Parent Directory
| 
Revision Log
Revision 1798 - (view) (download)
| 1 : | edgomez | 1382 | ;/************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * - 3dne Quantization/Dequantization - | ||
| 5 : | ; * | ||
| 6 : | ; * Copyright (C) 2002-2003 Peter Ross <pross@xvid.org> | ||
| 7 : | Isibaar | 1795 | ; * 2002-2008 Michael Militzer <michael@xvid.org> |
| 8 : | edgomez | 1382 | ; * 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 : | Isibaar | 1798 | ; * $Id: quantize_mpeg_mmx.asm,v 1.12 2008-11-26 02:21:02 Isibaar Exp $ |
| 25 : | edgomez | 1382 | ; * |
| 26 : | ; *************************************************************************/ | ||
| 27 : | |||
| 28 : | %define SATURATE | ||
| 29 : | |||
| 30 : | Isibaar | 1795 | %include "nasm.inc" |
| 31 : | edgomez | 1382 | |
| 32 : | ;============================================================================= | ||
| 33 : | ; Local data (Read Only) | ||
| 34 : | ;============================================================================= | ||
| 35 : | |||
| 36 : | Isibaar | 1795 | DATA |
| 37 : | edgomez | 1382 | |
| 38 : | mmx_one: | ||
| 39 : | times 4 dw 1 | ||
| 40 : | |||
| 41 : | ;----------------------------------------------------------------------------- | ||
| 42 : | ; divide by 2Q table | ||
| 43 : | ;----------------------------------------------------------------------------- | ||
| 44 : | |||
| 45 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 46 : | edgomez | 1382 | mmx_div: |
| 47 : | times 4 dw 65535 ; the div by 2 formula will overflow for the case | ||
| 48 : | ; quant=1 but we don't care much because quant=1 | ||
| 49 : | ; is handled by a different piece of code that | ||
| 50 : | ; doesn't use this table. | ||
| 51 : | %assign quant 2 | ||
| 52 : | %rep 30 | ||
| 53 : | times 4 dw (1<<17) / (quant*2) + 1 | ||
| 54 : | %assign quant quant+1 | ||
| 55 : | %endrep | ||
| 56 : | |||
| 57 : | %define VM18P 3 | ||
| 58 : | %define VM18Q 4 | ||
| 59 : | |||
| 60 : | |||
| 61 : | ;----------------------------------------------------------------------------- | ||
| 62 : | ; quantd table | ||
| 63 : | ;----------------------------------------------------------------------------- | ||
| 64 : | |||
| 65 : | quantd: | ||
| 66 : | %assign quant 1 | ||
| 67 : | %rep 31 | ||
| 68 : | times 4 dw ((VM18P*quant) + (VM18Q/2)) / VM18Q | ||
| 69 : | %assign quant quant+1 | ||
| 70 : | %endrep | ||
| 71 : | |||
| 72 : | ;----------------------------------------------------------------------------- | ||
| 73 : | ; multiple by 2Q table | ||
| 74 : | ;----------------------------------------------------------------------------- | ||
| 75 : | |||
| 76 : | mmx_mul_quant: | ||
| 77 : | %assign quant 1 | ||
| 78 : | %rep 31 | ||
| 79 : | times 4 dw quant | ||
| 80 : | %assign quant quant+1 | ||
| 81 : | %endrep | ||
| 82 : | |||
| 83 : | ;----------------------------------------------------------------------------- | ||
| 84 : | ; saturation limits | ||
| 85 : | ;----------------------------------------------------------------------------- | ||
| 86 : | |||
| 87 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 88 : | edgomez | 1382 | |
| 89 : | mmx_32767_minus_2047: | ||
| 90 : | times 4 dw (32767-2047) | ||
| 91 : | mmx_32768_minus_2048: | ||
| 92 : | times 4 dw (32768-2048) | ||
| 93 : | mmx_2047: | ||
| 94 : | times 4 dw 2047 | ||
| 95 : | mmx_minus_2048: | ||
| 96 : | times 4 dw (-2048) | ||
| 97 : | zero: | ||
| 98 : | times 4 dw 0 | ||
| 99 : | |||
| 100 : | ;============================================================================= | ||
| 101 : | syskin | 1713 | ; rounding |
| 102 : | ;============================================================================= | ||
| 103 : | |||
| 104 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 105 : | |||
| 106 : | syskin | 1713 | mmx_rounding: |
| 107 : | Isibaar | 1795 | dd (1<<13) |
| 108 : | dd (1<<13) | ||
| 109 : | syskin | 1713 | |
| 110 : | ;============================================================================= | ||
| 111 : | edgomez | 1382 | ; Code |
| 112 : | ;============================================================================= | ||
| 113 : | |||
| 114 : | Isibaar | 1795 | SECTION .rotext align=SECTION_ALIGN |
| 115 : | edgomez | 1382 | |
| 116 : | cglobal quant_mpeg_intra_mmx | ||
| 117 : | cglobal quant_mpeg_inter_mmx | ||
| 118 : | cglobal dequant_mpeg_intra_mmx | ||
| 119 : | cglobal dequant_mpeg_inter_mmx | ||
| 120 : | |||
| 121 : | syskin | 1713 | |
| 122 : | %macro QUANT_MMX 1 | ||
| 123 : | Isibaar | 1795 | movq mm0, [_EAX + 16*(%1)] ; data |
| 124 : | movq mm2, [TMP0 + 16*(%1) + 128] ; intra_matrix_rec | ||
| 125 : | movq mm4, [_EAX + 16*(%1) + 8] ; data | ||
| 126 : | movq mm6, [TMP0 + 16*(%1) + 128 + 8] ; intra_matrix_rec | ||
| 127 : | syskin | 1713 | |
| 128 : | movq mm1, mm0 | ||
| 129 : | movq mm5, mm4 | ||
| 130 : | |||
| 131 : | pmullw mm0, mm2 ; low results | ||
| 132 : | pmulhw mm1, mm2 ; high results | ||
| 133 : | pmullw mm4, mm6 ; low results | ||
| 134 : | pmulhw mm5, mm6 ; high results | ||
| 135 : | |||
| 136 : | movq mm2, mm0 | ||
| 137 : | movq mm6, mm4 | ||
| 138 : | |||
| 139 : | punpckhwd mm0, mm1 | ||
| 140 : | punpcklwd mm2, mm1 | ||
| 141 : | punpckhwd mm4, mm5 | ||
| 142 : | punpcklwd mm6, mm5 | ||
| 143 : | |||
| 144 : | paddd mm2, mm7 | ||
| 145 : | paddd mm0, mm7 | ||
| 146 : | paddd mm6, mm7 | ||
| 147 : | paddd mm4, mm7 | ||
| 148 : | |||
| 149 : | psrad mm2, 14 | ||
| 150 : | psrad mm0, 14 | ||
| 151 : | psrad mm6, 14 | ||
| 152 : | psrad mm4, 14 | ||
| 153 : | |||
| 154 : | packssdw mm2, mm0 | ||
| 155 : | packssdw mm6, mm4 | ||
| 156 : | |||
| 157 : | Isibaar | 1795 | movq [TMP1 + 16*(%1)], mm2 |
| 158 : | movq [TMP1 + 16*(%1)+8], mm6 | ||
| 159 : | syskin | 1713 | %endmacro |
| 160 : | |||
| 161 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
| 162 : | ; | ||
| 163 : | ; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, | ||
| 164 : | ; const int16_t const * data, | ||
| 165 : | ; const uint32_t quant, | ||
| 166 : | ; const uint32_t dcscalar, | ||
| 167 : | ; const uint16_t *mpeg_matrices); | ||
| 168 : | ; | ||
| 169 : | ;----------------------------------------------------------------------------- | ||
| 170 : | |||
| 171 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 172 : | edgomez | 1382 | quant_mpeg_intra_mmx: |
| 173 : | |||
| 174 : | Isibaar | 1795 | mov _EAX, prm2 ; data |
| 175 : | mov TMP0, prm5 ; mpeg_quant_matrices | ||
| 176 : | mov TMP1, prm1 ; coeff | ||
| 177 : | |||
| 178 : | syskin | 1713 | movq mm7, [mmx_rounding] |
| 179 : | edgomez | 1382 | |
| 180 : | Isibaar | 1798 | QUANT_MMX(0) |
| 181 : | QUANT_MMX(1) | ||
| 182 : | QUANT_MMX(2) | ||
| 183 : | QUANT_MMX(3) | ||
| 184 : | QUANT_MMX(4) | ||
| 185 : | QUANT_MMX(5) | ||
| 186 : | QUANT_MMX(6) | ||
| 187 : | QUANT_MMX(7) | ||
| 188 : | edgomez | 1382 | |
| 189 : | syskin | 1713 | ; calculate DC |
| 190 : | Isibaar | 1798 | movsx _EAX, word [_EAX] ; data[0] |
| 191 : | mov TMP0, prm4 ; dcscalar | ||
| 192 : | mov _EDX, _EAX | ||
| 193 : | shr TMP0, 1 ; TMP0 = dcscalar/2 | ||
| 194 : | sar _EDX, 31 ; TMP1 = sign extend of _EAX (ready for division too) | ||
| 195 : | xor TMP0, _EDX ; adjust TMP0 according to the sign of data[0] | ||
| 196 : | sub TMP0, _EDX | ||
| 197 : | add _EAX, TMP0 | ||
| 198 : | syskin | 1713 | |
| 199 : | Isibaar | 1798 | mov TMP0, prm4 ; dcscalar |
| 200 : | idiv TMP0 ; _EAX = _EDX:_EAX / dcscalar | ||
| 201 : | edgomez | 1382 | |
| 202 : | Isibaar | 1798 | mov _EDX, prm1 ; coeff again |
| 203 : | mov word [_EDX], ax ; coeff[0] = ax | ||
| 204 : | Isibaar | 1795 | |
| 205 : | xor _EAX, _EAX ; return(0); | ||
| 206 : | edgomez | 1382 | ret |
| 207 : | Isibaar | 1793 | ENDFUNC |
| 208 : | edgomez | 1382 | |
| 209 : | |||
| 210 : | ;----------------------------------------------------------------------------- | ||
| 211 : | ; | ||
| 212 : | ; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, | ||
| 213 : | ; const int16_t const * data, | ||
| 214 : | ; const uint32_t quant, | ||
| 215 : | ; const uint16_t *mpeg_matrices); | ||
| 216 : | ; | ||
| 217 : | ;----------------------------------------------------------------------------- | ||
| 218 : | |||
| 219 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 220 : | edgomez | 1382 | quant_mpeg_inter_mmx: |
| 221 : | |||
| 222 : | Isibaar | 1795 | mov TMP1, prm1 ; coeff |
| 223 : | mov _EAX, prm3 ; quant | ||
| 224 : | mov TMP0, prm4 ; mpeg_quant_matrices | ||
| 225 : | edgomez | 1382 | |
| 226 : | Isibaar | 1795 | push _ESI |
| 227 : | %ifdef ARCH_IS_X86_64 | ||
| 228 : | mov _ESI, prm2 ; data | ||
| 229 : | %else | ||
| 230 : | mov _ESI, [_ESP + 4 + 8] ; data | ||
| 231 : | %endif | ||
| 232 : | edgomez | 1382 | |
| 233 : | Isibaar | 1795 | push _EBX |
| 234 : | xor _EBX, _EBX | ||
| 235 : | edgomez | 1382 | |
| 236 : | pxor mm5, mm5 ; sum | ||
| 237 : | |||
| 238 : | cmp al, 1 | ||
| 239 : | jz near .q1loop | ||
| 240 : | |||
| 241 : | cmp al, 2 | ||
| 242 : | jz near .q2loop | ||
| 243 : | |||
| 244 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 245 : | lea r9, [mmx_div] | ||
| 246 : | movq mm7, [r9 + _EAX * 8 - 8] | ||
| 247 : | %else | ||
| 248 : | movq mm7, [mmx_div + _EAX * 8 - 8] ; divider | ||
| 249 : | %endif | ||
| 250 : | edgomez | 1382 | |
| 251 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 252 : | Isibaar | 1793 | .loop: |
| 253 : | Isibaar | 1795 | movq mm0, [_ESI + 8*_EBX] ; mm0 = [1st] |
| 254 : | movq mm3, [_ESI + 8*_EBX + 8] ; | ||
| 255 : | edgomez | 1382 | pxor mm1, mm1 ; mm1 = 0 |
| 256 : | pxor mm4, mm4 ; | ||
| 257 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 258 : | pcmpgtw mm4, mm3 ; | ||
| 259 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 260 : | pxor mm3, mm4 ; | ||
| 261 : | psubw mm0, mm1 ; displace | ||
| 262 : | psubw mm3, mm4 ; | ||
| 263 : | psllw mm0, 4 | ||
| 264 : | psllw mm3, 4 | ||
| 265 : | Isibaar | 1795 | movq mm2, [TMP0 + 512 + 8*_EBX] |
| 266 : | edgomez | 1382 | psrlw mm2, 1 |
| 267 : | paddw mm0, mm2 | ||
| 268 : | Isibaar | 1795 | movq mm2, [TMP0 + 768 + _EBX*8] |
| 269 : | edgomez | 1382 | pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
| 270 : | Isibaar | 1795 | movq mm2, [TMP0 + 512 + 8*_EBX + 8] |
| 271 : | edgomez | 1382 | psrlw mm2, 1 |
| 272 : | paddw mm3, mm2 | ||
| 273 : | Isibaar | 1795 | movq mm2, [TMP0 + 768 + _EBX*8 + 8] |
| 274 : | edgomez | 1382 | pmulhw mm3, mm2 |
| 275 : | pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 | ||
| 276 : | pmulhw mm3, mm7 ; | ||
| 277 : | psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 | ||
| 278 : | psrlw mm3, 1 | ||
| 279 : | paddw mm5, mm0 ; sum += mm0 | ||
| 280 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 281 : | paddw mm5, mm3 ; | ||
| 282 : | pxor mm3, mm4 ; | ||
| 283 : | psubw mm0, mm1 ; undisplace | ||
| 284 : | psubw mm3, mm4 | ||
| 285 : | Isibaar | 1795 | movq [TMP1 + 8*_EBX], mm0 |
| 286 : | movq [TMP1 + 8*_EBX + 8], mm3 | ||
| 287 : | edgomez | 1382 | |
| 288 : | Isibaar | 1795 | add _EBX, 2 |
| 289 : | cmp _EBX, 16 | ||
| 290 : | edgomez | 1382 | jnz near .loop |
| 291 : | |||
| 292 : | Isibaar | 1793 | .done: |
| 293 : | edgomez | 1382 | pmaddwd mm5, [mmx_one] |
| 294 : | movq mm0, mm5 | ||
| 295 : | psrlq mm5, 32 | ||
| 296 : | paddd mm0, mm5 | ||
| 297 : | movd eax, mm0 ; return sum | ||
| 298 : | |||
| 299 : | Isibaar | 1795 | pop _EBX |
| 300 : | pop _ESI | ||
| 301 : | edgomez | 1382 | |
| 302 : | ret | ||
| 303 : | |||
| 304 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 305 : | Isibaar | 1793 | .q1loop: |
| 306 : | Isibaar | 1795 | movq mm0, [_ESI + 8*_EBX] ; mm0 = [1st] |
| 307 : | movq mm3, [_ESI + 8*_EBX+ 8] | ||
| 308 : | edgomez | 1382 | pxor mm1, mm1 ; mm1 = 0 |
| 309 : | pxor mm4, mm4 ; | ||
| 310 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 311 : | pcmpgtw mm4, mm3 ; | ||
| 312 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 313 : | pxor mm3, mm4 ; | ||
| 314 : | psubw mm0, mm1 ; displace | ||
| 315 : | psubw mm3, mm4 ; | ||
| 316 : | psllw mm0, 4 | ||
| 317 : | psllw mm3, 4 | ||
| 318 : | Isibaar | 1795 | movq mm2, [TMP0 + 512 + 8*_EBX] |
| 319 : | edgomez | 1382 | psrlw mm2, 1 |
| 320 : | paddw mm0, mm2 | ||
| 321 : | Isibaar | 1795 | movq mm2, [TMP0 + 768 + _EBX*8] |
| 322 : | edgomez | 1382 | pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
| 323 : | Isibaar | 1795 | movq mm2, [TMP0 + 512 + 8*_EBX + 8] |
| 324 : | edgomez | 1382 | psrlw mm2, 1 |
| 325 : | paddw mm3, mm2 | ||
| 326 : | Isibaar | 1795 | movq mm2, [TMP0 + 768 + _EBX*8 + 8] |
| 327 : | edgomez | 1382 | pmulhw mm3, mm2 |
| 328 : | psrlw mm0, 1 ; mm0 >>= 1 (/2) | ||
| 329 : | psrlw mm3, 1 ; | ||
| 330 : | paddw mm5, mm0 ; sum += mm0 | ||
| 331 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 332 : | paddw mm5, mm3 ; | ||
| 333 : | pxor mm3, mm4 ; | ||
| 334 : | psubw mm0, mm1 ; undisplace | ||
| 335 : | psubw mm3, mm4 | ||
| 336 : | Isibaar | 1795 | movq [TMP1 + 8*_EBX], mm0 |
| 337 : | movq [TMP1 + 8*_EBX + 8], mm3 | ||
| 338 : | edgomez | 1382 | |
| 339 : | Isibaar | 1795 | add _EBX, 2 |
| 340 : | cmp _EBX, 16 | ||
| 341 : | edgomez | 1382 | jnz near .q1loop |
| 342 : | |||
| 343 : | jmp .done | ||
| 344 : | |||
| 345 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 346 : | Isibaar | 1793 | .q2loop: |
| 347 : | Isibaar | 1795 | movq mm0, [_ESI + 8*_EBX] ; mm0 = [1st] |
| 348 : | movq mm3, [_ESI + 8*_EBX+ 8] | ||
| 349 : | edgomez | 1382 | pxor mm1, mm1 ; mm1 = 0 |
| 350 : | pxor mm4, mm4 ; | ||
| 351 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 352 : | pcmpgtw mm4, mm3 ; | ||
| 353 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 354 : | pxor mm3, mm4 ; | ||
| 355 : | psubw mm0, mm1 ; displace | ||
| 356 : | psubw mm3, mm4 ; | ||
| 357 : | psllw mm0, 4 | ||
| 358 : | psllw mm3, 4 | ||
| 359 : | Isibaar | 1795 | movq mm2, [TMP0 + 512 + 8*_EBX] |
| 360 : | edgomez | 1382 | psrlw mm2, 1 |
| 361 : | paddw mm0, mm2 | ||
| 362 : | Isibaar | 1795 | movq mm2, [TMP0 + 768 + _EBX*8] |
| 363 : | edgomez | 1382 | pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
| 364 : | Isibaar | 1795 | movq mm2, [TMP0 + 512 + 8*_EBX + 8] |
| 365 : | edgomez | 1382 | psrlw mm2, 1 |
| 366 : | paddw mm3, mm2 | ||
| 367 : | Isibaar | 1795 | movq mm2, [TMP0 + 768 + _EBX*8 + 8] |
| 368 : | edgomez | 1382 | pmulhw mm3, mm2 |
| 369 : | psrlw mm0, 2 ; mm0 >>= 1 (/2) | ||
| 370 : | psrlw mm3, 2 ; | ||
| 371 : | paddw mm5, mm0 ; sum += mm0 | ||
| 372 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 373 : | paddw mm5, mm3 ; | ||
| 374 : | pxor mm3, mm4 ; | ||
| 375 : | psubw mm0, mm1 ; undisplace | ||
| 376 : | psubw mm3, mm4 | ||
| 377 : | Isibaar | 1795 | movq [TMP1 + 8*_EBX], mm0 |
| 378 : | movq [TMP1 + 8*_EBX + 8], mm3 | ||
| 379 : | edgomez | 1382 | |
| 380 : | Isibaar | 1795 | add _EBX, 2 |
| 381 : | cmp _EBX, 16 | ||
| 382 : | edgomez | 1382 | jnz near .q2loop |
| 383 : | |||
| 384 : | jmp .done | ||
| 385 : | Isibaar | 1793 | ENDFUNC |
| 386 : | edgomez | 1382 | |
| 387 : | |||
| 388 : | ;----------------------------------------------------------------------------- | ||
| 389 : | ; | ||
| 390 : | ; uint32_t dequant_mpeg_intra_mmx(int16_t *data, | ||
| 391 : | ; const int16_t const *coeff, | ||
| 392 : | ; const uint32_t quant, | ||
| 393 : | ; const uint32_t dcscalar, | ||
| 394 : | ; const uint16_t *mpeg_matrices); | ||
| 395 : | ; | ||
| 396 : | ;----------------------------------------------------------------------------- | ||
| 397 : | |||
| 398 : | ; Note: in order to saturate 'easily', we pre-shift the quantifier | ||
| 399 : | ; by 4. Then, the high-word of (coeff[]*matrix[i]*quant) are used to | ||
| 400 : | ; build a saturating mask. It is non-zero only when an overflow occured. | ||
| 401 : | ; We thus avoid packing/unpacking toward double-word. | ||
| 402 : | ; Moreover, we perform the mult (matrix[i]*quant) first, instead of, e.g., | ||
| 403 : | ; (coeff[i]*matrix[i]). This is less prone to overflow if coeff[] are not | ||
| 404 : | ; checked. Input ranges are: coeff in [-127,127], inter_matrix in [1..255],a | ||
| 405 : | ; and quant in [1..31]. | ||
| 406 : | ; | ||
| 407 : | ; The original loop is: | ||
| 408 : | ; | ||
| 409 : | %if 0 | ||
| 410 : | Isibaar | 1795 | movq mm0, [TMP0+8*_EAX + 8*16] ; mm0 = coeff[i] |
| 411 : | edgomez | 1382 | pxor mm1, mm1 |
| 412 : | pcmpgtw mm1, mm0 | ||
| 413 : | pxor mm0, mm1 ; change sign if negative | ||
| 414 : | psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] | ||
| 415 : | |||
| 416 : | movq mm2, mm7 ; mm2 = quant | ||
| 417 : | Isibaar | 1795 | pmullw mm2, [_EBX + 8*_EAX + 8*16 ] ; matrix[i]*quant. |
| 418 : | edgomez | 1382 | |
| 419 : | movq mm6, mm2 | ||
| 420 : | pmulhw mm2, mm0 ; high of coeff*(matrix*quant) (should be 0 if no overflow) | ||
| 421 : | pmullw mm0, mm6 ; low of coeff*(matrix*quant) | ||
| 422 : | |||
| 423 : | pxor mm5, mm5 | ||
| 424 : | pcmpgtw mm2, mm5 ; otherflow? | ||
| 425 : | psrlw mm2, 5 ; =0 if no clamp, 2047 otherwise | ||
| 426 : | psrlw mm0, 5 | ||
| 427 : | paddw mm0, mm1 ; start restoring sign | ||
| 428 : | por mm0, mm2 ; saturate to 2047 if needed | ||
| 429 : | pxor mm0, mm1 ; finish negating back | ||
| 430 : | |||
| 431 : | Isibaar | 1795 | movq [TMP1 + 8*_EAX + 8*16], mm0 ; data[i] |
| 432 : | add _EAX, 1 | ||
| 433 : | edgomez | 1382 | %endif |
| 434 : | |||
| 435 : | ;******************************************************************** | ||
| 436 : | |||
| 437 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 438 : | edgomez | 1382 | dequant_mpeg_intra_mmx: |
| 439 : | |||
| 440 : | Isibaar | 1795 | mov TMP1, prm1 ; data |
| 441 : | mov TMP0, prm2 ; coeff | ||
| 442 : | mov _EAX, prm5 ; mpeg_quant_matrices | ||
| 443 : | |||
| 444 : | push _EBX | ||
| 445 : | mov _EBX, _EAX | ||
| 446 : | %ifdef ARCH_IS_X86_64 | ||
| 447 : | mov _EAX, prm3 | ||
| 448 : | lea r9, [mmx_mul_quant] | ||
| 449 : | movq mm7, [r9 + _EAX*8 - 8] | ||
| 450 : | %else | ||
| 451 : | mov _EAX, [_ESP + 4 + 12] ; quant | ||
| 452 : | movq mm7, [mmx_mul_quant + _EAX*8 - 8] | ||
| 453 : | %endif | ||
| 454 : | edgomez | 1382 | |
| 455 : | Isibaar | 1795 | mov _EAX, -16 ; to keep ALIGNed, we regularly process coeff[0] |
| 456 : | edgomez | 1382 | psllw mm7, 2 ; << 2. See comment. |
| 457 : | pxor mm6, mm6 ; this is a NOP | ||
| 458 : | |||
| 459 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 460 : | Isibaar | 1793 | .loop: |
| 461 : | Isibaar | 1795 | movq mm0, [TMP0+8*_EAX + 8*16] ; mm0 = c = coeff[i] |
| 462 : | movq mm3, [TMP0+8*_EAX + 8*16 +8]; mm3 = c' = coeff[i+1] | ||
| 463 : | edgomez | 1382 | pxor mm1, mm1 |
| 464 : | pxor mm4, mm4 | ||
| 465 : | pcmpgtw mm1, mm0 ; mm1 = sgn(c) | ||
| 466 : | movq mm2, mm7 ; mm2 = quant | ||
| 467 : | |||
| 468 : | pcmpgtw mm4, mm3 ; mm4 = sgn(c') | ||
| 469 : | Isibaar | 1795 | pmullw mm2, [_EBX + 8*_EAX + 8*16 ] ; matrix[i]*quant |
| 470 : | edgomez | 1382 | |
| 471 : | pxor mm0, mm1 ; negate if negative | ||
| 472 : | pxor mm3, mm4 ; negate if negative | ||
| 473 : | |||
| 474 : | psubw mm0, mm1 | ||
| 475 : | psubw mm3, mm4 | ||
| 476 : | |||
| 477 : | ; we're short on register, here. Poor pairing... | ||
| 478 : | |||
| 479 : | movq mm5, mm2 | ||
| 480 : | pmullw mm2, mm0 ; low of coeff*(matrix*quant) | ||
| 481 : | |||
| 482 : | pmulhw mm0, mm5 ; high of coeff*(matrix*quant) | ||
| 483 : | movq mm5, mm7 ; mm2 = quant | ||
| 484 : | |||
| 485 : | Isibaar | 1795 | pmullw mm5, [_EBX + 8*_EAX + 8*16 +8] ; matrix[i+1]*quant |
| 486 : | edgomez | 1382 | |
| 487 : | movq mm6, mm5 | ||
| 488 : | Isibaar | 1795 | add _EAX,2 ; z-flag will be tested later |
| 489 : | edgomez | 1382 | |
| 490 : | pmullw mm6, mm3 ; low of coeff*(matrix*quant) | ||
| 491 : | pmulhw mm3, mm5 ; high of coeff*(matrix*quant) | ||
| 492 : | |||
| 493 : | pcmpgtw mm0, [zero] | ||
| 494 : | paddusw mm2, mm0 | ||
| 495 : | psrlw mm2, 5 | ||
| 496 : | |||
| 497 : | pcmpgtw mm3, [zero] | ||
| 498 : | paddusw mm6, mm3 | ||
| 499 : | psrlw mm6, 5 | ||
| 500 : | |||
| 501 : | pxor mm2, mm1 ; start negating back | ||
| 502 : | pxor mm6, mm4 ; start negating back | ||
| 503 : | |||
| 504 : | psubusw mm1, mm0 | ||
| 505 : | psubusw mm4, mm3 | ||
| 506 : | |||
| 507 : | psubw mm2, mm1 ; finish negating back | ||
| 508 : | psubw mm6, mm4 ; finish negating back | ||
| 509 : | |||
| 510 : | Isibaar | 1795 | movq [TMP1 + 8*_EAX + 8*16 -2*8 ], mm2 ; data[i] |
| 511 : | movq [TMP1 + 8*_EAX + 8*16 -2*8 +8], mm6 ; data[i+1] | ||
| 512 : | edgomez | 1382 | |
| 513 : | jnz near .loop | ||
| 514 : | |||
| 515 : | Isibaar | 1795 | pop _EBX |
| 516 : | |||
| 517 : | edgomez | 1382 | ; deal with DC |
| 518 : | Isibaar | 1795 | movd mm0, [TMP0] |
| 519 : | %ifdef ARCH_IS_X86_64 | ||
| 520 : | movq mm6, prm4 | ||
| 521 : | pmullw mm0, mm6 | ||
| 522 : | %else | ||
| 523 : | pmullw mm0, prm4 ; dcscalar | ||
| 524 : | %endif | ||
| 525 : | edgomez | 1382 | movq mm2, [mmx_32767_minus_2047] |
| 526 : | paddsw mm0, mm2 | ||
| 527 : | psubsw mm0, mm2 | ||
| 528 : | movq mm2, [mmx_32768_minus_2048] | ||
| 529 : | psubsw mm0, mm2 | ||
| 530 : | paddsw mm0, mm2 | ||
| 531 : | movd eax, mm0 | ||
| 532 : | Isibaar | 1795 | mov [TMP1], ax |
| 533 : | edgomez | 1382 | |
| 534 : | Isibaar | 1795 | xor _EAX, _EAX |
| 535 : | edgomez | 1382 | |
| 536 : | ret | ||
| 537 : | Isibaar | 1793 | ENDFUNC |
| 538 : | edgomez | 1382 | |
| 539 : | ;----------------------------------------------------------------------------- | ||
| 540 : | ; | ||
| 541 : | ; uint32_t dequant_mpeg_inter_mmx(int16_t * data, | ||
| 542 : | ; const int16_t * const coeff, | ||
| 543 : | ; const uint32_t quant, | ||
| 544 : | ; const uint16_t *mpeg_matrices); | ||
| 545 : | ; | ||
| 546 : | ;----------------------------------------------------------------------------- | ||
| 547 : | |||
| 548 : | ; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier | ||
| 549 : | ; so we handle the 3 cases: c<0, c==0, and c>0 in one shot. | ||
| 550 : | ; sgn(x) is the result of 'pcmpgtw 0,x': 0 if x>=0, -1 if x<0. | ||
| 551 : | ; It's mixed with the extraction of the absolute value. | ||
| 552 : | |||
| 553 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 554 : | edgomez | 1382 | dequant_mpeg_inter_mmx: |
| 555 : | |||
| 556 : | |||
| 557 : | Isibaar | 1795 | mov TMP1, prm1 ; data |
| 558 : | mov TMP0, prm2 ; coeff | ||
| 559 : | mov _EAX, prm3 ; quant | ||
| 560 : | edgomez | 1382 | |
| 561 : | Isibaar | 1795 | push _EBX |
| 562 : | %ifdef ARCH_IS_X86_64 | ||
| 563 : | mov _EBX, prm4 | ||
| 564 : | lea r9, [mmx_mul_quant] | ||
| 565 : | movq mm7, [r9 + _EAX*8 - 8] | ||
| 566 : | %else | ||
| 567 : | mov _EBX, [_ESP + 4 + 16] ; mpeg_quant_matrices | ||
| 568 : | movq mm7, [mmx_mul_quant + _EAX*8 - 8] | ||
| 569 : | %endif | ||
| 570 : | |||
| 571 : | mov _EAX, -16 | ||
| 572 : | edgomez | 1382 | paddw mm7, mm7 ; << 1 |
| 573 : | pxor mm6, mm6 ; mismatch sum | ||
| 574 : | |||
| 575 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 576 : | Isibaar | 1793 | .loop: |
| 577 : | Isibaar | 1795 | movq mm0, [TMP0+8*_EAX + 8*16 ] ; mm0 = coeff[i] |
| 578 : | movq mm2, [TMP0+8*_EAX + 8*16 +8] ; mm2 = coeff[i+1] | ||
| 579 : | add _EAX, 2 | ||
| 580 : | edgomez | 1382 | |
| 581 : | pxor mm1, mm1 | ||
| 582 : | pxor mm3, mm3 | ||
| 583 : | pcmpgtw mm1, mm0 ; mm1 = sgn(c) (preserved) | ||
| 584 : | pcmpgtw mm3, mm2 ; mm3 = sgn(c') (preserved) | ||
| 585 : | paddsw mm0, mm1 ; c += sgn(c) | ||
| 586 : | paddsw mm2, mm3 ; c += sgn(c') | ||
| 587 : | paddw mm0, mm0 ; c *= 2 | ||
| 588 : | paddw mm2, mm2 ; c'*= 2 | ||
| 589 : | |||
| 590 : | pxor mm4, mm4 | ||
| 591 : | pxor mm5, mm5 | ||
| 592 : | psubw mm4, mm0 ; -c | ||
| 593 : | psubw mm5, mm2 ; -c' | ||
| 594 : | psraw mm4, 16 ; mm4 = sgn(-c) | ||
| 595 : | psraw mm5, 16 ; mm5 = sgn(-c') | ||
| 596 : | psubsw mm0, mm4 ; c -= sgn(-c) | ||
| 597 : | psubsw mm2, mm5 ; c' -= sgn(-c') | ||
| 598 : | pxor mm0, mm1 ; finish changing sign if needed | ||
| 599 : | pxor mm2, mm3 ; finish changing sign if needed | ||
| 600 : | |||
| 601 : | ; we're short on register, here. Poor pairing... | ||
| 602 : | |||
| 603 : | movq mm4, mm7 ; (matrix*quant) | ||
| 604 : | Isibaar | 1795 | pmullw mm4, [_EBX + 512 + 8*_EAX + 8*16 -2*8] |
| 605 : | edgomez | 1382 | movq mm5, mm4 |
| 606 : | pmulhw mm5, mm0 ; high of c*(matrix*quant) | ||
| 607 : | pmullw mm0, mm4 ; low of c*(matrix*quant) | ||
| 608 : | |||
| 609 : | movq mm4, mm7 ; (matrix*quant) | ||
| 610 : | Isibaar | 1795 | pmullw mm4, [_EBX + 512 + 8*_EAX + 8*16 -2*8 + 8] |
| 611 : | edgomez | 1382 | |
| 612 : | pcmpgtw mm5, [zero] | ||
| 613 : | paddusw mm0, mm5 | ||
| 614 : | psrlw mm0, 5 | ||
| 615 : | pxor mm0, mm1 ; start restoring sign | ||
| 616 : | psubusw mm1, mm5 | ||
| 617 : | |||
| 618 : | movq mm5, mm4 | ||
| 619 : | pmulhw mm5, mm2 ; high of c*(matrix*quant) | ||
| 620 : | pmullw mm2, mm4 ; low of c*(matrix*quant) | ||
| 621 : | psubw mm0, mm1 ; finish restoring sign | ||
| 622 : | |||
| 623 : | pcmpgtw mm5, [zero] | ||
| 624 : | paddusw mm2, mm5 | ||
| 625 : | psrlw mm2, 5 | ||
| 626 : | pxor mm2, mm3 ; start restoring sign | ||
| 627 : | psubusw mm3, mm5 | ||
| 628 : | psubw mm2, mm3 ; finish restoring sign | ||
| 629 : | |||
| 630 : | pxor mm6, mm0 ; mismatch control | ||
| 631 : | Isibaar | 1795 | movq [TMP1 + 8*_EAX + 8*16 -2*8 ], mm0 ; data[i] |
| 632 : | edgomez | 1382 | pxor mm6, mm2 ; mismatch control |
| 633 : | Isibaar | 1795 | movq [TMP1 + 8*_EAX + 8*16 -2*8 +8], mm2 ; data[i+1] |
| 634 : | edgomez | 1382 | |
| 635 : | jnz near .loop | ||
| 636 : | |||
| 637 : | ; mismatch control | ||
| 638 : | |||
| 639 : | movq mm0, mm6 | ||
| 640 : | psrlq mm0, 48 | ||
| 641 : | movq mm1, mm6 | ||
| 642 : | movq mm2, mm6 | ||
| 643 : | psrlq mm1, 32 | ||
| 644 : | pxor mm6, mm0 | ||
| 645 : | psrlq mm2, 16 | ||
| 646 : | pxor mm6, mm1 | ||
| 647 : | pxor mm6, mm2 | ||
| 648 : | movd eax, mm6 | ||
| 649 : | Isibaar | 1795 | and _EAX, 1 |
| 650 : | xor _EAX, 1 | ||
| 651 : | xor word [TMP1 + 2*63], ax | ||
| 652 : | edgomez | 1382 | |
| 653 : | Isibaar | 1795 | xor _EAX, _EAX |
| 654 : | edgomez | 1382 | |
| 655 : | Isibaar | 1795 | pop _EBX |
| 656 : | edgomez | 1382 | |
| 657 : | ret | ||
| 658 : | Isibaar | 1793 | ENDFUNC |
| 659 : | edgomez | 1540 | |
| 660 : | Isibaar | 1790 | |
| 661 : | %ifidn __OUTPUT_FORMAT__,elf | ||
| 662 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
| 663 : | %endif | ||
| 664 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |