01 11 21 31 41 51 61 71), xmm5=(03 13 23 33 43 53 63 73) 378 379 movdqa xmm1,xmm6 ; transpose coefficients(phase 1) 380 punpcklwd xmm6,xmm2 ; xmm6=(00 01 10 11 20 21 30 31) 381 punpckhwd xmm1,xmm2 ; xmm1=(40 41 50 51 60 61 70 71) 382 movdqa xmm3,xmm0 ; transpose coefficients(phase 1) 383 punpcklwd xmm0,xmm5 ; xmm0=(02 03 12 13 22 23 32 33) 384 punpckhwd xmm3,xmm5 ; xmm3=(42 43 52 53 62 63 72 73) 385 386 movdqa xmm2, XMMWORD [wk(3)] ; xmm2=col4 387 movdqa xmm5, XMMWORD [wk(5)] ; xmm5=col6 388 389 ; xmm2=(04 14 24 34 44 54 64 74), xmm5=(06 16 26 36 46 56 66 76) 390 ; xmm4=(05 15 25 35 45 55 65 75), xmm7=(07 17 27 37 47 57 67 77) 391 392 movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=(02 03 12 13 22 23 32 33) 393 movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(42 43 52 53 62 63 72 73) 394 395 movdqa xmm0,xmm2 ; transpose coefficients(phase 1) 396 punpcklwd xmm2,xmm4 ; xmm2=(04 05 14 15 24 25 34 35) 397 punpckhwd xmm0,xmm4 ; xmm0=(44 45 54 55 64 65 74 75) 398 movdqa xmm3,xmm5 ; transpose coefficients(phase 1) 399 punpcklwd xmm5,xmm7 ; xmm5=(06 07 16 17 26 27 36 37) 400 punpckhwd xmm3,xmm7 ; xmm3=(46 47 56 57 66 67 76 77) 401 402 movdqa xmm4,xmm2 ; transpose coefficients(phase 2) 403 punpckldq xmm2,xmm5 ; xmm2=(04 05 06 07 14 15 16 17) 404 punpckhdq xmm4,xmm5 ; xmm4=(24 25 26 27 34 35 36 37) 405 movdqa xmm7,xmm0 ; transpose coefficients(phase 2) 406 punpckldq xmm0,xmm3 ; xmm0=(44 45 46 47 54 55 56 57) 407 punpckhdq xmm7,xmm3 ; xmm7=(64 65 66 67 74 75 76 77) 408 409 movdqa xmm5, XMMWORD [wk(0)] ; xmm5=(02 03 12 13 22 23 32 33) 410 movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53 62 63 72 73) 411 movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=(24 25 26 27 34 35 36 37) 412 movdqa XMMWORD [wk(3)], xmm0 ; wk(3)=(44 45 46 47 54 55 56 57) 413 414 movdqa xmm4,xmm6 ; transpose coefficients(phase 2) 415 punpckldq xmm6,xmm5 ; xmm6=(00 01 02 03 10 11 12 13) 416 punpckhdq xmm4,xmm5 ; xmm4=(20 21 22 23 30 31 32 33) 417 movdqa xmm0,xmm1 ; transpose coefficients(phase 2) 418 punpckldq xmm1,xmm3 ; xmm1=(40 41 42 43 50 51 52 53) 419 punpckhdq xmm0,xmm3 ; xmm0=(60 61 62 63 70 71 72 73) 420 421 movdqa xmm5,xmm6 ; transpose coefficients(phase 3) 422 punpcklqdq xmm6,xmm2 ; xmm6=(00 01 02 03 04 05 06 07)=data0 423 punpckhqdq xmm5,xmm2 ; xmm5=(10 11 12 13 14 15 16 17)=data1 424 movdqa xmm3,xmm0 ; transpose coefficients(phase 3) 425 punpcklqdq xmm0,xmm7 ; xmm0=(60 61 62 63 64 65 66 67)=data6 426 punpckhqdq xmm3,xmm7 ; xmm3=(70 71 72 73 74 75 76 77)=data7 427 428 movdqa xmm2,xmm5 429 movdqa xmm7,xmm6 430 psubw xmm5,xmm0 ; xmm5=data1-data6=tmp6 431 psubw xmm6,xmm3 ; xmm6=data0-data7=tmp7 432 paddw xmm2,xmm0 ; xmm2=data1+data6=tmp1 433 paddw xmm7,xmm3 ; xmm7=data0+data7=tmp0 434 435 movdqa xmm0, XMMWORD [wk(2)] ; xmm0=(24 25 26 27 34 35 36 37) 436 movdqa xmm3, XMMWORD [wk(3)] ; xmm3=(44 45 46 47 54 55 56 57) 437 movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=tmp6 438 movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7 439 440 movdqa xmm5,xmm4 ; transpose coefficients(phase 3) 441 punpcklqdq xmm4,xmm0 ; xmm4=(20 21 22 23 24 25 26 27)=data2 442 punpckhqdq xmm5,xmm0 ; xmm5=(30 31 32 33 34 35 36 37)=data3 443 movdqa xmm6,xmm1 ; transpose coefficients(phase 3) 444 punpcklqdq xmm1,xmm3 ; xmm1=(40 41 42 43 44 45 46 47)=data4 445 punpckhqdq xmm6,xmm3 ; xmm6=(50 51 52 53 54 55 56 57)=data5 446 447 movdqa xmm0,xmm5 448 movdqa xmm3,xmm4 449 paddw xmm5,xmm1 ; xmm5=data3+data4=tmp3 450 paddw xmm4,xmm6 ; xmm4=data2+data5=tmp2 451 psubw xmm0,xmm1 ; xmm0=data3-data4=tmp4 452 psubw xmm3,xmm6 ; xmm3=data2-data5=tmp5 453 454 ; -- Even part 455 456 movdqa xmm1,xmm7 457 movdqa xmm6,xmm2 458 paddw xmm7,xmm5 ; xmm7=tmp10 459 paddw xmm2,xmm4 ; xmm2=tmp11 460 psubw xmm1,xmm5 ; xmm1=tmp13 461 psubw xmm6,xmm4 ; xmm6=tmp12 462 463 movdqa xmm5,xmm7 464 paddw xmm7,xmm2 ; xmm7=tmp10+tmp11 465 psubw xmm5,xmm2 ; xmm5=tmp10-tmp11 466 467 paddw xmm7,[GOTOFF(ebx,PW_DESCALE_P2X)] 468 paddw xmm5,[GOTOFF(ebx,PW_DESCALE_P2X)] 469 psraw xmm7,PASS1_BITS ; xmm7=data0 470 psraw xmm5,PASS1_BITS ; xmm5=data4 471 472 movdqa XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)], xmm7 473 movdqa XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)], xmm5 474 475 ; (Original) 476 ; z1 = (tmp12 + tmp13) * 0.541196100; 477 ; data2 = z1 + tmp13 * 0.765366865; 478 ; data6 = z1 + tmp12 * -1.847759065; 479 ; 480 ; (This implementation) 481 ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; 482 ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); 483 484 movdqa xmm4,xmm1 ; xmm1=tmp13 485 movdqa xmm2,xmm1 486 punpcklwd xmm4,xmm6 ; xmm6=tmp12 487 punpckhwd xmm2,xmm6 488 movdqa xmm1,xmm4 489 movdqa xmm6,xmm2 490 pmaddwd xmm4,[GOTOFF(ebx,PW_F130_F054)] ; xmm4=data2L 491 pmaddwd xmm2,[GOTOFF(ebx,PW_F130_F054)] ; xmm2=data2H 492 pmaddwd xmm1,[GOTOFF(ebx,PW_F054_MF130)] ; xmm1=data6L 493 pmaddwd xmm6,[GOTOFF(ebx,PW_F054_MF130)] ; xmm6=data6H 494 495 paddd xmm4,[GOTOFF(ebx,PD_DESCALE_P2)] 496 paddd xmm2,[GOTOFF(ebx,PD_DESCALE_P2)] 497 psrad xmm4,DESCALE_P2 498 psrad xmm2,DESCALE_P2 499 paddd xmm1,[GOTOFF(ebx,PD_DESCALE_P2)] 500 paddd xmm6,[GOTOFF(ebx,PD_DESCALE_P2)] 501 psrad xmm1,DESCALE_P2 502 psrad xmm6,DESCALE_P2 503 504 packssdw xmm4,xmm2 ; xmm4=data2 505 packssdw xmm1,xmm6 ; xmm1=data6 506 507 movdqa XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)], xmm4 508 movdqa XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)], xmm1 509 510 ; -- Odd part 511 512 movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp6 513 movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7 514 515 movdqa xmm2,xmm0 ; xmm0=tmp4 516 movdqa xmm6,xmm3 ; xmm3=tmp5 517 paddw xmm2,xmm7 ; xmm2=z3 518 paddw xmm6,xmm5 ; xmm6=z4 519 520 ; (Original) 521 ; z5 = (z3 + z4) * 1.175875602; 522 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; 523 ; z3 += z5; z4 += z5; 524 ; 525 ; (This implementation) 526 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; 527 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); 528 529 movdqa xmm4,xmm2 530 movdqa xmm1,xmm2 531 punpcklwd xmm4,xmm6 532 punpckhwd xmm1,xmm6 533 movdqa xmm2,xmm4 534 movdqa xmm6,xmm1 535 pmaddwd xmm4,[GOTOFF(ebx,PW_MF078_F117)] ; xmm4=z3L 536 pmaddwd xmm1,[GOTOFF(ebx,PW_MF078_F117)] ; xmm1=z3H 537 pmaddwd xmm2,[GOTOFF(ebx,PW_F117_F078)] ; xmm2=z4L 538 pmaddwd xmm6,[GOTOFF(ebx,PW_F117_F078)] ; xmm6=z4H 539 540 movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=z3L 541 movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=z3H 542 543 ; (Original) 544 ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; 545 ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; 546 ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; 547 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; 548 ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; 549 ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; 550 ; 551 ; (This implementation) 552 ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; 553 ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; 554 ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); 555 ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); 556 ; data7 = tmp4 + z3; data5 = tmp5 + z4; 557 ; data3 = tmp6 + z3; data1 = tmp7 + z4; 558 559 movdqa xmm4,xmm0 560 movdqa xmm1,xmm0 561 punpcklwd xmm4,xmm5 562 punpckhwd xmm1,xmm5 563 movdqa xmm0,xmm4 564 movdqa xmm5,xmm1 565 pmaddwd xmm4,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm4=tmp4L 566 pmaddwd xmm1,[GOTOFF(ebx,PW_MF060_MF089)] ; xmm1=tmp4H 567 pmaddwd xmm0,[GOTOFF(ebx,PW_MF089_F060)] ; xmm0=tmp7L 568 pmaddwd xmm5,[GOTOFF(ebx,PW_MF089_F060)] ; xmm5=tmp7H 569 570 paddd xmm4, XMMWORD [wk(0)] ; xmm4=data7L 571 paddd xmm1, XMMWORD [wk(1)] ; xmm1=data7H 572 paddd xmm0,xmm2 ; xmm0=data1L 573 paddd xmm5,xmm6 ; xmm5=data1H 574 575 paddd xmm4,[GOTOFF(ebx,PD_DESCALE_P2)] 576 paddd xmm1,[GOTOFF(ebx,PD_DESCALE_P2)] 577 psrad xmm4,DESCALE_P2 578 psrad xmm1,DESCALE_P2 579 paddd xmm0,[GOTOFF(ebx,PD_DESCALE_P2)] 580 paddd xmm5,[GOTOFF(ebx,PD_DESCALE_P2)] 581 psrad xmm0,DESCALE_P2 582 psrad xmm5,DESCALE_P2 583 584 packssdw xmm4,xmm1 ; xmm4=data7 585 packssdw xmm0,xmm5 ; xmm0=data1 586 587 movdqa XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)], xmm4 588 movdqa XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)], xmm0 589 590 movdqa xmm1,xmm3 591 movdqa xmm5,xmm3 592 punpcklwd xmm1,xmm7 593 punpckhwd xmm5,xmm7 594 movdqa xmm3,xmm1 595 movdqa xmm7,xmm5 596 pmaddwd xmm1,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm1=tmp5L 597 pmaddwd xmm5,[GOTOFF(ebx,PW_MF050_MF256)] ; xmm5=tmp5H 598 pmaddwd xmm3,[GOTOFF(ebx,PW_MF256_F050)] ; xmm3=tmp6L 599 pmaddwd xmm7,[GOTOFF(ebx,PW_MF256_F050)] ; xmm7=tmp6H 600 601 paddd xmm1,xmm2 ; xmm1=data5L 602 paddd xmm5,xmm6 ; xmm5=data5H 603 paddd xmm3, XMMWORD [wk(0)] ; xmm3=data3L 604 paddd xmm7, XMMWORD [wk(1)] ; xmm7=data3H 605 606 paddd xmm1,[GOTOFF(ebx,PD_DESCALE_P2)] 607 paddd xmm5,[GOTOFF(ebx,PD_DESCALE_P2)] 608 psrad xmm1,DESCALE_P2 609 psrad xmm5,DESCALE_P2 610 paddd xmm3,[GOTOFF(ebx,PD_DESCALE_P2)] 611 paddd xmm7,[GOTOFF(ebx,PD_DESCALE_P2)] 612 psrad xmm3,DESCALE_P2 613 psrad xmm7,DESCALE_P2 614 615 packssdw xmm1,xmm5 ; xmm1=data5 616 packssdw xmm3,xmm7 ; xmm3=data3 617 618 movdqa XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)], xmm1 619 movdqa XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)], xmm3 620 621; pop edi ; unused 622; pop esi ; unused 623; pop edx ; need not be preserved 624; pop ecx ; unused 625 poppic ebx 626 mov esp,ebp ; esp <- aligned ebp 627 pop esp ; esp <- original ebp 628 pop ebp 629 ret 630 631; For some reason, the OS X linker does not honor the request to align the 632; segment unless we do this. 633 align 16 634