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 : | ; * 2002 Jaan Kalda | ||
| 8 : | ; * | ||
| 9 : | ; * This program is free software ; you can redistribute it and/or modify | ||
| 10 : | ; * it under the terms of the GNU General Public License as published by | ||
| 11 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
| 12 : | ; * (at your option) any later version. | ||
| 13 : | ; * | ||
| 14 : | ; * This program is distributed in the hope that it will be useful, | ||
| 15 : | ; * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
| 16 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 17 : | ; * GNU General Public License for more details. | ||
| 18 : | ; * | ||
| 19 : | ; * You should have received a copy of the GNU General Public License | ||
| 20 : | ; * along with this program ; if not, write to the Free Software | ||
| 21 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 22 : | ; * | ||
| 23 : | Isibaar | 1798 | ; * $Id: quantize_mpeg_xmm.asm,v 1.10 2008-11-26 02:21:02 Isibaar Exp $ |
| 24 : | edgomez | 1382 | ; * |
| 25 : | ; ***************************************************************************/ | ||
| 26 : | |||
| 27 : | ; _3dne functions are compatible with iSSE, but are optimized specifically | ||
| 28 : | ; for K7 pipelines | ||
| 29 : | |||
| 30 : | %define SATURATE | ||
| 31 : | |||
| 32 : | Isibaar | 1795 | %include "nasm.inc" |
| 33 : | edgomez | 1382 | |
| 34 : | ;============================================================================= | ||
| 35 : | ; Local data | ||
| 36 : | ;============================================================================= | ||
| 37 : | |||
| 38 : | Isibaar | 1795 | DATA |
| 39 : | edgomez | 1382 | |
| 40 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 41 : | edgomez | 1382 | mmzero: |
| 42 : | dd 0,0 | ||
| 43 : | mmx_one: | ||
| 44 : | times 4 dw 1 | ||
| 45 : | |||
| 46 : | ;----------------------------------------------------------------------------- | ||
| 47 : | ; divide by 2Q table | ||
| 48 : | ;----------------------------------------------------------------------------- | ||
| 49 : | |||
| 50 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 51 : | edgomez | 1382 | mmx_divs: ;i>2 |
| 52 : | %assign i 1 | ||
| 53 : | %rep 31 | ||
| 54 : | times 4 dw ((1 << 15) / i + 1) | ||
| 55 : | %assign i i+1 | ||
| 56 : | %endrep | ||
| 57 : | |||
| 58 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 59 : | edgomez | 1382 | mmx_div: ;quant>2 |
| 60 : | times 4 dw 65535 ; the div by 2 formula will overflow for the case | ||
| 61 : | ; quant=1 but we don't care much because quant=1 | ||
| 62 : | ; is handled by a different piece of code that | ||
| 63 : | ; doesn't use this table. | ||
| 64 : | %assign quant 2 | ||
| 65 : | %rep 31 | ||
| 66 : | times 4 dw ((1 << 16) / quant + 1) | ||
| 67 : | %assign quant quant+1 | ||
| 68 : | %endrep | ||
| 69 : | |||
| 70 : | %macro FIXX 1 | ||
| 71 : | dw (1 << 16) / (%1) + 1 | ||
| 72 : | %endmacro | ||
| 73 : | |||
| 74 : | Isibaar | 1795 | %ifndef ARCH_IS_X86_64 |
| 75 : | edgomez | 1382 | %define nop4 db 08Dh, 074h, 026h,0 |
| 76 : | Isibaar | 1795 | %define nop7 db 08dh, 02ch, 02dh,0,0,0,0 |
| 77 : | %else | ||
| 78 : | %define nop4 | ||
| 79 : | %define nop7 | ||
| 80 : | %endif | ||
| 81 : | %define nop3 add _ESP, byte 0 | ||
| 82 : | %define nop2 mov _ESP, _ESP | ||
| 83 : | %define nop6 add _EBP, dword 0 | ||
| 84 : | edgomez | 1382 | |
| 85 : | ;----------------------------------------------------------------------------- | ||
| 86 : | ; quantd table | ||
| 87 : | ;----------------------------------------------------------------------------- | ||
| 88 : | |||
| 89 : | %define VM18P 3 | ||
| 90 : | %define VM18Q 4 | ||
| 91 : | |||
| 92 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 93 : | edgomez | 1382 | quantd: |
| 94 : | %assign i 1 | ||
| 95 : | %rep 31 | ||
| 96 : | times 4 dw (((VM18P*i) + (VM18Q/2)) / VM18Q) | ||
| 97 : | %assign i i+1 | ||
| 98 : | %endrep | ||
| 99 : | |||
| 100 : | ;----------------------------------------------------------------------------- | ||
| 101 : | ; multiple by 2Q table | ||
| 102 : | ;----------------------------------------------------------------------------- | ||
| 103 : | |||
| 104 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 105 : | edgomez | 1382 | mmx_mul_quant: |
| 106 : | %assign i 1 | ||
| 107 : | %rep 31 | ||
| 108 : | times 4 dw i | ||
| 109 : | %assign i i+1 | ||
| 110 : | %endrep | ||
| 111 : | |||
| 112 : | ;----------------------------------------------------------------------------- | ||
| 113 : | ; saturation limits | ||
| 114 : | ;----------------------------------------------------------------------------- | ||
| 115 : | |||
| 116 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 117 : | edgomez | 1382 | mmx_32767_minus_2047: |
| 118 : | times 4 dw (32767-2047) | ||
| 119 : | mmx_32768_minus_2048: | ||
| 120 : | times 4 dw (32768-2048) | ||
| 121 : | mmx_2047: | ||
| 122 : | times 4 dw 2047 | ||
| 123 : | mmx_minus_2048: | ||
| 124 : | times 4 dw (-2048) | ||
| 125 : | zero: | ||
| 126 : | times 4 dw 0 | ||
| 127 : | |||
| 128 : | int_div: | ||
| 129 : | dd 0 | ||
| 130 : | %assign i 1 | ||
| 131 : | %rep 255 | ||
| 132 : | dd (1 << 17) / ( i) + 1 | ||
| 133 : | %assign i i+1 | ||
| 134 : | %endrep | ||
| 135 : | |||
| 136 : | ;============================================================================= | ||
| 137 : | ; Code | ||
| 138 : | ;============================================================================= | ||
| 139 : | |||
| 140 : | Isibaar | 1795 | SECTION .rotext align=SECTION_ALIGN |
| 141 : | edgomez | 1382 | |
| 142 : | cglobal quant_mpeg_inter_xmm | ||
| 143 : | cglobal dequant_mpeg_intra_3dne | ||
| 144 : | cglobal dequant_mpeg_inter_3dne | ||
| 145 : | |||
| 146 : | ;----------------------------------------------------------------------------- | ||
| 147 : | ; | ||
| 148 : | ; uint32_t quant_mpeg_inter_xmm(int16_t * coeff, | ||
| 149 : | ; const int16_t const * data, | ||
| 150 : | ; const uint32_t quant, | ||
| 151 : | ; const uint16_t *mpeg_matrices); | ||
| 152 : | ; | ||
| 153 : | ;----------------------------------------------------------------------------- | ||
| 154 : | |||
| 155 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 156 : | edgomez | 1382 | quant_mpeg_inter_xmm: |
| 157 : | Isibaar | 1795 | mov _EAX, prm2 ; data |
| 158 : | mov TMP0, prm3 ; quant | ||
| 159 : | mov TMP1, prm1 ; coeff | ||
| 160 : | push _ESI | ||
| 161 : | push _EDI | ||
| 162 : | push _EBX | ||
| 163 : | edgomez | 1382 | nop |
| 164 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 165 : | mov _EDI, prm4 | ||
| 166 : | %else | ||
| 167 : | mov _EDI, [_ESP + 12 + 16] | ||
| 168 : | %endif | ||
| 169 : | |||
| 170 : | mov _ESI, -14 | ||
| 171 : | mov _EBX, _ESP | ||
| 172 : | sub _ESP, byte 24 | ||
| 173 : | lea _EBX, [_ESP+8] | ||
| 174 : | and _EBX, byte -8 ;ALIGN 8 | ||
| 175 : | edgomez | 1382 | pxor mm0, mm0 |
| 176 : | pxor mm3, mm3 | ||
| 177 : | Isibaar | 1795 | movq [byte _EBX],mm0 |
| 178 : | movq [_EBX+8],mm0 | ||
| 179 : | Isibaar | 1798 | |
| 180 : | Isibaar | 1795 | cmp TMP0, byte 1 |
| 181 : | edgomez | 1382 | je near .q1loop |
| 182 : | Isibaar | 1795 | cmp TMP0, byte 19 |
| 183 : | edgomez | 1382 | jg near .lloop |
| 184 : | nop | ||
| 185 : | |||
| 186 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 187 : | Isibaar | 1793 | .loop: |
| 188 : | Isibaar | 1795 | movq mm1, [_EAX + 8*_ESI+112] ; mm0 = [1st] |
| 189 : | edgomez | 1382 | psubw mm0, mm1 ;-mm1 |
| 190 : | Isibaar | 1795 | movq mm4, [_EAX + 8*_ESI + 120] ; |
| 191 : | edgomez | 1382 | psubw mm3, mm4 ;-mm4 |
| 192 : | pmaxsw mm0, mm1 ;|src| | ||
| 193 : | pmaxsw mm3, mm4 | ||
| 194 : | nop2 | ||
| 195 : | psraw mm1, 15 ;sign src | ||
| 196 : | psraw mm4, 15 | ||
| 197 : | psllw mm0, 4 ; level << 4 | ||
| 198 : | psllw mm3, 4 ; | ||
| 199 : | Isibaar | 1795 | paddw mm0, [_EDI + 640 + 8*_ESI+112] |
| 200 : | paddw mm3, [_EDI + 640 + 8*_ESI+120] | ||
| 201 : | movq mm5, [_EDI + 896 + 8*_ESI+112] | ||
| 202 : | movq mm7, [_EDI + 896 + 8*_ESI+120] | ||
| 203 : | edgomez | 1382 | pmulhuw mm5, mm0 |
| 204 : | pmulhuw mm7, mm3 | ||
| 205 : | Isibaar | 1795 | mov _ESP, _ESP |
| 206 : | movq mm2, [_EDI + 512 + 8*_ESI+112] | ||
| 207 : | movq mm6, [_EDI + 512 + 8*_ESI+120] | ||
| 208 : | edgomez | 1382 | pmullw mm2, mm5 |
| 209 : | pmullw mm6, mm7 | ||
| 210 : | psubw mm0, mm2 | ||
| 211 : | psubw mm3, mm6 | ||
| 212 : | Isibaar | 1795 | movq mm2, [byte _EBX] |
| 213 : | %ifdef ARCH_IS_X86_64 | ||
| 214 : | lea r9, [mmx_divs] | ||
| 215 : | movq mm6, [r9 + TMP0 * 8 - 8] | ||
| 216 : | %else | ||
| 217 : | movq mm6, [mmx_divs + TMP0 * 8 - 8] | ||
| 218 : | %endif | ||
| 219 : | pmulhuw mm0, [_EDI + 768 + 8*_ESI+112] | ||
| 220 : | pmulhuw mm3, [_EDI + 768 + 8*_ESI+120] | ||
| 221 : | paddw mm2, [_EBX+8] ;sum | ||
| 222 : | edgomez | 1382 | paddw mm5, mm0 |
| 223 : | paddw mm7, mm3 | ||
| 224 : | pxor mm0, mm0 | ||
| 225 : | pxor mm3, mm3 | ||
| 226 : | pmulhuw mm5, mm6 ; mm0 = (mm0 / 2Q) >> 16 | ||
| 227 : | pmulhuw mm7, mm6 ; (level ) / quant (0<quant<32) | ||
| 228 : | Isibaar | 1795 | add _ESI, byte 2 |
| 229 : | edgomez | 1382 | paddw mm2, mm5 ;sum += x1 |
| 230 : | Isibaar | 1795 | movq [_EBX], mm7 ;store x2 |
| 231 : | edgomez | 1382 | pxor mm5, mm1 ; mm0 *= sign(mm0) |
| 232 : | pxor mm7, mm4 ; | ||
| 233 : | psubw mm5, mm1 ; undisplace | ||
| 234 : | psubw mm7, mm4 ; | ||
| 235 : | Isibaar | 1795 | movq [_EBX+8],mm2 ;store sum |
| 236 : | movq [TMP1 + 8*_ESI+112-16], mm5 | ||
| 237 : | movq [TMP1 + 8*_ESI +120-16], mm7 | ||
| 238 : | edgomez | 1382 | jng near .loop |
| 239 : | |||
| 240 : | Isibaar | 1793 | .done: |
| 241 : | edgomez | 1382 | ; calculate data[0] // (int32_t)dcscalar) |
| 242 : | Isibaar | 1795 | paddw mm2, [_EBX] |
| 243 : | mov _EBX, [_ESP+24] | ||
| 244 : | mov _EDI, [_ESP+PTR_SIZE+24] | ||
| 245 : | mov _ESI, [_ESP+2*PTR_SIZE+24] | ||
| 246 : | add _ESP, byte 3*PTR_SIZE+24 | ||
| 247 : | edgomez | 1382 | pmaddwd mm2, [mmx_one] |
| 248 : | punpckldq mm0, mm2 ;get low dw to mm0:high | ||
| 249 : | paddd mm0,mm2 | ||
| 250 : | punpckhdq mm0, mm0 ;get result to low | ||
| 251 : | movd eax, mm0 | ||
| 252 : | |||
| 253 : | ret | ||
| 254 : | |||
| 255 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 256 : | Isibaar | 1793 | .q1loop: |
| 257 : | Isibaar | 1795 | movq mm1, [_EAX + 8*_ESI+112] ; mm0 = [1st] |
| 258 : | edgomez | 1382 | psubw mm0, mm1 ;-mm1 |
| 259 : | Isibaar | 1795 | movq mm4, [_EAX + 8*_ESI+120] |
| 260 : | edgomez | 1382 | psubw mm3, mm4 ;-mm4 |
| 261 : | pmaxsw mm0, mm1 ;|src| | ||
| 262 : | pmaxsw mm3, mm4 | ||
| 263 : | nop2 | ||
| 264 : | psraw mm1, 15 ; sign src | ||
| 265 : | psraw mm4, 15 | ||
| 266 : | psllw mm0, 4 ; level << 4 | ||
| 267 : | psllw mm3, 4 | ||
| 268 : | Isibaar | 1795 | paddw mm0, [_EDI + 640 + 8*_ESI+112] ;mm0 is to be divided |
| 269 : | paddw mm3, [_EDI + 640 + 8*_ESI+120] ; inter1 contains fix for division by 1 | ||
| 270 : | movq mm5, [_EDI + 896 + 8*_ESI+112] ;with rounding down | ||
| 271 : | movq mm7, [_EDI + 896 + 8*_ESI+120] | ||
| 272 : | edgomez | 1382 | pmulhuw mm5, mm0 |
| 273 : | pmulhuw mm7, mm3 ;mm7: first approx of division | ||
| 274 : | Isibaar | 1795 | mov _ESP, _ESP |
| 275 : | movq mm2, [_EDI + 512 + 8*_ESI+112] | ||
| 276 : | movq mm6, [_EDI + 512 + 8*_ESI+120] ; divs for q<=16 | ||
| 277 : | edgomez | 1382 | pmullw mm2, mm5 ;test value <= original |
| 278 : | pmullw mm6, mm7 | ||
| 279 : | psubw mm0, mm2 ;mismatch | ||
| 280 : | psubw mm3, mm6 | ||
| 281 : | Isibaar | 1795 | movq mm2, [byte _EBX] |
| 282 : | pmulhuw mm0, [_EDI + 768 + 8*_ESI+112] ;correction | ||
| 283 : | pmulhuw mm3, [_EDI + 768 + 8*_ESI+120] | ||
| 284 : | paddw mm2, [_EBX+8] ;sum | ||
| 285 : | edgomez | 1382 | paddw mm5, mm0 ;final result |
| 286 : | paddw mm7, mm3 | ||
| 287 : | pxor mm0, mm0 | ||
| 288 : | pxor mm3, mm3 | ||
| 289 : | psrlw mm5, 1 ; (level ) /2 (quant = 1) | ||
| 290 : | psrlw mm7, 1 | ||
| 291 : | Isibaar | 1795 | add _ESI, byte 2 |
| 292 : | edgomez | 1382 | paddw mm2, mm5 ;sum += x1 |
| 293 : | Isibaar | 1795 | movq [_EBX], mm7 ;store x2 |
| 294 : | edgomez | 1382 | pxor mm5, mm1 ; mm0 *= sign(mm0) |
| 295 : | pxor mm7, mm4 ; | ||
| 296 : | psubw mm5, mm1 ; undisplace | ||
| 297 : | psubw mm7, mm4 ; | ||
| 298 : | Isibaar | 1795 | movq [_EBX+8], mm2 ;store sum |
| 299 : | movq [TMP1 + 8*_ESI+112-16], mm5 | ||
| 300 : | movq [TMP1 + 8*_ESI +120-16], mm7 | ||
| 301 : | edgomez | 1382 | jng near .q1loop |
| 302 : | jmp near .done | ||
| 303 : | |||
| 304 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 305 : | Isibaar | 1793 | .lloop: |
| 306 : | Isibaar | 1795 | movq mm1, [_EAX + 8*_ESI+112] ; mm0 = [1st] |
| 307 : | edgomez | 1382 | psubw mm0,mm1 ;-mm1 |
| 308 : | Isibaar | 1795 | movq mm4, [_EAX + 8*_ESI+120] |
| 309 : | edgomez | 1382 | psubw mm3,mm4 ;-mm4 |
| 310 : | pmaxsw mm0,mm1 ;|src| | ||
| 311 : | pmaxsw mm3,mm4 | ||
| 312 : | nop2 | ||
| 313 : | psraw mm1,15 ;sign src | ||
| 314 : | psraw mm4,15 | ||
| 315 : | psllw mm0, 4 ; level << 4 | ||
| 316 : | psllw mm3, 4 ; | ||
| 317 : | Isibaar | 1795 | paddw mm0, [_EDI + 640 + 8*_ESI+112] ;mm0 is to be divided inter1 contains fix for division by 1 |
| 318 : | paddw mm3, [_EDI + 640 + 8*_ESI+120] | ||
| 319 : | movq mm5,[_EDI + 896 + 8*_ESI+112] | ||
| 320 : | movq mm7,[_EDI + 896 + 8*_ESI+120] | ||
| 321 : | edgomez | 1382 | pmulhuw mm5,mm0 |
| 322 : | pmulhuw mm7,mm3 ;mm7: first approx of division | ||
| 323 : | Isibaar | 1795 | mov _ESP,_ESP |
| 324 : | movq mm2,[_EDI + 512 + 8*_ESI+112] | ||
| 325 : | movq mm6,[_EDI + 512 + 8*_ESI+120] | ||
| 326 : | edgomez | 1382 | pmullw mm2,mm5 ;test value <= original |
| 327 : | pmullw mm6,mm7 | ||
| 328 : | psubw mm0,mm2 ;mismatch | ||
| 329 : | psubw mm3,mm6 | ||
| 330 : | Isibaar | 1795 | movq mm2,[byte _EBX] |
| 331 : | %ifdef ARCH_IS_X86_64 | ||
| 332 : | lea r9, [mmx_div] | ||
| 333 : | movq mm6, [r9 + TMP0 * 8 - 8] | ||
| 334 : | %else | ||
| 335 : | movq mm6,[mmx_div + TMP0 * 8 - 8] ; divs for q<=16 | ||
| 336 : | %endif | ||
| 337 : | pmulhuw mm0,[_EDI + 768 + 8*_ESI+112] ;correction | ||
| 338 : | pmulhuw mm3,[_EDI + 768 + 8*_ESI+120] | ||
| 339 : | paddw mm2,[_EBX+8] ;sum | ||
| 340 : | edgomez | 1382 | paddw mm5,mm0 ;final result |
| 341 : | paddw mm7,mm3 | ||
| 342 : | pxor mm0,mm0 | ||
| 343 : | pxor mm3,mm3 | ||
| 344 : | pmulhuw mm5, mm6 ; mm0 = (mm0 / 2Q) >> 16 | ||
| 345 : | pmulhuw mm7, mm6 ; (level ) / quant (0<quant<32) | ||
| 346 : | Isibaar | 1795 | add _ESI,byte 2 |
| 347 : | edgomez | 1382 | psrlw mm5, 1 ; (level ) / (2*quant) |
| 348 : | paddw mm2,mm5 ;sum += x1 | ||
| 349 : | psrlw mm7, 1 | ||
| 350 : | Isibaar | 1795 | movq [_EBX],mm7 ;store x2 |
| 351 : | edgomez | 1382 | pxor mm5, mm1 ; mm0 *= sign(mm0) |
| 352 : | pxor mm7, mm4 ; | ||
| 353 : | psubw mm5, mm1 ; undisplace | ||
| 354 : | psubw mm7, mm4 ; | ||
| 355 : | Isibaar | 1795 | movq [_EBX+8], mm2 ;store sum |
| 356 : | movq [TMP1 + 8*_ESI+112-16], mm5 | ||
| 357 : | movq [TMP1 + 8*_ESI +120-16], mm7 | ||
| 358 : | edgomez | 1382 | jng near .lloop |
| 359 : | jmp near .done | ||
| 360 : | Isibaar | 1793 | ENDFUNC |
| 361 : | edgomez | 1382 | |
| 362 : | |||
| 363 : | ;----------------------------------------------------------------------------- | ||
| 364 : | ; | ||
| 365 : | ; uint32_t dequant_mpeg_intra_3dne(int16_t *data, | ||
| 366 : | ; const int16_t const *coeff, | ||
| 367 : | ; const uint32_t quant, | ||
| 368 : | ; const uint32_t dcscalar, | ||
| 369 : | ; const uint16_t *mpeg_matrices); | ||
| 370 : | ; | ||
| 371 : | ;----------------------------------------------------------------------------- | ||
| 372 : | |||
| 373 : | ; Note: in order to saturate 'easily', we pre-shift the quantifier | ||
| 374 : | ; by 4. Then, the high-word of (coeff[]*matrix[i]*quant) are used to | ||
| 375 : | ; build a saturating mask. It is non-zero only when an overflow occured. | ||
| 376 : | ; We thus avoid packing/unpacking toward double-word. | ||
| 377 : | ; Moreover, we perform the mult (matrix[i]*quant) first, instead of, e.g., | ||
| 378 : | ; (coeff[i]*matrix[i]). This is less prone to overflow if coeff[] are not | ||
| 379 : | ; checked. Input ranges are: coeff in [-127,127], inter_matrix in [1..255],a | ||
| 380 : | ; and quant in [1..31]. | ||
| 381 : | ; | ||
| 382 : | |||
| 383 : | %macro DEQUANT4INTRAMMX 1 | ||
| 384 : | Isibaar | 1795 | movq mm1, [byte TMP0+ 16 * %1] ; mm0 = c = coeff[i] |
| 385 : | movq mm4, [TMP0+ 16 * %1 +8] ; mm3 = c' = coeff[i+1] | ||
| 386 : | edgomez | 1382 | psubw mm0, mm1 |
| 387 : | psubw mm3, mm4 | ||
| 388 : | pmaxsw mm0, mm1 | ||
| 389 : | pmaxsw mm3, mm4 | ||
| 390 : | psraw mm1, 15 | ||
| 391 : | psraw mm4, 15 | ||
| 392 : | %if %1 | ||
| 393 : | Isibaar | 1795 | movq mm2, [_EAX+8] ;preshifted quant |
| 394 : | movq mm7, [_EAX+8] | ||
| 395 : | edgomez | 1382 | %endif |
| 396 : | Isibaar | 1795 | pmullw mm2, [TMP1 + 16 * %1 ] ; matrix[i]*quant |
| 397 : | pmullw mm7, [TMP1 + 16 * %1 +8] ; matrix[i+1]*quant | ||
| 398 : | edgomez | 1382 | movq mm5, mm0 |
| 399 : | movq mm6, mm3 | ||
| 400 : | pmulhw mm0, mm2 ; high of coeff*(matrix*quant) | ||
| 401 : | pmulhw mm3, mm7 ; high of coeff*(matrix*quant) | ||
| 402 : | pmullw mm2, mm5 ; low of coeff*(matrix*quant) | ||
| 403 : | pmullw mm7, mm6 ; low of coeff*(matrix*quant) | ||
| 404 : | Isibaar | 1795 | pcmpgtw mm0, [_EAX] |
| 405 : | pcmpgtw mm3, [_EAX] | ||
| 406 : | edgomez | 1382 | paddusw mm2, mm0 |
| 407 : | paddusw mm7, mm3 | ||
| 408 : | psrlw mm2, 5 | ||
| 409 : | psrlw mm7, 5 | ||
| 410 : | pxor mm2, mm1 ; start negating back | ||
| 411 : | pxor mm7, mm4 ; start negating back | ||
| 412 : | psubusw mm1, mm0 | ||
| 413 : | psubusw mm4, mm3 | ||
| 414 : | Isibaar | 1795 | movq mm0, [_EAX] ;zero |
| 415 : | movq mm3, [_EAX] ;zero | ||
| 416 : | edgomez | 1382 | psubw mm2, mm1 ; finish negating back |
| 417 : | psubw mm7, mm4 ; finish negating back | ||
| 418 : | Isibaar | 1795 | movq [byte _EDI + 16 * %1], mm2 ; data[i] |
| 419 : | movq [_EDI + 16 * %1 +8], mm7 ; data[i+1] | ||
| 420 : | edgomez | 1382 | %endmacro |
| 421 : | |||
| 422 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 423 : | edgomez | 1382 | dequant_mpeg_intra_3dne: |
| 424 : | Isibaar | 1795 | mov _EAX, prm3 ; quant |
| 425 : | %ifdef ARCH_IS_X86_64 | ||
| 426 : | lea TMP0, [mmx_mul_quant] | ||
| 427 : | movq mm7, [TMP0 + _EAX*8 - 8] | ||
| 428 : | %else | ||
| 429 : | movq mm7, [mmx_mul_quant + _EAX*8 - 8] | ||
| 430 : | %endif | ||
| 431 : | mov TMP0, prm2 ; coeff | ||
| 432 : | psllw mm7, 2 ; << 2. See comment. | ||
| 433 : | mov TMP1, prm5 ; mpeg_quant_matrices | ||
| 434 : | push _EBX | ||
| 435 : | movsx _EBX, word [TMP0] | ||
| 436 : | edgomez | 1382 | pxor mm0, mm0 |
| 437 : | pxor mm3, mm3 | ||
| 438 : | Isibaar | 1795 | push _ESI |
| 439 : | lea _EAX, [_ESP-28] | ||
| 440 : | sub _ESP, byte 32 | ||
| 441 : | and _EAX, byte -8 ;points to qword ALIGNed space on stack | ||
| 442 : | movq [_EAX], mm0 | ||
| 443 : | movq [_EAX+8], mm7 | ||
| 444 : | %ifdef ARCH_IS_X86_64 | ||
| 445 : | imul _EBX, prm4 ; dcscalar | ||
| 446 : | %else | ||
| 447 : | imul _EBX, [_ESP+16+8+32] ; dcscalar | ||
| 448 : | %endif | ||
| 449 : | edgomez | 1382 | movq mm2, mm7 |
| 450 : | Isibaar | 1795 | push _EDI |
| 451 : | edgomez | 1382 | |
| 452 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 453 : | mov _EDI, prm1 ; data | ||
| 454 : | %else | ||
| 455 : | mov _EDI, [_ESP+4+12+32] | ||
| 456 : | %endif | ||
| 457 : | |||
| 458 : | ALIGN SECTION_ALIGN | ||
| 459 : | |||
| 460 : | edgomez | 1382 | DEQUANT4INTRAMMX 0 |
| 461 : | |||
| 462 : | Isibaar | 1795 | mov _ESI, -2048 |
| 463 : | edgomez | 1382 | nop |
| 464 : | Isibaar | 1795 | cmp _EBX, _ESI |
| 465 : | edgomez | 1382 | |
| 466 : | DEQUANT4INTRAMMX 1 | ||
| 467 : | |||
| 468 : | Isibaar | 1795 | cmovl _EBX, _ESI |
| 469 : | neg _ESI | ||
| 470 : | sub _ESI, byte 1 ;2047 | ||
| 471 : | edgomez | 1382 | |
| 472 : | DEQUANT4INTRAMMX 2 | ||
| 473 : | |||
| 474 : | Isibaar | 1795 | cmp _EBX, _ESI |
| 475 : | cmovg _EBX, _ESI | ||
| 476 : | lea _EBP, [byte _EBP] | ||
| 477 : | edgomez | 1382 | |
| 478 : | DEQUANT4INTRAMMX 3 | ||
| 479 : | |||
| 480 : | Isibaar | 1795 | mov _ESI, [_ESP+32+PTR_SIZE] |
| 481 : | mov [byte _EDI], bx | ||
| 482 : | mov _EBX, [_ESP+32+2*PTR_SIZE] | ||
| 483 : | edgomez | 1382 | |
| 484 : | DEQUANT4INTRAMMX 4 | ||
| 485 : | DEQUANT4INTRAMMX 5 | ||
| 486 : | DEQUANT4INTRAMMX 6 | ||
| 487 : | DEQUANT4INTRAMMX 7 | ||
| 488 : | |||
| 489 : | Isibaar | 1795 | pop _EDI |
| 490 : | edgomez | 1382 | |
| 491 : | Isibaar | 1795 | add _ESP, byte 32+2*PTR_SIZE |
| 492 : | edgomez | 1382 | |
| 493 : | Isibaar | 1795 | xor _EAX, _EAX |
| 494 : | edgomez | 1382 | ret |
| 495 : | Isibaar | 1793 | ENDFUNC |
| 496 : | edgomez | 1382 | |
| 497 : | ;----------------------------------------------------------------------------- | ||
| 498 : | ; | ||
| 499 : | ; uint32_t dequant_mpeg_inter_3dne(int16_t * data, | ||
| 500 : | ; const int16_t * const coeff, | ||
| 501 : | ; const uint32_t quant, | ||
| 502 : | ; const uint16_t *mpeg_matrices); | ||
| 503 : | ; | ||
| 504 : | ;----------------------------------------------------------------------------- | ||
| 505 : | |||
| 506 : | ; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier | ||
| 507 : | ; so we handle the 3 cases: c<0, c==0, and c>0 in one shot. | ||
| 508 : | ; sgn(x) is the result of 'pcmpgtw 0,x': 0 if x>=0, -1 if x<0. | ||
| 509 : | ; It's mixed with the extraction of the absolute value. | ||
| 510 : | |||
| 511 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 512 : | edgomez | 1382 | dequant_mpeg_inter_3dne: |
| 513 : | Isibaar | 1795 | mov _EAX, prm3 ; quant |
| 514 : | %ifdef ARCH_IS_X86_64 | ||
| 515 : | lea TMP0, [mmx_mul_quant] | ||
| 516 : | movq mm7, [TMP0 + _EAX*8 - 8] | ||
| 517 : | %else | ||
| 518 : | movq mm7, [mmx_mul_quant + _EAX*8 - 8] | ||
| 519 : | %endif | ||
| 520 : | mov TMP1, prm1 ; data | ||
| 521 : | mov TMP0, prm2 ; coeff | ||
| 522 : | mov _EAX, -14 | ||
| 523 : | edgomez | 1382 | paddw mm7, mm7 ; << 1 |
| 524 : | pxor mm6, mm6 ; mismatch sum | ||
| 525 : | Isibaar | 1795 | push _ESI |
| 526 : | push _EDI | ||
| 527 : | mov _ESI, mmzero | ||
| 528 : | edgomez | 1382 | pxor mm1, mm1 |
| 529 : | pxor mm3, mm3 | ||
| 530 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 531 : | mov _EDI, prm4 | ||
| 532 : | %else | ||
| 533 : | mov _EDI, [_ESP + 8 + 16] ; mpeg_quant_matrices | ||
| 534 : | %endif | ||
| 535 : | edgomez | 1382 | nop |
| 536 : | nop4 | ||
| 537 : | |||
| 538 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 539 : | Isibaar | 1793 | .loop: |
| 540 : | Isibaar | 1795 | movq mm0, [TMP0+8*_EAX + 7*16 ] ; mm0 = coeff[i] |
| 541 : | edgomez | 1382 | pcmpgtw mm1, mm0 ; mm1 = sgn(c) (preserved) |
| 542 : | Isibaar | 1795 | movq mm2, [TMP0+8*_EAX + 7*16 +8] ; mm2 = coeff[i+1] |
| 543 : | edgomez | 1382 | pcmpgtw mm3, mm2 ; mm3 = sgn(c') (preserved) |
| 544 : | paddsw mm0, mm1 ; c += sgn(c) | ||
| 545 : | paddsw mm2, mm3 ; c += sgn(c') | ||
| 546 : | paddw mm0, mm0 ; c *= 2 | ||
| 547 : | paddw mm2, mm2 ; c'*= 2 | ||
| 548 : | |||
| 549 : | Isibaar | 1795 | movq mm4, [_ESI] |
| 550 : | movq mm5, [_ESI] | ||
| 551 : | edgomez | 1382 | psubw mm4, mm0 ; -c |
| 552 : | psubw mm5, mm2 ; -c' | ||
| 553 : | |||
| 554 : | psraw mm4, 16 ; mm4 = sgn(-c) | ||
| 555 : | psraw mm5, 16 ; mm5 = sgn(-c') | ||
| 556 : | psubsw mm0, mm4 ; c -= sgn(-c) | ||
| 557 : | psubsw mm2, mm5 ; c' -= sgn(-c') | ||
| 558 : | pxor mm0, mm1 ; finish changing sign if needed | ||
| 559 : | pxor mm2, mm3 ; finish changing sign if needed | ||
| 560 : | |||
| 561 : | ; we're short on register, here. Poor pairing... | ||
| 562 : | |||
| 563 : | movq mm4, mm7 ; (matrix*quant) | ||
| 564 : | nop | ||
| 565 : | Isibaar | 1795 | pmullw mm4, [_EDI + 512 + 8*_EAX + 7*16] |
| 566 : | edgomez | 1382 | movq mm5, mm4 |
| 567 : | pmulhw mm5, mm0 ; high of c*(matrix*quant) | ||
| 568 : | pmullw mm0, mm4 ; low of c*(matrix*quant) | ||
| 569 : | |||
| 570 : | movq mm4, mm7 ; (matrix*quant) | ||
| 571 : | Isibaar | 1795 | pmullw mm4, [_EDI + 512 + 8*_EAX + 7*16 + 8] |
| 572 : | add _EAX, byte 2 | ||
| 573 : | edgomez | 1382 | |
| 574 : | Isibaar | 1795 | pcmpgtw mm5, [_ESI] |
| 575 : | edgomez | 1382 | paddusw mm0, mm5 |
| 576 : | psrlw mm0, 5 | ||
| 577 : | pxor mm0, mm1 ; start restoring sign | ||
| 578 : | psubusw mm1, mm5 | ||
| 579 : | |||
| 580 : | movq mm5, mm4 | ||
| 581 : | pmulhw mm5, mm2 ; high of c*(matrix*quant) | ||
| 582 : | pmullw mm2, mm4 ; low of c*(matrix*quant) | ||
| 583 : | psubw mm0, mm1 ; finish restoring sign | ||
| 584 : | |||
| 585 : | Isibaar | 1795 | pcmpgtw mm5, [_ESI] |
| 586 : | edgomez | 1382 | paddusw mm2, mm5 |
| 587 : | psrlw mm2, 5 | ||
| 588 : | pxor mm2, mm3 ; start restoring sign | ||
| 589 : | psubusw mm3, mm5 | ||
| 590 : | psubw mm2, mm3 ; finish restoring sign | ||
| 591 : | Isibaar | 1795 | movq mm1, [_ESI] |
| 592 : | movq mm3, [byte _ESI] | ||
| 593 : | edgomez | 1382 | pxor mm6, mm0 ; mismatch control |
| 594 : | Isibaar | 1795 | movq [TMP1 + 8*_EAX + 7*16 -2*8 ], mm0 ; data[i] |
| 595 : | edgomez | 1382 | pxor mm6, mm2 ; mismatch control |
| 596 : | Isibaar | 1795 | movq [TMP1 + 8*_EAX + 7*16 -2*8 +8], mm2 ; data[i+1] |
| 597 : | edgomez | 1382 | |
| 598 : | jng .loop | ||
| 599 : | nop | ||
| 600 : | |||
| 601 : | ; mismatch control | ||
| 602 : | |||
| 603 : | pshufw mm0, mm6, 01010101b | ||
| 604 : | pshufw mm1, mm6, 10101010b | ||
| 605 : | pshufw mm2, mm6, 11111111b | ||
| 606 : | pxor mm6, mm0 | ||
| 607 : | pxor mm1, mm2 | ||
| 608 : | pxor mm6, mm1 | ||
| 609 : | movd eax, mm6 | ||
| 610 : | Isibaar | 1795 | pop _EDI |
| 611 : | and _EAX, byte 1 | ||
| 612 : | xor _EAX, byte 1 | ||
| 613 : | mov _ESI, [_ESP] | ||
| 614 : | add _ESP, byte PTR_SIZE | ||
| 615 : | xor word [TMP1 + 2*63], ax | ||
| 616 : | edgomez | 1382 | |
| 617 : | Isibaar | 1795 | xor _EAX, _EAX |
| 618 : | edgomez | 1382 | ret |
| 619 : | Isibaar | 1793 | ENDFUNC |
| 620 : | edgomez | 1540 | |
| 621 : | Isibaar | 1790 | |
| 622 : | %ifidn __OUTPUT_FORMAT__,elf | ||
| 623 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
| 624 : | %endif | ||
| 625 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |