Parent Directory
| 
Revision Log
Revision 908 - (view) (download)
| 1 : | edgomez | 851 | ;/************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * mmx quantization/dequantization | ||
| 5 : | ; * | ||
| 6 : | ; * This program is an implementation of a part of one or more MPEG-4 | ||
| 7 : | ; * Video tools as specified in ISO/IEC 14496-2 standard. Those intending | ||
| 8 : | ; * to use this software module in hardware or software products are | ||
| 9 : | ; * advised that its use may infringe existing patents or copyrights, and | ||
| 10 : | ; * any such use would be at such party's own risk. The original | ||
| 11 : | ; * developer of this software module and his/her company, and subsequent | ||
| 12 : | ; * editors and their companies, will have no liability for use of this | ||
| 13 : | ; * software or modifications or derivatives thereof. | ||
| 14 : | ; * | ||
| 15 : | ; * This program is free software; you can redistribute it and/or modify | ||
| 16 : | ; * it under the terms of the GNU General Public License as published by | ||
| 17 : | ; * the Free Software Foundation; either version 2 of the License, or | ||
| 18 : | ; * (at your option) any later version. | ||
| 19 : | ; * | ||
| 20 : | ; * This program is distributed in the hope that it will be useful, | ||
| 21 : | ; * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 22 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 23 : | ; * GNU General Public License for more details. | ||
| 24 : | ; * | ||
| 25 : | ; * You should have received a copy of the GNU General Public License | ||
| 26 : | ; * along with this program; if not, write to the Free Software | ||
| 27 : | ; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | ||
| 28 : | ; * | ||
| 29 : | ; *************************************************************************/ | ||
| 30 : | ; these 3dne functions are compatible with iSSE, but are optimized specifically for | ||
| 31 : | ; K7 pipelines | ||
| 32 : | ; | ||
| 33 : | ;------------------------------------------------------------------------------ | ||
| 34 : | ; 09.12.2002 Athlon optimizations contributed by Jaan Kalda | ||
| 35 : | ;------------------------------------------------------------------------------ | ||
| 36 : | |||
| 37 : | ; enable dequant saturate [-2048,2047], test purposes only. | ||
| 38 : | %define SATURATE | ||
| 39 : | |||
| 40 : | ; data/text alignment | ||
| 41 : | %define ALIGN 16 | ||
| 42 : | |||
| 43 : | bits 32 | ||
| 44 : | |||
| 45 : | %ifdef FORMAT_COFF | ||
| 46 : | section .data data | ||
| 47 : | %else | ||
| 48 : | section .data data align=16 | ||
| 49 : | %endif | ||
| 50 : | |||
| 51 : | %macro cglobal 1 | ||
| 52 : | %ifdef PREFIX | ||
| 53 : | global _%1 | ||
| 54 : | %define %1 _%1 | ||
| 55 : | %else | ||
| 56 : | global %1 | ||
| 57 : | %endif | ||
| 58 : | %endmacro | ||
| 59 : | align 4 | ||
| 60 : | int_div | ||
| 61 : | dd 0 | ||
| 62 : | %assign i 1 | ||
| 63 : | %rep 255 | ||
| 64 : | dd (1 << 16) / ( i) + 1 | ||
| 65 : | %assign i i+1 | ||
| 66 : | %endrep | ||
| 67 : | |||
| 68 : | align 16 | ||
| 69 : | |||
| 70 : | plus_one times 8 dw 1 | ||
| 71 : | |||
| 72 : | ;=========================================================================== | ||
| 73 : | ; | ||
| 74 : | ; subtract by Q/2 table | ||
| 75 : | ; | ||
| 76 : | ;=========================================================================== | ||
| 77 : | |||
| 78 : | %macro MMX_SUB 1 | ||
| 79 : | times 4 dw %1 / 2 | ||
| 80 : | %endmacro | ||
| 81 : | |||
| 82 : | |||
| 83 : | align 16 | ||
| 84 : | mmx_sub | ||
| 85 : | %assign i 1 | ||
| 86 : | %rep 31 | ||
| 87 : | times 4 dw i / 2 | ||
| 88 : | %assign i i+1 | ||
| 89 : | %endrep | ||
| 90 : | |||
| 91 : | |||
| 92 : | ;=========================================================================== | ||
| 93 : | ; | ||
| 94 : | ; divide by 2Q table | ||
| 95 : | ; | ||
| 96 : | ; use a shift of 16 to take full advantage of _pmulhw_ | ||
| 97 : | ; for q=1, _pmulhw_ will overflow so it is treated seperately | ||
| 98 : | ; (3dnow2 provides _pmulhuw_ which wont cause overflow) | ||
| 99 : | ; | ||
| 100 : | ;=========================================================================== | ||
| 101 : | |||
| 102 : | align 16 | ||
| 103 : | mmx_div | ||
| 104 : | |||
| 105 : | %assign i 1 | ||
| 106 : | %rep 31 | ||
| 107 : | times 4 dw (1 << 16) / (i * 2) + 1 | ||
| 108 : | %assign i i+1 | ||
| 109 : | %endrep | ||
| 110 : | |||
| 111 : | ;=========================================================================== | ||
| 112 : | ; | ||
| 113 : | ; add by (odd(Q) ? Q : Q - 1) table | ||
| 114 : | ; | ||
| 115 : | ;=========================================================================== | ||
| 116 : | |||
| 117 : | %macro MMX_ADD 1 | ||
| 118 : | %if %1 % 2 != 0 | ||
| 119 : | times 4 dw %1 | ||
| 120 : | %else | ||
| 121 : | times 4 dw %1 - 1 | ||
| 122 : | %endif | ||
| 123 : | %endmacro | ||
| 124 : | |||
| 125 : | align 16 | ||
| 126 : | mmx_add | ||
| 127 : | |||
| 128 : | %assign i 1 | ||
| 129 : | %rep 31 | ||
| 130 : | MMX_ADD i | ||
| 131 : | %assign i i+1 | ||
| 132 : | %endrep | ||
| 133 : | |||
| 134 : | ;=========================================================================== | ||
| 135 : | ; | ||
| 136 : | ; multiple by 2Q table | ||
| 137 : | ; | ||
| 138 : | ;=========================================================================== | ||
| 139 : | |||
| 140 : | %macro MMX_MUL 1 | ||
| 141 : | times 4 dw %1 * 2 | ||
| 142 : | %endmacro | ||
| 143 : | |||
| 144 : | align 16 | ||
| 145 : | mmx_mul | ||
| 146 : | |||
| 147 : | %assign i 1 | ||
| 148 : | %rep 31 | ||
| 149 : | times 4 dw i * 2 | ||
| 150 : | %assign i i+1 | ||
| 151 : | %endrep | ||
| 152 : | |||
| 153 : | ;=========================================================================== | ||
| 154 : | ; | ||
| 155 : | ; saturation limits | ||
| 156 : | ; | ||
| 157 : | ;=========================================================================== | ||
| 158 : | |||
| 159 : | align 8 | ||
| 160 : | mmx_32768_minus_2048 times 4 dw (32768-2048) | ||
| 161 : | mmx_32767_minus_2047 times 4 dw (32767-2047) | ||
| 162 : | |||
| 163 : | align 16 | ||
| 164 : | mmx_2047 times 4 dw 2047 | ||
| 165 : | |||
| 166 : | align 8 | ||
| 167 : | mmzero dd 0, 0 | ||
| 168 : | int2047 dd 2047 | ||
| 169 : | int_2048 dd -2048 | ||
| 170 : | |||
| 171 : | section .text | ||
| 172 : | |||
| 173 : | |||
| 174 : | ;=========================================================================== | ||
| 175 : | ; | ||
| 176 : | ; void quant_intra_3dne(int16_t * coeff, | ||
| 177 : | ; const int16_t const * data, | ||
| 178 : | ; const uint32_t quant, | ||
| 179 : | ; const uint32_t dcscalar); | ||
| 180 : | ; | ||
| 181 : | ;=========================================================================== | ||
| 182 : | ;This is Athlon-optimized code (ca 70 clk per call) | ||
| 183 : | ;Optimized by Jaan, 30 Nov 2002 | ||
| 184 : | |||
| 185 : | %macro quant_intra1 1 | ||
| 186 : | psubw mm1,mm0 ;A3 | ||
| 187 : | psubw mm3,mm2 ;B3 | ||
| 188 : | %if (%1) | ||
| 189 : | psubw mm5, mm4 ;C8 | ||
| 190 : | psubw mm7, mm6 ;D8 | ||
| 191 : | %endif | ||
| 192 : | |||
| 193 : | align 8 | ||
| 194 : | db 0Fh, 6Fh, 64h, 21h, (%1 * 32 +16) ;movq mm4, [ecx + %1 * 32 +16+32] ;C1 | ||
| 195 : | pmaxsw mm1,mm0 ;A4 | ||
| 196 : | db 0Fh, 6Fh, 74h, 21h, (%1 * 32 +24);movq mm6, [ecx + %1 * 32 +24+32] ;D1 | ||
| 197 : | pmaxsw mm3,mm2 ;B4 | ||
| 198 : | |||
| 199 : | |||
| 200 : | psraw mm0,15 ;A5 | ||
| 201 : | psraw mm2,15 ;B5 | ||
| 202 : | %if (%1) | ||
| 203 : | movq [edx + %1 * 32 + 16-32], mm5 ;C9 | ||
| 204 : | movq [edx + %1 * 32 + 24-32], mm7 ;D9 | ||
| 205 : | %endif | ||
| 206 : | |||
| 207 : | psrlw mm1, 1 ;A6 | ||
| 208 : | psrlw mm3, 1 ;B6 | ||
| 209 : | movq mm5, [ebx] ;C2 | ||
| 210 : | movq mm7, [ebx] ;D2 | ||
| 211 : | |||
| 212 : | pxor mm1, mm0 ;A7 | ||
| 213 : | pxor mm3, mm2 ;B7 | ||
| 214 : | |||
| 215 : | psubw mm5,mm4 ;C3 | ||
| 216 : | psubw mm7,mm6 ;D3 | ||
| 217 : | psubw mm1, mm0 ;A8 | ||
| 218 : | psubw mm3, mm2 ;B8 | ||
| 219 : | |||
| 220 : | %if (%1 == 0) | ||
| 221 : | push ebp | ||
| 222 : | movq mm0, [ecx + %1 * 32 +32] | ||
| 223 : | %elif (%1 < 3) | ||
| 224 : | db 0Fh, 6Fh, 44h, 21h, (%1 * 32 +32) ;movq mm0, [ecx + %1 * 32 +32] ;A1 | ||
| 225 : | %endif | ||
| 226 : | pmaxsw mm5,mm4 ;C4 | ||
| 227 : | %if (%1 < 3) | ||
| 228 : | db 0Fh, 6Fh, 54h, 21h, ( %1 * 32 +8+32) ;movq mm2, [ecx + %1 * 32 +8+32] ;B1 | ||
| 229 : | %else | ||
| 230 : | cmp esp,esp | ||
| 231 : | %endif | ||
| 232 : | pmaxsw mm7,mm6 ;D4 | ||
| 233 : | |||
| 234 : | psraw mm4,15 ;C5 | ||
| 235 : | psraw mm6,15 ;D5 | ||
| 236 : | movq [byte edx + %1 * 32], mm1 ;A9 | ||
| 237 : | movq [edx + %1 * 32+8], mm3 ;B9 | ||
| 238 : | |||
| 239 : | |||
| 240 : | psrlw mm5, 1 ;C6 | ||
| 241 : | psrlw mm7, 1 ;D6 | ||
| 242 : | %if (%1 < 3) | ||
| 243 : | movq mm1, [ebx] ;A2 | ||
| 244 : | movq mm3, [ebx] ;B2 | ||
| 245 : | %endif | ||
| 246 : | %if (%1 == 3) | ||
| 247 : | imul eax,[int_div+4*edi] | ||
| 248 : | %endif | ||
| 249 : | pxor mm5, mm4 ;C7 | ||
| 250 : | pxor mm7, mm6 ;D7 | ||
| 251 : | %endm | ||
| 252 : | |||
| 253 : | |||
| 254 : | %macro quant_intra 1 ;rules for athlon: 1) schedule latencies, 2) add/mul and load/store in 2:1 proportion, | ||
| 255 : | ; 3) avoid spliting >3byte instructions over 8byte boundaries | ||
| 256 : | psubw mm1,mm0 ;A3 | ||
| 257 : | psubw mm3,mm2 ;B3 | ||
| 258 : | %if (%1) | ||
| 259 : | psubw mm5, mm4 ;C8 | ||
| 260 : | psubw mm7, mm6 ;D8 | ||
| 261 : | %endif | ||
| 262 : | |||
| 263 : | align 8 | ||
| 264 : | db 0Fh, 6Fh, 64h, 21h, (%1 * 32 +16) ;movq mm4, [ecx + %1 * 32 +16+32] ;C1 | ||
| 265 : | pmaxsw mm1,mm0 ;A4 | ||
| 266 : | db 0Fh, 6Fh, 74h, 21h, (%1 * 32 +24);movq mm6, [ecx + %1 * 32 +24+32] ;D1 | ||
| 267 : | pmaxsw mm3,mm2 ;B4 | ||
| 268 : | |||
| 269 : | |||
| 270 : | psraw mm0,15 ;A5 | ||
| 271 : | psraw mm2,15 ;B5 | ||
| 272 : | %if (%1) | ||
| 273 : | movq [edx + %1 * 32 + 16-32], mm5 ;C9 | ||
| 274 : | movq [edx + %1 * 32 + 24-32], mm7 ;D9 | ||
| 275 : | %endif | ||
| 276 : | |||
| 277 : | pmulhw mm1, [esi] ;A6 | ||
| 278 : | pmulhw mm3, [esi] ;B6 | ||
| 279 : | movq mm5, [ebx] ;C2 | ||
| 280 : | movq mm7, [ebx] ;D2 | ||
| 281 : | |||
| 282 : | nop | ||
| 283 : | nop | ||
| 284 : | pxor mm1, mm0 ;A7 | ||
| 285 : | pxor mm3, mm2 ;B7 | ||
| 286 : | |||
| 287 : | psubw mm5,mm4 ;C3 | ||
| 288 : | psubw mm7,mm6 ;D3 | ||
| 289 : | psubw mm1, mm0 ;A8 | ||
| 290 : | psubw mm3, mm2 ;B8 | ||
| 291 : | |||
| 292 : | |||
| 293 : | %if (%1 < 3) | ||
| 294 : | db 0Fh, 6Fh, 44h, 21h, (%1 * 32 +32) ;movq mm0, [ecx + %1 * 32 +32] ;A1 | ||
| 295 : | %endif | ||
| 296 : | pmaxsw mm5,mm4 ;C4 | ||
| 297 : | %if (%1 < 3) | ||
| 298 : | db 0Fh, 6Fh, 54h, 21h, ( %1 * 32 +8+32) ;movq mm2, [ecx + %1 * 32 +8+32] ;B1 | ||
| 299 : | %else | ||
| 300 : | cmp esp,esp | ||
| 301 : | %endif | ||
| 302 : | pmaxsw mm7,mm6 ;D4 | ||
| 303 : | |||
| 304 : | psraw mm4,15 ;C5 | ||
| 305 : | psraw mm6,15 ;D5 | ||
| 306 : | movq [byte edx + %1 * 32], mm1 ;A9 | ||
| 307 : | movq [edx + %1 * 32+8], mm3 ;B9 | ||
| 308 : | |||
| 309 : | |||
| 310 : | pmulhw mm5, [esi] ;C6 | ||
| 311 : | pmulhw mm7, [esi] ;D6 | ||
| 312 : | %if (%1 < 3) | ||
| 313 : | movq mm1, [ebx] ;A2 | ||
| 314 : | movq mm3, [ebx] ;B2 | ||
| 315 : | %endif | ||
| 316 : | %if (%1 == 0) | ||
| 317 : | push ebp | ||
| 318 : | %elif (%1 < 3) | ||
| 319 : | nop | ||
| 320 : | %endif | ||
| 321 : | nop | ||
| 322 : | %if (%1 == 3) | ||
| 323 : | imul eax,[int_div+4*edi] | ||
| 324 : | %endif | ||
| 325 : | pxor mm5, mm4 ;C7 | ||
| 326 : | pxor mm7, mm6 ;D7 | ||
| 327 : | %endmacro | ||
| 328 : | |||
| 329 : | |||
| 330 : | align ALIGN | ||
| 331 : | cglobal quant_intra_3dne | ||
| 332 : | quant_intra_3dne: | ||
| 333 : | |||
| 334 : | mov eax, [esp + 12] ; quant | ||
| 335 : | mov ecx, [esp + 8] ; data | ||
| 336 : | mov edx, [esp + 4] ; coeff | ||
| 337 : | cmp al, 1 | ||
| 338 : | pxor mm1,mm1 | ||
| 339 : | pxor mm3,mm3 | ||
| 340 : | movq mm0, [ecx ] ; mm0 = [1st] | ||
| 341 : | movq mm2, [ecx +8] | ||
| 342 : | push esi | ||
| 343 : | lea esi, [mmx_div + eax * 8 - 8] | ||
| 344 : | |||
| 345 : | push ebx | ||
| 346 : | mov ebx,mmzero | ||
| 347 : | push edi | ||
| 348 : | jz near .q1loop | ||
| 349 : | quant_intra 0 | ||
| 350 : | mov ebp, [esp + 16 + 16] ; dcscalar | ||
| 351 : | movsx eax, word [byte ecx] ;x | ||
| 352 : | quant_intra 1 | ||
| 353 : | mov edi,eax | ||
| 354 : | sar edi,31 ;sign(x) | ||
| 355 : | shr ebp,byte 1 ; ebp = dcscalar /2 | ||
| 356 : | quant_intra 2 | ||
| 357 : | sub eax,edi ; x (+1) | ||
| 358 : | xor ebp,edi ;sign(x) dcscalar /2 (-1) | ||
| 359 : | mov edi,[esp + 16 + 16] | ||
| 360 : | lea eax,[byte eax+ebp] ;x + sign(x) dcscalar /2 | ||
| 361 : | mov ebp,[byte esp] | ||
| 362 : | quant_intra 3 | ||
| 363 : | psubw mm5, mm4 ;C8 | ||
| 364 : | mov esi,[esp+12] | ||
| 365 : | mov edi,[esp+4] | ||
| 366 : | mov ebx,[esp+8] | ||
| 367 : | add esp,byte 16 | ||
| 368 : | sar eax,16 | ||
| 369 : | mov [edx], ax ; coeff[0] = ax | ||
| 370 : | psubw mm7, mm6 ;D8 | ||
| 371 : | movq [edx + 3 * 32 + 16], mm5 ;C9 | ||
| 372 : | movq [edx + 3 * 32 + 24], mm7 ;D9 | ||
| 373 : | ret | ||
| 374 : | align 16 | ||
| 375 : | .q1loop | ||
| 376 : | quant_intra1 0 | ||
| 377 : | mov ebp, [esp + 16 + 16] ; dcscalar | ||
| 378 : | movsx eax, word [byte ecx] ;x | ||
| 379 : | quant_intra1 1 | ||
| 380 : | mov edi,eax | ||
| 381 : | sar edi,31 ;sign(x) | ||
| 382 : | shr ebp,byte 1 ; ebp = dcscalar /2 | ||
| 383 : | quant_intra1 2 | ||
| 384 : | sub eax,edi ; x (+1) | ||
| 385 : | xor ebp,edi ;sign(x) dcscalar /2 (-1) | ||
| 386 : | mov edi,[esp + 16 + 16] | ||
| 387 : | lea eax,[byte eax+ebp] ;x + sign(x) dcscalar /2 | ||
| 388 : | mov ebp,[byte esp] | ||
| 389 : | quant_intra1 3 | ||
| 390 : | psubw mm5, mm4 ;C8 | ||
| 391 : | mov esi,[dword esp+12] | ||
| 392 : | mov edi,[esp+4] | ||
| 393 : | mov ebx,[esp+8] | ||
| 394 : | add esp,byte 16 | ||
| 395 : | sar eax,16 | ||
| 396 : | mov [edx], ax ; coeff[0] = ax | ||
| 397 : | psubw mm7, mm6 ;D8 | ||
| 398 : | movq [edx + 3 * 32 + 16], mm5 ;C9 | ||
| 399 : | movq [edx + 3 * 32 + 24], mm7 ;D9 | ||
| 400 : | ret | ||
| 401 : | |||
| 402 : | |||
| 403 : | |||
| 404 : | |||
| 405 : | ;=========================================================================== | ||
| 406 : | ; | ||
| 407 : | ; uint32_t quant_inter_3dne(int16_t * coeff, | ||
| 408 : | ; const int16_t const * data, | ||
| 409 : | ; const uint32_t quant); | ||
| 410 : | ; | ||
| 411 : | ;=========================================================================== | ||
| 412 : | ;This is Athlon-optimized code (ca 90 clk per call) | ||
| 413 : | ;Optimized by Jaan, 30 Nov 2002 | ||
| 414 : | |||
| 415 : | |||
| 416 : | %macro quantinter 1 | ||
| 417 : | movq mm1, [eax] ;A2 | ||
| 418 : | psraw mm3,15 ;B6 | ||
| 419 : | %if (%1) | ||
| 420 : | psubw mm2, mm6 ;C10 | ||
| 421 : | %endif | ||
| 422 : | psubw mm1,mm0 ;A3 | ||
| 423 : | pmulhw mm4, mm7 ; B7 | ||
| 424 : | movq mm6, [ecx + %1*24+16] ; C1 | ||
| 425 : | pmaxsw mm1,mm0 ;A4 | ||
| 426 : | paddw mm5, mm4 ;B8 | ||
| 427 : | %if (%1) | ||
| 428 : | movq [edx + %1*24+16-24], mm2 ;C11 | ||
| 429 : | %endif | ||
| 430 : | psubusw mm1, [ebx] ; A5 mm0 -= sub (unsigned, dont go < 0) | ||
| 431 : | pxor mm4, mm3 ;B9 | ||
| 432 : | movq mm2, [eax] ;C2 | ||
| 433 : | psraw mm0,15 ;A6 | ||
| 434 : | psubw mm4, mm3 ;B10 | ||
| 435 : | psubw mm2,mm6 ;C3 | ||
| 436 : | pmulhw mm1, mm7 ; A7 mm0 = (mm0 / 2Q) >> 24 | ||
| 437 : | movq mm3, [ecx + %1*24+8] ; B1 | ||
| 438 : | pmaxsw mm2,mm6 ;C4 | ||
| 439 : | paddw mm5, mm1 ; A8 sum += mm0 | ||
| 440 : | %if (%1) | ||
| 441 : | movq [edx + %1*24+8-24], mm4 ;B11 | ||
| 442 : | %else | ||
| 443 : | movq [edx + 120], mm4 ;B11 | ||
| 444 : | %endif | ||
| 445 : | psubusw mm2, [ebx] ;C5 | ||
| 446 : | pxor mm1, mm0 ; A9 mm0 *= sign(mm0) | ||
| 447 : | movq mm4, [eax] ;B2 | ||
| 448 : | psraw mm6,15 ;C6 | ||
| 449 : | psubw mm1, mm0 ;A10 undisplace | ||
| 450 : | psubw mm4,mm3 ;B3 | ||
| 451 : | pmulhw mm2, mm7 ; C7 | ||
| 452 : | movq mm0, [ecx + %1*24+24] ;A1 mm0 = [1st] | ||
| 453 : | pmaxsw mm4,mm3 ;B4 | ||
| 454 : | paddw mm5, mm2 ;C8 | ||
| 455 : | movq [byte edx + %1*24], mm1 ;A11 | ||
| 456 : | psubusw mm4, [ebx] ;B5 | ||
| 457 : | pxor mm2, mm6 ;C9 | ||
| 458 : | %endmacro | ||
| 459 : | |||
| 460 : | %macro quantinter1 1 | ||
| 461 : | movq mm0, [byte ecx + %1*16] ; mm0 = [1st] | ||
| 462 : | movq mm3, [ecx + %1*16+8] ; | ||
| 463 : | movq mm1, [eax] | ||
| 464 : | movq mm4, [eax] | ||
| 465 : | psubw mm1,mm0 | ||
| 466 : | psubw mm4,mm3 | ||
| 467 : | pmaxsw mm1,mm0 | ||
| 468 : | pmaxsw mm4,mm3 | ||
| 469 : | psubusw mm1, mm6 ; mm0 -= sub (unsigned, dont go < 0) | ||
| 470 : | psubusw mm4, mm6 ; | ||
| 471 : | psraw mm0,15 | ||
| 472 : | psraw mm3,15 | ||
| 473 : | psrlw mm1, 1 ; mm0 = (mm0 / 2Q) >> 16 | ||
| 474 : | psrlw mm4, 1 ; | ||
| 475 : | paddw mm5, mm1 ; sum += mm0 | ||
| 476 : | pxor mm1, mm0 ; mm0 *= sign(mm0) | ||
| 477 : | paddw mm5, mm4 | ||
| 478 : | pxor mm4, mm3 ; | ||
| 479 : | psubw mm1, mm0 ; undisplace | ||
| 480 : | psubw mm4, mm3 | ||
| 481 : | cmp esp,esp | ||
| 482 : | movq [byte edx + %1*16], mm1 | ||
| 483 : | movq [edx + %1*16+8], mm4 | ||
| 484 : | %endmacro | ||
| 485 : | |||
| 486 : | align ALIGN | ||
| 487 : | cglobal quant_inter_3dne | ||
| 488 : | quant_inter_3dne | ||
| 489 : | |||
| 490 : | mov edx, [esp + 4] ; coeff | ||
| 491 : | mov ecx, [esp + 8] ; data | ||
| 492 : | mov eax, [esp + 12] ; quant | ||
| 493 : | push ebx | ||
| 494 : | |||
| 495 : | pxor mm5, mm5 ; sum | ||
| 496 : | nop | ||
| 497 : | lea ebx,[mmx_sub + eax * 8 - 8] ; sub | ||
| 498 : | movq mm7, [mmx_div + eax * 8 - 8] ; divider | ||
| 499 : | |||
| 500 : | cmp al, 1 | ||
| 501 : | lea eax,[mmzero] | ||
| 502 : | jz near .q1loop | ||
| 503 : | cmp esp,esp | ||
| 504 : | align 8 | ||
| 505 : | movq mm3, [ecx + 120] ; B1 | ||
| 506 : | pxor mm4,mm4 ;B2 | ||
| 507 : | psubw mm4,mm3 ;B3 | ||
| 508 : | movq mm0, [ecx] ;A1 mm0 = [1st] | ||
| 509 : | pmaxsw mm4,mm3 ;B4 | ||
| 510 : | psubusw mm4, [ebx] ;B5 | ||
| 511 : | |||
| 512 : | quantinter 0 | ||
| 513 : | quantinter 1 | ||
| 514 : | quantinter 2 | ||
| 515 : | quantinter 3 | ||
| 516 : | quantinter 4 | ||
| 517 : | psraw mm3,15 ;B6 | ||
| 518 : | psubw mm2, mm6 ;C10 | ||
| 519 : | pmulhw mm4, mm7 ; B7 | ||
| 520 : | paddw mm5, mm4 ;B8 | ||
| 521 : | pxor mm4, mm3 ;B9 | ||
| 522 : | psubw mm4, mm3 ;B10 | ||
| 523 : | movq [edx + 4*24+16], mm2 ;C11 | ||
| 524 : | pop ebx | ||
| 525 : | movq [edx + 4*24+8], mm4 ;B11 | ||
| 526 : | pmaddwd mm5, [plus_one] | ||
| 527 : | movq mm0, mm5 | ||
| 528 : | punpckhdq mm5, mm5 | ||
| 529 : | paddd mm0, mm5 | ||
| 530 : | movd eax, mm0 ; return sum | ||
| 531 : | ret | ||
| 532 : | |||
| 533 : | align ALIGN | ||
| 534 : | .q1loop | ||
| 535 : | movq mm6,[byte ebx] | ||
| 536 : | quantinter1 0 | ||
| 537 : | quantinter1 1 | ||
| 538 : | quantinter1 2 | ||
| 539 : | quantinter1 3 | ||
| 540 : | quantinter1 4 | ||
| 541 : | quantinter1 5 | ||
| 542 : | quantinter1 6 | ||
| 543 : | quantinter1 7 | ||
| 544 : | |||
| 545 : | pmaddwd mm5, [plus_one] | ||
| 546 : | movq mm0, mm5 | ||
| 547 : | psrlq mm5, 32 | ||
| 548 : | paddd mm0, mm5 | ||
| 549 : | movd eax, mm0 ; return sum | ||
| 550 : | |||
| 551 : | pop ebx | ||
| 552 : | |||
| 553 : | ret | ||
| 554 : | |||
| 555 : | ;=========================================================================== | ||
| 556 : | ; | ||
| 557 : | ; void dequant_intra_3dne(int16_t *data, | ||
| 558 : | ; const int16_t const *coeff, | ||
| 559 : | ; const uint32_t quant, | ||
| 560 : | ; const uint32_t dcscalar); | ||
| 561 : | ; | ||
| 562 : | ;=========================================================================== | ||
| 563 : | |||
| 564 : | ; this is the same as dequant_inter_3dne, except that we're | ||
| 565 : | ; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) | ||
| 566 : | |||
| 567 : | ;This is Athlon-optimized code (ca 106 clk per call) | ||
| 568 : | |||
| 569 : | %macro dequant 1 | ||
| 570 : | movq mm1, [ecx+%1*24] ;A2 ; c = coeff[i] | ||
| 571 : | psubw mm0,mm1 ;-c ;A3 (1st dep) | ||
| 572 : | %if (%1) | ||
| 573 : | paddw mm4,mm6 ; C11 mm6 free (4th+) | ||
| 574 : | %endif | ||
| 575 : | pmaxsw mm0,mm1 ;|c| ;A4 (2nd) | ||
| 576 : | %if (%1) | ||
| 577 : | mov ebp,ebp | ||
| 578 : | pminsw mm4,[ebx] ; C12 saturates to +2047 (5th+) 1ater | ||
| 579 : | %endif | ||
| 580 : | movq mm6,[esi] ;0 ;A5 mm6 in use | ||
| 581 : | pandn mm7,[eax] ; B9 offset = isZero ? 0 : quant_add (2nd) | ||
| 582 : | %if (%1) | ||
| 583 : | pxor mm5, mm4 ; C13 (6th+) 1later | ||
| 584 : | %endif | ||
| 585 : | movq mm4,[esi] ; C1 ;0 | ||
| 586 : | mov esp,esp | ||
| 587 : | pcmpeqw mm6, [ecx+%1*24] ;A6 (c ==0) ? -1 : 0 (1st) | ||
| 588 : | align 4 | ||
| 589 : | psraw mm1,15 ; sign(c) ;A7 (2nd) | ||
| 590 : | %if (%1) | ||
| 591 : | movq [edx+%1*24+16-24], mm5 ; C14 (7th) 2later | ||
| 592 : | %endif | ||
| 593 : | paddw mm7,mm3 ; B10 offset +negate back (3rd) | ||
| 594 : | pmullw mm0, [edi] ;*= 2Q ;A8 (3rd+) | ||
| 595 : | paddw mm2,mm7 ; B11 mm7 free (4th+) | ||
| 596 : | lea ebp,[byte ebp] | ||
| 597 : | movq mm5, [ecx+%1*24+16] ;C2 ; c = coeff[i] | ||
| 598 : | psubw mm4,mm5 ;-c ;C3 (1st dep) | ||
| 599 : | pandn mm6,[eax] ; A9 offset = isZero ? 0 : quant_add (2nd) | ||
| 600 : | pminsw mm2,[ebx] ; B12 saturates to +2047 (5th+) | ||
| 601 : | pxor mm3, mm2 ; B13 (6th+) | ||
| 602 : | movq mm2,[byte esi] ; B1 ;0 | ||
| 603 : | %if (%1) | ||
| 604 : | movq [edx+%1*24+8-24], mm3 ; B14 (7th) | ||
| 605 : | %else | ||
| 606 : | movq [edx+120], mm3 | ||
| 607 : | %endif | ||
| 608 : | pmaxsw mm4,mm5 ;|c| ;C4 (2nd) | ||
| 609 : | paddw mm6,mm1 ; A10 offset +negate back (3rd) | ||
| 610 : | movq mm3, [ecx+%1*24 + 8] ;B2 ; c = coeff[i] | ||
| 611 : | psubw mm2,mm3 ;-c ;B3 (1st dep) | ||
| 612 : | paddw mm0,mm6 ; A11 mm6 free (4th+) | ||
| 613 : | movq mm6,[byte esi] ;0 ;C5 mm6 in use | ||
| 614 : | pcmpeqw mm6, [ecx+%1*24+16] ;C6 (c ==0) ? -1 : 0 (1st) | ||
| 615 : | pminsw mm0,[ebx] ; A12 saturates to +2047 (5th+) | ||
| 616 : | pmaxsw mm2,mm3 ;|c| ;B4 (2nd) | ||
| 617 : | pxor mm1, mm0 ; A13 (6th+) | ||
| 618 : | pmullw mm4, [edi] ;*= 2Q ;C8 (3rd+) | ||
| 619 : | psraw mm5,15 ; sign(c) ;C7 (2nd) | ||
| 620 : | movq mm7,[byte esi] ;0 ;B5 mm7 in use | ||
| 621 : | pcmpeqw mm7, [ecx+%1*24 + 8] ;B6 (c ==0) ? -1 : 0 (1st) | ||
| 622 : | %if (%1 < 4) | ||
| 623 : | movq mm0,[byte esi] ; A1 ;0 | ||
| 624 : | %endif | ||
| 625 : | pandn mm6,[byte eax] ; C9 offset = isZero ? 0 : quant_add (2nd) | ||
| 626 : | psraw mm3,15 ; sign(c) ;B7 (2nd) | ||
| 627 : | movq [byte edx+%1*24], mm1 ; A14 (7th) | ||
| 628 : | paddw mm6,mm5 ; C10 offset +negate back (3rd) | ||
| 629 : | pmullw mm2, [edi] ;*= 2Q ;B8 (3rd+) | ||
| 630 : | mov esp,esp | ||
| 631 : | %endmacro | ||
| 632 : | |||
| 633 : | |||
| 634 : | align ALIGN | ||
| 635 : | cglobal dequant_intra_3dne | ||
| 636 : | dequant_intra_3dne: | ||
| 637 : | mov ecx, [esp+ 8] ; coeff | ||
| 638 : | mov eax, [esp+12] ; quant | ||
| 639 : | pxor mm0,mm0 | ||
| 640 : | pxor mm2,mm2 | ||
| 641 : | push edi | ||
| 642 : | push ebx | ||
| 643 : | lea edi,[mmx_mul + eax*8 - 8] ; 2*quant | ||
| 644 : | push ebp | ||
| 645 : | mov ebx,mmx_2047 | ||
| 646 : | movsx ebp,word [ecx] | ||
| 647 : | lea eax,[mmx_add + eax*8 - 8] ; quant or quant-1 | ||
| 648 : | push esi | ||
| 649 : | mov esi,mmzero | ||
| 650 : | pxor mm7,mm7 | ||
| 651 : | movq mm3, [ecx+120] ;B2 ; c = coeff[i] | ||
| 652 : | pcmpeqw mm7, [ecx+120] ;B6 (c ==0) ? -1 : 0 (1st) | ||
| 653 : | |||
| 654 : | imul ebp,[esp+16+16] ; dcscalar | ||
| 655 : | psubw mm2,mm3 ;-c ;B3 (1st dep) | ||
| 656 : | pmaxsw mm2,mm3 ;|c| ;B4 (2nd) | ||
| 657 : | pmullw mm2, [edi] ;*= 2Q ;B8 (3rd+) | ||
| 658 : | psraw mm3,15 ; sign(c) ;B7 (2nd) | ||
| 659 : | mov edx, [esp+ 4+16] ; data | ||
| 660 : | align 8 | ||
| 661 : | dequant 0 | ||
| 662 : | cmp ebp,-2048 | ||
| 663 : | mov esp,esp | ||
| 664 : | dequant 1 | ||
| 665 : | cmovl ebp,[int_2048] | ||
| 666 : | nop | ||
| 667 : | dequant 2 | ||
| 668 : | cmp ebp,2047 | ||
| 669 : | mov esp,esp | ||
| 670 : | dequant 3 | ||
| 671 : | cmovg ebp,[int2047] | ||
| 672 : | nop | ||
| 673 : | dequant 4 | ||
| 674 : | |||
| 675 : | paddw mm4,mm6 ; C11 mm6 free (4th+) | ||
| 676 : | pminsw mm4,[ebx] ; C12 saturates to +2047 (5th+) | ||
| 677 : | pandn mm7,[eax] ; B9 offset = isZero ? 0 : quant_add (2nd) | ||
| 678 : | mov eax,ebp | ||
| 679 : | mov esi,[esp] | ||
| 680 : | mov ebp,[esp+4] | ||
| 681 : | pxor mm5, mm4 ; C13 (6th+) | ||
| 682 : | paddw mm7,mm3 ; B10 offset +negate back (3rd) | ||
| 683 : | movq [edx+4*24+16], mm5 ; C14 (7th) | ||
| 684 : | paddw mm2,mm7 ; B11 mm7 free (4th+) | ||
| 685 : | pminsw mm2,[ebx] ; B12 saturates to +2047 (5th+) | ||
| 686 : | mov ebx,[esp+8] | ||
| 687 : | mov edi,[esp+12] | ||
| 688 : | add esp,byte 16 | ||
| 689 : | pxor mm3, mm2 ; B13 (6th+) | ||
| 690 : | movq [edx+4*24+8], mm3 ; B14 (7th) | ||
| 691 : | mov [edx], ax | ||
| 692 : | ret | ||
| 693 : | |||
| 694 : | ;=========================================================================== | ||
| 695 : | ; | ||
| 696 : | ; void dequant_inter_3dne(int16_t * data, | ||
| 697 : | ; const int16_t * const coeff, | ||
| 698 : | ; const uint32_t quant); | ||
| 699 : | ; | ||
| 700 : | ;=========================================================================== | ||
| 701 : | |||
| 702 : | ; this is the same as dequant_inter_3dne, | ||
| 703 : | ; except that we're saturating using 'pminsw' (saves 2 cycles/loop) | ||
| 704 : | ;This is Athlon-optimized code (ca 100 clk per call) | ||
| 705 : | ;Optimized by Jaan, 30 Nov 2002 | ||
| 706 : | |||
| 707 : | align ALIGN | ||
| 708 : | cglobal dequant_inter_3dne | ||
| 709 : | dequant_inter_3dne: | ||
| 710 : | |||
| 711 : | mov ecx, [esp+ 8] ; coeff | ||
| 712 : | mov eax, [esp+12] ; quant | ||
| 713 : | pxor mm0,mm0 | ||
| 714 : | pxor mm2,mm2 | ||
| 715 : | push edi | ||
| 716 : | push ebx | ||
| 717 : | push esi | ||
| 718 : | lea edi,[mmx_mul + eax*8 - 8] ; 2*quant | ||
| 719 : | mov ebx,mmx_2047 | ||
| 720 : | pxor mm7,mm7 | ||
| 721 : | movq mm3, [ecx+120] ;B2 ; c = coeff[i] | ||
| 722 : | pcmpeqw mm7, [ecx+120] ;B6 (c ==0) ? -1 : 0 (1st) | ||
| 723 : | lea eax,[mmx_add + eax*8 - 8] ; quant or quant-1 | ||
| 724 : | psubw mm2,mm3 ;-c ;B3 (1st dep) | ||
| 725 : | mov esi,mmzero | ||
| 726 : | pmaxsw mm2,mm3 ;|c| ;B4 (2nd) | ||
| 727 : | pmullw mm2, [edi] ;*= 2Q ;B8 (3rd+) | ||
| 728 : | psraw mm3,15 ; sign(c) ;B7 (2nd) | ||
| 729 : | mov edx, [dword esp+ 4+12] ; data | ||
| 730 : | align 8 | ||
| 731 : | dequant 0 | ||
| 732 : | dequant 1 | ||
| 733 : | dequant 2 | ||
| 734 : | dequant 3 | ||
| 735 : | dequant 4 | ||
| 736 : | |||
| 737 : | paddw mm4,mm6 ; C11 mm6 free (4th+) | ||
| 738 : | pminsw mm4,[ebx] ; C12 saturates to +2047 (5th+) | ||
| 739 : | pandn mm7,[eax] ; B9 offset = isZero ? 0 : quant_add (2nd) | ||
| 740 : | mov esi,[esp] | ||
| 741 : | pxor mm5, mm4 ; C13 (6th+) | ||
| 742 : | paddw mm7,mm3 ; B10 offset +negate back (3rd) | ||
| 743 : | movq [edx+4*24+16], mm5 ; C14 (7th) | ||
| 744 : | paddw mm2,mm7 ; B11 mm7 free (4th+) | ||
| 745 : | pminsw mm2,[ebx] ; B12 saturates to +2047 (5th+) | ||
| 746 : | mov ebx,[esp+4] | ||
| 747 : | mov edi,[esp+8] | ||
| 748 : | add esp,byte 12 | ||
| 749 : | pxor mm3, mm2 ; B13 (6th+) | ||
| 750 : | movq [edx+4*24+8], mm3 ; B14 (7th) | ||
| 751 : | ret |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |