Parent Directory
| 
Revision Log
Revision 1793 - (view) (download)
| 1 : | edgomez | 1382 | ;/**************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * - SSE2 forward discrete cosine transform - | ||
| 5 : | ; * | ||
| 6 : | ; * Copyright(C) 2003 Pascal Massimino <skal@planet-d.net> | ||
| 7 : | ; * | ||
| 8 : | ; * This program is free software; you can redistribute it and/or modify it | ||
| 9 : | ; * under the terms of the GNU General Public License as published by | ||
| 10 : | ; * the Free Software Foundation; either version 2 of the License, or | ||
| 11 : | ; * (at your option) any later version. | ||
| 12 : | ; * | ||
| 13 : | ; * This program is distributed in the hope that it will be useful, | ||
| 14 : | ; * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 15 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 16 : | ; * GNU General Public License for more details. | ||
| 17 : | ; * | ||
| 18 : | ; * You should have received a copy of the GNU General Public License | ||
| 19 : | ; * along with this program; if not, write to the Free Software | ||
| 20 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 21 : | ; * | ||
| 22 : | Isibaar | 1793 | ; * $Id: fdct_sse2_skal.asm,v 1.9 2008-11-11 20:46:24 Isibaar Exp $ |
| 23 : | edgomez | 1382 | ; * |
| 24 : | ; ***************************************************************************/ | ||
| 25 : | |||
| 26 : | BITS 32 | ||
| 27 : | |||
| 28 : | %macro cglobal 1 | ||
| 29 : | edgomez | 1535 | %ifdef PREFIX |
| 30 : | %ifdef MARK_FUNCS | ||
| 31 : | edgomez | 1540 | global _%1:function %1.endfunc-%1 |
| 32 : | %define %1 _%1:function %1.endfunc-%1 | ||
| 33 : | Isibaar | 1793 | %define ENDFUNC .endfunc |
| 34 : | edgomez | 1535 | %else |
| 35 : | global _%1 | ||
| 36 : | %define %1 _%1 | ||
| 37 : | Isibaar | 1793 | %define ENDFUNC |
| 38 : | edgomez | 1535 | %endif |
| 39 : | %else | ||
| 40 : | %ifdef MARK_FUNCS | ||
| 41 : | edgomez | 1540 | global %1:function %1.endfunc-%1 |
| 42 : | Isibaar | 1793 | %define ENDFUNC .endfunc |
| 43 : | edgomez | 1535 | %else |
| 44 : | global %1 | ||
| 45 : | Isibaar | 1793 | %define ENDFUNC |
| 46 : | edgomez | 1535 | %endif |
| 47 : | %endif | ||
| 48 : | edgomez | 1382 | %endmacro |
| 49 : | |||
| 50 : | ;----------------------------------------------------------------------------- | ||
| 51 : | ; | ||
| 52 : | ; -=FDCT=- | ||
| 53 : | ; | ||
| 54 : | ; Vertical pass is an implementation of the scheme: | ||
| 55 : | ; Loeffler C., Ligtenberg A., and Moschytz C.S.: | ||
| 56 : | ; Practical Fast 1D DCT Algorithm with Eleven Multiplications, | ||
| 57 : | ; Proc. ICASSP 1989, 988-991. | ||
| 58 : | ; | ||
| 59 : | ; Horizontal pass is a double 4x4 vector/matrix multiplication, | ||
| 60 : | ; (see also Intel's Application Note 922: | ||
| 61 : | ; http://developer.intel.com/vtune/cbts/strmsimd/922down.htm | ||
| 62 : | ; Copyright (C) 1999 Intel Corporation) | ||
| 63 : | ; | ||
| 64 : | ; Notes: | ||
| 65 : | ; * tan(3pi/16) is greater than 0.5, and would use the | ||
| 66 : | ; sign bit when turned into 16b fixed-point precision. So, | ||
| 67 : | ; we use the trick: x*tan3 = x*(tan3-1)+x | ||
| 68 : | ; | ||
| 69 : | ; * There's only one SSE-specific instruction (pshufw). | ||
| 70 : | ; | ||
| 71 : | ; * There's still 1 or 2 ticks to save in fLLM_PASS, but | ||
| 72 : | ; I prefer having a readable code, instead of a tightly | ||
| 73 : | ; scheduled one... | ||
| 74 : | ; | ||
| 75 : | ; * Quantization stage (as well as pre-transposition for the | ||
| 76 : | ; idct way back) can be included in the fTab* constants | ||
| 77 : | ; (with induced loss of precision, somehow) | ||
| 78 : | ; | ||
| 79 : | ; * Some more details at: http://skal.planet-d.net/coding/dct.html | ||
| 80 : | ; | ||
| 81 : | ;----------------------------------------------------------------------------- | ||
| 82 : | ; | ||
| 83 : | ; -=IDCT=- | ||
| 84 : | ; | ||
| 85 : | ; A little slower than fdct, because the final stages (butterflies and | ||
| 86 : | ; descaling) require some unpairable shifting and packing, all on | ||
| 87 : | ; the same CPU unit. | ||
| 88 : | ; | ||
| 89 : | ;----------------------------------------------------------------------------- | ||
| 90 : | |||
| 91 : | ;============================================================================= | ||
| 92 : | ; Read only data | ||
| 93 : | ;============================================================================= | ||
| 94 : | |||
| 95 : | %ifdef FORMAT_COFF | ||
| 96 : | edgomez | 1519 | SECTION .rodata |
| 97 : | edgomez | 1382 | %else |
| 98 : | edgomez | 1519 | SECTION .rodata align=16 |
| 99 : | edgomez | 1382 | %endif |
| 100 : | |||
| 101 : | ALIGN 16 | ||
| 102 : | tan1: times 8 dw 0x32ec ; tan( pi/16) | ||
| 103 : | tan2: times 8 dw 0x6a0a ; tan(2pi/16) (=sqrt(2)-1) | ||
| 104 : | tan3: times 8 dw 0xab0e ; tan(3pi/16)-1 | ||
| 105 : | sqrt2: times 8 dw 0x5a82 ; 0.5/sqrt(2) | ||
| 106 : | |||
| 107 : | ;----------------------------------------------------------------------------- | ||
| 108 : | ; Inverse DCT tables | ||
| 109 : | ;----------------------------------------------------------------------------- | ||
| 110 : | |||
| 111 : | Isibaar | 1627 | ALIGN 16 |
| 112 : | edgomez | 1382 | iTab1: |
| 113 : | dw 0x4000, 0x539f, 0x4000, 0x22a3 | ||
| 114 : | dw 0x4000, 0xdd5d, 0x4000, 0xac61 | ||
| 115 : | dw 0x4000, 0x22a3, 0xc000, 0xac61 | ||
| 116 : | dw 0xc000, 0x539f, 0x4000, 0xdd5d | ||
| 117 : | dw 0x58c5, 0x4b42, 0x4b42, 0xee58 | ||
| 118 : | dw 0x3249, 0xa73b, 0x11a8, 0xcdb7 | ||
| 119 : | dw 0x3249, 0x11a8, 0xa73b, 0xcdb7 | ||
| 120 : | dw 0x11a8, 0x4b42, 0x4b42, 0xa73b | ||
| 121 : | |||
| 122 : | iTab2: | ||
| 123 : | dw 0x58c5, 0x73fc, 0x58c5, 0x300b | ||
| 124 : | dw 0x58c5, 0xcff5, 0x58c5, 0x8c04 | ||
| 125 : | dw 0x58c5, 0x300b, 0xa73b, 0x8c04 | ||
| 126 : | dw 0xa73b, 0x73fc, 0x58c5, 0xcff5 | ||
| 127 : | dw 0x7b21, 0x6862, 0x6862, 0xe782 | ||
| 128 : | dw 0x45bf, 0x84df, 0x187e, 0xba41 | ||
| 129 : | dw 0x45bf, 0x187e, 0x84df, 0xba41 | ||
| 130 : | dw 0x187e, 0x6862, 0x6862, 0x84df | ||
| 131 : | |||
| 132 : | iTab3: | ||
| 133 : | dw 0x539f, 0x6d41, 0x539f, 0x2d41 | ||
| 134 : | dw 0x539f, 0xd2bf, 0x539f, 0x92bf | ||
| 135 : | dw 0x539f, 0x2d41, 0xac61, 0x92bf | ||
| 136 : | dw 0xac61, 0x6d41, 0x539f, 0xd2bf | ||
| 137 : | dw 0x73fc, 0x6254, 0x6254, 0xe8ee | ||
| 138 : | dw 0x41b3, 0x8c04, 0x1712, 0xbe4d | ||
| 139 : | dw 0x41b3, 0x1712, 0x8c04, 0xbe4d | ||
| 140 : | dw 0x1712, 0x6254, 0x6254, 0x8c04 | ||
| 141 : | |||
| 142 : | iTab4: | ||
| 143 : | dw 0x4b42, 0x6254, 0x4b42, 0x28ba | ||
| 144 : | dw 0x4b42, 0xd746, 0x4b42, 0x9dac | ||
| 145 : | dw 0x4b42, 0x28ba, 0xb4be, 0x9dac | ||
| 146 : | dw 0xb4be, 0x6254, 0x4b42, 0xd746 | ||
| 147 : | dw 0x6862, 0x587e, 0x587e, 0xeb3d | ||
| 148 : | dw 0x3b21, 0x979e, 0x14c3, 0xc4df | ||
| 149 : | dw 0x3b21, 0x14c3, 0x979e, 0xc4df | ||
| 150 : | dw 0x14c3, 0x587e, 0x587e, 0x979e | ||
| 151 : | |||
| 152 : | Isibaar | 1627 | ALIGN 16 |
| 153 : | Skal | 1618 | Walken_Idct_Rounders: |
| 154 : | dd 65536, 65536, 65536, 65536 | ||
| 155 : | dd 3597, 3597, 3597, 3597 | ||
| 156 : | dd 2260, 2260, 2260, 2260 | ||
| 157 : | dd 1203, 1203, 1203, 1203 | ||
| 158 : | dd 0, 0, 0, 0 | ||
| 159 : | dd 120, 120, 120, 120 | ||
| 160 : | dd 512, 512, 512, 512 | ||
| 161 : | dd 512, 512, 512, 512 | ||
| 162 : | edgomez | 1382 | |
| 163 : | Skal | 1618 | times 8 dw (65536>>11) |
| 164 : | times 8 dw ( 3597>>11) | ||
| 165 : | times 8 dw ( 2260>>11) | ||
| 166 : | edgomez | 1382 | ; other rounders are zero... |
| 167 : | |||
| 168 : | ;----------------------------------------------------------------------------- | ||
| 169 : | ; Forward DCT tables | ||
| 170 : | ;----------------------------------------------------------------------------- | ||
| 171 : | |||
| 172 : | ALIGN 16 | ||
| 173 : | fTab1: | ||
| 174 : | dw 0x4000, 0x4000, 0x58c5, 0x4b42, | ||
| 175 : | dw 0xdd5d, 0xac61, 0xa73b, 0xcdb7, | ||
| 176 : | dw 0x4000, 0x4000, 0x3249, 0x11a8, | ||
| 177 : | dw 0x539f, 0x22a3, 0x4b42, 0xee58, | ||
| 178 : | dw 0x4000, 0xc000, 0x3249, 0xa73b, | ||
| 179 : | dw 0x539f, 0xdd5d, 0x4b42, 0xa73b, | ||
| 180 : | dw 0xc000, 0x4000, 0x11a8, 0x4b42, | ||
| 181 : | dw 0x22a3, 0xac61, 0x11a8, 0xcdb7 | ||
| 182 : | |||
| 183 : | fTab2: | ||
| 184 : | dw 0x58c5, 0x58c5, 0x7b21, 0x6862, | ||
| 185 : | dw 0xcff5, 0x8c04, 0x84df, 0xba41, | ||
| 186 : | dw 0x58c5, 0x58c5, 0x45bf, 0x187e, | ||
| 187 : | dw 0x73fc, 0x300b, 0x6862, 0xe782, | ||
| 188 : | dw 0x58c5, 0xa73b, 0x45bf, 0x84df, | ||
| 189 : | dw 0x73fc, 0xcff5, 0x6862, 0x84df, | ||
| 190 : | dw 0xa73b, 0x58c5, 0x187e, 0x6862, | ||
| 191 : | dw 0x300b, 0x8c04, 0x187e, 0xba41 | ||
| 192 : | |||
| 193 : | fTab3: | ||
| 194 : | dw 0x539f, 0x539f, 0x73fc, 0x6254, | ||
| 195 : | dw 0xd2bf, 0x92bf, 0x8c04, 0xbe4d, | ||
| 196 : | dw 0x539f, 0x539f, 0x41b3, 0x1712, | ||
| 197 : | dw 0x6d41, 0x2d41, 0x6254, 0xe8ee, | ||
| 198 : | dw 0x539f, 0xac61, 0x41b3, 0x8c04, | ||
| 199 : | dw 0x6d41, 0xd2bf, 0x6254, 0x8c04, | ||
| 200 : | dw 0xac61, 0x539f, 0x1712, 0x6254, | ||
| 201 : | dw 0x2d41, 0x92bf, 0x1712, 0xbe4d | ||
| 202 : | |||
| 203 : | fTab4: | ||
| 204 : | dw 0x4b42, 0x4b42, 0x6862, 0x587e, | ||
| 205 : | dw 0xd746, 0x9dac, 0x979e, 0xc4df, | ||
| 206 : | dw 0x4b42, 0x4b42, 0x3b21, 0x14c3, | ||
| 207 : | dw 0x6254, 0x28ba, 0x587e, 0xeb3d, | ||
| 208 : | dw 0x4b42, 0xb4be, 0x3b21, 0x979e, | ||
| 209 : | dw 0x6254, 0xd746, 0x587e, 0x979e, | ||
| 210 : | dw 0xb4be, 0x4b42, 0x14c3, 0x587e, | ||
| 211 : | dw 0x28ba, 0x9dac, 0x14c3, 0xc4df | ||
| 212 : | |||
| 213 : | |||
| 214 : | ALIGN 16 | ||
| 215 : | Fdct_Rnd0: dw 6,8,8,8, 6,8,8,8 | ||
| 216 : | Fdct_Rnd1: dw 8,8,8,8, 8,8,8,8 | ||
| 217 : | Fdct_Rnd2: dw 10,8,8,8, 8,8,8,8 | ||
| 218 : | Rounder1: dw 1,1,1,1, 1,1,1,1 | ||
| 219 : | |||
| 220 : | ;============================================================================= | ||
| 221 : | ; Code | ||
| 222 : | ;============================================================================= | ||
| 223 : | |||
| 224 : | SECTION .text | ||
| 225 : | |||
| 226 : | cglobal idct_sse2_skal | ||
| 227 : | cglobal fdct_sse2_skal | ||
| 228 : | |||
| 229 : | ;----------------------------------------------------------------------------- | ||
| 230 : | ; Helper macro iMTX_MULT | ||
| 231 : | ;----------------------------------------------------------------------------- | ||
| 232 : | |||
| 233 : | %macro iMTX_MULT 4 ; %1=src, %2 = Table to use, %3=rounder, %4=Shift | ||
| 234 : | |||
| 235 : | movdqa xmm0, [ecx+%1*16] ; xmm0 = [01234567] | ||
| 236 : | |||
| 237 : | Skal | 1618 | pshuflw xmm0, xmm0, 11011000b ; [02134567] ; these two shufflings could be |
| 238 : | pshufhw xmm0, xmm0, 11011000b ; [02134657] ; integrated in zig-zag orders | ||
| 239 : | |||
| 240 : | edgomez | 1382 | pshufd xmm4, xmm0, 00000000b ; [02020202] |
| 241 : | pshufd xmm5, xmm0, 10101010b ; [46464646] | ||
| 242 : | pshufd xmm6, xmm0, 01010101b ; [13131313] | ||
| 243 : | pshufd xmm7, xmm0, 11111111b ; [57575757] | ||
| 244 : | |||
| 245 : | pmaddwd xmm4, [%2+ 0] ; dot [M00,M01][M04,M05][M08,M09][M12,M13] | ||
| 246 : | pmaddwd xmm5, [%2+16] ; dot [M02,M03][M06,M07][M10,M11][M14,M15] | ||
| 247 : | pmaddwd xmm6, [%2+32] ; dot [M16,M17][M20,M21][M24,M25][M28,M29] | ||
| 248 : | pmaddwd xmm7, [%2+48] ; dot [M18,M19][M22,M23][M26,M27][M30,M31] | ||
| 249 : | paddd xmm4, [%3] ; Round | ||
| 250 : | |||
| 251 : | paddd xmm6, xmm7 ; [b0|b1|b2|b3] | ||
| 252 : | paddd xmm4, xmm5 ; [a0|a1|a2|a3] | ||
| 253 : | |||
| 254 : | movdqa xmm7, xmm6 | ||
| 255 : | paddd xmm6, xmm4 ; mm6=a+b | ||
| 256 : | psubd xmm4, xmm7 ; mm4=a-b | ||
| 257 : | psrad xmm6, %4 ; => out [0123] | ||
| 258 : | psrad xmm4, %4 ; => out [7654] | ||
| 259 : | |||
| 260 : | packssdw xmm6, xmm4 ; [01237654] | ||
| 261 : | Skal | 1618 | |
| 262 : | edgomez | 1382 | pshufhw xmm6, xmm6, 00011011b ; [01234567] |
| 263 : | |||
| 264 : | movdqa [ecx+%1*16], xmm6 | ||
| 265 : | |||
| 266 : | %endmacro | ||
| 267 : | |||
| 268 : | ;----------------------------------------------------------------------------- | ||
| 269 : | ; Helper macro iLLM_PASS | ||
| 270 : | ;----------------------------------------------------------------------------- | ||
| 271 : | |||
| 272 : | %macro iLLM_PASS 1 ; %1: src/dst | ||
| 273 : | |||
| 274 : | Skal | 1618 | movdqa xmm0, [tan3] ; t3-1 |
| 275 : | movdqa xmm3, [%1+16*3] ; x3 | ||
| 276 : | edgomez | 1382 | movdqa xmm1, xmm0 ; t3-1 |
| 277 : | Skal | 1618 | movdqa xmm5, [%1+16*5] ; x5 |
| 278 : | edgomez | 1382 | |
| 279 : | Skal | 1618 | movdqa xmm4, [tan1] ; t1 |
| 280 : | movdqa xmm6, [%1+16*1] ; x1 | ||
| 281 : | movdqa xmm7, [%1+16*7] ; x7 | ||
| 282 : | edgomez | 1382 | movdqa xmm2, xmm4 ; t1 |
| 283 : | |||
| 284 : | pmulhw xmm0, xmm3 ; x3*(t3-1) | ||
| 285 : | pmulhw xmm1, xmm5 ; x5*(t3-1) | ||
| 286 : | paddsw xmm0, xmm3 ; x3*t3 | ||
| 287 : | paddsw xmm1, xmm5 ; x5*t3 | ||
| 288 : | psubsw xmm0, xmm5 ; x3*t3-x5 = tm35 | ||
| 289 : | paddsw xmm1, xmm3 ; x3+x5*t3 = tp35 | ||
| 290 : | |||
| 291 : | pmulhw xmm4, xmm7 ; x7*t1 | ||
| 292 : | pmulhw xmm2, xmm6 ; x1*t1 | ||
| 293 : | paddsw xmm4, xmm6 ; x1+t1*x7 = tp17 | ||
| 294 : | psubsw xmm2, xmm7 ; x1*t1-x7 = tm17 | ||
| 295 : | |||
| 296 : | |||
| 297 : | movdqa xmm3, [sqrt2] | ||
| 298 : | movdqa xmm7, xmm4 | ||
| 299 : | movdqa xmm6, xmm2 | ||
| 300 : | psubsw xmm4, xmm1 ; tp17-tp35 = t1 | ||
| 301 : | psubsw xmm2, xmm0 ; tm17-tm35 = b3 | ||
| 302 : | paddsw xmm1, xmm7 ; tp17+tp35 = b0 | ||
| 303 : | paddsw xmm0, xmm6 ; tm17+tm35 = t2 | ||
| 304 : | |||
| 305 : | ; xmm1 = b0, xmm2 = b3. preserved | ||
| 306 : | |||
| 307 : | movdqa xmm6, xmm4 | ||
| 308 : | psubsw xmm4, xmm0 ; t1-t2 | ||
| 309 : | paddsw xmm0, xmm6 ; t1+t2 | ||
| 310 : | |||
| 311 : | pmulhw xmm4, xmm3 ; (t1-t2)/(2.sqrt2) | ||
| 312 : | pmulhw xmm0, xmm3 ; (t1+t2)/(2.sqrt2) | ||
| 313 : | |||
| 314 : | paddsw xmm0, xmm0 ; 2.(t1+t2) = b1 | ||
| 315 : | paddsw xmm4, xmm4 ; 2.(t1-t2) = b2 | ||
| 316 : | |||
| 317 : | Skal | 1618 | movdqa xmm7, [tan2] ; t2 |
| 318 : | movdqa xmm3, [%1+2*16] ; x2 | ||
| 319 : | movdqa xmm6, [%1+6*16] ; x6 | ||
| 320 : | edgomez | 1382 | movdqa xmm5, xmm7 ; t2 |
| 321 : | |||
| 322 : | pmulhw xmm7, xmm6 ; x6*t2 | ||
| 323 : | pmulhw xmm5, xmm3 ; x2*t2 | ||
| 324 : | |||
| 325 : | paddsw xmm7, xmm3 ; x2+x6*t2 = tp26 | ||
| 326 : | psubsw xmm5, xmm6 ; x2*t2-x6 = tm26 | ||
| 327 : | |||
| 328 : | |||
| 329 : | Skal | 1618 | ; use:xmm3,xmm5,xmm6,xmm7 frozen: xmm0,xmm4,xmm1,xmm2 |
| 330 : | edgomez | 1382 | |
| 331 : | movdqa xmm3, [%1+0*16] ; x0 | ||
| 332 : | movdqa xmm6, [%1+4*16] ; x4 | ||
| 333 : | |||
| 334 : | Skal | 1618 | movdqa [%1 ], xmm2 ; we spill 1 reg to perform safe butterflies |
| 335 : | |||
| 336 : | movdqa xmm2, xmm3 | ||
| 337 : | edgomez | 1382 | psubsw xmm3, xmm6 ; x0-x4 = tm04 |
| 338 : | Skal | 1618 | paddsw xmm6, xmm2 ; x0+x4 = tp04 |
| 339 : | edgomez | 1382 | |
| 340 : | Skal | 1618 | movdqa xmm2, xmm6 |
| 341 : | psubsw xmm6, xmm7 | ||
| 342 : | paddsw xmm7, xmm2 | ||
| 343 : | movdqa xmm2, xmm3 | ||
| 344 : | psubsw xmm3, xmm5 | ||
| 345 : | paddsw xmm5, xmm2 | ||
| 346 : | edgomez | 1382 | |
| 347 : | Skal | 1618 | movdqa xmm2, xmm5 |
| 348 : | psubsw xmm5, xmm0 | ||
| 349 : | paddsw xmm0, xmm2 | ||
| 350 : | movdqa xmm2, xmm3 | ||
| 351 : | psubsw xmm3, xmm4 | ||
| 352 : | paddsw xmm4, xmm2 | ||
| 353 : | edgomez | 1382 | |
| 354 : | Skal | 1618 | movdqa xmm2, [%1] |
| 355 : | edgomez | 1382 | |
| 356 : | Skal | 1618 | psraw xmm5, 6 ; out6 |
| 357 : | psraw xmm3, 6 ; out5 | ||
| 358 : | psraw xmm0, 6 ; out1 | ||
| 359 : | psraw xmm4, 6 ; out2 | ||
| 360 : | |||
| 361 : | edgomez | 1382 | movdqa [%1+6*16], xmm5 |
| 362 : | movdqa [%1+5*16], xmm3 | ||
| 363 : | movdqa [%1+1*16], xmm0 | ||
| 364 : | movdqa [%1+2*16], xmm4 | ||
| 365 : | |||
| 366 : | ; reminder: xmm1=b0, xmm2=b3, xmm7=a0, xmm6=a3 | ||
| 367 : | |||
| 368 : | movdqa xmm0, xmm7 | ||
| 369 : | movdqa xmm4, xmm6 | ||
| 370 : | psubsw xmm7, xmm1 ; a0-b0 | ||
| 371 : | psubsw xmm6, xmm2 ; a3-b3 | ||
| 372 : | paddsw xmm1, xmm0 ; a0+b0 | ||
| 373 : | paddsw xmm2, xmm4 ; a3+b3 | ||
| 374 : | |||
| 375 : | Skal | 1618 | psraw xmm1, 6 ; out0 |
| 376 : | psraw xmm7, 6 ; out7 | ||
| 377 : | psraw xmm2, 6 ; out3 | ||
| 378 : | psraw xmm6, 6 ; out4 | ||
| 379 : | edgomez | 1382 | |
| 380 : | Skal | 1618 | ; store result |
| 381 : | |||
| 382 : | edgomez | 1382 | movdqa [%1+0*16], xmm1 |
| 383 : | movdqa [%1+3*16], xmm2 | ||
| 384 : | movdqa [%1+4*16], xmm6 | ||
| 385 : | movdqa [%1+7*16], xmm7 | ||
| 386 : | Skal | 1618 | |
| 387 : | edgomez | 1382 | %endmacro |
| 388 : | |||
| 389 : | ;----------------------------------------------------------------------------- | ||
| 390 : | ; Helper macro TEST_ROW (test a null row) | ||
| 391 : | ;----------------------------------------------------------------------------- | ||
| 392 : | |||
| 393 : | %macro TEST_ROW 2 ; %1:src, %2:label x8 | ||
| 394 : | mov eax, [%1 ] | ||
| 395 : | mov edx, [%1+ 8] | ||
| 396 : | or eax, [%1+ 4] | ||
| 397 : | or edx, [%1+12] | ||
| 398 : | or eax, edx | ||
| 399 : | jz near %2 | ||
| 400 : | %endmacro | ||
| 401 : | |||
| 402 : | ;----------------------------------------------------------------------------- | ||
| 403 : | ; Function idct (this one skips null rows) | ||
| 404 : | ;----------------------------------------------------------------------------- | ||
| 405 : | Skal | 1618 | ; IEEE1180 and Walken compatible version |
| 406 : | edgomez | 1382 | |
| 407 : | Isibaar | 1627 | ALIGN 16 |
| 408 : | Skal | 1618 | idct_sse2_skal: |
| 409 : | edgomez | 1382 | |
| 410 : | mov ecx, [esp+ 4] ; Src | ||
| 411 : | |||
| 412 : | TEST_ROW ecx, .Row0_Round | ||
| 413 : | Skal | 1618 | iMTX_MULT 0, iTab1, Walken_Idct_Rounders + 16*0, 11 |
| 414 : | edgomez | 1382 | jmp .Row1 |
| 415 : | Isibaar | 1793 | .Row0_Round: |
| 416 : | Skal | 1618 | movdqa xmm0, [Walken_Idct_Rounders + 16*8 + 8*0] |
| 417 : | movdqa [ecx ], xmm0 | ||
| 418 : | edgomez | 1382 | |
| 419 : | Isibaar | 1793 | .Row1: |
| 420 : | edgomez | 1382 | TEST_ROW ecx+16, .Row1_Round |
| 421 : | Skal | 1618 | iMTX_MULT 1, iTab2, Walken_Idct_Rounders + 16*1, 11 |
| 422 : | edgomez | 1382 | jmp .Row2 |
| 423 : | Isibaar | 1793 | .Row1_Round: |
| 424 : | Skal | 1618 | movdqa xmm0, [Walken_Idct_Rounders + 16*8 + 16*1] |
| 425 : | movdqa [ecx+16 ], xmm0 | ||
| 426 : | edgomez | 1382 | |
| 427 : | Isibaar | 1793 | .Row2: |
| 428 : | edgomez | 1382 | TEST_ROW ecx+32, .Row2_Round |
| 429 : | Skal | 1618 | iMTX_MULT 2, iTab3, Walken_Idct_Rounders + 16*2, 11 |
| 430 : | edgomez | 1382 | jmp .Row3 |
| 431 : | Isibaar | 1793 | .Row2_Round: |
| 432 : | Skal | 1618 | movdqa xmm0, [Walken_Idct_Rounders + 16*8 + 16*2] |
| 433 : | movdqa [ecx+32 ], xmm0 | ||
| 434 : | edgomez | 1382 | |
| 435 : | Isibaar | 1793 | .Row3: |
| 436 : | edgomez | 1382 | TEST_ROW ecx+48, .Row4 |
| 437 : | Skal | 1618 | iMTX_MULT 3, iTab4, Walken_Idct_Rounders + 16*3, 11 |
| 438 : | edgomez | 1382 | |
| 439 : | Isibaar | 1793 | .Row4: |
| 440 : | edgomez | 1382 | TEST_ROW ecx+64, .Row5 |
| 441 : | Skal | 1618 | iMTX_MULT 4, iTab1, Walken_Idct_Rounders + 16*4, 11 |
| 442 : | edgomez | 1382 | |
| 443 : | Isibaar | 1793 | .Row5: |
| 444 : | edgomez | 1382 | TEST_ROW ecx+80, .Row6 |
| 445 : | Skal | 1618 | iMTX_MULT 5, iTab4, Walken_Idct_Rounders + 16*5, 11 |
| 446 : | edgomez | 1382 | |
| 447 : | Isibaar | 1793 | .Row6: |
| 448 : | edgomez | 1382 | TEST_ROW ecx+96, .Row7 |
| 449 : | Skal | 1618 | iMTX_MULT 6, iTab3, Walken_Idct_Rounders + 16*6, 11 |
| 450 : | edgomez | 1382 | |
| 451 : | Isibaar | 1793 | .Row7: |
| 452 : | edgomez | 1382 | TEST_ROW ecx+112, .End |
| 453 : | Skal | 1618 | iMTX_MULT 7, iTab2, Walken_Idct_Rounders + 16*7, 11 |
| 454 : | Isibaar | 1793 | .End: |
| 455 : | edgomez | 1382 | |
| 456 : | Skal | 1618 | iLLM_PASS ecx |
| 457 : | |||
| 458 : | edgomez | 1382 | ret |
| 459 : | Isibaar | 1793 | ENDFUNC |
| 460 : | edgomez | 1382 | |
| 461 : | ;----------------------------------------------------------------------------- | ||
| 462 : | ; Helper macro fLLM_PASS | ||
| 463 : | ;----------------------------------------------------------------------------- | ||
| 464 : | |||
| 465 : | %macro fLLM_PASS 2 ; %1: src/dst, %2:Shift | ||
| 466 : | |||
| 467 : | movdqa xmm0, [%1+0*16] ; In0 | ||
| 468 : | movdqa xmm2, [%1+2*16] ; In2 | ||
| 469 : | movdqa xmm3, xmm0 | ||
| 470 : | movdqa xmm4, xmm2 | ||
| 471 : | movdqa xmm7, [%1+7*16] ; In7 | ||
| 472 : | movdqa xmm5, [%1+5*16] ; In5 | ||
| 473 : | |||
| 474 : | psubsw xmm0, xmm7 ; t7 = In0-In7 | ||
| 475 : | paddsw xmm7, xmm3 ; t0 = In0+In7 | ||
| 476 : | psubsw xmm2, xmm5 ; t5 = In2-In5 | ||
| 477 : | paddsw xmm5, xmm4 ; t2 = In2+In5 | ||
| 478 : | |||
| 479 : | movdqa xmm3, [%1+3*16] ; In3 | ||
| 480 : | movdqa xmm4, [%1+4*16] ; In4 | ||
| 481 : | movdqa xmm1, xmm3 | ||
| 482 : | psubsw xmm3, xmm4 ; t4 = In3-In4 | ||
| 483 : | paddsw xmm4, xmm1 ; t3 = In3+In4 | ||
| 484 : | movdqa xmm6, [%1+6*16] ; In6 | ||
| 485 : | movdqa xmm1, [%1+1*16] ; In1 | ||
| 486 : | psubsw xmm1, xmm6 ; t6 = In1-In6 | ||
| 487 : | paddsw xmm6, [%1+1*16] ; t1 = In1+In6 | ||
| 488 : | |||
| 489 : | psubsw xmm7, xmm4 ; tm03 = t0-t3 | ||
| 490 : | psubsw xmm6, xmm5 ; tm12 = t1-t2 | ||
| 491 : | paddsw xmm4, xmm4 ; 2.t3 | ||
| 492 : | paddsw xmm5, xmm5 ; 2.t2 | ||
| 493 : | paddsw xmm4, xmm7 ; tp03 = t0+t3 | ||
| 494 : | paddsw xmm5, xmm6 ; tp12 = t1+t2 | ||
| 495 : | |||
| 496 : | psllw xmm2, %2+1 ; shift t5 (shift +1 to.. | ||
| 497 : | psllw xmm1, %2+1 ; shift t6 ..compensate cos4/2) | ||
| 498 : | psllw xmm4, %2 ; shift t3 | ||
| 499 : | psllw xmm5, %2 ; shift t2 | ||
| 500 : | psllw xmm7, %2 ; shift t0 | ||
| 501 : | psllw xmm6, %2 ; shift t1 | ||
| 502 : | psllw xmm3, %2 ; shift t4 | ||
| 503 : | psllw xmm0, %2 ; shift t7 | ||
| 504 : | |||
| 505 : | psubsw xmm4, xmm5 ; out4 = tp03-tp12 | ||
| 506 : | psubsw xmm1, xmm2 ; xmm1: t6-t5 | ||
| 507 : | paddsw xmm5, xmm5 | ||
| 508 : | paddsw xmm2, xmm2 | ||
| 509 : | paddsw xmm5, xmm4 ; out0 = tp03+tp12 | ||
| 510 : | movdqa [%1+4*16], xmm4 ; => out4 | ||
| 511 : | paddsw xmm2, xmm1 ; xmm2: t6+t5 | ||
| 512 : | movdqa [%1+0*16], xmm5 ; => out0 | ||
| 513 : | |||
| 514 : | movdqa xmm4, [tan2] ; xmm4 <= tan2 | ||
| 515 : | pmulhw xmm4, xmm7 ; tm03*tan2 | ||
| 516 : | movdqa xmm5, [tan2] ; xmm5 <= tan2 | ||
| 517 : | psubsw xmm4, xmm6 ; out6 = tm03*tan2 - tm12 | ||
| 518 : | pmulhw xmm5, xmm6 ; tm12*tan2 | ||
| 519 : | paddsw xmm5, xmm7 ; out2 = tm12*tan2 + tm03 | ||
| 520 : | |||
| 521 : | movdqa xmm6, [sqrt2] | ||
| 522 : | movdqa xmm7, [Rounder1] | ||
| 523 : | |||
| 524 : | pmulhw xmm2, xmm6 ; xmm2: tp65 = (t6 + t5)*cos4 | ||
| 525 : | por xmm5, xmm7 ; correct out2 | ||
| 526 : | por xmm4, xmm7 ; correct out6 | ||
| 527 : | pmulhw xmm1, xmm6 ; xmm1: tm65 = (t6 - t5)*cos4 | ||
| 528 : | por xmm2, xmm7 ; correct tp65 | ||
| 529 : | |||
| 530 : | movdqa [%1+2*16], xmm5 ; => out2 | ||
| 531 : | movdqa xmm5, xmm3 ; save t4 | ||
| 532 : | movdqa [%1+6*16], xmm4 ; => out6 | ||
| 533 : | movdqa xmm4, xmm0 ; save t7 | ||
| 534 : | |||
| 535 : | psubsw xmm3, xmm1 ; xmm3: tm465 = t4 - tm65 | ||
| 536 : | psubsw xmm0, xmm2 ; xmm0: tm765 = t7 - tp65 | ||
| 537 : | paddsw xmm2, xmm4 ; xmm2: tp765 = t7 + tp65 | ||
| 538 : | paddsw xmm1, xmm5 ; xmm1: tp465 = t4 + tm65 | ||
| 539 : | |||
| 540 : | movdqa xmm4, [tan3] ; tan3 - 1 | ||
| 541 : | movdqa xmm5, [tan1] ; tan1 | ||
| 542 : | |||
| 543 : | movdqa xmm7, xmm3 ; save tm465 | ||
| 544 : | pmulhw xmm3, xmm4 ; tm465*(tan3-1) | ||
| 545 : | movdqa xmm6, xmm1 ; save tp465 | ||
| 546 : | pmulhw xmm1, xmm5 ; tp465*tan1 | ||
| 547 : | |||
| 548 : | paddsw xmm3, xmm7 ; tm465*tan3 | ||
| 549 : | pmulhw xmm4, xmm0 ; tm765*(tan3-1) | ||
| 550 : | paddsw xmm4, xmm0 ; tm765*tan3 | ||
| 551 : | pmulhw xmm5, xmm2 ; tp765*tan1 | ||
| 552 : | |||
| 553 : | paddsw xmm1, xmm2 ; out1 = tp765 + tp465*tan1 | ||
| 554 : | psubsw xmm0, xmm3 ; out3 = tm765 - tm465*tan3 | ||
| 555 : | paddsw xmm7, xmm4 ; out5 = tm465 + tm765*tan3 | ||
| 556 : | psubsw xmm5, xmm6 ; out7 =-tp465 + tp765*tan1 | ||
| 557 : | |||
| 558 : | movdqa [%1+1*16], xmm1 ; => out1 | ||
| 559 : | movdqa [%1+3*16], xmm0 ; => out3 | ||
| 560 : | movdqa [%1+5*16], xmm7 ; => out5 | ||
| 561 : | movdqa [%1+7*16], xmm5 ; => out7 | ||
| 562 : | |||
| 563 : | %endmacro | ||
| 564 : | |||
| 565 : | ;----------------------------------------------------------------------------- | ||
| 566 : | ;Helper macro fMTX_MULT | ||
| 567 : | ;----------------------------------------------------------------------------- | ||
| 568 : | |||
| 569 : | %macro fMTX_MULT 3 ; %1=src, %2 = Coeffs, %3=rounders | ||
| 570 : | |||
| 571 : | movdqa xmm0, [ecx+%1*16+0] ; xmm0 = [0123][4567] | ||
| 572 : | pshufhw xmm1, xmm0, 00011011b ; xmm1 = [----][7654] | ||
| 573 : | pshufd xmm0, xmm0, 01000100b | ||
| 574 : | pshufd xmm1, xmm1, 11101110b | ||
| 575 : | |||
| 576 : | movdqa xmm2, xmm0 | ||
| 577 : | paddsw xmm0, xmm1 ; xmm0 = [a0 a1 a2 a3] | ||
| 578 : | psubsw xmm2, xmm1 ; xmm2 = [b0 b1 b2 b3] | ||
| 579 : | |||
| 580 : | punpckldq xmm0, xmm2 ; xmm0 = [a0 a1 b0 b1][a2 a3 b2 b3] | ||
| 581 : | pshufd xmm2, xmm0, 01001110b ; xmm2 = [a2 a3 b2 b3][a0 a1 b0 b1] | ||
| 582 : | |||
| 583 : | ; [M00 M01 M16 M17] [M06 M07 M22 M23] x mm0 = [0 /1 /2'/3'] | ||
| 584 : | ; [M02 M03 M18 M19] [M04 M05 M20 M21] x mm2 = [0'/1'/2 /3 ] | ||
| 585 : | ; [M08 M09 M24 M25] [M14 M15 M30 M31] x mm0 = [4 /5 /6'/7'] | ||
| 586 : | ; [M10 M11 M26 M27] [M12 M13 M28 M29] x mm2 = [4'/5'/6 /7 ] | ||
| 587 : | |||
| 588 : | movdqa xmm1, [%2+16] | ||
| 589 : | movdqa xmm3, [%2+32] | ||
| 590 : | pmaddwd xmm1, xmm2 | ||
| 591 : | pmaddwd xmm3, xmm0 | ||
| 592 : | pmaddwd xmm2, [%2+48] | ||
| 593 : | pmaddwd xmm0, [%2+ 0] | ||
| 594 : | |||
| 595 : | paddd xmm0, xmm1 ; [ out0 | out1 ][ out2 | out3 ] | ||
| 596 : | paddd xmm2, xmm3 ; [ out4 | out5 ][ out6 | out7 ] | ||
| 597 : | psrad xmm0, 16 | ||
| 598 : | psrad xmm2, 16 | ||
| 599 : | |||
| 600 : | packssdw xmm0, xmm2 ; [ out0 .. out7 ] | ||
| 601 : | paddsw xmm0, [%3] ; Round | ||
| 602 : | |||
| 603 : | psraw xmm0, 4 ; => [-2048, 2047] | ||
| 604 : | |||
| 605 : | movdqa [ecx+%1*16+0], xmm0 | ||
| 606 : | %endmacro | ||
| 607 : | |||
| 608 : | ;----------------------------------------------------------------------------- | ||
| 609 : | ; Function Forward DCT | ||
| 610 : | ;----------------------------------------------------------------------------- | ||
| 611 : | |||
| 612 : | ALIGN 16 | ||
| 613 : | fdct_sse2_skal: | ||
| 614 : | mov ecx, [esp+4] | ||
| 615 : | fLLM_PASS ecx+0, 3 | ||
| 616 : | fMTX_MULT 0, fTab1, Fdct_Rnd0 | ||
| 617 : | fMTX_MULT 1, fTab2, Fdct_Rnd2 | ||
| 618 : | fMTX_MULT 2, fTab3, Fdct_Rnd1 | ||
| 619 : | fMTX_MULT 3, fTab4, Fdct_Rnd1 | ||
| 620 : | fMTX_MULT 4, fTab1, Fdct_Rnd0 | ||
| 621 : | fMTX_MULT 5, fTab4, Fdct_Rnd1 | ||
| 622 : | fMTX_MULT 6, fTab3, Fdct_Rnd1 | ||
| 623 : | fMTX_MULT 7, fTab2, Fdct_Rnd1 | ||
| 624 : | ret | ||
| 625 : | Isibaar | 1793 | ENDFUNC |
| 626 : | edgomez | 1540 | |
| 627 : | Isibaar | 1790 | |
| 628 : | %ifidn __OUTPUT_FORMAT__,elf | ||
| 629 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
| 630 : | %endif | ||
| 631 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |