1; 2; jfdctint.asm - accurate integer FDCT (SSE2) 3; 4; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB 5; Copyright (C) 2016, D. R. Commander. 6; 7; Based on the x86 SIMD extension for IJG JPEG library 8; Copyright (C) 1999-2006, MIYASAKA Masaru. 9; For conditions of distribution and use, see copyright notice in jsimdext.inc 10; 11; This file should be assembled with NASM (Netwide Assembler), 12; can *not* be assembled with Microsoft's MASM or any compatible 13; assembler (including Borland's Turbo Assembler). 14; NASM is available from http://nasm.sourceforge.net/ or 15; http://sourceforge.net/project/showfiles.php?group_id=6208 16; 17; This file contains a slow-but-accurate integer implementation of the 18; forward DCT (Discrete Cosine Transform). The following code is based 19; directly on the IJG's original jfdctint.c; see the jfdctint.c for 20; more details. 21; 22; [TAB8] 23 24%include "jsimdext.inc" 25%include "jdct.inc" 26 27; -------------------------------------------------------------------------- 28 29%define CONST_BITS 13 30%define PASS1_BITS 2 31 32%define DESCALE_P1 (CONST_BITS - PASS1_BITS) 33%define DESCALE_P2 (CONST_BITS + PASS1_BITS) 34 35%if CONST_BITS == 13 36F_0_298 equ 2446 ; FIX(0.298631336) 37F_0_390 equ 3196 ; FIX(0.390180644) 38F_0_541 equ 4433 ; FIX(0.541196100) 39F_0_765 equ 6270 ; FIX(0.765366865) 40F_0_899 equ 7373 ; FIX(0.899976223) 41F_1_175 equ 9633 ; FIX(1.175875602) 42F_1_501 equ 12299 ; FIX(1.501321110) 43F_1_847 equ 15137 ; FIX(1.847759065) 44F_1_961 equ 16069 ; FIX(1.961570560) 45F_2_053 equ 16819 ; FIX(2.053119869) 46F_2_562 equ 20995 ; FIX(2.562915447) 47F_3_072 equ 25172 ; FIX(3.072711026) 48%else 49; NASM cannot do compile-time arithmetic on floating-point constants. 50%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) 51F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) 52F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) 53F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) 54F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) 55F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) 56F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) 57F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) 58F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) 59F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) 60F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) 61F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) 62F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) 63%endif 64 65; -------------------------------------------------------------------------- 66 SECTION SEG_CONST 67 68 alignz 32 69 GLOBAL_DATA(jconst_fdct_islow_sse2) 70 71EXTN(jconst_fdct_islow_sse2): 72 73PW_F130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541 74PW_F054_MF130 times 4 dw F_0_541, (F_0_541 - F_1_847) 75PW_MF078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175 76PW_F117_F078 times 4 dw F_1_175, (F_1_175 - F_0_390) 77PW_MF060_MF089 times 4 dw (F_0_298 - F_0_899), -F_0_899 78PW_MF089_F060 times 4 dw -F_0_899, (F_1_501 - F_0_899) 79PW_MF050_MF256 times 4 dw (F_2_053 - F_2_562), -F_2_562 80PW_MF256_F050 times 4 dw -F_2_562, (F_3_072 - F_2_562) 81PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1 - 1) 82PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2 - 1) 83PW_DESCALE_P2X times 8 dw 1 << (PASS1_BITS - 1) 84 85 alignz 32 86 87; -------------------------------------------------------------------------- 88 SECTION SEG_TEXT 89 BITS 32 90; 91; Perform the forward DCT on one block of samples. 92; 93; GLOBAL(void) 94; jsimd_fdct_islow_sse2(DCTELEM *data) 95; 96 97%define data(b) (b) + 8 ; DCTELEM *data 98 99%define original_ebp ebp + 0 100%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD 101 ; xmmword wk[WK_NUM] 102%define WK_NUM 6 103 104 align 32 105 GLOBAL_FUNCTION(jsimd_fdct_islow_sse2) 106 107EXTN(jsimd_fdct_islow_sse2): 108 push ebp 109 mov eax, esp ; eax = original ebp 110 sub esp, byte 4 111 and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits 112 mov [esp], eax 113 mov ebp, esp ; ebp = aligned ebp 114 lea esp, [wk(0)] 115 pushpic ebx 116; push ecx ; unused 117; push edx ; need not be preserved 118; push esi ; unused 119; push edi ; unused 120 121 get_GOT ebx ; get GOT address 122 123 ; ---- Pass 1: process rows. 124 125 mov edx, POINTER [data(eax)] ; (DCTELEM *) 126 127 movdqa xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)] 128 movdqa xmm1, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)] 129 movdqa xmm2, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)] 130 movdqa xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)] 131 132 ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27) 133 ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37) 134 135 movdqa xmm4, xmm0 ; transpose coefficients(phase 1) 136 punpcklwd xmm0, xmm1 ; xmm0=(00 10 01 11 02 12 03 13) 137 punpckhwd xmm4, xmm1 ; xmm4=(04 14 05 15 06 16 07 17) 138 movdqa xmm5, xmm2 ; transpose coefficients(phase 1) 139 punpcklwd xmm2, xmm3 ; xmm2=(20 30 21 31 22 32 23 33) 140 punpckhwd xmm5, xmm3 ; xmm5=(24 34 25 35 26 36 27 37) 141 142 movdqa xmm6, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)] 143 movdqa xmm7, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)] 144 movdqa xmm1, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)] 145 movdqa xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)] 146 147 ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62) 148 ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63) 149 150 movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33) 151 movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37) 152 153 movdqa xmm2, xmm6 ; transpose coefficients(phase 1) 154 punpcklwd xmm6, xmm7 ; xmm6=(40 50 41 51 42 52 43 53) 155 punpckhwd xmm2, xmm7 ; xmm2=(44 54 45 55 46 56 47 57) 156 movdqa xmm5, xmm1 ; transpose coefficients(phase 1) 157 punpcklwd xmm1, xmm3 ; xmm1=(60 70 61 71 62 72 63 73) 158 punpckhwd xmm5, xmm3 ; xmm5=(64 74 65 75 66 76 67 77) 159 160 movdqa xmm7, xmm6 ; transpose coefficients(phase 2) 161 punpckldq xmm6, xmm1 ; xmm6=(40 50 60 70 41 51 61 71) 162 punpckhdq xmm7, xmm1 ; xmm7=(42 52 62 72 43 53 63 73) 163 movdqa xmm3, xmm2 ; transpose coefficients(phase 2) 164 punpckldq xmm2, xmm5 ; xmm2=(44 54 64 74 45 55 65 75) 165 punpckhdq xmm3, xmm5 ; xmm3=(46 56 66 76 47 57 67 77) 166 167 movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33) 168 movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37) 169 movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=(42 52 62 72 43 53 63 73) 170 movdqa XMMWORD [wk(3)], xmm2 ; wk(3)=(44 54 64 74 45 55 65 75) 171 172 movdqa xmm7, xmm0 ; transpose coefficients(phase 2) 173 punpckldq xmm0, xmm1 ; xmm0=(00 10 20 30 01 11 21 31) 174 punpckhdq xmm7, xmm1 ; xmm7=(02 12 22 32 03 13 23 33) 175 movdqa xmm2, xmm4 ; transpose coefficients(phase 2) 176 punpckldq xmm4, xmm5 ; xmm4=(04 14 24 34 05 15 25 35) 177 punpckhdq xmm2, xmm5 ; xmm2=(06 16 26 36 07 17 27 37) 178 179 movdqa xmm1, xmm0 ; transpose coefficients(phase 3) 180 punpcklqdq xmm0, xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0 181 punpckhqdq xmm1, xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1 182 movdqa xmm5, xmm2 ; transpose coefficients(phase 3) 183 punpcklqdq xmm2, xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6 184 punpckhqdq xmm5, xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7 185 186 movdqa xmm6, xmm1 187 movdqa xmm3, xmm0 188 psubw xmm1, xmm2 ; xmm1=data1-data6=tmp6 189 psubw xmm0, xmm5 ; xmm0=data0-data7=tmp7 190 paddw xmm6, xmm2 ; xmm6=data1+data6=tmp1 191 paddw xmm3, xmm5 ; xmm3=data0+data7=tmp0 192 193 movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(42 52 62 72 43 53 63 73) 194 movdqa xmm5, XMMWORD [wk(3)] ; xmm5=(44 54 64 74 45 55 65 75) 195 movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6 196 movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7 197 198 movdqa xmm1, xmm7 ; transpose coefficients(phase 3) 199 punpcklqdq xmm7, xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2 200 punpckhqdq xmm1, xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3 201 movdqa xmm0, xmm4 ; transpose coefficients(phase 3) 202 punpcklqdq xmm4, xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4 203 punpckhqdq xmm0, xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5 204 205 movdqa xmm2, xmm1 206 movdqa xmm5, xmm7 207 paddw xmm1, xmm4 ; xmm1=data3+data4=tmp3 208 paddw xmm7, xmm0 ; xmm7=data2+data5=tmp2 209 psubw xmm2, xmm4 ; xmm2=data3-data4=tmp4 210 psubw xmm5, xmm0 ; xmm5=data2-data5=tmp5 211 212 ; -- Even part 213 214 movdqa xmm4, xmm3 215 movdqa xmm0, xmm6 216 paddw xmm3, xmm1 ; xmm3=tmp10 217 paddw xmm6, xmm7 ; xmm6=tmp11 218 psubw xmm4, xmm1 ; xmm4=tmp13 219 psubw xmm0, xmm7 ; xmm0=tmp12 220 221 movdqa xmm1, xmm3 222 paddw xmm3, xmm6 ; xmm3=tmp10+tmp11 223 psubw xmm1, xmm6 ; xmm1=tmp10-tmp11 224 225 psllw xmm3, PASS1_BITS ; xmm3=data0 226 psllw xmm1, PASS1_BITS ; xmm1=data4 227 228 movdqa XMMWORD [wk(2)], xmm3 ; wk(2)=data0 229 movdqa XMMWORD [wk(3)], xmm1 ; wk(3)=data4 230 231 ; (Original) 232 ; z1 = (tmp12 + tmp13) * 0.541196100; 233 ; data2 = z1 + tmp13 * 0.765366865; 234 ; data6 = z1 + tmp12 * -1.847759065; 235 ; 236 ; (This implementation) 237 ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; 238 ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); 239 240 movdqa xmm7, xmm4 ; xmm4=tmp13 241 movdqa xmm6, xmm4 242 punpcklwd xmm7, xmm0 ; xmm0=tmp12 243 punpckhwd xmm6, xmm0 244 movdqa xmm4, xmm7 245 movdqa xmm0, xmm6 246 pmaddwd xmm7, [GOTOFF(ebx,PW_F130_F054)] ; xmm7=data2L 247 pmaddwd xmm6, [GOTOFF(ebx,PW_F130_F054)] ; xmm6=data2H 248 pmaddwd xmm4, [GOTOFF(ebx,PW_F054_MF130)] ; xmm4=data6L 249 pmaddwd xmm0, [GOTOFF(ebx,PW_F054_MF130)] ; xmm0=data6H 250 251 paddd xmm7, [GOTOFF(ebx,PD_DESCALE_P1)] 252 paddd xmm6, [GOTOFF(ebx,PD_DESCALE_P1)] 253 psrad xmm7, DESCALE_P1 254 psrad xmm6, DESCALE_P1 255 paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P1)] 256 paddd xmm0, [GOTOFF(ebx,PD_DESCALE_P1)] 257 psrad xmm4, DESCALE_P1 258 psrad xmm0, DESCALE_P1 259 260 packssdw xmm7, xmm6 ; xmm7=data2 261 packssdw xmm4, xmm0 ; xmm4=data6 262 263 movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=data2 264 movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=data6 265 266 ; -- Odd part 267 268 movdqa xmm3, XMMWORD [wk(0)] ; xmm3=tmp6 269 movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp7 270 271 movdqa xmm6, xmm2 ; xmm2=tmp4 272 movdqa xmm0, xmm5 ; xmm5=tmp5 273 paddw xmm6, xmm3 ; xmm6=z3 274 paddw xmm0, xmm1 ; xmm0=z4 275 276 ; (Original) 277 ; z5 = (z3 + z4) * 1.175875602; 278 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; 279 ; z3 += z5; z4 += z5; 280 ; 281 ; (This implementation) 282 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; 283 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); 284 285 movdqa xmm7, xmm6 286 movdqa xmm4, xmm6 287 punpcklwd xmm7, xmm0 288 punpckhwd xmm4, xmm0 289 movdqa xmm6, xmm7 290 movdqa xmm0, xmm4 291 pmaddwd xmm7, [GOTOFF(ebx,PW_MF078_F117)] ; xmm7=z3L 292 pmaddwd xmm4, [GOTOFF(ebx,PW_MF078_F117)] ; xmm4=z3H 293 pmaddwd xmm6, [GOTOFF(ebx,PW_F117_F078)] ; xmm6=z4L 294 pmaddwd xmm0, [GOTOFF(ebx,PW_F117_F078)] ; xmm0=z4H 295 296 movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=z3L 297 movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=z3H 298 299 ; (Original) 300 ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; 301 ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; 302 ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; 303 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; 304 ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; 305 ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; 306 ; 307 ; (This implementation) 308 ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; 309 ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; 310 ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); 311 ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); 312 ; data7 = tmp4 + z3; data5 = tmp5 + z4; 313 ; data3 = tmp6 + z3; data1 = tmp7 + z4; 314 315 movdqa xmm7, xmm2 316 movdqa xmm4, xmm2 317 punpcklwd xmm7, xmm1 318 punpckhwd xmm4, xmm1 319 movdqa xmm2, xmm7 320 movdqa xmm1, xmm4 321 pmaddwd xmm7, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm7=tmp4L 322 pmaddwd xmm4, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm4=tmp4H 323 pmaddwd xmm2, [GOTOFF(ebx,PW_MF089_F060)] ; xmm2=tmp7L 324 pmaddwd xmm1, [GOTOFF(ebx,PW_MF089_F060)] ; xmm1=tmp7H 325 326 paddd xmm7, XMMWORD [wk(0)] ; xmm7=data7L 327 paddd xmm4, XMMWORD [wk(1)] ; xmm4=data7H 328 paddd xmm2, xmm6 ; xmm2=data1L 329 paddd xmm1, xmm0 ; xmm1=data1H 330 331 paddd xmm7, [GOTOFF(ebx,PD_DESCALE_P1)] 332 paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P1)] 333 psrad xmm7, DESCALE_P1 334 psrad xmm4, DESCALE_P1 335 paddd xmm2, [GOTOFF(ebx,PD_DESCALE_P1)] 336 paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P1)] 337 psrad xmm2, DESCALE_P1 338 psrad xmm1, DESCALE_P1 339 340 packssdw xmm7, xmm4 ; xmm7=data7 341 packssdw xmm2, xmm1 ; xmm2=data1 342 343 movdqa xmm4, xmm5 344 movdqa xmm1, xmm5 345 punpcklwd xmm4, xmm3 346 punpckhwd xmm1, xmm3 347 movdqa xmm5, xmm4 348 movdqa xmm3, xmm1 349 pmaddwd xmm4, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm4=tmp5L 350 pmaddwd xmm1, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm1=tmp5H 351 pmaddwd xmm5, [GOTOFF(ebx,PW_MF256_F050)] ; xmm5=tmp6L 352 pmaddwd xmm3, [GOTOFF(ebx,PW_MF256_F050)] ; xmm3=tmp6H 353 354 paddd xmm4, xmm6 ; xmm4=data5L 355 paddd xmm1, xmm0 ; xmm1=data5H 356 paddd xmm5, XMMWORD [wk(0)] ; xmm5=data3L 357 paddd xmm3, XMMWORD [wk(1)] ; xmm3=data3H 358 359 paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P1)] 360 paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P1)] 361 psrad xmm4, DESCALE_P1 362 psrad xmm1, DESCALE_P1 363 paddd xmm5, [GOTOFF(ebx,PD_DESCALE_P1)] 364 paddd xmm3, [GOTOFF(ebx,PD_DESCALE_P1)] 365 psrad xmm5, DESCALE_P1 366 psrad xmm3, DESCALE_P1 367 368 packssdw xmm4, xmm1 ; xmm4=data5 369 packssdw xmm5, xmm3 ; xmm5=data3 370 371 ; ---- Pass 2: process columns. 372 373; mov edx, POINTER [data(eax)] ; (DCTELEM *) 374 375 movdqa xmm6, XMMWORD [wk(2)] ; xmm6=col0 376 movdqa xmm0, XMMWORD [wk(4)] ; xmm0=col2 377 378 ; xmm6=(00 10 20 30 40 50 60 70), xmm0=(02 12 22 32 42 52 62 72) 379 ; xmm2=(01 11 21 31 41 51 61 71), xmm5=(03 13 23 33 43 53 63 73) 380 381 movdqa xmm1, xmm6 ; transpose coefficients(phase 1) 382 punpcklwd xmm6, xmm2 ; xmm6=(00 01 10 11 20 21 30 31) 383 punpckhwd xmm1, xmm2 ; xmm1=(40 41 50 51 60 61 70 71) 384 movdqa xmm3, xmm0 ; transpose coefficients(phase 1) 385 punpcklwd xmm0, xmm5 ; xmm0=(02 03 12 13 22 23 32 33) 386 punpckhwd xmm3, xmm5 ; xmm3=(42 43 52 53 62 63 72 73) 387 388 movdqa xmm2, XMMWORD [wk(3)] ; xmm2=col4 389 movdqa xmm5, XMMWORD [wk(5)] ; xmm5=col6 390 391 ; xmm2=(04 14 24 34 44 54 64 74), xmm5=(06 16 26 36 46 56 66 76) 392 ; xmm4=(05 15 25 35 45 55 65 75), xmm7=(07 17 27 37 47 57 67 77) 393 394 movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=(02 03 12 13 22 23 32 33) 395 movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(42 43 52 53 62 63 72 73) 396 397 movdqa xmm0, xmm2 ; transpose coefficients(phase 1) 398 punpcklwd xmm2, xmm4 ; xmm2=(04 05 14 15 24 25 34 35) 399 punpckhwd xmm0, xmm4 ; xmm0=(44 45 54 55 64 65 74 75) 400 movdqa xmm3, xmm5 ; transpose coefficients(phase 1) 401 punpcklwd xmm5, xmm7 ; xmm5=(06 07 16 17 26 27 36 37) 402 punpckhwd xmm3, xmm7 ; xmm3=(46 47 56 57 66 67 76 77) 403 404 movdqa xmm4, xmm2 ; transpose coefficients(phase 2) 405 punpckldq xmm2, xmm5 ; xmm2=(04 05 06 07 14 15 16 17) 406 punpckhdq xmm4, xmm5 ; xmm4=(24 25 26 27 34 35 36 37) 407 movdqa xmm7, xmm0 ; transpose coefficients(phase 2) 408 punpckldq xmm0, xmm3 ; xmm0=(44 45 46 47 54 55 56 57) 409 punpckhdq xmm7, xmm3 ; xmm7=(64 65 66 67 74 75 76 77) 410 411 movdqa xmm5, XMMWORD [wk(0)] ; xmm5=(02 03 12 13 22 23 32 33) 412 movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53 62 63 72 73) 413 movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=(24 25 26 27 34 35 36 37) 414 movdqa XMMWORD [wk(3)], xmm0 ; wk(3)=(44 45 46 47 54 55 56 57) 415 416 movdqa xmm4, xmm6 ; transpose coefficients(phase 2) 417 punpckldq xmm6, xmm5 ; xmm6=(00 01 02 03 10 11 12 13) 418 punpckhdq xmm4, xmm5 ; xmm4=(20 21 22 23 30 31 32 33) 419 movdqa xmm0, xmm1 ; transpose coefficients(phase 2) 420 punpckldq xmm1, xmm3 ; xmm1=(40 41 42 43 50 51 52 53) 421 punpckhdq xmm0, xmm3 ; xmm0=(60 61 62 63 70 71 72 73) 422 423 movdqa xmm5, xmm6 ; transpose coefficients(phase 3) 424 punpcklqdq xmm6, xmm2 ; xmm6=(00 01 02 03 04 05 06 07)=data0 425 punpckhqdq xmm5, xmm2 ; xmm5=(10 11 12 13 14 15 16 17)=data1 426 movdqa xmm3, xmm0 ; transpose coefficients(phase 3) 427 punpcklqdq xmm0, xmm7 ; xmm0=(60 61 62 63 64 65 66 67)=data6 428 punpckhqdq xmm3, xmm7 ; xmm3=(70 71 72 73 74 75 76 77)=data7 429 430 movdqa xmm2, xmm5 431 movdqa xmm7, xmm6 432 psubw xmm5, xmm0 ; xmm5=data1-data6=tmp6 433 psubw xmm6, xmm3 ; xmm6=data0-data7=tmp7 434 paddw xmm2, xmm0 ; xmm2=data1+data6=tmp1 435 paddw xmm7, xmm3 ; xmm7=data0+data7=tmp0 436 437 movdqa xmm0, XMMWORD [wk(2)] ; xmm0=(24 25 26 27 34 35 36 37) 438 movdqa xmm3, XMMWORD [wk(3)] ; xmm3=(44 45 46 47 54 55 56 57) 439 movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=tmp6 440 movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7 441 442 movdqa xmm5, xmm4 ; transpose coefficients(phase 3) 443 punpcklqdq xmm4, xmm0 ; xmm4=(20 21 22 23 24 25 26 27)=data2 444 punpckhqdq xmm5, xmm0 ; xmm5=(30 31 32 33 34 35 36 37)=data3 445 movdqa xmm6, xmm1 ; transpose coefficients(phase 3) 446 punpcklqdq xmm1, xmm3 ; xmm1=(40 41 42 43 44 45 46 47)=data4 447 punpckhqdq xmm6, xmm3 ; xmm6=(50 51 52 53 54 55 56 57)=data5 448 449 movdqa xmm0, xmm5 450 movdqa xmm3, xmm4 451 paddw xmm5, xmm1 ; xmm5=data3+data4=tmp3 452 paddw xmm4, xmm6 ; xmm4=data2+data5=tmp2 453 psubw xmm0, xmm1 ; xmm0=data3-data4=tmp4 454 psubw xmm3, xmm6 ; xmm3=data2-data5=tmp5 455 456 ; -- Even part 457 458 movdqa xmm1, xmm7 459 movdqa xmm6, xmm2 460 paddw xmm7, xmm5 ; xmm7=tmp10 461 paddw xmm2, xmm4 ; xmm2=tmp11 462 psubw xmm1, xmm5 ; xmm1=tmp13 463 psubw xmm6, xmm4 ; xmm6=tmp12 464 465 movdqa xmm5, xmm7 466 paddw xmm7, xmm2 ; xmm7=tmp10+tmp11 467 psubw xmm5, xmm2 ; xmm5=tmp10-tmp11 468 469 paddw xmm7, [GOTOFF(ebx,PW_DESCALE_P2X)] 470 paddw xmm5, [GOTOFF(ebx,PW_DESCALE_P2X)] 471 psraw xmm7, PASS1_BITS ; xmm7=data0 472 psraw xmm5, PASS1_BITS ; xmm5=data4 473 474 movdqa XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)], xmm7 475 movdqa XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)], xmm5 476 477 ; (Original) 478 ; z1 = (tmp12 + tmp13) * 0.541196100; 479 ; data2 = z1 + tmp13 * 0.765366865; 480 ; data6 = z1 + tmp12 * -1.847759065; 481 ; 482 ; (This implementation) 483 ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; 484 ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); 485 486 movdqa xmm4, xmm1 ; xmm1=tmp13 487 movdqa xmm2, xmm1 488 punpcklwd xmm4, xmm6 ; xmm6=tmp12 489 punpckhwd xmm2, xmm6 490 movdqa xmm1, xmm4 491 movdqa xmm6, xmm2 492 pmaddwd xmm4, [GOTOFF(ebx,PW_F130_F054)] ; xmm4=data2L 493 pmaddwd xmm2, [GOTOFF(ebx,PW_F130_F054)] ; xmm2=data2H 494 pmaddwd xmm1, [GOTOFF(ebx,PW_F054_MF130)] ; xmm1=data6L 495 pmaddwd xmm6, [GOTOFF(ebx,PW_F054_MF130)] ; xmm6=data6H 496 497 paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P2)] 498 paddd xmm2, [GOTOFF(ebx,PD_DESCALE_P2)] 499 psrad xmm4, DESCALE_P2 500 psrad xmm2, DESCALE_P2 501 paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P2)] 502 paddd xmm6, [GOTOFF(ebx,PD_DESCALE_P2)] 503 psrad xmm1, DESCALE_P2 504 psrad xmm6, DESCALE_P2 505 506 packssdw xmm4, xmm2 ; xmm4=data2 507 packssdw xmm1, xmm6 ; xmm1=data6 508 509 movdqa XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)], xmm4 510 movdqa XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)], xmm1 511 512 ; -- Odd part 513 514 movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp6 515 movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7 516 517 movdqa xmm2, xmm0 ; xmm0=tmp4 518 movdqa xmm6, xmm3 ; xmm3=tmp5 519 paddw xmm2, xmm7 ; xmm2=z3 520 paddw xmm6, xmm5 ; xmm6=z4 521 522 ; (Original) 523 ; z5 = (z3 + z4) * 1.175875602; 524 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; 525 ; z3 += z5; z4 += z5; 526 ; 527 ; (This implementation) 528 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; 529 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); 530 531 movdqa xmm4, xmm2 532 movdqa xmm1, xmm2 533 punpcklwd xmm4, xmm6 534 punpckhwd xmm1, xmm6 535 movdqa xmm2, xmm4 536 movdqa xmm6, xmm1 537 pmaddwd xmm4, [GOTOFF(ebx,PW_MF078_F117)] ; xmm4=z3L 538 pmaddwd xmm1, [GOTOFF(ebx,PW_MF078_F117)] ; xmm1=z3H 539 pmaddwd xmm2, [GOTOFF(ebx,PW_F117_F078)] ; xmm2=z4L 540 pmaddwd xmm6, [GOTOFF(ebx,PW_F117_F078)] ; xmm6=z4H 541 542 movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=z3L 543 movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=z3H 544 545 ; (Original) 546 ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; 547 ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; 548 ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; 549 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; 550 ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; 551 ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; 552 ; 553 ; (This implementation) 554 ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; 555 ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; 556 ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); 557 ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); 558 ; data7 = tmp4 + z3; data5 = tmp5 + z4; 559 ; data3 = tmp6 + z3; data1 = tmp7 + z4; 560 561 movdqa xmm4, xmm0 562 movdqa xmm1, xmm0 563 punpcklwd xmm4, xmm5 564 punpckhwd xmm1, xmm5 565 movdqa xmm0, xmm4 566 movdqa xmm5, xmm1 567 pmaddwd xmm4, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm4=tmp4L 568 pmaddwd xmm1, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm1=tmp4H 569 pmaddwd xmm0, [GOTOFF(ebx,PW_MF089_F060)] ; xmm0=tmp7L 570 pmaddwd xmm5, [GOTOFF(ebx,PW_MF089_F060)] ; xmm5=tmp7H 571 572 paddd xmm4, XMMWORD [wk(0)] ; xmm4=data7L 573 paddd xmm1, XMMWORD [wk(1)] ; xmm1=data7H 574 paddd xmm0, xmm2 ; xmm0=data1L 575 paddd xmm5, xmm6 ; xmm5=data1H 576 577 paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P2)] 578 paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P2)] 579 psrad xmm4, DESCALE_P2 580 psrad xmm1, DESCALE_P2 581 paddd xmm0, [GOTOFF(ebx,PD_DESCALE_P2)] 582 paddd xmm5, [GOTOFF(ebx,PD_DESCALE_P2)] 583 psrad xmm0, DESCALE_P2 584 psrad xmm5, DESCALE_P2 585 586 packssdw xmm4, xmm1 ; xmm4=data7 587 packssdw xmm0, xmm5 ; xmm0=data1 588 589 movdqa XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)], xmm4 590 movdqa XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)], xmm0 591 592 movdqa xmm1, xmm3 593 movdqa xmm5, xmm3 594 punpcklwd xmm1, xmm7 595 punpckhwd xmm5, xmm7 596 movdqa xmm3, xmm1 597 movdqa xmm7, xmm5 598 pmaddwd xmm1, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm1=tmp5L 599 pmaddwd xmm5, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm5=tmp5H 600 pmaddwd xmm3, [GOTOFF(ebx,PW_MF256_F050)] ; xmm3=tmp6L 601 pmaddwd xmm7, [GOTOFF(ebx,PW_MF256_F050)] ; xmm7=tmp6H 602 603 paddd xmm1, xmm2 ; xmm1=data5L 604 paddd xmm5, xmm6 ; xmm5=data5H 605 paddd xmm3, XMMWORD [wk(0)] ; xmm3=data3L 606 paddd xmm7, XMMWORD [wk(1)] ; xmm7=data3H 607 608 paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P2)] 609 paddd xmm5, [GOTOFF(ebx,PD_DESCALE_P2)] 610 psrad xmm1, DESCALE_P2 611 psrad xmm5, DESCALE_P2 612 paddd xmm3, [GOTOFF(ebx,PD_DESCALE_P2)] 613 paddd xmm7, [GOTOFF(ebx,PD_DESCALE_P2)] 614 psrad xmm3, DESCALE_P2 615 psrad xmm7, DESCALE_P2 616 617 packssdw xmm1, xmm5 ; xmm1=data5 618 packssdw xmm3, xmm7 ; xmm3=data3 619 620 movdqa XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)], xmm1 621 movdqa XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)], xmm3 622 623; pop edi ; unused 624; pop esi ; unused 625; pop edx ; need not be preserved 626; pop ecx ; unused 627 poppic ebx 628 mov esp, ebp ; esp <- aligned ebp 629 pop esp ; esp <- original ebp 630 pop ebp 631 ret 632 633; For some reason, the OS X linker does not honor the request to align the 634; segment unless we do this. 635 align 32 636