1; 2; jidctint.asm - accurate integer IDCT (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; inverse DCT (Discrete Cosine Transform). The following code is based 19; directly on the IJG's original jidctint.c; see the jidctint.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 + 3) 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_idct_islow_sse2) 70 71EXTN(jconst_idct_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) 83PB_CENTERJSAMP times 16 db CENTERJSAMPLE 84 85 alignz 32 86 87; -------------------------------------------------------------------------- 88 SECTION SEG_TEXT 89 BITS 32 90; 91; Perform dequantization and inverse DCT on one block of coefficients. 92; 93; GLOBAL(void) 94; jsimd_idct_islow_sse2(void *dct_table, JCOEFPTR coef_block, 95; JSAMPARRAY output_buf, JDIMENSION output_col) 96; 97 98%define dct_table(b) (b) + 8 ; jpeg_component_info *compptr 99%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block 100%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf 101%define output_col(b) (b) + 20 ; JDIMENSION output_col 102 103%define original_ebp ebp + 0 104%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD 105 ; xmmword wk[WK_NUM] 106%define WK_NUM 12 107 108 align 32 109 GLOBAL_FUNCTION(jsimd_idct_islow_sse2) 110 111EXTN(jsimd_idct_islow_sse2): 112 push ebp 113 mov eax, esp ; eax = original ebp 114 sub esp, byte 4 115 and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits 116 mov [esp], eax 117 mov ebp, esp ; ebp = aligned ebp 118 lea esp, [wk(0)] 119 pushpic ebx 120; push ecx ; unused 121; push edx ; need not be preserved 122 push esi 123 push edi 124 125 get_GOT ebx ; get GOT address 126 127 ; ---- Pass 1: process columns from input. 128 129; mov eax, [original_ebp] 130 mov edx, POINTER [dct_table(eax)] ; quantptr 131 mov esi, JCOEFPTR [coef_block(eax)] ; inptr 132 133%ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2 134 mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] 135 or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] 136 jnz near .columnDCT 137 138 movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] 139 movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] 140 por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] 141 por xmm1, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)] 142 por xmm0, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] 143 por xmm1, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] 144 por xmm0, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] 145 por xmm1, xmm0 146 packsswb xmm1, xmm1 147 packsswb xmm1, xmm1 148 movd eax, xmm1 149 test eax, eax 150 jnz short .columnDCT 151 152 ; -- AC terms all zero 153 154 movdqa xmm5, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] 155 pmullw xmm5, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] 156 157 psllw xmm5, PASS1_BITS 158 159 movdqa xmm4, xmm5 ; xmm5=in0=(00 01 02 03 04 05 06 07) 160 punpcklwd xmm5, xmm5 ; xmm5=(00 00 01 01 02 02 03 03) 161 punpckhwd xmm4, xmm4 ; xmm4=(04 04 05 05 06 06 07 07) 162 163 pshufd xmm7, xmm5, 0x00 ; xmm7=col0=(00 00 00 00 00 00 00 00) 164 pshufd xmm6, xmm5, 0x55 ; xmm6=col1=(01 01 01 01 01 01 01 01) 165 pshufd xmm1, xmm5, 0xAA ; xmm1=col2=(02 02 02 02 02 02 02 02) 166 pshufd xmm5, xmm5, 0xFF ; xmm5=col3=(03 03 03 03 03 03 03 03) 167 pshufd xmm0, xmm4, 0x00 ; xmm0=col4=(04 04 04 04 04 04 04 04) 168 pshufd xmm3, xmm4, 0x55 ; xmm3=col5=(05 05 05 05 05 05 05 05) 169 pshufd xmm2, xmm4, 0xAA ; xmm2=col6=(06 06 06 06 06 06 06 06) 170 pshufd xmm4, xmm4, 0xFF ; xmm4=col7=(07 07 07 07 07 07 07 07) 171 172 movdqa XMMWORD [wk(8)], xmm6 ; wk(8)=col1 173 movdqa XMMWORD [wk(9)], xmm5 ; wk(9)=col3 174 movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5 175 movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7 176 jmp near .column_end 177 alignx 16, 7 178%endif 179.columnDCT: 180 181 ; -- Even part 182 183 movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] 184 movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] 185 pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] 186 pmullw xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)] 187 movdqa xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)] 188 movdqa xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] 189 pmullw xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_ISLOW_MULT_TYPE)] 190 pmullw xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)] 191 192 ; (Original) 193 ; z1 = (z2 + z3) * 0.541196100; 194 ; tmp2 = z1 + z3 * -1.847759065; 195 ; tmp3 = z1 + z2 * 0.765366865; 196 ; 197 ; (This implementation) 198 ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); 199 ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; 200 201 movdqa xmm4, xmm1 ; xmm1=in2=z2 202 movdqa xmm5, xmm1 203 punpcklwd xmm4, xmm3 ; xmm3=in6=z3 204 punpckhwd xmm5, xmm3 205 movdqa xmm1, xmm4 206 movdqa xmm3, xmm5 207 pmaddwd xmm4, [GOTOFF(ebx,PW_F130_F054)] ; xmm4=tmp3L 208 pmaddwd xmm5, [GOTOFF(ebx,PW_F130_F054)] ; xmm5=tmp3H 209 pmaddwd xmm1, [GOTOFF(ebx,PW_F054_MF130)] ; xmm1=tmp2L 210 pmaddwd xmm3, [GOTOFF(ebx,PW_F054_MF130)] ; xmm3=tmp2H 211 212 movdqa xmm6, xmm0 213 paddw xmm0, xmm2 ; xmm0=in0+in4 214 psubw xmm6, xmm2 ; xmm6=in0-in4 215 216 pxor xmm7, xmm7 217 pxor xmm2, xmm2 218 punpcklwd xmm7, xmm0 ; xmm7=tmp0L 219 punpckhwd xmm2, xmm0 ; xmm2=tmp0H 220 psrad xmm7, (16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS 221 psrad xmm2, (16-CONST_BITS) ; psrad xmm2,16 & pslld xmm2,CONST_BITS 222 223 movdqa xmm0, xmm7 224 paddd xmm7, xmm4 ; xmm7=tmp10L 225 psubd xmm0, xmm4 ; xmm0=tmp13L 226 movdqa xmm4, xmm2 227 paddd xmm2, xmm5 ; xmm2=tmp10H 228 psubd xmm4, xmm5 ; xmm4=tmp13H 229 230 movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=tmp10L 231 movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=tmp10H 232 movdqa XMMWORD [wk(2)], xmm0 ; wk(2)=tmp13L 233 movdqa XMMWORD [wk(3)], xmm4 ; wk(3)=tmp13H 234 235 pxor xmm5, xmm5 236 pxor xmm7, xmm7 237 punpcklwd xmm5, xmm6 ; xmm5=tmp1L 238 punpckhwd xmm7, xmm6 ; xmm7=tmp1H 239 psrad xmm5, (16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS 240 psrad xmm7, (16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS 241 242 movdqa xmm2, xmm5 243 paddd xmm5, xmm1 ; xmm5=tmp11L 244 psubd xmm2, xmm1 ; xmm2=tmp12L 245 movdqa xmm0, xmm7 246 paddd xmm7, xmm3 ; xmm7=tmp11H 247 psubd xmm0, xmm3 ; xmm0=tmp12H 248 249 movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L 250 movdqa XMMWORD [wk(5)], xmm7 ; wk(5)=tmp11H 251 movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=tmp12L 252 movdqa XMMWORD [wk(7)], xmm0 ; wk(7)=tmp12H 253 254 ; -- Odd part 255 256 movdqa xmm4, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] 257 movdqa xmm6, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] 258 pmullw xmm4, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)] 259 pmullw xmm6, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)] 260 movdqa xmm1, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] 261 movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] 262 pmullw xmm1, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)] 263 pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)] 264 265 movdqa xmm5, xmm6 266 movdqa xmm7, xmm4 267 paddw xmm5, xmm3 ; xmm5=z3 268 paddw xmm7, xmm1 ; xmm7=z4 269 270 ; (Original) 271 ; z5 = (z3 + z4) * 1.175875602; 272 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; 273 ; z3 += z5; z4 += z5; 274 ; 275 ; (This implementation) 276 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; 277 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); 278 279 movdqa xmm2, xmm5 280 movdqa xmm0, xmm5 281 punpcklwd xmm2, xmm7 282 punpckhwd xmm0, xmm7 283 movdqa xmm5, xmm2 284 movdqa xmm7, xmm0 285 pmaddwd xmm2, [GOTOFF(ebx,PW_MF078_F117)] ; xmm2=z3L 286 pmaddwd xmm0, [GOTOFF(ebx,PW_MF078_F117)] ; xmm0=z3H 287 pmaddwd xmm5, [GOTOFF(ebx,PW_F117_F078)] ; xmm5=z4L 288 pmaddwd xmm7, [GOTOFF(ebx,PW_F117_F078)] ; xmm7=z4H 289 290 movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=z3L 291 movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=z3H 292 293 ; (Original) 294 ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; 295 ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; 296 ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; 297 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; 298 ; tmp0 += z1 + z3; tmp1 += z2 + z4; 299 ; tmp2 += z2 + z3; tmp3 += z1 + z4; 300 ; 301 ; (This implementation) 302 ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; 303 ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; 304 ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); 305 ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); 306 ; tmp0 += z3; tmp1 += z4; 307 ; tmp2 += z3; tmp3 += z4; 308 309 movdqa xmm2, xmm3 310 movdqa xmm0, xmm3 311 punpcklwd xmm2, xmm4 312 punpckhwd xmm0, xmm4 313 movdqa xmm3, xmm2 314 movdqa xmm4, xmm0 315 pmaddwd xmm2, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm2=tmp0L 316 pmaddwd xmm0, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm0=tmp0H 317 pmaddwd xmm3, [GOTOFF(ebx,PW_MF089_F060)] ; xmm3=tmp3L 318 pmaddwd xmm4, [GOTOFF(ebx,PW_MF089_F060)] ; xmm4=tmp3H 319 320 paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp0L 321 paddd xmm0, XMMWORD [wk(11)] ; xmm0=tmp0H 322 paddd xmm3, xmm5 ; xmm3=tmp3L 323 paddd xmm4, xmm7 ; xmm4=tmp3H 324 325 movdqa XMMWORD [wk(8)], xmm2 ; wk(8)=tmp0L 326 movdqa XMMWORD [wk(9)], xmm0 ; wk(9)=tmp0H 327 328 movdqa xmm2, xmm1 329 movdqa xmm0, xmm1 330 punpcklwd xmm2, xmm6 331 punpckhwd xmm0, xmm6 332 movdqa xmm1, xmm2 333 movdqa xmm6, xmm0 334 pmaddwd xmm2, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm2=tmp1L 335 pmaddwd xmm0, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm0=tmp1H 336 pmaddwd xmm1, [GOTOFF(ebx,PW_MF256_F050)] ; xmm1=tmp2L 337 pmaddwd xmm6, [GOTOFF(ebx,PW_MF256_F050)] ; xmm6=tmp2H 338 339 paddd xmm2, xmm5 ; xmm2=tmp1L 340 paddd xmm0, xmm7 ; xmm0=tmp1H 341 paddd xmm1, XMMWORD [wk(10)] ; xmm1=tmp2L 342 paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H 343 344 movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=tmp1L 345 movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=tmp1H 346 347 ; -- Final output stage 348 349 movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L 350 movdqa xmm7, XMMWORD [wk(1)] ; xmm7=tmp10H 351 352 movdqa xmm2, xmm5 353 movdqa xmm0, xmm7 354 paddd xmm5, xmm3 ; xmm5=data0L 355 paddd xmm7, xmm4 ; xmm7=data0H 356 psubd xmm2, xmm3 ; xmm2=data7L 357 psubd xmm0, xmm4 ; xmm0=data7H 358 359 movdqa xmm3, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm3=[PD_DESCALE_P1] 360 361 paddd xmm5, xmm3 362 paddd xmm7, xmm3 363 psrad xmm5, DESCALE_P1 364 psrad xmm7, DESCALE_P1 365 paddd xmm2, xmm3 366 paddd xmm0, xmm3 367 psrad xmm2, DESCALE_P1 368 psrad xmm0, DESCALE_P1 369 370 packssdw xmm5, xmm7 ; xmm5=data0=(00 01 02 03 04 05 06 07) 371 packssdw xmm2, xmm0 ; xmm2=data7=(70 71 72 73 74 75 76 77) 372 373 movdqa xmm4, XMMWORD [wk(4)] ; xmm4=tmp11L 374 movdqa xmm3, XMMWORD [wk(5)] ; xmm3=tmp11H 375 376 movdqa xmm7, xmm4 377 movdqa xmm0, xmm3 378 paddd xmm4, xmm1 ; xmm4=data1L 379 paddd xmm3, xmm6 ; xmm3=data1H 380 psubd xmm7, xmm1 ; xmm7=data6L 381 psubd xmm0, xmm6 ; xmm0=data6H 382 383 movdqa xmm1, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm1=[PD_DESCALE_P1] 384 385 paddd xmm4, xmm1 386 paddd xmm3, xmm1 387 psrad xmm4, DESCALE_P1 388 psrad xmm3, DESCALE_P1 389 paddd xmm7, xmm1 390 paddd xmm0, xmm1 391 psrad xmm7, DESCALE_P1 392 psrad xmm0, DESCALE_P1 393 394 packssdw xmm4, xmm3 ; xmm4=data1=(10 11 12 13 14 15 16 17) 395 packssdw xmm7, xmm0 ; xmm7=data6=(60 61 62 63 64 65 66 67) 396 397 movdqa xmm6, xmm5 ; transpose coefficients(phase 1) 398 punpcklwd xmm5, xmm4 ; xmm5=(00 10 01 11 02 12 03 13) 399 punpckhwd xmm6, xmm4 ; xmm6=(04 14 05 15 06 16 07 17) 400 movdqa xmm1, xmm7 ; transpose coefficients(phase 1) 401 punpcklwd xmm7, xmm2 ; xmm7=(60 70 61 71 62 72 63 73) 402 punpckhwd xmm1, xmm2 ; xmm1=(64 74 65 75 66 76 67 77) 403 404 movdqa xmm3, XMMWORD [wk(6)] ; xmm3=tmp12L 405 movdqa xmm0, XMMWORD [wk(7)] ; xmm0=tmp12H 406 movdqa xmm4, XMMWORD [wk(10)] ; xmm4=tmp1L 407 movdqa xmm2, XMMWORD [wk(11)] ; xmm2=tmp1H 408 409 movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 01 11 02 12 03 13) 410 movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=(04 14 05 15 06 16 07 17) 411 movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=(60 70 61 71 62 72 63 73) 412 movdqa XMMWORD [wk(5)], xmm1 ; wk(5)=(64 74 65 75 66 76 67 77) 413 414 movdqa xmm5, xmm3 415 movdqa xmm6, xmm0 416 paddd xmm3, xmm4 ; xmm3=data2L 417 paddd xmm0, xmm2 ; xmm0=data2H 418 psubd xmm5, xmm4 ; xmm5=data5L 419 psubd xmm6, xmm2 ; xmm6=data5H 420 421 movdqa xmm7, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm7=[PD_DESCALE_P1] 422 423 paddd xmm3, xmm7 424 paddd xmm0, xmm7 425 psrad xmm3, DESCALE_P1 426 psrad xmm0, DESCALE_P1 427 paddd xmm5, xmm7 428 paddd xmm6, xmm7 429 psrad xmm5, DESCALE_P1 430 psrad xmm6, DESCALE_P1 431 432 packssdw xmm3, xmm0 ; xmm3=data2=(20 21 22 23 24 25 26 27) 433 packssdw xmm5, xmm6 ; xmm5=data5=(50 51 52 53 54 55 56 57) 434 435 movdqa xmm1, XMMWORD [wk(2)] ; xmm1=tmp13L 436 movdqa xmm4, XMMWORD [wk(3)] ; xmm4=tmp13H 437 movdqa xmm2, XMMWORD [wk(8)] ; xmm2=tmp0L 438 movdqa xmm7, XMMWORD [wk(9)] ; xmm7=tmp0H 439 440 movdqa xmm0, xmm1 441 movdqa xmm6, xmm4 442 paddd xmm1, xmm2 ; xmm1=data3L 443 paddd xmm4, xmm7 ; xmm4=data3H 444 psubd xmm0, xmm2 ; xmm0=data4L 445 psubd xmm6, xmm7 ; xmm6=data4H 446 447 movdqa xmm2, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm2=[PD_DESCALE_P1] 448 449 paddd xmm1, xmm2 450 paddd xmm4, xmm2 451 psrad xmm1, DESCALE_P1 452 psrad xmm4, DESCALE_P1 453 paddd xmm0, xmm2 454 paddd xmm6, xmm2 455 psrad xmm0, DESCALE_P1 456 psrad xmm6, DESCALE_P1 457 458 packssdw xmm1, xmm4 ; xmm1=data3=(30 31 32 33 34 35 36 37) 459 packssdw xmm0, xmm6 ; xmm0=data4=(40 41 42 43 44 45 46 47) 460 461 movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 01 11 02 12 03 13) 462 movdqa xmm2, XMMWORD [wk(1)] ; xmm2=(04 14 05 15 06 16 07 17) 463 464 movdqa xmm4, xmm3 ; transpose coefficients(phase 1) 465 punpcklwd xmm3, xmm1 ; xmm3=(20 30 21 31 22 32 23 33) 466 punpckhwd xmm4, xmm1 ; xmm4=(24 34 25 35 26 36 27 37) 467 movdqa xmm6, xmm0 ; transpose coefficients(phase 1) 468 punpcklwd xmm0, xmm5 ; xmm0=(40 50 41 51 42 52 43 53) 469 punpckhwd xmm6, xmm5 ; xmm6=(44 54 45 55 46 56 47 57) 470 471 movdqa xmm1, xmm7 ; transpose coefficients(phase 2) 472 punpckldq xmm7, xmm3 ; xmm7=(00 10 20 30 01 11 21 31) 473 punpckhdq xmm1, xmm3 ; xmm1=(02 12 22 32 03 13 23 33) 474 movdqa xmm5, xmm2 ; transpose coefficients(phase 2) 475 punpckldq xmm2, xmm4 ; xmm2=(04 14 24 34 05 15 25 35) 476 punpckhdq xmm5, xmm4 ; xmm5=(06 16 26 36 07 17 27 37) 477 478 movdqa xmm3, XMMWORD [wk(4)] ; xmm3=(60 70 61 71 62 72 63 73) 479 movdqa xmm4, XMMWORD [wk(5)] ; xmm4=(64 74 65 75 66 76 67 77) 480 481 movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=(04 14 24 34 05 15 25 35) 482 movdqa XMMWORD [wk(7)], xmm5 ; wk(7)=(06 16 26 36 07 17 27 37) 483 484 movdqa xmm2, xmm0 ; transpose coefficients(phase 2) 485 punpckldq xmm0, xmm3 ; xmm0=(40 50 60 70 41 51 61 71) 486 punpckhdq xmm2, xmm3 ; xmm2=(42 52 62 72 43 53 63 73) 487 movdqa xmm5, xmm6 ; transpose coefficients(phase 2) 488 punpckldq xmm6, xmm4 ; xmm6=(44 54 64 74 45 55 65 75) 489 punpckhdq xmm5, xmm4 ; xmm5=(46 56 66 76 47 57 67 77) 490 491 movdqa xmm3, xmm7 ; transpose coefficients(phase 3) 492 punpcklqdq xmm7, xmm0 ; xmm7=col0=(00 10 20 30 40 50 60 70) 493 punpckhqdq xmm3, xmm0 ; xmm3=col1=(01 11 21 31 41 51 61 71) 494 movdqa xmm4, xmm1 ; transpose coefficients(phase 3) 495 punpcklqdq xmm1, xmm2 ; xmm1=col2=(02 12 22 32 42 52 62 72) 496 punpckhqdq xmm4, xmm2 ; xmm4=col3=(03 13 23 33 43 53 63 73) 497 498 movdqa xmm0, XMMWORD [wk(6)] ; xmm0=(04 14 24 34 05 15 25 35) 499 movdqa xmm2, XMMWORD [wk(7)] ; xmm2=(06 16 26 36 07 17 27 37) 500 501 movdqa XMMWORD [wk(8)], xmm3 ; wk(8)=col1 502 movdqa XMMWORD [wk(9)], xmm4 ; wk(9)=col3 503 504 movdqa xmm3, xmm0 ; transpose coefficients(phase 3) 505 punpcklqdq xmm0, xmm6 ; xmm0=col4=(04 14 24 34 44 54 64 74) 506 punpckhqdq xmm3, xmm6 ; xmm3=col5=(05 15 25 35 45 55 65 75) 507 movdqa xmm4, xmm2 ; transpose coefficients(phase 3) 508 punpcklqdq xmm2, xmm5 ; xmm2=col6=(06 16 26 36 46 56 66 76) 509 punpckhqdq xmm4, xmm5 ; xmm4=col7=(07 17 27 37 47 57 67 77) 510 511 movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5 512 movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7 513.column_end: 514 515 ; -- Prefetch the next coefficient block 516 517 prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32] 518 prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32] 519 prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32] 520 prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32] 521 522 ; ---- Pass 2: process rows from work array, store into output array. 523 524 mov eax, [original_ebp] 525 mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) 526 mov eax, JDIMENSION [output_col(eax)] 527 528 ; -- Even part 529 530 ; xmm7=col0, xmm1=col2, xmm0=col4, xmm2=col6 531 532 ; (Original) 533 ; z1 = (z2 + z3) * 0.541196100; 534 ; tmp2 = z1 + z3 * -1.847759065; 535 ; tmp3 = z1 + z2 * 0.765366865; 536 ; 537 ; (This implementation) 538 ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); 539 ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; 540 541 movdqa xmm6, xmm1 ; xmm1=in2=z2 542 movdqa xmm5, xmm1 543 punpcklwd xmm6, xmm2 ; xmm2=in6=z3 544 punpckhwd xmm5, xmm2 545 movdqa xmm1, xmm6 546 movdqa xmm2, xmm5 547 pmaddwd xmm6, [GOTOFF(ebx,PW_F130_F054)] ; xmm6=tmp3L 548 pmaddwd xmm5, [GOTOFF(ebx,PW_F130_F054)] ; xmm5=tmp3H 549 pmaddwd xmm1, [GOTOFF(ebx,PW_F054_MF130)] ; xmm1=tmp2L 550 pmaddwd xmm2, [GOTOFF(ebx,PW_F054_MF130)] ; xmm2=tmp2H 551 552 movdqa xmm3, xmm7 553 paddw xmm7, xmm0 ; xmm7=in0+in4 554 psubw xmm3, xmm0 ; xmm3=in0-in4 555 556 pxor xmm4, xmm4 557 pxor xmm0, xmm0 558 punpcklwd xmm4, xmm7 ; xmm4=tmp0L 559 punpckhwd xmm0, xmm7 ; xmm0=tmp0H 560 psrad xmm4, (16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS 561 psrad xmm0, (16-CONST_BITS) ; psrad xmm0,16 & pslld xmm0,CONST_BITS 562 563 movdqa xmm7, xmm4 564 paddd xmm4, xmm6 ; xmm4=tmp10L 565 psubd xmm7, xmm6 ; xmm7=tmp13L 566 movdqa xmm6, xmm0 567 paddd xmm0, xmm5 ; xmm0=tmp10H 568 psubd xmm6, xmm5 ; xmm6=tmp13H 569 570 movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=tmp10L 571 movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp10H 572 movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=tmp13L 573 movdqa XMMWORD [wk(3)], xmm6 ; wk(3)=tmp13H 574 575 pxor xmm5, xmm5 576 pxor xmm4, xmm4 577 punpcklwd xmm5, xmm3 ; xmm5=tmp1L 578 punpckhwd xmm4, xmm3 ; xmm4=tmp1H 579 psrad xmm5, (16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS 580 psrad xmm4, (16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS 581 582 movdqa xmm0, xmm5 583 paddd xmm5, xmm1 ; xmm5=tmp11L 584 psubd xmm0, xmm1 ; xmm0=tmp12L 585 movdqa xmm7, xmm4 586 paddd xmm4, xmm2 ; xmm4=tmp11H 587 psubd xmm7, xmm2 ; xmm7=tmp12H 588 589 movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L 590 movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=tmp11H 591 movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=tmp12L 592 movdqa XMMWORD [wk(7)], xmm7 ; wk(7)=tmp12H 593 594 ; -- Odd part 595 596 movdqa xmm6, XMMWORD [wk(9)] ; xmm6=col3 597 movdqa xmm3, XMMWORD [wk(8)] ; xmm3=col1 598 movdqa xmm1, XMMWORD [wk(11)] ; xmm1=col7 599 movdqa xmm2, XMMWORD [wk(10)] ; xmm2=col5 600 601 movdqa xmm5, xmm6 602 movdqa xmm4, xmm3 603 paddw xmm5, xmm1 ; xmm5=z3 604 paddw xmm4, xmm2 ; xmm4=z4 605 606 ; (Original) 607 ; z5 = (z3 + z4) * 1.175875602; 608 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; 609 ; z3 += z5; z4 += z5; 610 ; 611 ; (This implementation) 612 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; 613 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); 614 615 movdqa xmm0, xmm5 616 movdqa xmm7, xmm5 617 punpcklwd xmm0, xmm4 618 punpckhwd xmm7, xmm4 619 movdqa xmm5, xmm0 620 movdqa xmm4, xmm7 621 pmaddwd xmm0, [GOTOFF(ebx,PW_MF078_F117)] ; xmm0=z3L 622 pmaddwd xmm7, [GOTOFF(ebx,PW_MF078_F117)] ; xmm7=z3H 623 pmaddwd xmm5, [GOTOFF(ebx,PW_F117_F078)] ; xmm5=z4L 624 pmaddwd xmm4, [GOTOFF(ebx,PW_F117_F078)] ; xmm4=z4H 625 626 movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=z3L 627 movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=z3H 628 629 ; (Original) 630 ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; 631 ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; 632 ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; 633 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; 634 ; tmp0 += z1 + z3; tmp1 += z2 + z4; 635 ; tmp2 += z2 + z3; tmp3 += z1 + z4; 636 ; 637 ; (This implementation) 638 ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; 639 ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; 640 ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); 641 ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); 642 ; tmp0 += z3; tmp1 += z4; 643 ; tmp2 += z3; tmp3 += z4; 644 645 movdqa xmm0, xmm1 646 movdqa xmm7, xmm1 647 punpcklwd xmm0, xmm3 648 punpckhwd xmm7, xmm3 649 movdqa xmm1, xmm0 650 movdqa xmm3, xmm7 651 pmaddwd xmm0, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm0=tmp0L 652 pmaddwd xmm7, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm7=tmp0H 653 pmaddwd xmm1, [GOTOFF(ebx,PW_MF089_F060)] ; xmm1=tmp3L 654 pmaddwd xmm3, [GOTOFF(ebx,PW_MF089_F060)] ; xmm3=tmp3H 655 656 paddd xmm0, XMMWORD [wk(10)] ; xmm0=tmp0L 657 paddd xmm7, XMMWORD [wk(11)] ; xmm7=tmp0H 658 paddd xmm1, xmm5 ; xmm1=tmp3L 659 paddd xmm3, xmm4 ; xmm3=tmp3H 660 661 movdqa XMMWORD [wk(8)], xmm0 ; wk(8)=tmp0L 662 movdqa XMMWORD [wk(9)], xmm7 ; wk(9)=tmp0H 663 664 movdqa xmm0, xmm2 665 movdqa xmm7, xmm2 666 punpcklwd xmm0, xmm6 667 punpckhwd xmm7, xmm6 668 movdqa xmm2, xmm0 669 movdqa xmm6, xmm7 670 pmaddwd xmm0, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm0=tmp1L 671 pmaddwd xmm7, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm7=tmp1H 672 pmaddwd xmm2, [GOTOFF(ebx,PW_MF256_F050)] ; xmm2=tmp2L 673 pmaddwd xmm6, [GOTOFF(ebx,PW_MF256_F050)] ; xmm6=tmp2H 674 675 paddd xmm0, xmm5 ; xmm0=tmp1L 676 paddd xmm7, xmm4 ; xmm7=tmp1H 677 paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp2L 678 paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H 679 680 movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=tmp1L 681 movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=tmp1H 682 683 ; -- Final output stage 684 685 movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L 686 movdqa xmm4, XMMWORD [wk(1)] ; xmm4=tmp10H 687 688 movdqa xmm0, xmm5 689 movdqa xmm7, xmm4 690 paddd xmm5, xmm1 ; xmm5=data0L 691 paddd xmm4, xmm3 ; xmm4=data0H 692 psubd xmm0, xmm1 ; xmm0=data7L 693 psubd xmm7, xmm3 ; xmm7=data7H 694 695 movdqa xmm1, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm1=[PD_DESCALE_P2] 696 697 paddd xmm5, xmm1 698 paddd xmm4, xmm1 699 psrad xmm5, DESCALE_P2 700 psrad xmm4, DESCALE_P2 701 paddd xmm0, xmm1 702 paddd xmm7, xmm1 703 psrad xmm0, DESCALE_P2 704 psrad xmm7, DESCALE_P2 705 706 packssdw xmm5, xmm4 ; xmm5=data0=(00 10 20 30 40 50 60 70) 707 packssdw xmm0, xmm7 ; xmm0=data7=(07 17 27 37 47 57 67 77) 708 709 movdqa xmm3, XMMWORD [wk(4)] ; xmm3=tmp11L 710 movdqa xmm1, XMMWORD [wk(5)] ; xmm1=tmp11H 711 712 movdqa xmm4, xmm3 713 movdqa xmm7, xmm1 714 paddd xmm3, xmm2 ; xmm3=data1L 715 paddd xmm1, xmm6 ; xmm1=data1H 716 psubd xmm4, xmm2 ; xmm4=data6L 717 psubd xmm7, xmm6 ; xmm7=data6H 718 719 movdqa xmm2, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm2=[PD_DESCALE_P2] 720 721 paddd xmm3, xmm2 722 paddd xmm1, xmm2 723 psrad xmm3, DESCALE_P2 724 psrad xmm1, DESCALE_P2 725 paddd xmm4, xmm2 726 paddd xmm7, xmm2 727 psrad xmm4, DESCALE_P2 728 psrad xmm7, DESCALE_P2 729 730 packssdw xmm3, xmm1 ; xmm3=data1=(01 11 21 31 41 51 61 71) 731 packssdw xmm4, xmm7 ; xmm4=data6=(06 16 26 36 46 56 66 76) 732 733 packsswb xmm5, xmm4 ; xmm5=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) 734 packsswb xmm3, xmm0 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) 735 736 movdqa xmm6, XMMWORD [wk(6)] ; xmm6=tmp12L 737 movdqa xmm2, XMMWORD [wk(7)] ; xmm2=tmp12H 738 movdqa xmm1, XMMWORD [wk(10)] ; xmm1=tmp1L 739 movdqa xmm7, XMMWORD [wk(11)] ; xmm7=tmp1H 740 741 movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) 742 movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) 743 744 movdqa xmm4, xmm6 745 movdqa xmm0, xmm2 746 paddd xmm6, xmm1 ; xmm6=data2L 747 paddd xmm2, xmm7 ; xmm2=data2H 748 psubd xmm4, xmm1 ; xmm4=data5L 749 psubd xmm0, xmm7 ; xmm0=data5H 750 751 movdqa xmm5, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm5=[PD_DESCALE_P2] 752 753 paddd xmm6, xmm5 754 paddd xmm2, xmm5 755 psrad xmm6, DESCALE_P2 756 psrad xmm2, DESCALE_P2 757 paddd xmm4, xmm5 758 paddd xmm0, xmm5 759 psrad xmm4, DESCALE_P2 760 psrad xmm0, DESCALE_P2 761 762 packssdw xmm6, xmm2 ; xmm6=data2=(02 12 22 32 42 52 62 72) 763 packssdw xmm4, xmm0 ; xmm4=data5=(05 15 25 35 45 55 65 75) 764 765 movdqa xmm3, XMMWORD [wk(2)] ; xmm3=tmp13L 766 movdqa xmm1, XMMWORD [wk(3)] ; xmm1=tmp13H 767 movdqa xmm7, XMMWORD [wk(8)] ; xmm7=tmp0L 768 movdqa xmm5, XMMWORD [wk(9)] ; xmm5=tmp0H 769 770 movdqa xmm2, xmm3 771 movdqa xmm0, xmm1 772 paddd xmm3, xmm7 ; xmm3=data3L 773 paddd xmm1, xmm5 ; xmm1=data3H 774 psubd xmm2, xmm7 ; xmm2=data4L 775 psubd xmm0, xmm5 ; xmm0=data4H 776 777 movdqa xmm7, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm7=[PD_DESCALE_P2] 778 779 paddd xmm3, xmm7 780 paddd xmm1, xmm7 781 psrad xmm3, DESCALE_P2 782 psrad xmm1, DESCALE_P2 783 paddd xmm2, xmm7 784 paddd xmm0, xmm7 785 psrad xmm2, DESCALE_P2 786 psrad xmm0, DESCALE_P2 787 788 movdqa xmm5, [GOTOFF(ebx,PB_CENTERJSAMP)] ; xmm5=[PB_CENTERJSAMP] 789 790 packssdw xmm3, xmm1 ; xmm3=data3=(03 13 23 33 43 53 63 73) 791 packssdw xmm2, xmm0 ; xmm2=data4=(04 14 24 34 44 54 64 74) 792 793 movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) 794 movdqa xmm1, XMMWORD [wk(1)] ; xmm1=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) 795 796 packsswb xmm6, xmm2 ; xmm6=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74) 797 packsswb xmm3, xmm4 ; xmm3=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75) 798 799 paddb xmm7, xmm5 800 paddb xmm1, xmm5 801 paddb xmm6, xmm5 802 paddb xmm3, xmm5 803 804 movdqa xmm0, xmm7 ; transpose coefficients(phase 1) 805 punpcklbw xmm7, xmm1 ; xmm7=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71) 806 punpckhbw xmm0, xmm1 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77) 807 movdqa xmm2, xmm6 ; transpose coefficients(phase 1) 808 punpcklbw xmm6, xmm3 ; xmm6=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73) 809 punpckhbw xmm2, xmm3 ; xmm2=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75) 810 811 movdqa xmm4, xmm7 ; transpose coefficients(phase 2) 812 punpcklwd xmm7, xmm6 ; xmm7=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33) 813 punpckhwd xmm4, xmm6 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73) 814 movdqa xmm5, xmm2 ; transpose coefficients(phase 2) 815 punpcklwd xmm2, xmm0 ; xmm2=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37) 816 punpckhwd xmm5, xmm0 ; xmm5=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77) 817 818 movdqa xmm1, xmm7 ; transpose coefficients(phase 3) 819 punpckldq xmm7, xmm2 ; xmm7=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) 820 punpckhdq xmm1, xmm2 ; xmm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) 821 movdqa xmm3, xmm4 ; transpose coefficients(phase 3) 822 punpckldq xmm4, xmm5 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57) 823 punpckhdq xmm3, xmm5 ; xmm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77) 824 825 pshufd xmm6, xmm7, 0x4E ; xmm6=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07) 826 pshufd xmm0, xmm1, 0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27) 827 pshufd xmm2, xmm4, 0x4E ; xmm2=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47) 828 pshufd xmm5, xmm3, 0x4E ; xmm5=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67) 829 830 mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] 831 mov esi, JSAMPROW [edi+2*SIZEOF_JSAMPROW] 832 movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm7 833 movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm1 834 mov edx, JSAMPROW [edi+4*SIZEOF_JSAMPROW] 835 mov esi, JSAMPROW [edi+6*SIZEOF_JSAMPROW] 836 movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm4 837 movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm3 838 839 mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] 840 mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW] 841 movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6 842 movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm0 843 mov edx, JSAMPROW [edi+5*SIZEOF_JSAMPROW] 844 mov esi, JSAMPROW [edi+7*SIZEOF_JSAMPROW] 845 movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm2 846 movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm5 847 848 pop edi 849 pop esi 850; pop edx ; need not be preserved 851; pop ecx ; unused 852 poppic ebx 853 mov esp, ebp ; esp <- aligned ebp 854 pop esp ; esp <- original ebp 855 pop ebp 856 ret 857 858; For some reason, the OS X linker does not honor the request to align the 859; segment unless we do this. 860 align 32 861