Parent Directory
| 
Revision Log
Revision 1176 - (view) (download)
| 1 : | edgomez | 1176 | ;/************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * - 3dne Quantization/Dequantization - | ||
| 5 : | ; * | ||
| 6 : | ; * Copyright (C) 2002-2003 Peter Ross <pross@xvid.org> | ||
| 7 : | ; * 2002-2003 Michael Militzer <isibaar@xvid.org> | ||
| 8 : | ; * 2002-2003 Pascal Massimino <skal@planet-d.net> | ||
| 9 : | ; * | ||
| 10 : | ; * This program is free software ; you can redistribute it and/or modify | ||
| 11 : | ; * it under the terms of the GNU General Public License as published by | ||
| 12 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
| 13 : | ; * (at your option) any later version. | ||
| 14 : | ; * | ||
| 15 : | ; * This program is distributed in the hope that it will be useful, | ||
| 16 : | ; * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
| 17 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 18 : | ; * GNU General Public License for more details. | ||
| 19 : | ; * | ||
| 20 : | ; * You should have received a copy of the GNU General Public License | ||
| 21 : | ; * along with this program ; if not, write to the Free Software | ||
| 22 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 23 : | ; * | ||
| 24 : | ; * $Id: quantize_mpeg_mmx.asm,v 1.1.2.2 2003-10-09 18:50:22 edgomez Exp $ | ||
| 25 : | ; * | ||
| 26 : | ; *************************************************************************/ | ||
| 27 : | edgomez | 1174 | |
| 28 : | ; data/text alignment | ||
| 29 : | %define ALIGN 8 | ||
| 30 : | |||
| 31 : | %define SATURATE | ||
| 32 : | |||
| 33 : | bits 32 | ||
| 34 : | |||
| 35 : | %macro cglobal 1 | ||
| 36 : | %ifdef PREFIX | ||
| 37 : | global _%1 | ||
| 38 : | %define %1 _%1 | ||
| 39 : | %else | ||
| 40 : | global %1 | ||
| 41 : | %endif | ||
| 42 : | %endmacro | ||
| 43 : | |||
| 44 : | %macro cextern 1 | ||
| 45 : | %ifdef PREFIX | ||
| 46 : | extern _%1 | ||
| 47 : | %define %1 _%1 | ||
| 48 : | %else | ||
| 49 : | extern %1 | ||
| 50 : | %endif | ||
| 51 : | %endmacro | ||
| 52 : | |||
| 53 : | edgomez | 1176 | ;*************************************************************************** |
| 54 : | ; Local data | ||
| 55 : | ;*************************************************************************** | ||
| 56 : | edgomez | 1174 | |
| 57 : | edgomez | 1176 | %ifdef FORMAT_COFF |
| 58 : | SECTION .data data | ||
| 59 : | %else | ||
| 60 : | SECTION .data data align=8 | ||
| 61 : | %endif | ||
| 62 : | |||
| 63 : | mmx_one: | ||
| 64 : | times 4 dw 1 | ||
| 65 : | |||
| 66 : | edgomez | 1174 | ;=========================================================================== |
| 67 : | ; | ||
| 68 : | ; divide by 2Q table | ||
| 69 : | ; | ||
| 70 : | ;=========================================================================== | ||
| 71 : | |||
| 72 : | edgomez | 1176 | align 16 |
| 73 : | mmx_div: | ||
| 74 : | %assign quant 1 | ||
| 75 : | %rep 31 | ||
| 76 : | times 4 dw (1<<17) / (quant*2) + 1 | ||
| 77 : | %assign quant quant+1 | ||
| 78 : | %endrep | ||
| 79 : | edgomez | 1174 | |
| 80 : | ;=========================================================================== | ||
| 81 : | ; | ||
| 82 : | ; intra matrix | ||
| 83 : | ; | ||
| 84 : | ;=========================================================================== | ||
| 85 : | |||
| 86 : | cextern intra_matrix | ||
| 87 : | cextern intra_matrix_fix | ||
| 88 : | |||
| 89 : | ;=========================================================================== | ||
| 90 : | ; | ||
| 91 : | ; inter matrix | ||
| 92 : | ; | ||
| 93 : | ;=========================================================================== | ||
| 94 : | |||
| 95 : | cextern inter_matrix | ||
| 96 : | cextern inter_matrix_fix | ||
| 97 : | |||
| 98 : | |||
| 99 : | %define VM18P 3 | ||
| 100 : | %define VM18Q 4 | ||
| 101 : | |||
| 102 : | |||
| 103 : | ;=========================================================================== | ||
| 104 : | ; | ||
| 105 : | ; quantd table | ||
| 106 : | ; | ||
| 107 : | ;=========================================================================== | ||
| 108 : | |||
| 109 : | edgomez | 1176 | quantd: |
| 110 : | %assign quant 1 | ||
| 111 : | %rep 31 | ||
| 112 : | times 4 dw ((VM18P*quant) + (VM18Q/2)) / VM18Q | ||
| 113 : | %assign quant quant+1 | ||
| 114 : | %endrep | ||
| 115 : | edgomez | 1174 | |
| 116 : | ;=========================================================================== | ||
| 117 : | ; | ||
| 118 : | ; multiple by 2Q table | ||
| 119 : | ; | ||
| 120 : | ;=========================================================================== | ||
| 121 : | |||
| 122 : | edgomez | 1176 | mmx_mul_quant: |
| 123 : | %assign quant 1 | ||
| 124 : | %rep 31 | ||
| 125 : | times 4 dw quant | ||
| 126 : | %assign quant quant+1 | ||
| 127 : | %endrep | ||
| 128 : | edgomez | 1174 | |
| 129 : | ;=========================================================================== | ||
| 130 : | ; | ||
| 131 : | ; saturation limits | ||
| 132 : | ; | ||
| 133 : | ;=========================================================================== | ||
| 134 : | |||
| 135 : | align 16 | ||
| 136 : | |||
| 137 : | edgomez | 1176 | mmx_32767_minus_2047: |
| 138 : | times 4 dw (32767-2047) | ||
| 139 : | mmx_32768_minus_2048: | ||
| 140 : | times 4 dw (32768-2048) | ||
| 141 : | mmx_2047: | ||
| 142 : | times 4 dw 2047 | ||
| 143 : | mmx_minus_2048: | ||
| 144 : | times 4 dw (-2048) | ||
| 145 : | zero: | ||
| 146 : | times 4 dw 0 | ||
| 147 : | edgomez | 1174 | |
| 148 : | edgomez | 1176 | ;*************************************************************************** |
| 149 : | ; Local data | ||
| 150 : | ;*************************************************************************** | ||
| 151 : | |||
| 152 : | edgomez | 1174 | section .text |
| 153 : | |||
| 154 : | ;=========================================================================== | ||
| 155 : | ; | ||
| 156 : | edgomez | 1176 | ; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
| 157 : | ; const int16_t const * data, | ||
| 158 : | ; const uint32_t quant, | ||
| 159 : | ; const uint32_t dcscalar); | ||
| 160 : | edgomez | 1174 | ; |
| 161 : | ;=========================================================================== | ||
| 162 : | |||
| 163 : | align ALIGN | ||
| 164 : | cglobal quant_mpeg_intra_mmx | ||
| 165 : | quant_mpeg_intra_mmx: | ||
| 166 : | |||
| 167 : | edgomez | 1176 | push ecx |
| 168 : | push esi | ||
| 169 : | push edi | ||
| 170 : | edgomez | 1174 | |
| 171 : | edgomez | 1176 | mov edi, [esp + 12 + 4] ; coeff |
| 172 : | mov esi, [esp + 12 + 8] ; data | ||
| 173 : | mov eax, [esp + 12 + 12] ; quant | ||
| 174 : | edgomez | 1174 | |
| 175 : | edgomez | 1176 | movq mm5, [quantd + eax * 8 - 8] ; quantd -> mm5 |
| 176 : | edgomez | 1174 | |
| 177 : | edgomez | 1176 | xor ecx, ecx |
| 178 : | cmp al, 1 | ||
| 179 : | jz near .q1loop | ||
| 180 : | edgomez | 1174 | |
| 181 : | edgomez | 1176 | cmp al, 2 |
| 182 : | jz near .q2loop | ||
| 183 : | edgomez | 1174 | |
| 184 : | edgomez | 1176 | movq mm7, [mmx_div + eax * 8 - 8] ; multipliers[quant] -> mm7 |
| 185 : | edgomez | 1174 | |
| 186 : | align ALIGN | ||
| 187 : | .loop | ||
| 188 : | edgomez | 1176 | movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
| 189 : | movq mm3, [esi + 8*ecx + 8] ; | ||
| 190 : | edgomez | 1174 | |
| 191 : | edgomez | 1176 | pxor mm1, mm1 ; mm1 = 0 |
| 192 : | pxor mm4, mm4 | ||
| 193 : | edgomez | 1174 | |
| 194 : | edgomez | 1176 | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
| 195 : | pcmpgtw mm4, mm3 | ||
| 196 : | edgomez | 1174 | |
| 197 : | edgomez | 1176 | pxor mm0, mm1 ; mm0 = |mm0| |
| 198 : | pxor mm3, mm4 ; | ||
| 199 : | psubw mm0, mm1 ; displace | ||
| 200 : | psubw mm3, mm4 ; | ||
| 201 : | edgomez | 1174 | |
| 202 : | edgomez | 1176 | psllw mm0, 4 ; level << 4 |
| 203 : | psllw mm3, 4 ; | ||
| 204 : | edgomez | 1174 | |
| 205 : | edgomez | 1176 | movq mm2, [intra_matrix + 8*ecx] |
| 206 : | psrlw mm2, 1 ; intra_matrix[i]>>1 | ||
| 207 : | paddw mm0, mm2 | ||
| 208 : | edgomez | 1174 | |
| 209 : | edgomez | 1176 | movq mm2, [intra_matrix_fix + ecx*8] |
| 210 : | pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] | ||
| 211 : | edgomez | 1174 | |
| 212 : | edgomez | 1176 | movq mm2, [intra_matrix + 8*ecx + 8] |
| 213 : | psrlw mm2, 1 | ||
| 214 : | paddw mm3, mm2 | ||
| 215 : | edgomez | 1174 | |
| 216 : | edgomez | 1176 | movq mm2, [intra_matrix_fix + ecx*8 + 8] |
| 217 : | pmulhw mm3, mm2 | ||
| 218 : | edgomez | 1174 | |
| 219 : | edgomez | 1176 | paddw mm0, mm5 ; + quantd |
| 220 : | paddw mm3, mm5 | ||
| 221 : | pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 | ||
| 222 : | pmulhw mm3, mm7 ; | ||
| 223 : | psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 | ||
| 224 : | psrlw mm3, 1 | ||
| 225 : | edgomez | 1174 | |
| 226 : | edgomez | 1176 | pxor mm0, mm1 ; mm0 *= sign(mm0) |
| 227 : | pxor mm3, mm4 ; | ||
| 228 : | psubw mm0, mm1 ; undisplace | ||
| 229 : | psubw mm3, mm4 ; | ||
| 230 : | edgomez | 1174 | |
| 231 : | edgomez | 1176 | movq [edi + 8*ecx], mm0 |
| 232 : | movq [edi + 8*ecx + 8], mm3 | ||
| 233 : | edgomez | 1174 | |
| 234 : | edgomez | 1176 | add ecx,2 |
| 235 : | cmp ecx,16 | ||
| 236 : | jnz near .loop | ||
| 237 : | edgomez | 1174 | |
| 238 : | .done | ||
| 239 : | ; caclulate data[0] // (int32_t)dcscalar) | ||
| 240 : | |||
| 241 : | edgomez | 1176 | mov ecx, [esp + 12 + 16] ; dcscalar |
| 242 : | mov edx, ecx | ||
| 243 : | movsx eax, word [esi] ; data[0] | ||
| 244 : | shr edx, 1 ; edx = dcscalar /2 | ||
| 245 : | cmp eax, 0 | ||
| 246 : | jg .gtzero | ||
| 247 : | edgomez | 1174 | |
| 248 : | edgomez | 1176 | sub eax, edx |
| 249 : | jmp short .mul | ||
| 250 : | edgomez | 1174 | .gtzero |
| 251 : | edgomez | 1176 | add eax, edx |
| 252 : | edgomez | 1174 | .mul |
| 253 : | edgomez | 1176 | cdq ; expand eax -> edx:eax |
| 254 : | idiv ecx ; eax = edx:eax / dcscalar | ||
| 255 : | edgomez | 1174 | |
| 256 : | edgomez | 1176 | mov [edi], ax ; coeff[0] = ax |
| 257 : | edgomez | 1174 | |
| 258 : | edgomez | 1176 | pop edi |
| 259 : | pop esi | ||
| 260 : | pop ecx | ||
| 261 : | edgomez | 1174 | |
| 262 : | edgomez | 1176 | xor eax, eax ; return(0); |
| 263 : | ret | ||
| 264 : | edgomez | 1174 | |
| 265 : | align ALIGN | ||
| 266 : | .q1loop | ||
| 267 : | edgomez | 1176 | movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
| 268 : | movq mm3, [esi + 8*ecx + 8] ; | ||
| 269 : | edgomez | 1174 | |
| 270 : | edgomez | 1176 | pxor mm1, mm1 ; mm1 = 0 |
| 271 : | pxor mm4, mm4 ; | ||
| 272 : | edgomez | 1174 | |
| 273 : | edgomez | 1176 | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
| 274 : | pcmpgtw mm4, mm3 ; | ||
| 275 : | edgomez | 1174 | |
| 276 : | edgomez | 1176 | pxor mm0, mm1 ; mm0 = |mm0| |
| 277 : | pxor mm3, mm4 ; | ||
| 278 : | psubw mm0, mm1 ; displace | ||
| 279 : | psubw mm3, mm4 ; | ||
| 280 : | edgomez | 1174 | |
| 281 : | edgomez | 1176 | psllw mm0, 4 |
| 282 : | psllw mm3, 4 | ||
| 283 : | edgomez | 1174 | |
| 284 : | edgomez | 1176 | movq mm2, [intra_matrix + 8*ecx] |
| 285 : | psrlw mm2, 1 | ||
| 286 : | paddw mm0, mm2 | ||
| 287 : | edgomez | 1174 | |
| 288 : | edgomez | 1176 | movq mm2, [intra_matrix_fix + ecx*8] |
| 289 : | pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] | ||
| 290 : | edgomez | 1174 | |
| 291 : | edgomez | 1176 | movq mm2, [intra_matrix + 8*ecx + 8] |
| 292 : | psrlw mm2, 1 | ||
| 293 : | paddw mm3, mm2 | ||
| 294 : | edgomez | 1174 | |
| 295 : | edgomez | 1176 | movq mm2, [intra_matrix_fix + ecx*8 + 8] |
| 296 : | pmulhw mm3, mm2 | ||
| 297 : | edgomez | 1174 | |
| 298 : | edgomez | 1176 | paddw mm0, mm5 |
| 299 : | paddw mm3, mm5 | ||
| 300 : | edgomez | 1174 | |
| 301 : | edgomez | 1176 | psrlw mm0, 1 ; mm0 >>= 1 (/2) |
| 302 : | psrlw mm3, 1 ; | ||
| 303 : | edgomez | 1174 | |
| 304 : | edgomez | 1176 | pxor mm0, mm1 ; mm0 *= sign(mm0) |
| 305 : | pxor mm3, mm4 ; | ||
| 306 : | psubw mm0, mm1 ; undisplace | ||
| 307 : | psubw mm3, mm4 ; | ||
| 308 : | edgomez | 1174 | |
| 309 : | edgomez | 1176 | movq [edi + 8*ecx], mm0 |
| 310 : | movq [edi + 8*ecx + 8], mm3 | ||
| 311 : | edgomez | 1174 | |
| 312 : | edgomez | 1176 | add ecx, 2 |
| 313 : | cmp ecx, 16 | ||
| 314 : | jnz near .q1loop | ||
| 315 : | jmp near .done | ||
| 316 : | edgomez | 1174 | |
| 317 : | |||
| 318 : | align ALIGN | ||
| 319 : | .q2loop | ||
| 320 : | edgomez | 1176 | movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
| 321 : | movq mm3, [esi + 8*ecx + 8] ; | ||
| 322 : | edgomez | 1174 | |
| 323 : | edgomez | 1176 | pxor mm1, mm1 ; mm1 = 0 |
| 324 : | pxor mm4, mm4 ; | ||
| 325 : | edgomez | 1174 | |
| 326 : | edgomez | 1176 | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
| 327 : | pcmpgtw mm4, mm3 ; | ||
| 328 : | edgomez | 1174 | |
| 329 : | edgomez | 1176 | pxor mm0, mm1 ; mm0 = |mm0| |
| 330 : | pxor mm3, mm4 ; | ||
| 331 : | psubw mm0, mm1 ; displace | ||
| 332 : | psubw mm3, mm4 ; | ||
| 333 : | edgomez | 1174 | |
| 334 : | edgomez | 1176 | psllw mm0, 4 |
| 335 : | psllw mm3, 4 | ||
| 336 : | edgomez | 1174 | |
| 337 : | edgomez | 1176 | movq mm2, [intra_matrix + 8*ecx] |
| 338 : | psrlw mm2, 1 | ||
| 339 : | paddw mm0, mm2 | ||
| 340 : | edgomez | 1174 | |
| 341 : | edgomez | 1176 | movq mm2, [intra_matrix_fix + ecx*8] |
| 342 : | pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] | ||
| 343 : | edgomez | 1174 | |
| 344 : | edgomez | 1176 | movq mm2, [intra_matrix + 8*ecx + 8] |
| 345 : | psrlw mm2, 1 | ||
| 346 : | paddw mm3, mm2 | ||
| 347 : | edgomez | 1174 | |
| 348 : | edgomez | 1176 | movq mm2, [intra_matrix_fix + ecx*8 + 8] |
| 349 : | pmulhw mm3, mm2 | ||
| 350 : | edgomez | 1174 | |
| 351 : | edgomez | 1176 | paddw mm0, mm5 |
| 352 : | paddw mm3, mm5 | ||
| 353 : | edgomez | 1174 | |
| 354 : | edgomez | 1176 | psrlw mm0, 2 ; mm0 >>= 1 (/4) |
| 355 : | psrlw mm3, 2 ; | ||
| 356 : | edgomez | 1174 | |
| 357 : | edgomez | 1176 | pxor mm0, mm1 ; mm0 *= sign(mm0) |
| 358 : | pxor mm3, mm4 ; | ||
| 359 : | psubw mm0, mm1 ; undisplace | ||
| 360 : | psubw mm3, mm4 ; | ||
| 361 : | edgomez | 1174 | |
| 362 : | edgomez | 1176 | movq [edi + 8*ecx], mm0 |
| 363 : | movq [edi + 8*ecx + 8], mm3 | ||
| 364 : | edgomez | 1174 | |
| 365 : | edgomez | 1176 | add ecx,2 |
| 366 : | cmp ecx,16 | ||
| 367 : | jnz near .q2loop | ||
| 368 : | jmp near .done | ||
| 369 : | edgomez | 1174 | |
| 370 : | |||
| 371 : | ;=========================================================================== | ||
| 372 : | ; | ||
| 373 : | edgomez | 1176 | ; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
| 374 : | ; const int16_t const * data, | ||
| 375 : | ; const uint32_t quant); | ||
| 376 : | edgomez | 1174 | ; |
| 377 : | ;=========================================================================== | ||
| 378 : | |||
| 379 : | align ALIGN | ||
| 380 : | cglobal quant_mpeg_inter_mmx | ||
| 381 : | quant_mpeg_inter_mmx: | ||
| 382 : | |||
| 383 : | edgomez | 1176 | push ecx |
| 384 : | push esi | ||
| 385 : | push edi | ||
| 386 : | edgomez | 1174 | |
| 387 : | edgomez | 1176 | mov edi, [esp + 12 + 4] ; coeff |
| 388 : | mov esi, [esp + 12 + 8] ; data | ||
| 389 : | mov eax, [esp + 12 + 12] ; quant | ||
| 390 : | edgomez | 1174 | |
| 391 : | edgomez | 1176 | xor ecx, ecx |
| 392 : | edgomez | 1174 | |
| 393 : | edgomez | 1176 | pxor mm5, mm5 ; sum |
| 394 : | edgomez | 1174 | |
| 395 : | edgomez | 1176 | cmp al, 1 |
| 396 : | jz near .q1loop | ||
| 397 : | edgomez | 1174 | |
| 398 : | edgomez | 1176 | cmp al, 2 |
| 399 : | jz near .q2loop | ||
| 400 : | edgomez | 1174 | |
| 401 : | edgomez | 1176 | movq mm7, [mmx_div + eax * 8 - 8] ; divider |
| 402 : | edgomez | 1174 | |
| 403 : | align ALIGN | ||
| 404 : | .loop | ||
| 405 : | edgomez | 1176 | movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
| 406 : | movq mm3, [esi + 8*ecx + 8] ; | ||
| 407 : | pxor mm1, mm1 ; mm1 = 0 | ||
| 408 : | pxor mm4, mm4 ; | ||
| 409 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 410 : | pcmpgtw mm4, mm3 ; | ||
| 411 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 412 : | pxor mm3, mm4 ; | ||
| 413 : | psubw mm0, mm1 ; displace | ||
| 414 : | psubw mm3, mm4 ; | ||
| 415 : | edgomez | 1174 | |
| 416 : | edgomez | 1176 | psllw mm0, 4 |
| 417 : | psllw mm3, 4 | ||
| 418 : | edgomez | 1174 | |
| 419 : | edgomez | 1176 | movq mm2, [inter_matrix + 8*ecx] |
| 420 : | psrlw mm2, 1 | ||
| 421 : | paddw mm0, mm2 | ||
| 422 : | edgomez | 1174 | |
| 423 : | edgomez | 1176 | movq mm2, [inter_matrix_fix + ecx*8] |
| 424 : | pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] | ||
| 425 : | edgomez | 1174 | |
| 426 : | edgomez | 1176 | movq mm2, [inter_matrix + 8*ecx + 8] |
| 427 : | psrlw mm2, 1 | ||
| 428 : | paddw mm3, mm2 | ||
| 429 : | edgomez | 1174 | |
| 430 : | edgomez | 1176 | movq mm2, [inter_matrix_fix + ecx*8 + 8] |
| 431 : | pmulhw mm3, mm2 | ||
| 432 : | edgomez | 1174 | |
| 433 : | edgomez | 1176 | pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
| 434 : | pmulhw mm3, mm7 ; | ||
| 435 : | psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 | ||
| 436 : | psrlw mm3, 1 | ||
| 437 : | edgomez | 1174 | |
| 438 : | edgomez | 1176 | paddw mm5, mm0 ; sum += mm0 |
| 439 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 440 : | paddw mm5, mm3 ; | ||
| 441 : | pxor mm3, mm4 ; | ||
| 442 : | psubw mm0, mm1 ; undisplace | ||
| 443 : | psubw mm3, mm4 | ||
| 444 : | movq [edi + 8*ecx], mm0 | ||
| 445 : | movq [edi + 8*ecx + 8], mm3 | ||
| 446 : | edgomez | 1174 | |
| 447 : | edgomez | 1176 | add ecx, 2 |
| 448 : | cmp ecx, 16 | ||
| 449 : | jnz near .loop | ||
| 450 : | edgomez | 1174 | |
| 451 : | .done | ||
| 452 : | edgomez | 1176 | pmaddwd mm5, [mmx_one] |
| 453 : | movq mm0, mm5 | ||
| 454 : | psrlq mm5, 32 | ||
| 455 : | paddd mm0, mm5 | ||
| 456 : | movd eax, mm0 ; return sum | ||
| 457 : | edgomez | 1174 | |
| 458 : | edgomez | 1176 | pop edi |
| 459 : | pop esi | ||
| 460 : | pop ecx | ||
| 461 : | edgomez | 1174 | |
| 462 : | edgomez | 1176 | ret |
| 463 : | edgomez | 1174 | |
| 464 : | align ALIGN | ||
| 465 : | .q1loop | ||
| 466 : | edgomez | 1176 | movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
| 467 : | movq mm3, [esi + 8*ecx+ 8] | ||
| 468 : | edgomez | 1174 | ; |
| 469 : | edgomez | 1176 | pxor mm1, mm1 ; mm1 = 0 |
| 470 : | pxor mm4, mm4 ; | ||
| 471 : | edgomez | 1174 | |
| 472 : | edgomez | 1176 | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
| 473 : | pcmpgtw mm4, mm3 ; | ||
| 474 : | edgomez | 1174 | |
| 475 : | edgomez | 1176 | pxor mm0, mm1 ; mm0 = |mm0| |
| 476 : | pxor mm3, mm4 ; | ||
| 477 : | psubw mm0, mm1 ; displace | ||
| 478 : | psubw mm3, mm4 ; | ||
| 479 : | edgomez | 1174 | |
| 480 : | edgomez | 1176 | psllw mm0, 4 |
| 481 : | psllw mm3, 4 | ||
| 482 : | edgomez | 1174 | |
| 483 : | edgomez | 1176 | movq mm2, [inter_matrix + 8*ecx] |
| 484 : | psrlw mm2, 1 | ||
| 485 : | paddw mm0, mm2 | ||
| 486 : | edgomez | 1174 | |
| 487 : | edgomez | 1176 | movq mm2, [inter_matrix_fix + ecx*8] |
| 488 : | pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] | ||
| 489 : | edgomez | 1174 | |
| 490 : | edgomez | 1176 | movq mm2, [inter_matrix + 8*ecx + 8] |
| 491 : | psrlw mm2, 1 | ||
| 492 : | paddw mm3, mm2 | ||
| 493 : | |||
| 494 : | movq mm2, [inter_matrix_fix + ecx*8 + 8] | ||
| 495 : | pmulhw mm3, mm2 | ||
| 496 : | edgomez | 1174 | |
| 497 : | edgomez | 1176 | psrlw mm0, 1 ; mm0 >>= 1 (/2) |
| 498 : | psrlw mm3, 1 ; | ||
| 499 : | edgomez | 1174 | |
| 500 : | edgomez | 1176 | paddw mm5, mm0 ; sum += mm0 |
| 501 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 502 : | paddw mm5, mm3 ; | ||
| 503 : | pxor mm3, mm4 ; | ||
| 504 : | psubw mm0, mm1 ; undisplace | ||
| 505 : | psubw mm3, mm4 | ||
| 506 : | edgomez | 1174 | |
| 507 : | edgomez | 1176 | movq [edi + 8*ecx], mm0 |
| 508 : | movq [edi + 8*ecx + 8], mm3 | ||
| 509 : | edgomez | 1174 | |
| 510 : | edgomez | 1176 | add ecx, 2 |
| 511 : | cmp ecx, 16 | ||
| 512 : | jnz near .q1loop | ||
| 513 : | edgomez | 1174 | |
| 514 : | edgomez | 1176 | jmp .done |
| 515 : | edgomez | 1174 | |
| 516 : | |||
| 517 : | align ALIGN | ||
| 518 : | .q2loop | ||
| 519 : | edgomez | 1176 | movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
| 520 : | movq mm3, [esi + 8*ecx+ 8] | ||
| 521 : | edgomez | 1174 | ; |
| 522 : | edgomez | 1176 | pxor mm1, mm1 ; mm1 = 0 |
| 523 : | pxor mm4, mm4 ; | ||
| 524 : | edgomez | 1174 | |
| 525 : | edgomez | 1176 | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
| 526 : | pcmpgtw mm4, mm3 ; | ||
| 527 : | edgomez | 1174 | |
| 528 : | edgomez | 1176 | pxor mm0, mm1 ; mm0 = |mm0| |
| 529 : | pxor mm3, mm4 ; | ||
| 530 : | psubw mm0, mm1 ; displace | ||
| 531 : | psubw mm3, mm4 ; | ||
| 532 : | edgomez | 1174 | |
| 533 : | edgomez | 1176 | psllw mm0, 4 |
| 534 : | psllw mm3, 4 | ||
| 535 : | edgomez | 1174 | |
| 536 : | edgomez | 1176 | movq mm2, [inter_matrix + 8*ecx] |
| 537 : | psrlw mm2, 1 | ||
| 538 : | paddw mm0, mm2 | ||
| 539 : | edgomez | 1174 | |
| 540 : | edgomez | 1176 | movq mm2, [inter_matrix_fix + ecx*8] |
| 541 : | pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] | ||
| 542 : | edgomez | 1174 | |
| 543 : | edgomez | 1176 | movq mm2, [inter_matrix + 8*ecx + 8] |
| 544 : | psrlw mm2, 1 | ||
| 545 : | paddw mm3, mm2 | ||
| 546 : | edgomez | 1174 | |
| 547 : | edgomez | 1176 | movq mm2, [inter_matrix_fix + ecx*8 + 8] |
| 548 : | pmulhw mm3, mm2 | ||
| 549 : | edgomez | 1174 | |
| 550 : | edgomez | 1176 | psrlw mm0, 2 ; mm0 >>= 1 (/2) |
| 551 : | psrlw mm3, 2 ; | ||
| 552 : | edgomez | 1174 | |
| 553 : | edgomez | 1176 | paddw mm5, mm0 ; sum += mm0 |
| 554 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 555 : | paddw mm5, mm3 ; | ||
| 556 : | pxor mm3, mm4 ; | ||
| 557 : | psubw mm0, mm1 ; undisplace | ||
| 558 : | psubw mm3, mm4 | ||
| 559 : | edgomez | 1174 | |
| 560 : | edgomez | 1176 | movq [edi + 8*ecx], mm0 |
| 561 : | movq [edi + 8*ecx + 8], mm3 | ||
| 562 : | edgomez | 1174 | |
| 563 : | edgomez | 1176 | add ecx, 2 |
| 564 : | cmp ecx, 16 | ||
| 565 : | jnz near .q2loop | ||
| 566 : | edgomez | 1174 | |
| 567 : | edgomez | 1176 | jmp .done |
| 568 : | edgomez | 1174 | |
| 569 : | |||
| 570 : | ;=========================================================================== | ||
| 571 : | ; | ||
| 572 : | edgomez | 1176 | ; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
| 573 : | ; const int16_t const *coeff, | ||
| 574 : | ; const uint32_t quant, | ||
| 575 : | ; const uint32_t dcscalar); | ||
| 576 : | edgomez | 1174 | ; |
| 577 : | ;=========================================================================== | ||
| 578 : | |||
| 579 : | ; Note: in order to saturate 'easily', we pre-shift the quantifier | ||
| 580 : | ; by 4. Then, the high-word of (coeff[]*matrix[i]*quant) are used to | ||
| 581 : | ; build a saturating mask. It is non-zero only when an overflow occured. | ||
| 582 : | ; We thus avoid packing/unpacking toward double-word. | ||
| 583 : | ; Moreover, we perform the mult (matrix[i]*quant) first, instead of, e.g., | ||
| 584 : | ; (coeff[i]*matrix[i]). This is less prone to overflow if coeff[] are not | ||
| 585 : | ; checked. Input ranges are: coeff in [-127,127], inter_matrix in [1..255],a | ||
| 586 : | ; and quant in [1..31]. | ||
| 587 : | ; | ||
| 588 : | ; The original loop is: | ||
| 589 : | ; | ||
| 590 : | %if 0 | ||
| 591 : | edgomez | 1176 | movq mm0, [ecx+8*eax + 8*16] ; mm0 = coeff[i] |
| 592 : | pxor mm1, mm1 | ||
| 593 : | pcmpgtw mm1, mm0 | ||
| 594 : | pxor mm0, mm1 ; change sign if negative | ||
| 595 : | psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] | ||
| 596 : | edgomez | 1174 | |
| 597 : | edgomez | 1176 | movq mm2, mm7 ; mm2 = quant |
| 598 : | pmullw mm2, [intra_matrix + 8*eax + 8*16 ] ; matrix[i]*quant. | ||
| 599 : | edgomez | 1174 | |
| 600 : | edgomez | 1176 | movq mm6, mm2 |
| 601 : | pmulhw mm2, mm0 ; high of coeff*(matrix*quant) (should be 0 if no overflow) | ||
| 602 : | pmullw mm0, mm6 ; low of coeff*(matrix*quant) | ||
| 603 : | edgomez | 1174 | |
| 604 : | edgomez | 1176 | pxor mm5, mm5 |
| 605 : | pcmpgtw mm2, mm5 ; otherflow? | ||
| 606 : | psrlw mm2, 5 ; =0 if no clamp, 2047 otherwise | ||
| 607 : | psrlw mm0, 5 | ||
| 608 : | paddw mm0, mm1 ; start restoring sign | ||
| 609 : | por mm0, mm2 ; saturate to 2047 if needed | ||
| 610 : | pxor mm0, mm1 ; finish negating back | ||
| 611 : | edgomez | 1174 | |
| 612 : | edgomez | 1176 | movq [edx + 8*eax + 8*16], mm0 ; data[i] |
| 613 : | add eax, 1 | ||
| 614 : | edgomez | 1174 | %endif |
| 615 : | |||
| 616 : | ;******************************************************************** | ||
| 617 : | |||
| 618 : | align 16 | ||
| 619 : | cglobal dequant_mpeg_intra_mmx | ||
| 620 : | dequant_mpeg_intra_mmx: | ||
| 621 : | |||
| 622 : | edgomez | 1176 | mov edx, [esp+4] ; data |
| 623 : | mov ecx, [esp+8] ; coeff | ||
| 624 : | mov eax, [esp+12] ; quant | ||
| 625 : | edgomez | 1174 | |
| 626 : | edgomez | 1176 | movq mm7, [mmx_mul_quant + eax*8 - 8] |
| 627 : | mov eax, -16 ; to keep aligned, we regularly process coeff[0] | ||
| 628 : | psllw mm7, 2 ; << 2. See comment. | ||
| 629 : | pxor mm6, mm6 ; this is a NOP | ||
| 630 : | edgomez | 1174 | |
| 631 : | align 16 | ||
| 632 : | .loop | ||
| 633 : | edgomez | 1176 | movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
| 634 : | movq mm3, [ecx+8*eax + 8*16 +8]; mm3 = c' = coeff[i+1] | ||
| 635 : | pxor mm1, mm1 | ||
| 636 : | pxor mm4, mm4 | ||
| 637 : | pcmpgtw mm1, mm0 ; mm1 = sgn(c) | ||
| 638 : | movq mm2, mm7 ; mm2 = quant | ||
| 639 : | edgomez | 1174 | |
| 640 : | edgomez | 1176 | pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
| 641 : | pmullw mm2, [intra_matrix + 8*eax + 8*16 ] ; matrix[i]*quant | ||
| 642 : | edgomez | 1174 | |
| 643 : | edgomez | 1176 | pxor mm0, mm1 ; negate if negative |
| 644 : | pxor mm3, mm4 ; negate if negative | ||
| 645 : | edgomez | 1174 | |
| 646 : | edgomez | 1176 | psubw mm0, mm1 |
| 647 : | psubw mm3, mm4 | ||
| 648 : | edgomez | 1174 | |
| 649 : | ; we're short on register, here. Poor pairing... | ||
| 650 : | |||
| 651 : | edgomez | 1176 | movq mm5, mm2 |
| 652 : | pmullw mm2, mm0 ; low of coeff*(matrix*quant) | ||
| 653 : | edgomez | 1174 | |
| 654 : | edgomez | 1176 | pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
| 655 : | movq mm5, mm7 ; mm2 = quant | ||
| 656 : | edgomez | 1174 | |
| 657 : | edgomez | 1176 | pmullw mm5, [intra_matrix + 8*eax + 8*16 +8] ; matrix[i+1]*quant |
| 658 : | edgomez | 1174 | |
| 659 : | edgomez | 1176 | movq mm6, mm5 |
| 660 : | add eax,2 ; z-flag will be tested later | ||
| 661 : | edgomez | 1174 | |
| 662 : | edgomez | 1176 | pmullw mm6, mm3 ; low of coeff*(matrix*quant) |
| 663 : | pmulhw mm3, mm5 ; high of coeff*(matrix*quant) | ||
| 664 : | edgomez | 1174 | |
| 665 : | edgomez | 1176 | pcmpgtw mm0, [zero] |
| 666 : | paddusw mm2, mm0 | ||
| 667 : | psrlw mm2, 5 | ||
| 668 : | edgomez | 1174 | |
| 669 : | edgomez | 1176 | pcmpgtw mm3, [zero] |
| 670 : | paddusw mm6, mm3 | ||
| 671 : | psrlw mm6, 5 | ||
| 672 : | edgomez | 1174 | |
| 673 : | edgomez | 1176 | pxor mm2, mm1 ; start negating back |
| 674 : | pxor mm6, mm4 ; start negating back | ||
| 675 : | edgomez | 1174 | |
| 676 : | edgomez | 1176 | psubusw mm1, mm0 |
| 677 : | psubusw mm4, mm3 | ||
| 678 : | edgomez | 1174 | |
| 679 : | edgomez | 1176 | psubw mm2, mm1 ; finish negating back |
| 680 : | psubw mm6, mm4 ; finish negating back | ||
| 681 : | edgomez | 1174 | |
| 682 : | edgomez | 1176 | movq [edx + 8*eax + 8*16 -2*8 ], mm2 ; data[i] |
| 683 : | movq [edx + 8*eax + 8*16 -2*8 +8], mm6 ; data[i+1] | ||
| 684 : | edgomez | 1174 | |
| 685 : | edgomez | 1176 | jnz near .loop |
| 686 : | edgomez | 1174 | |
| 687 : | ; deal with DC | ||
| 688 : | |||
| 689 : | edgomez | 1176 | movd mm0, [ecx] |
| 690 : | pmullw mm0, [esp+16] ; dcscalar | ||
| 691 : | movq mm2, [mmx_32767_minus_2047] | ||
| 692 : | paddsw mm0, mm2 | ||
| 693 : | psubsw mm0, mm2 | ||
| 694 : | movq mm2, [mmx_32768_minus_2048] | ||
| 695 : | psubsw mm0, mm2 | ||
| 696 : | paddsw mm0, mm2 | ||
| 697 : | movd eax, mm0 | ||
| 698 : | mov [edx], ax | ||
| 699 : | edgomez | 1174 | |
| 700 : | edgomez | 1176 | xor eax, eax |
| 701 : | ret | ||
| 702 : | edgomez | 1174 | |
| 703 : | ;=========================================================================== | ||
| 704 : | ; | ||
| 705 : | edgomez | 1176 | ; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
| 706 : | ; const int16_t * const coeff, | ||
| 707 : | ; const uint32_t quant); | ||
| 708 : | edgomez | 1174 | ; |
| 709 : | ;=========================================================================== | ||
| 710 : | |||
| 711 : | ; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier | ||
| 712 : | ; so we handle the 3 cases: c<0, c==0, and c>0 in one shot. | ||
| 713 : | ; sgn(x) is the result of 'pcmpgtw 0,x': 0 if x>=0, -1 if x<0. | ||
| 714 : | ; It's mixed with the extraction of the absolute value. | ||
| 715 : | |||
| 716 : | align 16 | ||
| 717 : | cglobal dequant_mpeg_inter_mmx | ||
| 718 : | dequant_mpeg_inter_mmx: | ||
| 719 : | |||
| 720 : | edgomez | 1176 | mov edx, [esp+ 4] ; data |
| 721 : | mov ecx, [esp+ 8] ; coeff | ||
| 722 : | mov eax, [esp+12] ; quant | ||
| 723 : | movq mm7, [mmx_mul_quant + eax*8 - 8] | ||
| 724 : | mov eax, -16 | ||
| 725 : | paddw mm7, mm7 ; << 1 | ||
| 726 : | pxor mm6, mm6 ; mismatch sum | ||
| 727 : | edgomez | 1174 | |
| 728 : | align 16 | ||
| 729 : | .loop | ||
| 730 : | edgomez | 1176 | movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
| 731 : | movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] | ||
| 732 : | add eax, 2 | ||
| 733 : | edgomez | 1174 | |
| 734 : | edgomez | 1176 | pxor mm1, mm1 |
| 735 : | pxor mm3, mm3 | ||
| 736 : | pcmpgtw mm1, mm0 ; mm1 = sgn(c) (preserved) | ||
| 737 : | pcmpgtw mm3, mm2 ; mm3 = sgn(c') (preserved) | ||
| 738 : | paddsw mm0, mm1 ; c += sgn(c) | ||
| 739 : | paddsw mm2, mm3 ; c += sgn(c') | ||
| 740 : | paddw mm0, mm0 ; c *= 2 | ||
| 741 : | paddw mm2, mm2 ; c'*= 2 | ||
| 742 : | edgomez | 1174 | |
| 743 : | edgomez | 1176 | pxor mm4, mm4 |
| 744 : | pxor mm5, mm5 | ||
| 745 : | psubw mm4, mm0 ; -c | ||
| 746 : | psubw mm5, mm2 ; -c' | ||
| 747 : | psraw mm4, 16 ; mm4 = sgn(-c) | ||
| 748 : | psraw mm5, 16 ; mm5 = sgn(-c') | ||
| 749 : | psubsw mm0, mm4 ; c -= sgn(-c) | ||
| 750 : | psubsw mm2, mm5 ; c' -= sgn(-c') | ||
| 751 : | pxor mm0, mm1 ; finish changing sign if needed | ||
| 752 : | pxor mm2, mm3 ; finish changing sign if needed | ||
| 753 : | edgomez | 1174 | |
| 754 : | ; we're short on register, here. Poor pairing... | ||
| 755 : | |||
| 756 : | edgomez | 1176 | movq mm4, mm7 ; (matrix*quant) |
| 757 : | pmullw mm4, [inter_matrix + 8*eax + 8*16 -2*8] | ||
| 758 : | movq mm5, mm4 | ||
| 759 : | pmulhw mm5, mm0 ; high of c*(matrix*quant) | ||
| 760 : | pmullw mm0, mm4 ; low of c*(matrix*quant) | ||
| 761 : | edgomez | 1174 | |
| 762 : | edgomez | 1176 | movq mm4, mm7 ; (matrix*quant) |
| 763 : | pmullw mm4, [inter_matrix + 8*eax + 8*16 -2*8 + 8] | ||
| 764 : | edgomez | 1174 | |
| 765 : | edgomez | 1176 | pcmpgtw mm5, [zero] |
| 766 : | paddusw mm0, mm5 | ||
| 767 : | psrlw mm0, 5 | ||
| 768 : | pxor mm0, mm1 ; start restoring sign | ||
| 769 : | psubusw mm1, mm5 | ||
| 770 : | edgomez | 1174 | |
| 771 : | edgomez | 1176 | movq mm5, mm4 |
| 772 : | pmulhw mm5, mm2 ; high of c*(matrix*quant) | ||
| 773 : | pmullw mm2, mm4 ; low of c*(matrix*quant) | ||
| 774 : | psubw mm0, mm1 ; finish restoring sign | ||
| 775 : | edgomez | 1174 | |
| 776 : | edgomez | 1176 | pcmpgtw mm5, [zero] |
| 777 : | paddusw mm2, mm5 | ||
| 778 : | psrlw mm2, 5 | ||
| 779 : | pxor mm2, mm3 ; start restoring sign | ||
| 780 : | psubusw mm3, mm5 | ||
| 781 : | psubw mm2, mm3 ; finish restoring sign | ||
| 782 : | edgomez | 1174 | |
| 783 : | edgomez | 1176 | pxor mm6, mm0 ; mismatch control |
| 784 : | movq [edx + 8*eax + 8*16 -2*8 ], mm0 ; data[i] | ||
| 785 : | pxor mm6, mm2 ; mismatch control | ||
| 786 : | movq [edx + 8*eax + 8*16 -2*8 +8], mm2 ; data[i+1] | ||
| 787 : | edgomez | 1174 | |
| 788 : | edgomez | 1176 | jnz near .loop |
| 789 : | edgomez | 1174 | |
| 790 : | ; mismatch control | ||
| 791 : | |||
| 792 : | edgomez | 1176 | movq mm0, mm6 |
| 793 : | psrlq mm0, 48 | ||
| 794 : | movq mm1, mm6 | ||
| 795 : | movq mm2, mm6 | ||
| 796 : | psrlq mm1, 32 | ||
| 797 : | pxor mm6, mm0 | ||
| 798 : | psrlq mm2, 16 | ||
| 799 : | pxor mm6, mm1 | ||
| 800 : | pxor mm6, mm2 | ||
| 801 : | movd eax, mm6 | ||
| 802 : | and eax, 1 | ||
| 803 : | xor eax, 1 | ||
| 804 : | xor word [edx + 2*63], ax | ||
| 805 : | edgomez | 1174 | |
| 806 : | edgomez | 1176 | xor eax, eax |
| 807 : | ret |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |