Parent Directory
| 
Revision Log
Revision 1.1 - (view) (download)
| 1 : | edgomez | 1.1 | ;/**************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * - MMX and XMM forward discrete cosine transform - | ||
| 5 : | ; * | ||
| 6 : | ; * Copyright(C) 2001 Peter Ross <pross@xvid.org> | ||
| 7 : | ; * 2004 Andre Werthmann <wertmann@aei.mpg.de> | ||
| 8 : | ; * | ||
| 9 : | ; * This program is free software; you can redistribute it and/or modify it | ||
| 10 : | ; * under the terms of the GNU General Public License as published by | ||
| 11 : | ; * the Free Software Foundation; either version 2 of the License, or | ||
| 12 : | ; * (at your option) any later version. | ||
| 13 : | ; * | ||
| 14 : | ; * This program is distributed in the hope that it will be useful, | ||
| 15 : | ; * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 16 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 17 : | ; * GNU General Public License for more details. | ||
| 18 : | ; * | ||
| 19 : | ; * You should have received a copy of the GNU General Public License | ||
| 20 : | ; * along with this program; if not, write to the Free Software | ||
| 21 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 22 : | ; * | ||
| 23 : | ; * $Id$ | ||
| 24 : | ; * | ||
| 25 : | ; ***************************************************************************/ | ||
| 26 : | |||
| 27 : | ; **************************************************************************** | ||
| 28 : | ; | ||
| 29 : | ; Originally provided by Intel at AP-922 | ||
| 30 : | ; http://developer.intel.com/vtune/cbts/strmsimd/922down.htm | ||
| 31 : | ; (See more app notes at http://developer.intel.com/vtune/cbts/strmsimd/appnotes.htm) | ||
| 32 : | ; but in a limited edition. | ||
| 33 : | ; New macro implements a column part for precise iDCT | ||
| 34 : | ; The routine precision now satisfies IEEE standard 1180-1990. | ||
| 35 : | ; | ||
| 36 : | ; Copyright(C) 2000-2001 Peter Gubanov <peter@elecard.net.ru> | ||
| 37 : | ; Rounding trick Copyright(C) 2000 Michel Lespinasse <walken@zoy.org> | ||
| 38 : | ; | ||
| 39 : | ; http://www.elecard.com/peter/idct.html | ||
| 40 : | ; http://www.linuxvideo.org/mpeg2dec/ | ||
| 41 : | ; | ||
| 42 : | ; ***************************************************************************/ | ||
| 43 : | ; | ||
| 44 : | ; These examples contain code fragments for first stage iDCT 8x8 | ||
| 45 : | ; (for rows) and first stage DCT 8x8 (for columns) | ||
| 46 : | ; | ||
| 47 : | |||
| 48 : | BITS 64 | ||
| 49 : | |||
| 50 : | ;============================================================================= | ||
| 51 : | ; Macros and other preprocessor constants | ||
| 52 : | ;============================================================================= | ||
| 53 : | |||
| 54 : | %macro cglobal 1 | ||
| 55 : | %ifdef PREFIX | ||
| 56 : | %ifdef MARK_FUNCS | ||
| 57 : | global _%1:function %1.endfunc-%1 | ||
| 58 : | %define %1 _%1:function %1.endfunc-%1 | ||
| 59 : | %else | ||
| 60 : | global _%1 | ||
| 61 : | %define %1 _%1 | ||
| 62 : | %endif | ||
| 63 : | %else | ||
| 64 : | %ifdef MARK_FUNCS | ||
| 65 : | global %1:function %1.endfunc-%1 | ||
| 66 : | %else | ||
| 67 : | global %1 | ||
| 68 : | %endif | ||
| 69 : | %endif | ||
| 70 : | %endmacro | ||
| 71 : | |||
| 72 : | %define BITS_INV_ACC 5 ; 4 or 5 for IEEE | ||
| 73 : | %define SHIFT_INV_ROW 16 - BITS_INV_ACC | ||
| 74 : | %define SHIFT_INV_COL 1 + BITS_INV_ACC | ||
| 75 : | %define RND_INV_ROW 1024 * (6 - BITS_INV_ACC) ; 1 << (SHIFT_INV_ROW-1) | ||
| 76 : | %define RND_INV_COL 16 * (BITS_INV_ACC - 3) ; 1 << (SHIFT_INV_COL-1) | ||
| 77 : | %define RND_INV_CORR RND_INV_COL - 1 ; correction -1.0 and round | ||
| 78 : | |||
| 79 : | %define BITS_FRW_ACC 3 ; 2 or 3 for accuracy | ||
| 80 : | %define SHIFT_FRW_COL BITS_FRW_ACC | ||
| 81 : | %define SHIFT_FRW_ROW BITS_FRW_ACC + 17 | ||
| 82 : | %define RND_FRW_ROW 262144*(BITS_FRW_ACC - 1) ; 1 << (SHIFT_FRW_ROW-1) | ||
| 83 : | |||
| 84 : | ;============================================================================= | ||
| 85 : | ; Local Data (Read Only) | ||
| 86 : | ;============================================================================= | ||
| 87 : | |||
| 88 : | %ifdef FORMAT_COFF | ||
| 89 : | SECTION .rodata | ||
| 90 : | %else | ||
| 91 : | SECTION .rodata align=16 | ||
| 92 : | %endif | ||
| 93 : | |||
| 94 : | ;----------------------------------------------------------------------------- | ||
| 95 : | ; Various memory constants (trigonometric values or rounding values) | ||
| 96 : | ;----------------------------------------------------------------------------- | ||
| 97 : | |||
| 98 : | ALIGN 16 | ||
| 99 : | one_corr: | ||
| 100 : | dw 1, 1, 1, 1 | ||
| 101 : | round_inv_row: | ||
| 102 : | dd RND_INV_ROW, RND_INV_ROW | ||
| 103 : | round_inv_col: | ||
| 104 : | dw RND_INV_COL, RND_INV_COL, RND_INV_COL, RND_INV_COL | ||
| 105 : | round_inv_corr: | ||
| 106 : | dw RND_INV_CORR, RND_INV_CORR, RND_INV_CORR, RND_INV_CORR | ||
| 107 : | round_frw_row: | ||
| 108 : | dd RND_FRW_ROW, RND_FRW_ROW | ||
| 109 : | tg_1_16: | ||
| 110 : | dw 13036, 13036, 13036, 13036 ; tg * (2<<16) + 0.5 | ||
| 111 : | tg_2_16: | ||
| 112 : | dw 27146, 27146, 27146, 27146 ; tg * (2<<16) + 0.5 | ||
| 113 : | tg_3_16: | ||
| 114 : | dw -21746, -21746, -21746, -21746 ; tg * (2<<16) + 0.5 | ||
| 115 : | cos_4_16: | ||
| 116 : | dw -19195, -19195, -19195, -19195 ; cos * (2<<16) + 0.5 | ||
| 117 : | ocos_4_16: | ||
| 118 : | dw 23170, 23170, 23170, 23170 ; cos * (2<<15) + 0.5 | ||
| 119 : | otg_3_16: | ||
| 120 : | dw 21895, 21895, 21895, 21895 ; tg * (2<<16) + 0.5 | ||
| 121 : | |||
| 122 : | %if SHIFT_INV_ROW == 12 ; assume SHIFT_INV_ROW == 12 | ||
| 123 : | rounder_0: | ||
| 124 : | dd 65536, 65536 | ||
| 125 : | rounder_4: | ||
| 126 : | dd 0, 0 | ||
| 127 : | rounder_1: | ||
| 128 : | dd 7195, 7195 | ||
| 129 : | rounder_7 | ||
| 130 : | dd 1024, 1024 | ||
| 131 : | rounder_2: | ||
| 132 : | dd 4520, 4520 | ||
| 133 : | rounder_6: | ||
| 134 : | dd 1024, 1024 | ||
| 135 : | rounder_3: | ||
| 136 : | dd 2407, 2407 | ||
| 137 : | rounder_5: | ||
| 138 : | dd 240, 240 | ||
| 139 : | |||
| 140 : | %elif SHIFT_INV_ROW == 11 ; assume SHIFT_INV_ROW == 11 | ||
| 141 : | rounder_0: | ||
| 142 : | dd 65536, 65536 | ||
| 143 : | rounder_4: | ||
| 144 : | dd 0, 0 | ||
| 145 : | rounder_1: | ||
| 146 : | dd 3597, 3597 | ||
| 147 : | rounder_7: | ||
| 148 : | dd 512, 512 | ||
| 149 : | rounder_2: | ||
| 150 : | dd 2260, 2260 | ||
| 151 : | rounder_6: | ||
| 152 : | dd 512, 512 | ||
| 153 : | rounder_3: | ||
| 154 : | dd 1203, 1203 | ||
| 155 : | rounder_5: | ||
| 156 : | dd 120, 120 | ||
| 157 : | %else | ||
| 158 : | |||
| 159 : | %error invalid SHIFT_INV_ROW | ||
| 160 : | |||
| 161 : | %endif | ||
| 162 : | |||
| 163 : | ;----------------------------------------------------------------------------- | ||
| 164 : | ; | ||
| 165 : | ; The first stage iDCT 8x8 - inverse DCTs of rows | ||
| 166 : | ; | ||
| 167 : | ;----------------------------------------------------------------------------- | ||
| 168 : | ; The 8-point inverse DCT direct algorithm | ||
| 169 : | ;----------------------------------------------------------------------------- | ||
| 170 : | ; | ||
| 171 : | ; static const short w[32] = { | ||
| 172 : | ; FIX(cos_4_16), FIX(cos_2_16), FIX(cos_4_16), FIX(cos_6_16), | ||
| 173 : | ; FIX(cos_4_16), FIX(cos_6_16), -FIX(cos_4_16), -FIX(cos_2_16), | ||
| 174 : | ; FIX(cos_4_16), -FIX(cos_6_16), -FIX(cos_4_16), FIX(cos_2_16), | ||
| 175 : | ; FIX(cos_4_16), -FIX(cos_2_16), FIX(cos_4_16), -FIX(cos_6_16), | ||
| 176 : | ; FIX(cos_1_16), FIX(cos_3_16), FIX(cos_5_16), FIX(cos_7_16), | ||
| 177 : | ; FIX(cos_3_16), -FIX(cos_7_16), -FIX(cos_1_16), -FIX(cos_5_16), | ||
| 178 : | ; FIX(cos_5_16), -FIX(cos_1_16), FIX(cos_7_16), FIX(cos_3_16), | ||
| 179 : | ; FIX(cos_7_16), -FIX(cos_5_16), FIX(cos_3_16), -FIX(cos_1_16) }; | ||
| 180 : | ; | ||
| 181 : | ; #define DCT_8_INV_ROW(x, y) | ||
| 182 : | ; { | ||
| 183 : | ; int a0, a1, a2, a3, b0, b1, b2, b3; | ||
| 184 : | ; | ||
| 185 : | ; a0 =x[0]*w[0]+x[2]*w[1]+x[4]*w[2]+x[6]*w[3]; | ||
| 186 : | ; a1 =x[0]*w[4]+x[2]*w[5]+x[4]*w[6]+x[6]*w[7]; | ||
| 187 : | ; a2 = x[0] * w[ 8] + x[2] * w[ 9] + x[4] * w[10] + x[6] * w[11]; | ||
| 188 : | ; a3 = x[0] * w[12] + x[2] * w[13] + x[4] * w[14] + x[6] * w[15]; | ||
| 189 : | ; b0 = x[1] * w[16] + x[3] * w[17] + x[5] * w[18] + x[7] * w[19]; | ||
| 190 : | ; b1 = x[1] * w[20] + x[3] * w[21] + x[5] * w[22] + x[7] * w[23]; | ||
| 191 : | ; b2 = x[1] * w[24] + x[3] * w[25] + x[5] * w[26] + x[7] * w[27]; | ||
| 192 : | ; b3 = x[1] * w[28] + x[3] * w[29] + x[5] * w[30] + x[7] * w[31]; | ||
| 193 : | ; | ||
| 194 : | ; y[0] = SHIFT_ROUND ( a0 + b0 ); | ||
| 195 : | ; y[1] = SHIFT_ROUND ( a1 + b1 ); | ||
| 196 : | ; y[2] = SHIFT_ROUND ( a2 + b2 ); | ||
| 197 : | ; y[3] = SHIFT_ROUND ( a3 + b3 ); | ||
| 198 : | ; y[4] = SHIFT_ROUND ( a3 - b3 ); | ||
| 199 : | ; y[5] = SHIFT_ROUND ( a2 - b2 ); | ||
| 200 : | ; y[6] = SHIFT_ROUND ( a1 - b1 ); | ||
| 201 : | ; y[7] = SHIFT_ROUND ( a0 - b0 ); | ||
| 202 : | ; } | ||
| 203 : | ; | ||
| 204 : | ;----------------------------------------------------------------------------- | ||
| 205 : | ; | ||
| 206 : | ; In this implementation the outputs of the iDCT-1D are multiplied | ||
| 207 : | ; for rows 0,4 - by cos_4_16, | ||
| 208 : | ; for rows 1,7 - by cos_1_16, | ||
| 209 : | ; for rows 2,6 - by cos_2_16, | ||
| 210 : | ; for rows 3,5 - by cos_3_16 | ||
| 211 : | ; and are shifted to the left for better accuracy | ||
| 212 : | ; | ||
| 213 : | ; For the constants used, | ||
| 214 : | ; FIX(float_const) = (short) (float_const * (1<<15) + 0.5) | ||
| 215 : | ; | ||
| 216 : | ;----------------------------------------------------------------------------- | ||
| 217 : | |||
| 218 : | ;----------------------------------------------------------------------------- | ||
| 219 : | ; Tables for xmm processors | ||
| 220 : | ;----------------------------------------------------------------------------- | ||
| 221 : | |||
| 222 : | ; %3 for rows 0,4 - constants are multiplied by cos_4_16 | ||
| 223 : | tab_i_04_xmm: | ||
| 224 : | dw 16384, 21407, 16384, 8867 ; movq-> w05 w04 w01 w00 | ||
| 225 : | dw 16384, 8867, -16384, -21407 ; w07 w06 w03 w02 | ||
| 226 : | dw 16384, -8867, 16384, -21407 ; w13 w12 w09 w08 | ||
| 227 : | dw -16384, 21407, 16384, -8867 ; w15 w14 w11 w10 | ||
| 228 : | dw 22725, 19266, 19266, -4520 ; w21 w20 w17 w16 | ||
| 229 : | dw 12873, 4520, -22725, -12873 ; w23 w22 w19 w18 | ||
| 230 : | dw 12873, -22725, 4520, -12873 ; w29 w28 w25 w24 | ||
| 231 : | dw 4520, 19266, 19266, -22725 ; w31 w30 w27 w26 | ||
| 232 : | |||
| 233 : | ; %3 for rows 1,7 - constants are multiplied by cos_1_16 | ||
| 234 : | tab_i_17_xmm: | ||
| 235 : | dw 22725, 29692, 22725, 12299 ; movq-> w05 w04 w01 w00 | ||
| 236 : | dw 22725, 12299, -22725, -29692 ; w07 w06 w03 w02 | ||
| 237 : | dw 22725, -12299, 22725, -29692 ; w13 w12 w09 w08 | ||
| 238 : | dw -22725, 29692, 22725, -12299 ; w15 w14 w11 w10 | ||
| 239 : | dw 31521, 26722, 26722, -6270 ; w21 w20 w17 w16 | ||
| 240 : | dw 17855, 6270, -31521, -17855 ; w23 w22 w19 w18 | ||
| 241 : | dw 17855, -31521, 6270, -17855 ; w29 w28 w25 w24 | ||
| 242 : | dw 6270, 26722, 26722, -31521 ; w31 w30 w27 w26 | ||
| 243 : | |||
| 244 : | ; %3 for rows 2,6 - constants are multiplied by cos_2_16 | ||
| 245 : | tab_i_26_xmm: | ||
| 246 : | dw 21407, 27969, 21407, 11585 ; movq-> w05 w04 w01 w00 | ||
| 247 : | dw 21407, 11585, -21407, -27969 ; w07 w06 w03 w02 | ||
| 248 : | dw 21407, -11585, 21407, -27969 ; w13 w12 w09 w08 | ||
| 249 : | dw -21407, 27969, 21407, -11585 ; w15 w14 w11 w10 | ||
| 250 : | dw 29692, 25172, 25172, -5906 ; w21 w20 w17 w16 | ||
| 251 : | dw 16819, 5906, -29692, -16819 ; w23 w22 w19 w18 | ||
| 252 : | dw 16819, -29692, 5906, -16819 ; w29 w28 w25 w24 | ||
| 253 : | dw 5906, 25172, 25172, -29692 ; w31 w30 w27 w26 | ||
| 254 : | |||
| 255 : | ; %3 for rows 3,5 - constants are multiplied by cos_3_16 | ||
| 256 : | tab_i_35_xmm: | ||
| 257 : | dw 19266, 25172, 19266, 10426 ; movq-> w05 w04 w01 w00 | ||
| 258 : | dw 19266, 10426, -19266, -25172 ; w07 w06 w03 w02 | ||
| 259 : | dw 19266, -10426, 19266, -25172 ; w13 w12 w09 w08 | ||
| 260 : | dw -19266, 25172, 19266, -10426 ; w15 w14 w11 w10 | ||
| 261 : | dw 26722, 22654, 22654, -5315 ; w21 w20 w17 w16 | ||
| 262 : | dw 15137, 5315, -26722, -15137 ; w23 w22 w19 w18 | ||
| 263 : | dw 15137, -26722, 5315, -15137 ; w29 w28 w25 w24 | ||
| 264 : | dw 5315, 22654, 22654, -26722 ; w31 w30 w27 w26 | ||
| 265 : | |||
| 266 : | ;============================================================================= | ||
| 267 : | ; Helper macros for the code | ||
| 268 : | ;============================================================================= | ||
| 269 : | |||
| 270 : | ;----------------------------------------------------------------------------- | ||
| 271 : | ; DCT_8_INV_ROW_XMM INP, OUT, TABLE, ROUNDER | ||
| 272 : | ;----------------------------------------------------------------------------- | ||
| 273 : | |||
| 274 : | %macro DCT_8_INV_ROW_XMM 4 | ||
| 275 : | movq mm0, [%1] ; 0 ; x3 x2 x1 x0 | ||
| 276 : | movq mm1, [%1+8] ; 1 ; x7 x6 x5 x4 | ||
| 277 : | movq mm2, mm0 ; 2 ; x3 x2 x1 x0 | ||
| 278 : | movq mm3, [%3] ; 3 ; w05 w04 w01 w00 | ||
| 279 : | pshufw mm0, mm0, 10001000b ; x2 x0 x2 x0 | ||
| 280 : | movq mm4, [%3+8] ; 4 ; w07 w06 w03 w02 | ||
| 281 : | movq mm5, mm1 ; 5 ; x7 x6 x5 x4 | ||
| 282 : | pmaddwd mm3, mm0 ; x2*w05+x0*w04 x2*w01+x0*w00 | ||
| 283 : | movq mm6, [%3+32] ; 6 ; w21 w20 w17 w16 | ||
| 284 : | pshufw mm1, mm1, 10001000b ; x6 x4 x6 x4 | ||
| 285 : | pmaddwd mm4, mm1 ; x6*w07+x4*w06 x6*w03+x4*w02 | ||
| 286 : | movq mm7, [%3+40] ; 7 ; w23 w22 w19 w18 | ||
| 287 : | pshufw mm2, mm2, 11011101b ; x3 x1 x3 x1 | ||
| 288 : | pmaddwd mm6, mm2 ; x3*w21+x1*w20 x3*w17+x1*w16 | ||
| 289 : | pshufw mm5, mm5, 11011101b ; x7 x5 x7 x5 | ||
| 290 : | pmaddwd mm7, mm5 ; x7*w23+x5*w22 x7*w19+x5*w18 | ||
| 291 : | paddd mm3, [%4] ; +%4 | ||
| 292 : | pmaddwd mm0, [%3+16] ; x2*w13+x0*w12 x2*w09+x0*w08 | ||
| 293 : | paddd mm3, mm4 ; 4 ; a1=sum(even1) a0=sum(even0) | ||
| 294 : | pmaddwd mm1, [%3+24] ; x6*w15+x4*w14 x6*w11+x4*w10 | ||
| 295 : | movq mm4, mm3 ; 4 ; a1 a0 | ||
| 296 : | pmaddwd mm2, [%3+48] ; x3*w29+x1*w28 x3*w25+x1*w24 | ||
| 297 : | paddd mm6, mm7 ; 7 ; b1=sum(odd1) b0=sum(odd0) | ||
| 298 : | pmaddwd mm5, [%3+56] ; x7*w31+x5*w30 x7*w27+x5*w26 | ||
| 299 : | paddd mm3, mm6 ; a1+b1 a0+b0 | ||
| 300 : | paddd mm0, [%4] ; +%4 | ||
| 301 : | psrad mm3, SHIFT_INV_ROW ; y1=a1+b1 y0=a0+b0 | ||
| 302 : | paddd mm0, mm1 ; 1 ; a3=sum(even3) a2=sum(even2) | ||
| 303 : | psubd mm4, mm6 ; 6 ; a1-b1 a0-b0 | ||
| 304 : | movq mm7, mm0 ; 7 ; a3 a2 | ||
| 305 : | paddd mm2, mm5 ; 5 ; b3=sum(odd3) b2=sum(odd2) | ||
| 306 : | paddd mm0, mm2 ; a3+b3 a2+b2 | ||
| 307 : | psrad mm4, SHIFT_INV_ROW ; y6=a1-b1 y7=a0-b0 | ||
| 308 : | psubd mm7, mm2 ; 2 ; a3-b3 a2-b2 | ||
| 309 : | psrad mm0, SHIFT_INV_ROW ; y3=a3+b3 y2=a2+b2 | ||
| 310 : | psrad mm7, SHIFT_INV_ROW ; y4=a3-b3 y5=a2-b2 | ||
| 311 : | packssdw mm3, mm0 ; 0 ; y3 y2 y1 y0 | ||
| 312 : | packssdw mm7, mm4 ; 4 ; y6 y7 y4 y5 | ||
| 313 : | movq [%2], mm3 ; 3 ; save y3 y2 y1 y0 | ||
| 314 : | pshufw mm7, mm7, 10110001b ; y7 y6 y5 y4 | ||
| 315 : | movq [%2+8], mm7 ; 7 ; save y7 y6 y5 y4 | ||
| 316 : | %endmacro | ||
| 317 : | |||
| 318 : | ;----------------------------------------------------------------------------- | ||
| 319 : | ; | ||
| 320 : | ; The first stage DCT 8x8 - forward DCTs of columns | ||
| 321 : | ; | ||
| 322 : | ; The %2puts are multiplied | ||
| 323 : | ; for rows 0,4 - on cos_4_16, | ||
| 324 : | ; for rows 1,7 - on cos_1_16, | ||
| 325 : | ; for rows 2,6 - on cos_2_16, | ||
| 326 : | ; for rows 3,5 - on cos_3_16 | ||
| 327 : | ; and are shifted to the left for rise of accuracy | ||
| 328 : | ; | ||
| 329 : | ;----------------------------------------------------------------------------- | ||
| 330 : | ; | ||
| 331 : | ; The 8-point scaled forward DCT algorithm (26a8m) | ||
| 332 : | ; | ||
| 333 : | ;----------------------------------------------------------------------------- | ||
| 334 : | ; | ||
| 335 : | ; #define DCT_8_FRW_COL(x, y) | ||
| 336 : | ;{ | ||
| 337 : | ; short t0, t1, t2, t3, t4, t5, t6, t7; | ||
| 338 : | ; short tp03, tm03, tp12, tm12, tp65, tm65; | ||
| 339 : | ; short tp465, tm465, tp765, tm765; | ||
| 340 : | ; | ||
| 341 : | ; t0 = LEFT_SHIFT ( x[0] + x[7] ); | ||
| 342 : | ; t1 = LEFT_SHIFT ( x[1] + x[6] ); | ||
| 343 : | ; t2 = LEFT_SHIFT ( x[2] + x[5] ); | ||
| 344 : | ; t3 = LEFT_SHIFT ( x[3] + x[4] ); | ||
| 345 : | ; t4 = LEFT_SHIFT ( x[3] - x[4] ); | ||
| 346 : | ; t5 = LEFT_SHIFT ( x[2] - x[5] ); | ||
| 347 : | ; t6 = LEFT_SHIFT ( x[1] - x[6] ); | ||
| 348 : | ; t7 = LEFT_SHIFT ( x[0] - x[7] ); | ||
| 349 : | ; | ||
| 350 : | ; tp03 = t0 + t3; | ||
| 351 : | ; tm03 = t0 - t3; | ||
| 352 : | ; tp12 = t1 + t2; | ||
| 353 : | ; tm12 = t1 - t2; | ||
| 354 : | ; | ||
| 355 : | ; y[0] = tp03 + tp12; | ||
| 356 : | ; y[4] = tp03 - tp12; | ||
| 357 : | ; | ||
| 358 : | ; y[2] = tm03 + tm12 * tg_2_16; | ||
| 359 : | ; y[6] = tm03 * tg_2_16 - tm12; | ||
| 360 : | ; | ||
| 361 : | ; tp65 =(t6 +t5 )*cos_4_16; | ||
| 362 : | ; tm65 =(t6 -t5 )*cos_4_16; | ||
| 363 : | ; | ||
| 364 : | ; tp765 = t7 + tp65; | ||
| 365 : | ; tm765 = t7 - tp65; | ||
| 366 : | ; tp465 = t4 + tm65; | ||
| 367 : | ; tm465 = t4 - tm65; | ||
| 368 : | ; | ||
| 369 : | ; y[1] = tp765 + tp465 * tg_1_16; | ||
| 370 : | ; y[7] = tp765 * tg_1_16 - tp465; | ||
| 371 : | ; y[5] = tm765 * tg_3_16 + tm465; | ||
| 372 : | ; y[3] = tm765 - tm465 * tg_3_16; | ||
| 373 : | ;} | ||
| 374 : | ; | ||
| 375 : | ;----------------------------------------------------------------------------- | ||
| 376 : | |||
| 377 : | ;----------------------------------------------------------------------------- | ||
| 378 : | ; DCT_8_INV_COL_4 INP,OUT | ||
| 379 : | ;----------------------------------------------------------------------------- | ||
| 380 : | |||
| 381 : | %macro DCT_8_INV_COL 2 | ||
| 382 : | movq mm0, [tg_3_16 wrt rip] | ||
| 383 : | movq mm3, [%1+16*3] | ||
| 384 : | movq mm1, mm0 ; tg_3_16 | ||
| 385 : | movq mm5, [%1+16*5] | ||
| 386 : | pmulhw mm0, mm3 ; x3*(tg_3_16-1) | ||
| 387 : | movq mm4, [tg_1_16 wrt rip] | ||
| 388 : | pmulhw mm1, mm5 ; x5*(tg_3_16-1) | ||
| 389 : | movq mm7, [%1+16*7] | ||
| 390 : | movq mm2, mm4 ; tg_1_16 | ||
| 391 : | movq mm6, [%1+16*1] | ||
| 392 : | pmulhw mm4, mm7 ; x7*tg_1_16 | ||
| 393 : | paddsw mm0, mm3 ; x3*tg_3_16 | ||
| 394 : | pmulhw mm2, mm6 ; x1*tg_1_16 | ||
| 395 : | paddsw mm1, mm3 ; x3+x5*(tg_3_16-1) | ||
| 396 : | psubsw mm0, mm5 ; x3*tg_3_16-x5 = tm35 | ||
| 397 : | movq mm3, [ocos_4_16 wrt rip] | ||
| 398 : | paddsw mm1, mm5 ; x3+x5*tg_3_16 = tp35 | ||
| 399 : | paddsw mm4, mm6 ; x1+tg_1_16*x7 = tp17 | ||
| 400 : | psubsw mm2, mm7 ; x1*tg_1_16-x7 = tm17 | ||
| 401 : | movq mm5, mm4 ; tp17 | ||
| 402 : | movq mm6, mm2 ; tm17 | ||
| 403 : | paddsw mm5, mm1 ; tp17+tp35 = b0 | ||
| 404 : | psubsw mm6, mm0 ; tm17-tm35 = b3 | ||
| 405 : | psubsw mm4, mm1 ; tp17-tp35 = t1 | ||
| 406 : | paddsw mm2, mm0 ; tm17+tm35 = t2 | ||
| 407 : | movq mm7, [tg_2_16 wrt rip] | ||
| 408 : | movq mm1, mm4 ; t1 | ||
| 409 : | movq [%2+3*16], mm5 ; save b0 | ||
| 410 : | paddsw mm1, mm2 ; t1+t2 | ||
| 411 : | movq [%2+5*16], mm6 ; save b3 | ||
| 412 : | psubsw mm4, mm2 ; t1-t2 | ||
| 413 : | movq mm5, [%1+2*16] | ||
| 414 : | movq mm0, mm7 ; tg_2_16 | ||
| 415 : | movq mm6, [%1+6*16] | ||
| 416 : | pmulhw mm0, mm5 ; x2*tg_2_16 | ||
| 417 : | pmulhw mm7, mm6 ; x6*tg_2_16 | ||
| 418 : | ; slot | ||
| 419 : | pmulhw mm1, mm3 ; ocos_4_16*(t1+t2) = b1/2 | ||
| 420 : | ; slot | ||
| 421 : | movq mm2, [%1+0*16] | ||
| 422 : | pmulhw mm4, mm3 ; ocos_4_16*(t1-t2) = b2/2 | ||
| 423 : | psubsw mm0, mm6 ; t2*tg_2_16-x6 = tm26 | ||
| 424 : | movq mm3, mm2 ; x0 | ||
| 425 : | movq mm6, [%1+4*16] | ||
| 426 : | paddsw mm7, mm5 ; x2+x6*tg_2_16 = tp26 | ||
| 427 : | paddsw mm2, mm6 ; x0+x4 = tp04 | ||
| 428 : | psubsw mm3, mm6 ; x0-x4 = tm04 | ||
| 429 : | movq mm5, mm2 ; tp04 | ||
| 430 : | movq mm6, mm3 ; tm04 | ||
| 431 : | psubsw mm2, mm7 ; tp04-tp26 = a3 | ||
| 432 : | paddsw mm3, mm0 ; tm04+tm26 = a1 | ||
| 433 : | paddsw mm1, mm1 ; b1 | ||
| 434 : | paddsw mm4, mm4 ; b2 | ||
| 435 : | paddsw mm5, mm7 ; tp04+tp26 = a0 | ||
| 436 : | psubsw mm6, mm0 ; tm04-tm26 = a2 | ||
| 437 : | movq mm7, mm3 ; a1 | ||
| 438 : | movq mm0, mm6 ; a2 | ||
| 439 : | paddsw mm3, mm1 ; a1+b1 | ||
| 440 : | paddsw mm6, mm4 ; a2+b2 | ||
| 441 : | psraw mm3, SHIFT_INV_COL ; dst1 | ||
| 442 : | psubsw mm7, mm1 ; a1-b1 | ||
| 443 : | psraw mm6, SHIFT_INV_COL ; dst2 | ||
| 444 : | psubsw mm0, mm4 ; a2-b2 | ||
| 445 : | movq mm1, [%2+3*16] ; load b0 | ||
| 446 : | psraw mm7, SHIFT_INV_COL ; dst6 | ||
| 447 : | movq mm4, mm5 ; a0 | ||
| 448 : | psraw mm0, SHIFT_INV_COL ; dst5 | ||
| 449 : | movq [%2+1*16], mm3 | ||
| 450 : | paddsw mm5, mm1 ; a0+b0 | ||
| 451 : | movq [%2+2*16], mm6 | ||
| 452 : | psubsw mm4, mm1 ; a0-b0 | ||
| 453 : | movq mm3, [%2+5*16] ; load b3 | ||
| 454 : | psraw mm5, SHIFT_INV_COL ; dst0 | ||
| 455 : | movq mm6, mm2 ; a3 | ||
| 456 : | psraw mm4, SHIFT_INV_COL ; dst7 | ||
| 457 : | movq [%2+5*16], mm0 | ||
| 458 : | paddsw mm2, mm3 ; a3+b3 | ||
| 459 : | movq [%2+6*16], mm7 | ||
| 460 : | psubsw mm6, mm3 ; a3-b3 | ||
| 461 : | movq [%2+0*16], mm5 | ||
| 462 : | psraw mm2, SHIFT_INV_COL ; dst3 | ||
| 463 : | movq [%2+7*16], mm4 | ||
| 464 : | psraw mm6, SHIFT_INV_COL ; dst4 | ||
| 465 : | movq [%2+3*16], mm2 | ||
| 466 : | movq [%2+4*16], mm6 | ||
| 467 : | %endmacro | ||
| 468 : | |||
| 469 : | ;============================================================================= | ||
| 470 : | ; Code | ||
| 471 : | ;============================================================================= | ||
| 472 : | |||
| 473 : | SECTION .text align=16 | ||
| 474 : | |||
| 475 : | cglobal idct_x86_64 | ||
| 476 : | |||
| 477 : | ;----------------------------------------------------------------------------- | ||
| 478 : | ; void idct_xmm(uint16_t block[64]); | ||
| 479 : | ;----------------------------------------------------------------------------- | ||
| 480 : | |||
| 481 : | ALIGN 16 | ||
| 482 : | idct_x86_64: | ||
| 483 : | mov rax, rdi ; block[] | ||
| 484 : | |||
| 485 : | ;; Process each row | ||
| 486 : | lea rsi, [tab_i_04_xmm wrt rip] | ||
| 487 : | lea rdi, [rounder_0 wrt rip] | ||
| 488 : | DCT_8_INV_ROW_XMM rax+0*16, rax+0*16, rsi, rdi | ||
| 489 : | |||
| 490 : | lea rsi, [tab_i_17_xmm wrt rip] | ||
| 491 : | lea rdi, [rounder_1 wrt rip] | ||
| 492 : | DCT_8_INV_ROW_XMM rax+1*16, rax+1*16, rsi, rdi | ||
| 493 : | |||
| 494 : | lea rsi, [tab_i_26_xmm wrt rip] | ||
| 495 : | lea rdi, [rounder_2 wrt rip] | ||
| 496 : | DCT_8_INV_ROW_XMM rax+2*16, rax+2*16, rsi, rdi | ||
| 497 : | |||
| 498 : | lea rsi, [tab_i_35_xmm wrt rip] | ||
| 499 : | lea rdi, [rounder_3 wrt rip] | ||
| 500 : | DCT_8_INV_ROW_XMM rax+3*16, rax+3*16, rsi, rdi | ||
| 501 : | |||
| 502 : | lea rsi, [tab_i_04_xmm wrt rip] | ||
| 503 : | lea rdi, [rounder_4 wrt rip] | ||
| 504 : | DCT_8_INV_ROW_XMM rax+4*16, rax+4*16, rsi, rdi | ||
| 505 : | |||
| 506 : | lea rsi, [tab_i_35_xmm wrt rip] | ||
| 507 : | lea rdi, [rounder_5 wrt rip] | ||
| 508 : | DCT_8_INV_ROW_XMM rax+5*16, rax+5*16, rsi, rdi | ||
| 509 : | |||
| 510 : | lea rsi, [tab_i_26_xmm wrt rip] | ||
| 511 : | lea rdi, [rounder_6 wrt rip] | ||
| 512 : | DCT_8_INV_ROW_XMM rax+6*16, rax+6*16, rsi, rdi | ||
| 513 : | |||
| 514 : | lea rsi, [tab_i_17_xmm wrt rip] | ||
| 515 : | lea rdi, [rounder_7 wrt rip] | ||
| 516 : | DCT_8_INV_ROW_XMM rax+7*16, rax+7*16, rsi, rdi | ||
| 517 : | |||
| 518 : | |||
| 519 : | ;; Process the columns (4 at a time) | ||
| 520 : | DCT_8_INV_COL rax+0, rax+0 | ||
| 521 : | DCT_8_INV_COL rax+8, rax+8 | ||
| 522 : | |||
| 523 : | ret | ||
| 524 : | .endfunc | ||
| 525 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |