Parent Directory
| 
Revision Log
Revision 1230 - (view) (download)
| 1 : | edgomez | 1176 | ;/***************************************************************************** |
| 2 : | edgomez | 1174 | ; * |
| 3 : | edgomez | 1176 | ; * XVID MPEG-4 VIDEO CODEC |
| 4 : | ; * - MPEG4 Quantization H263 implementation / MMX optimized - | ||
| 5 : | edgomez | 1174 | ; * |
| 6 : | edgomez | 1176 | ; * Copyright(C) 2001-2003 Peter Ross <pross@xvid.org> |
| 7 : | ; * 2002-2003 Pascal Massimino <skal@planet-d.net> | ||
| 8 : | edgomez | 1174 | ; * |
| 9 : | edgomez | 1176 | ; * 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 : | edgomez | 1174 | ; * |
| 14 : | edgomez | 1176 | ; * 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 : | edgomez | 1174 | ; * |
| 19 : | edgomez | 1176 | ; * 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 : | edgomez | 1174 | ; * |
| 23 : | edgomez | 1230 | ; * $Id: quantize_h263_mmx.asm,v 1.1.2.5 2003-11-30 16:13:16 edgomez Exp $ |
| 24 : | edgomez | 1174 | ; * |
| 25 : | edgomez | 1176 | ; ****************************************************************************/ |
| 26 : | edgomez | 1174 | |
| 27 : | ; enable dequant saturate [-2048,2047], test purposes only. | ||
| 28 : | %define SATURATE | ||
| 29 : | |||
| 30 : | edgomez | 1192 | BITS 32 |
| 31 : | edgomez | 1174 | |
| 32 : | edgomez | 1192 | %macro cglobal 1 |
| 33 : | edgomez | 1176 | %ifdef PREFIX |
| 34 : | edgomez | 1192 | global _%1 |
| 35 : | edgomez | 1174 | %define %1 _%1 |
| 36 : | %else | ||
| 37 : | global %1 | ||
| 38 : | %endif | ||
| 39 : | %endmacro | ||
| 40 : | |||
| 41 : | edgomez | 1192 | ;============================================================================= |
| 42 : | ; Read only Local data | ||
| 43 : | ;============================================================================= | ||
| 44 : | edgomez | 1174 | |
| 45 : | edgomez | 1198 | %ifdef FORMAT_COFF |
| 46 : | SECTION .rodata data | ||
| 47 : | %else | ||
| 48 : | SECTION .rodata data align=16 | ||
| 49 : | %endif | ||
| 50 : | edgomez | 1174 | |
| 51 : | edgomez | 1192 | ALIGN 16 |
| 52 : | edgomez | 1176 | plus_one: |
| 53 : | times 8 dw 1 | ||
| 54 : | |||
| 55 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 56 : | edgomez | 1174 | ; |
| 57 : | ; subtract by Q/2 table | ||
| 58 : | ; | ||
| 59 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 60 : | edgomez | 1174 | |
| 61 : | edgomez | 1192 | ALIGN 16 |
| 62 : | edgomez | 1176 | mmx_sub: |
| 63 : | %assign quant 1 | ||
| 64 : | %rep 31 | ||
| 65 : | edgomez | 1192 | times 4 dw quant / 2 |
| 66 : | edgomez | 1176 | %assign quant quant+1 |
| 67 : | %endrep | ||
| 68 : | edgomez | 1174 | |
| 69 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 70 : | edgomez | 1174 | ; |
| 71 : | edgomez | 1192 | ; divide by 2Q table |
| 72 : | edgomez | 1174 | ; |
| 73 : | ; use a shift of 16 to take full advantage of _pmulhw_ | ||
| 74 : | ; for q=1, _pmulhw_ will overflow so it is treated seperately | ||
| 75 : | ; (3dnow2 provides _pmulhuw_ which wont cause overflow) | ||
| 76 : | ; | ||
| 77 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 78 : | edgomez | 1174 | |
| 79 : | edgomez | 1192 | ALIGN 16 |
| 80 : | edgomez | 1176 | mmx_div: |
| 81 : | %assign quant 1 | ||
| 82 : | %rep 31 | ||
| 83 : | times 4 dw (1<<16) / (quant*2) + 1 | ||
| 84 : | %assign quant quant+1 | ||
| 85 : | %endrep | ||
| 86 : | edgomez | 1174 | |
| 87 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 88 : | edgomez | 1174 | ; |
| 89 : | ; add by (odd(Q) ? Q : Q - 1) table | ||
| 90 : | ; | ||
| 91 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 92 : | edgomez | 1174 | |
| 93 : | edgomez | 1192 | ALIGN 16 |
| 94 : | edgomez | 1176 | mmx_add: |
| 95 : | %assign quant 1 | ||
| 96 : | %rep 31 | ||
| 97 : | %if quant % 2 != 0 | ||
| 98 : | times 4 dw quant | ||
| 99 : | %else | ||
| 100 : | times 4 dw quant - 1 | ||
| 101 : | %endif | ||
| 102 : | %assign quant quant+1 | ||
| 103 : | %endrep | ||
| 104 : | edgomez | 1174 | |
| 105 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 106 : | edgomez | 1174 | ; |
| 107 : | ; multiple by 2Q table | ||
| 108 : | ; | ||
| 109 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 110 : | edgomez | 1174 | |
| 111 : | edgomez | 1192 | ALIGN 16 |
| 112 : | edgomez | 1176 | mmx_mul: |
| 113 : | %assign quant 1 | ||
| 114 : | %rep 31 | ||
| 115 : | times 4 dw quant*2 | ||
| 116 : | %assign quant quant+1 | ||
| 117 : | %endrep | ||
| 118 : | edgomez | 1174 | |
| 119 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 120 : | edgomez | 1174 | ; |
| 121 : | edgomez | 1192 | ; saturation limits |
| 122 : | edgomez | 1174 | ; |
| 123 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 124 : | edgomez | 1174 | |
| 125 : | edgomez | 1192 | ALIGN 16 |
| 126 : | edgomez | 1176 | sse2_2047: |
| 127 : | times 8 dw 2047 | ||
| 128 : | edgomez | 1174 | |
| 129 : | edgomez | 1192 | ALIGN 16 |
| 130 : | edgomez | 1176 | mmx_2047: |
| 131 : | times 4 dw 2047 | ||
| 132 : | edgomez | 1174 | |
| 133 : | edgomez | 1192 | ALIGN 8 |
| 134 : | edgomez | 1176 | mmx_32768_minus_2048: |
| 135 : | times 4 dw (32768-2048) | ||
| 136 : | edgomez | 1174 | |
| 137 : | edgomez | 1176 | mmx_32767_minus_2047: |
| 138 : | times 4 dw (32767-2047) | ||
| 139 : | edgomez | 1174 | |
| 140 : | edgomez | 1176 | |
| 141 : | edgomez | 1192 | ;============================================================================= |
| 142 : | edgomez | 1176 | ; Code |
| 143 : | edgomez | 1192 | ;============================================================================= |
| 144 : | edgomez | 1176 | |
| 145 : | edgomez | 1192 | SECTION .text |
| 146 : | edgomez | 1174 | |
| 147 : | edgomez | 1192 | cglobal quant_h263_intra_mmx |
| 148 : | cglobal quant_h263_intra_sse2 | ||
| 149 : | cglobal quant_h263_inter_mmx | ||
| 150 : | cglobal quant_h263_inter_sse2 | ||
| 151 : | cglobal dequant_h263_intra_mmx | ||
| 152 : | cglobal dequant_h263_intra_xmm | ||
| 153 : | cglobal dequant_h263_intra_sse2 | ||
| 154 : | cglobal dequant_h263_inter_mmx | ||
| 155 : | cglobal dequant_h263_inter_xmm | ||
| 156 : | cglobal dequant_h263_inter_sse2 | ||
| 157 : | edgomez | 1174 | |
| 158 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 159 : | edgomez | 1174 | ; |
| 160 : | edgomez | 1176 | ; uint32_t quant_h263_intra_mmx(int16_t * coeff, |
| 161 : | ; const int16_t const * data, | ||
| 162 : | ; const uint32_t quant, | ||
| 163 : | edgomez | 1230 | ; const uint32_t dcscalar, |
| 164 : | ; const uint16_t *mpeg_matrices); | ||
| 165 : | edgomez | 1174 | ; |
| 166 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 167 : | edgomez | 1174 | |
| 168 : | edgomez | 1192 | ALIGN 16 |
| 169 : | edgomez | 1174 | quant_h263_intra_mmx: |
| 170 : | |||
| 171 : | edgomez | 1192 | push ecx |
| 172 : | push esi | ||
| 173 : | push edi | ||
| 174 : | edgomez | 1174 | |
| 175 : | edgomez | 1192 | mov edi, [esp + 12 + 4] ; coeff |
| 176 : | mov esi, [esp + 12 + 8] ; data | ||
| 177 : | mov eax, [esp + 12 + 12] ; quant | ||
| 178 : | edgomez | 1174 | |
| 179 : | edgomez | 1192 | xor ecx, ecx |
| 180 : | cmp al, 1 | ||
| 181 : | jz .q1loop | ||
| 182 : | edgomez | 1174 | |
| 183 : | edgomez | 1192 | movq mm7, [mmx_div + eax * 8 - 8] |
| 184 : | edgomez | 1176 | |
| 185 : | edgomez | 1192 | ALIGN 16 |
| 186 : | edgomez | 1174 | .loop |
| 187 : | edgomez | 1192 | movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
| 188 : | movq mm3, [esi + 8*ecx + 8] | ||
| 189 : | pxor mm1, mm1 ; mm1 = 0 | ||
| 190 : | pxor mm4, mm4 ; | ||
| 191 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 192 : | pcmpgtw mm4, mm3 ; | ||
| 193 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 194 : | pxor mm3, mm4 ; | ||
| 195 : | psubw mm0, mm1 ; displace | ||
| 196 : | psubw mm3, mm4 ; | ||
| 197 : | pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 | ||
| 198 : | pmulhw mm3, mm7 ; | ||
| 199 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 200 : | pxor mm3, mm4 ; | ||
| 201 : | psubw mm0, mm1 ; undisplace | ||
| 202 : | psubw mm3, mm4 ; | ||
| 203 : | movq [edi + 8*ecx], mm0 | ||
| 204 : | movq [edi + 8*ecx + 8], mm3 | ||
| 205 : | edgomez | 1174 | |
| 206 : | edgomez | 1192 | add ecx, 2 |
| 207 : | cmp ecx, 16 | ||
| 208 : | jnz .loop | ||
| 209 : | edgomez | 1176 | |
| 210 : | edgomez | 1192 | .done |
| 211 : | edgomez | 1176 | |
| 212 : | edgomez | 1192 | ; caclulate data[0] // (int32_t)dcscalar) |
| 213 : | mov ecx, [esp + 12 + 16] ; dcscalar | ||
| 214 : | mov edx, ecx | ||
| 215 : | movsx eax, word [esi] ; data[0] | ||
| 216 : | shr edx, 1 ; edx = dcscalar /2 | ||
| 217 : | cmp eax, 0 | ||
| 218 : | jg .gtzero | ||
| 219 : | edgomez | 1174 | |
| 220 : | edgomez | 1192 | sub eax, edx |
| 221 : | jmp short .mul | ||
| 222 : | edgomez | 1174 | |
| 223 : | .gtzero | ||
| 224 : | edgomez | 1192 | add eax, edx |
| 225 : | edgomez | 1174 | .mul |
| 226 : | edgomez | 1192 | cdq ; expand eax -> edx:eax |
| 227 : | idiv ecx ; eax = edx:eax / dcscalar | ||
| 228 : | mov [edi], ax ; coeff[0] = ax | ||
| 229 : | edgomez | 1174 | |
| 230 : | edgomez | 1192 | xor eax, eax ; return(0); |
| 231 : | pop edi | ||
| 232 : | pop esi | ||
| 233 : | pop ecx | ||
| 234 : | edgomez | 1174 | |
| 235 : | edgomez | 1192 | ret |
| 236 : | edgomez | 1174 | |
| 237 : | edgomez | 1192 | ALIGN 16 |
| 238 : | edgomez | 1174 | .q1loop |
| 239 : | edgomez | 1192 | movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
| 240 : | movq mm3, [esi + 8*ecx + 8] | ||
| 241 : | pxor mm1, mm1 ; mm1 = 0 | ||
| 242 : | pxor mm4, mm4 ; | ||
| 243 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 244 : | pcmpgtw mm4, mm3 ; | ||
| 245 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 246 : | pxor mm3, mm4 ; | ||
| 247 : | psubw mm0, mm1 ; displace | ||
| 248 : | psubw mm3, mm4 ; | ||
| 249 : | psrlw mm0, 1 ; mm0 >>= 1 (/2) | ||
| 250 : | psrlw mm3, 1 ; | ||
| 251 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 252 : | pxor mm3, mm4 | ||
| 253 : | psubw mm0, mm1 ; undisplace | ||
| 254 : | psubw mm3, mm4 ; | ||
| 255 : | movq [edi + 8*ecx], mm0 | ||
| 256 : | movq [edi + 8*ecx + 8], mm3 | ||
| 257 : | edgomez | 1174 | |
| 258 : | edgomez | 1192 | add ecx, 2 |
| 259 : | cmp ecx, 16 | ||
| 260 : | jnz .q1loop | ||
| 261 : | jmp short .done | ||
| 262 : | edgomez | 1174 | |
| 263 : | |||
| 264 : | |||
| 265 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 266 : | edgomez | 1174 | ; |
| 267 : | edgomez | 1192 | ; uint32_t quant_h263_intra_sse2(int16_t * coeff, |
| 268 : | edgomez | 1176 | ; const int16_t const * data, |
| 269 : | ; const uint32_t quant, | ||
| 270 : | edgomez | 1230 | ; const uint32_t dcscalar, |
| 271 : | ; const uint16_t *mpeg_matrices); | ||
| 272 : | edgomez | 1174 | ; |
| 273 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 274 : | edgomez | 1174 | |
| 275 : | edgomez | 1192 | ALIGN 16 |
| 276 : | edgomez | 1174 | quant_h263_intra_sse2: |
| 277 : | |||
| 278 : | edgomez | 1192 | push esi |
| 279 : | push edi | ||
| 280 : | edgomez | 1174 | |
| 281 : | edgomez | 1192 | mov edi, [esp + 8 + 4] ; coeff |
| 282 : | mov esi, [esp + 8 + 8] ; data | ||
| 283 : | mov eax, [esp + 8 + 12] ; quant | ||
| 284 : | edgomez | 1174 | |
| 285 : | edgomez | 1192 | xor ecx, ecx |
| 286 : | cmp al, 1 | ||
| 287 : | jz near .qas2_q1loop | ||
| 288 : | edgomez | 1174 | |
| 289 : | .qas2_not1 | ||
| 290 : | edgomez | 1192 | movq mm7, [mmx_div + eax*8 - 8] |
| 291 : | movq2dq xmm7, mm7 | ||
| 292 : | movlhps xmm7, xmm7 | ||
| 293 : | edgomez | 1174 | |
| 294 : | edgomez | 1192 | ALIGN 16 |
| 295 : | edgomez | 1174 | .qas2_loop |
| 296 : | edgomez | 1192 | movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
| 297 : | movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] | ||
| 298 : | pxor xmm1, xmm1 | ||
| 299 : | pxor xmm4, xmm4 | ||
| 300 : | pcmpgtw xmm1, xmm0 | ||
| 301 : | pcmpgtw xmm4, xmm3 | ||
| 302 : | pxor xmm0, xmm1 | ||
| 303 : | pxor xmm3, xmm4 | ||
| 304 : | psubw xmm0, xmm1 | ||
| 305 : | psubw xmm3, xmm4 | ||
| 306 : | pmulhw xmm0, xmm7 | ||
| 307 : | pmulhw xmm3, xmm7 | ||
| 308 : | pxor xmm0, xmm1 | ||
| 309 : | pxor xmm3, xmm4 | ||
| 310 : | psubw xmm0, xmm1 | ||
| 311 : | psubw xmm3, xmm4 | ||
| 312 : | movdqa [edi + ecx*8], xmm0 | ||
| 313 : | movdqa [edi + ecx*8 + 16], xmm3 | ||
| 314 : | edgomez | 1174 | |
| 315 : | edgomez | 1192 | add ecx, 4 |
| 316 : | cmp ecx, 16 | ||
| 317 : | jnz .qas2_loop | ||
| 318 : | edgomez | 1174 | |
| 319 : | edgomez | 1192 | .qas2_done |
| 320 : | mov ecx, [esp + 8 + 16] ; dcscalar | ||
| 321 : | mov edx, ecx | ||
| 322 : | movsx eax, word [esi] | ||
| 323 : | shr edx, 1 | ||
| 324 : | cmp eax, 0 | ||
| 325 : | jg .qas2_gtzero | ||
| 326 : | edgomez | 1176 | |
| 327 : | edgomez | 1192 | sub eax, edx |
| 328 : | jmp short .qas2_mul | ||
| 329 : | |||
| 330 : | edgomez | 1174 | .qas2_gtzero |
| 331 : | edgomez | 1192 | add eax, edx |
| 332 : | edgomez | 1176 | |
| 333 : | edgomez | 1174 | .qas2_mul |
| 334 : | edgomez | 1192 | cdq |
| 335 : | idiv ecx | ||
| 336 : | edgomez | 1174 | |
| 337 : | edgomez | 1192 | mov [edi], ax |
| 338 : | edgomez | 1174 | |
| 339 : | edgomez | 1192 | xor eax, eax ; return(0); |
| 340 : | pop edi | ||
| 341 : | pop esi | ||
| 342 : | edgomez | 1174 | |
| 343 : | edgomez | 1192 | ret |
| 344 : | |||
| 345 : | ALIGN 16 | ||
| 346 : | edgomez | 1174 | .qas2_q1loop |
| 347 : | edgomez | 1192 | movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
| 348 : | movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] | ||
| 349 : | pxor xmm1, xmm1 | ||
| 350 : | pxor xmm4, xmm4 | ||
| 351 : | pcmpgtw xmm1, xmm0 | ||
| 352 : | pcmpgtw xmm4, xmm3 | ||
| 353 : | pxor xmm0, xmm1 | ||
| 354 : | pxor xmm3, xmm4 | ||
| 355 : | psubw xmm0, xmm1 | ||
| 356 : | psubw xmm3, xmm4 | ||
| 357 : | psrlw xmm0, 1 | ||
| 358 : | psrlw xmm3, 1 | ||
| 359 : | pxor xmm0, xmm1 | ||
| 360 : | pxor xmm3, xmm4 | ||
| 361 : | psubw xmm0, xmm1 | ||
| 362 : | psubw xmm3, xmm4 | ||
| 363 : | movdqa [edi + ecx*8], xmm0 | ||
| 364 : | movdqa [edi + ecx*8 + 16], xmm3 | ||
| 365 : | edgomez | 1174 | |
| 366 : | edgomez | 1192 | add ecx, 4 |
| 367 : | cmp ecx, 16 | ||
| 368 : | jnz .qas2_q1loop | ||
| 369 : | jmp near .qas2_done | ||
| 370 : | edgomez | 1174 | |
| 371 : | |||
| 372 : | |||
| 373 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 374 : | edgomez | 1174 | ; |
| 375 : | edgomez | 1176 | ; uint32_t quant_h263_inter_mmx(int16_t * coeff, |
| 376 : | ; const int16_t const * data, | ||
| 377 : | edgomez | 1230 | ; const uint32_t quant, |
| 378 : | ; const uint16_t *mpeg_matrices); | ||
| 379 : | edgomez | 1174 | ; |
| 380 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 381 : | edgomez | 1174 | |
| 382 : | edgomez | 1192 | ALIGN 16 |
| 383 : | edgomez | 1174 | quant_h263_inter_mmx: |
| 384 : | |||
| 385 : | edgomez | 1192 | push ecx |
| 386 : | push esi | ||
| 387 : | push edi | ||
| 388 : | edgomez | 1174 | |
| 389 : | edgomez | 1192 | mov edi, [esp + 12 + 4] ; coeff |
| 390 : | mov esi, [esp + 12 + 8] ; data | ||
| 391 : | mov eax, [esp + 12 + 12] ; quant | ||
| 392 : | edgomez | 1174 | |
| 393 : | edgomez | 1192 | xor ecx, ecx |
| 394 : | edgomez | 1174 | |
| 395 : | edgomez | 1192 | pxor mm5, mm5 ; sum |
| 396 : | movq mm6, [mmx_sub + eax * 8 - 8] ; sub | ||
| 397 : | edgomez | 1174 | |
| 398 : | edgomez | 1192 | cmp al, 1 |
| 399 : | jz .q1loop | ||
| 400 : | edgomez | 1174 | |
| 401 : | edgomez | 1192 | movq mm7, [mmx_div + eax * 8 - 8] ; divider |
| 402 : | edgomez | 1174 | |
| 403 : | edgomez | 1192 | ALIGN 8 |
| 404 : | edgomez | 1174 | .loop |
| 405 : | edgomez | 1192 | 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 : | psubusw mm0, mm6 ; mm0 -= sub (unsigned, dont go < 0) | ||
| 416 : | psubusw mm3, mm6 ; | ||
| 417 : | pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 | ||
| 418 : | pmulhw mm3, mm7 ; | ||
| 419 : | paddw mm5, mm0 ; sum += mm0 | ||
| 420 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 421 : | paddw mm5, mm3 ; | ||
| 422 : | pxor mm3, mm4 ; | ||
| 423 : | psubw mm0, mm1 ; undisplace | ||
| 424 : | psubw mm3, mm4 | ||
| 425 : | movq [edi + 8*ecx], mm0 | ||
| 426 : | movq [edi + 8*ecx + 8], mm3 | ||
| 427 : | edgomez | 1174 | |
| 428 : | edgomez | 1192 | add ecx, 2 |
| 429 : | cmp ecx, 16 | ||
| 430 : | jnz .loop | ||
| 431 : | edgomez | 1174 | |
| 432 : | .done | ||
| 433 : | edgomez | 1192 | pmaddwd mm5, [plus_one] |
| 434 : | movq mm0, mm5 | ||
| 435 : | psrlq mm5, 32 | ||
| 436 : | paddd mm0, mm5 | ||
| 437 : | edgomez | 1174 | |
| 438 : | edgomez | 1192 | movd eax, mm0 ; return sum |
| 439 : | pop edi | ||
| 440 : | pop esi | ||
| 441 : | pop ecx | ||
| 442 : | edgomez | 1174 | |
| 443 : | edgomez | 1192 | ret |
| 444 : | edgomez | 1174 | |
| 445 : | edgomez | 1192 | ALIGN 8 |
| 446 : | edgomez | 1174 | .q1loop |
| 447 : | edgomez | 1192 | movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
| 448 : | movq mm3, [esi + 8*ecx+ 8] ; | ||
| 449 : | pxor mm1, mm1 ; mm1 = 0 | ||
| 450 : | pxor mm4, mm4 ; | ||
| 451 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 452 : | pcmpgtw mm4, mm3 ; | ||
| 453 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 454 : | pxor mm3, mm4 ; | ||
| 455 : | psubw mm0, mm1 ; displace | ||
| 456 : | psubw mm3, mm4 ; | ||
| 457 : | psubusw mm0, mm6 ; mm0 -= sub (unsigned, dont go < 0) | ||
| 458 : | psubusw mm3, mm6 ; | ||
| 459 : | psrlw mm0, 1 ; mm0 >>= 1 (/2) | ||
| 460 : | psrlw mm3, 1 ; | ||
| 461 : | paddw mm5, mm0 ; sum += mm0 | ||
| 462 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 463 : | paddw mm5, mm3 ; | ||
| 464 : | pxor mm3, mm4 ; | ||
| 465 : | psubw mm0, mm1 ; undisplace | ||
| 466 : | psubw mm3, mm4 | ||
| 467 : | movq [edi + 8*ecx], mm0 | ||
| 468 : | movq [edi + 8*ecx + 8], mm3 | ||
| 469 : | edgomez | 1174 | |
| 470 : | edgomez | 1192 | add ecx, 2 |
| 471 : | cmp ecx, 16 | ||
| 472 : | jnz .q1loop | ||
| 473 : | edgomez | 1174 | |
| 474 : | edgomez | 1192 | jmp .done |
| 475 : | edgomez | 1174 | |
| 476 : | |||
| 477 : | edgomez | 1192 | |
| 478 : | ;----------------------------------------------------------------------------- | ||
| 479 : | edgomez | 1174 | ; |
| 480 : | edgomez | 1176 | ; uint32_t quant_h263_inter_sse2(int16_t * coeff, |
| 481 : | ; const int16_t const * data, | ||
| 482 : | edgomez | 1230 | ; const uint32_t quant, |
| 483 : | ; const uint16_t *mpeg_matrices); | ||
| 484 : | edgomez | 1174 | ; |
| 485 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 486 : | edgomez | 1174 | |
| 487 : | edgomez | 1192 | ALIGN 16 |
| 488 : | edgomez | 1174 | quant_h263_inter_sse2: |
| 489 : | |||
| 490 : | edgomez | 1192 | push esi |
| 491 : | push edi | ||
| 492 : | edgomez | 1174 | |
| 493 : | edgomez | 1192 | mov edi, [esp + 8 + 4] ; coeff |
| 494 : | mov esi, [esp + 8 + 8] ; data | ||
| 495 : | mov eax, [esp + 8 + 12] ; quant | ||
| 496 : | edgomez | 1174 | |
| 497 : | edgomez | 1192 | xor ecx, ecx |
| 498 : | edgomez | 1174 | |
| 499 : | edgomez | 1192 | pxor xmm5, xmm5 ; sum |
| 500 : | edgomez | 1174 | |
| 501 : | edgomez | 1192 | movq mm0, [mmx_sub + eax*8 - 8] ; sub |
| 502 : | movq2dq xmm6, mm0 ; load into low 8 bytes | ||
| 503 : | movlhps xmm6, xmm6 ; duplicate into high 8 bytes | ||
| 504 : | edgomez | 1174 | |
| 505 : | edgomez | 1192 | cmp al, 1 |
| 506 : | jz near .qes2_q1loop | ||
| 507 : | edgomez | 1174 | |
| 508 : | .qes2_not1 | ||
| 509 : | edgomez | 1192 | movq mm0, [mmx_div + eax*8 - 8] ; divider |
| 510 : | movq2dq xmm7, mm0 | ||
| 511 : | movlhps xmm7, xmm7 | ||
| 512 : | edgomez | 1174 | |
| 513 : | edgomez | 1192 | ALIGN 16 |
| 514 : | edgomez | 1174 | .qes2_loop |
| 515 : | edgomez | 1192 | movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
| 516 : | movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] | ||
| 517 : | pxor xmm1, xmm1 | ||
| 518 : | pxor xmm4, xmm4 | ||
| 519 : | pcmpgtw xmm1, xmm0 | ||
| 520 : | pcmpgtw xmm4, xmm3 | ||
| 521 : | pxor xmm0, xmm1 | ||
| 522 : | pxor xmm3, xmm4 | ||
| 523 : | psubw xmm0, xmm1 | ||
| 524 : | psubw xmm3, xmm4 | ||
| 525 : | psubusw xmm0, xmm6 | ||
| 526 : | psubusw xmm3, xmm6 | ||
| 527 : | pmulhw xmm0, xmm7 | ||
| 528 : | pmulhw xmm3, xmm7 | ||
| 529 : | paddw xmm5, xmm0 | ||
| 530 : | pxor xmm0, xmm1 | ||
| 531 : | paddw xmm5, xmm3 | ||
| 532 : | pxor xmm3, xmm4 | ||
| 533 : | psubw xmm0, xmm1 | ||
| 534 : | psubw xmm3, xmm4 | ||
| 535 : | movdqa [edi + ecx*8], xmm0 | ||
| 536 : | movdqa [edi + ecx*8 + 16], xmm3 | ||
| 537 : | edgomez | 1174 | |
| 538 : | edgomez | 1192 | add ecx, 4 |
| 539 : | cmp ecx, 16 | ||
| 540 : | jnz .qes2_loop | ||
| 541 : | edgomez | 1174 | |
| 542 : | .qes2_done | ||
| 543 : | edgomez | 1192 | movdqu xmm6, [plus_one] |
| 544 : | pmaddwd xmm5, xmm6 | ||
| 545 : | movhlps xmm6, xmm5 | ||
| 546 : | paddd xmm5, xmm6 | ||
| 547 : | movdq2q mm0, xmm5 | ||
| 548 : | edgomez | 1174 | |
| 549 : | edgomez | 1192 | movq mm5, mm0 |
| 550 : | psrlq mm5, 32 | ||
| 551 : | paddd mm0, mm5 | ||
| 552 : | edgomez | 1174 | |
| 553 : | edgomez | 1192 | movd eax, mm0 ; return sum |
| 554 : | edgomez | 1174 | |
| 555 : | edgomez | 1192 | pop edi |
| 556 : | pop esi | ||
| 557 : | edgomez | 1174 | |
| 558 : | edgomez | 1192 | ret |
| 559 : | edgomez | 1176 | |
| 560 : | edgomez | 1192 | ALIGN 16 |
| 561 : | edgomez | 1174 | .qes2_q1loop |
| 562 : | edgomez | 1192 | movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
| 563 : | movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] | ||
| 564 : | pxor xmm1, xmm1 | ||
| 565 : | pxor xmm4, xmm4 | ||
| 566 : | pcmpgtw xmm1, xmm0 | ||
| 567 : | pcmpgtw xmm4, xmm3 | ||
| 568 : | pxor xmm0, xmm1 | ||
| 569 : | pxor xmm3, xmm4 | ||
| 570 : | psubw xmm0, xmm1 | ||
| 571 : | psubw xmm3, xmm4 | ||
| 572 : | psubusw xmm0, xmm6 | ||
| 573 : | psubusw xmm3, xmm6 | ||
| 574 : | psrlw xmm0, 1 | ||
| 575 : | psrlw xmm3, 1 | ||
| 576 : | paddw xmm5, xmm0 | ||
| 577 : | pxor xmm0, xmm1 | ||
| 578 : | paddw xmm5, xmm3 | ||
| 579 : | pxor xmm3, xmm4 | ||
| 580 : | psubw xmm0, xmm1 | ||
| 581 : | psubw xmm3, xmm4 | ||
| 582 : | movdqa [edi + ecx*8], xmm0 | ||
| 583 : | movdqa [edi + ecx*8 + 16], xmm3 | ||
| 584 : | |||
| 585 : | add ecx, 4 | ||
| 586 : | cmp ecx, 16 | ||
| 587 : | jnz .qes2_q1loop | ||
| 588 : | jmp .qes2_done | ||
| 589 : | |||
| 590 : | |||
| 591 : | ;----------------------------------------------------------------------------- | ||
| 592 : | edgomez | 1174 | ; |
| 593 : | edgomez | 1176 | ; uint32_t dequant_h263_intra_mmx(int16_t *data, |
| 594 : | ; const int16_t const *coeff, | ||
| 595 : | ; const uint32_t quant, | ||
| 596 : | edgomez | 1230 | ; const uint32_t dcscalar, |
| 597 : | ; const uint16_t *mpeg_matrices); | ||
| 598 : | edgomez | 1174 | ; |
| 599 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 600 : | edgomez | 1174 | |
| 601 : | ; note: we only saturate to +2047 *before* restoring the sign. | ||
| 602 : | ; Hence, final clamp really is [-2048,2047] | ||
| 603 : | |||
| 604 : | edgomez | 1192 | ALIGN 16 |
| 605 : | edgomez | 1174 | dequant_h263_intra_mmx: |
| 606 : | |||
| 607 : | edgomez | 1192 | mov edx, [esp+ 4] ; data |
| 608 : | mov ecx, [esp+ 8] ; coeff | ||
| 609 : | mov eax, [esp+12] ; quant | ||
| 610 : | movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 | ||
| 611 : | movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant | ||
| 612 : | mov eax, -16 | ||
| 613 : | edgomez | 1174 | |
| 614 : | edgomez | 1192 | ALIGN 16 |
| 615 : | edgomez | 1174 | .loop |
| 616 : | edgomez | 1192 | movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
| 617 : | movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] | ||
| 618 : | pxor mm1, mm1 | ||
| 619 : | pxor mm4, mm4 | ||
| 620 : | pcmpgtw mm1, mm0 ; sign(c) | ||
| 621 : | pcmpgtw mm4, mm3 ; sign(c') | ||
| 622 : | pxor mm2, mm2 | ||
| 623 : | pxor mm5, mm5 | ||
| 624 : | pcmpeqw mm2, mm0 ; c is zero | ||
| 625 : | pcmpeqw mm5, mm3 ; c' is zero | ||
| 626 : | pandn mm2, mm6 ; offset = isZero ? 0 : quant_add | ||
| 627 : | pandn mm5, mm6 | ||
| 628 : | pxor mm0, mm1 ; negate if negative | ||
| 629 : | pxor mm3, mm4 ; negate if negative | ||
| 630 : | psubw mm0, mm1 | ||
| 631 : | psubw mm3, mm4 | ||
| 632 : | pmullw mm0, mm7 ; *= 2Q | ||
| 633 : | pmullw mm3, mm7 ; *= 2Q | ||
| 634 : | paddw mm0, mm2 ; + offset | ||
| 635 : | paddw mm3, mm5 ; + offset | ||
| 636 : | paddw mm0, mm1 ; negate back | ||
| 637 : | paddw mm3, mm4 ; negate back | ||
| 638 : | edgomez | 1174 | |
| 639 : | edgomez | 1192 | ; saturates to +2047 |
| 640 : | movq mm2, [mmx_32767_minus_2047] | ||
| 641 : | add eax, 2 | ||
| 642 : | paddsw mm0, mm2 | ||
| 643 : | paddsw mm3, mm2 | ||
| 644 : | psubsw mm0, mm2 | ||
| 645 : | psubsw mm3, mm2 | ||
| 646 : | edgomez | 1174 | |
| 647 : | edgomez | 1192 | pxor mm0, mm1 |
| 648 : | pxor mm3, mm4 | ||
| 649 : | movq [edx + 8*eax + 8*16 - 2*8], mm0 | ||
| 650 : | movq [edx + 8*eax + 8*16+8 - 2*8], mm3 | ||
| 651 : | jnz near .loop | ||
| 652 : | edgomez | 1174 | |
| 653 : | edgomez | 1192 | ; deal with DC |
| 654 : | movd mm0, [ecx] | ||
| 655 : | pmullw mm0, [esp+16] ; dcscalar | ||
| 656 : | movq mm2, [mmx_32767_minus_2047] | ||
| 657 : | paddsw mm0, mm2 | ||
| 658 : | psubsw mm0, mm2 | ||
| 659 : | movq mm3, [mmx_32768_minus_2048] | ||
| 660 : | psubsw mm0, mm3 | ||
| 661 : | paddsw mm0, mm3 | ||
| 662 : | movd eax, mm0 | ||
| 663 : | mov [edx], ax | ||
| 664 : | edgomez | 1174 | |
| 665 : | edgomez | 1192 | xor eax, eax ; return(0); |
| 666 : | ret | ||
| 667 : | edgomez | 1174 | |
| 668 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 669 : | edgomez | 1174 | ; |
| 670 : | edgomez | 1176 | ; uint32_t dequant_h263_intra_xmm(int16_t *data, |
| 671 : | ; const int16_t const *coeff, | ||
| 672 : | ; const uint32_t quant, | ||
| 673 : | edgomez | 1230 | ; const uint32_t dcscalar, |
| 674 : | ; const uint16_t *mpeg_matrices); | ||
| 675 : | edgomez | 1174 | ; |
| 676 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 677 : | edgomez | 1174 | |
| 678 : | ; this is the same as dequant_inter_mmx, except that we're | ||
| 679 : | ; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) | ||
| 680 : | |||
| 681 : | edgomez | 1192 | ALIGN 16 |
| 682 : | edgomez | 1174 | dequant_h263_intra_xmm: |
| 683 : | |||
| 684 : | edgomez | 1192 | mov edx, [esp+ 4] ; data |
| 685 : | mov ecx, [esp+ 8] ; coeff | ||
| 686 : | mov eax, [esp+12] ; quant | ||
| 687 : | movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 | ||
| 688 : | movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant | ||
| 689 : | mov eax, -16 | ||
| 690 : | edgomez | 1174 | |
| 691 : | edgomez | 1192 | ALIGN 16 |
| 692 : | edgomez | 1174 | .loop |
| 693 : | edgomez | 1192 | movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
| 694 : | movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] | ||
| 695 : | pxor mm1, mm1 | ||
| 696 : | pxor mm4, mm4 | ||
| 697 : | pcmpgtw mm1, mm0 ; sign(c) | ||
| 698 : | pcmpgtw mm4, mm3 ; sign(c') | ||
| 699 : | pxor mm2, mm2 | ||
| 700 : | pxor mm5, mm5 | ||
| 701 : | pcmpeqw mm2, mm0 ; c is zero | ||
| 702 : | pcmpeqw mm5, mm3 ; c' is zero | ||
| 703 : | pandn mm2, mm6 ; offset = isZero ? 0 : quant_add | ||
| 704 : | pandn mm5, mm6 | ||
| 705 : | pxor mm0, mm1 ; negate if negative | ||
| 706 : | pxor mm3, mm4 ; negate if negative | ||
| 707 : | psubw mm0, mm1 | ||
| 708 : | psubw mm3, mm4 | ||
| 709 : | pmullw mm0, mm7 ; *= 2Q | ||
| 710 : | pmullw mm3, mm7 ; *= 2Q | ||
| 711 : | paddw mm0, mm2 ; + offset | ||
| 712 : | paddw mm3, mm5 ; + offset | ||
| 713 : | paddw mm0, mm1 ; negate back | ||
| 714 : | paddw mm3, mm4 ; negate back | ||
| 715 : | edgomez | 1174 | |
| 716 : | edgomez | 1176 | ; saturates to +2047 |
| 717 : | edgomez | 1192 | movq mm2, [mmx_2047] |
| 718 : | pminsw mm0, mm2 | ||
| 719 : | add eax, 2 | ||
| 720 : | pminsw mm3, mm2 | ||
| 721 : | edgomez | 1174 | |
| 722 : | edgomez | 1192 | pxor mm0, mm1 |
| 723 : | pxor mm3, mm4 | ||
| 724 : | movq [edx + 8*eax + 8*16 - 2*8], mm0 | ||
| 725 : | movq [edx + 8*eax + 8*16+8 - 2*8], mm3 | ||
| 726 : | jnz near .loop | ||
| 727 : | edgomez | 1174 | |
| 728 : | ; deal with DC | ||
| 729 : | edgomez | 1192 | movd mm0, [ecx] |
| 730 : | pmullw mm0, [esp+16] ; dcscalar | ||
| 731 : | movq mm2, [mmx_32767_minus_2047] | ||
| 732 : | paddsw mm0, mm2 | ||
| 733 : | psubsw mm0, mm2 | ||
| 734 : | movq mm2, [mmx_32768_minus_2048] | ||
| 735 : | psubsw mm0, mm2 | ||
| 736 : | paddsw mm0, mm2 | ||
| 737 : | movd eax, mm0 | ||
| 738 : | mov [edx], ax | ||
| 739 : | edgomez | 1174 | |
| 740 : | edgomez | 1192 | xor eax, eax |
| 741 : | ret | ||
| 742 : | edgomez | 1174 | |
| 743 : | |||
| 744 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 745 : | edgomez | 1174 | ; |
| 746 : | edgomez | 1176 | ; uint32_t dequant_h263_intra_sse2(int16_t *data, |
| 747 : | ; const int16_t const *coeff, | ||
| 748 : | ; const uint32_t quant, | ||
| 749 : | edgomez | 1230 | ; const uint32_t dcscalar, |
| 750 : | ; const uint16_t *mpeg_matrices); | ||
| 751 : | edgomez | 1174 | ; |
| 752 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 753 : | edgomez | 1176 | |
| 754 : | edgomez | 1192 | ALIGN 16 |
| 755 : | edgomez | 1174 | dequant_h263_intra_sse2: |
| 756 : | edgomez | 1192 | mov edx, [esp+ 4] ; data |
| 757 : | mov ecx, [esp+ 8] ; coeff | ||
| 758 : | mov eax, [esp+12] ; quant | ||
| 759 : | movq mm6, [mmx_add + eax * 8 - 8] | ||
| 760 : | movq mm7, [mmx_mul + eax * 8 - 8] | ||
| 761 : | movq2dq xmm6, mm6 | ||
| 762 : | movq2dq xmm7, mm7 | ||
| 763 : | movlhps xmm6, xmm6 | ||
| 764 : | movlhps xmm7, xmm7 | ||
| 765 : | mov eax, -16 | ||
| 766 : | edgomez | 1174 | |
| 767 : | edgomez | 1192 | ALIGN 16 |
| 768 : | edgomez | 1174 | .loop |
| 769 : | edgomez | 1192 | movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
| 770 : | movdqa xmm3, [ecx + 8*16 + 8*eax+ 16] | ||
| 771 : | pxor xmm1, xmm1 | ||
| 772 : | pxor xmm4, xmm4 | ||
| 773 : | pcmpgtw xmm1, xmm0 ; sign(c) | ||
| 774 : | pcmpgtw xmm4, xmm3 | ||
| 775 : | pxor xmm2, xmm2 | ||
| 776 : | pxor xmm5, xmm5 | ||
| 777 : | pcmpeqw xmm2, xmm0 ; c is zero | ||
| 778 : | pcmpeqw xmm5, xmm3 | ||
| 779 : | pandn xmm2, xmm6 ; offset = isZero ? 0 : quant_add | ||
| 780 : | pandn xmm5, xmm6 | ||
| 781 : | pxor xmm0, xmm1 ; negate if negative | ||
| 782 : | pxor xmm3, xmm4 | ||
| 783 : | psubw xmm0, xmm1 | ||
| 784 : | psubw xmm3, xmm4 | ||
| 785 : | pmullw xmm0, xmm7 ; *= 2Q | ||
| 786 : | pmullw xmm3, xmm7 | ||
| 787 : | paddw xmm0, xmm2 ; + offset | ||
| 788 : | paddw xmm3, xmm5 | ||
| 789 : | paddw xmm0, xmm1 ; negate back | ||
| 790 : | paddw xmm3, xmm4 | ||
| 791 : | edgomez | 1174 | |
| 792 : | edgomez | 1192 | ; saturates to +2047 |
| 793 : | movdqa xmm2, [sse2_2047] | ||
| 794 : | pminsw xmm0, xmm2 | ||
| 795 : | add eax, 4 | ||
| 796 : | pminsw xmm3, xmm2 | ||
| 797 : | edgomez | 1174 | |
| 798 : | edgomez | 1192 | pxor xmm0, xmm1 |
| 799 : | pxor xmm3, xmm4 | ||
| 800 : | movdqa [edx + 8*16 - 8*4 + 8*eax], xmm0 | ||
| 801 : | movdqa [edx + 8*16 - 8*4 + 8*eax + 16], xmm3 | ||
| 802 : | jnz near .loop | ||
| 803 : | edgomez | 1174 | |
| 804 : | edgomez | 1192 | ; deal with DC |
| 805 : | movd mm0, [ecx] | ||
| 806 : | pmullw mm0, [esp+16] ; dcscalar | ||
| 807 : | movq mm2, [mmx_32767_minus_2047] | ||
| 808 : | paddsw mm0, mm2 | ||
| 809 : | psubsw mm0, mm2 | ||
| 810 : | movq mm2, [mmx_32768_minus_2048] | ||
| 811 : | psubsw mm0, mm2 | ||
| 812 : | paddsw mm0, mm2 | ||
| 813 : | movd eax, mm0 | ||
| 814 : | mov [edx], ax | ||
| 815 : | edgomez | 1174 | |
| 816 : | edgomez | 1192 | xor eax, eax |
| 817 : | ret | ||
| 818 : | edgomez | 1174 | |
| 819 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 820 : | edgomez | 1174 | ; |
| 821 : | edgomez | 1176 | ; uint32t dequant_h263_inter_mmx(int16_t * data, |
| 822 : | ; const int16_t * const coeff, | ||
| 823 : | edgomez | 1230 | ; const uint32_t quant, |
| 824 : | ; const uint16_t *mpeg_matrices); | ||
| 825 : | edgomez | 1174 | ; |
| 826 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 827 : | edgomez | 1174 | |
| 828 : | edgomez | 1192 | ALIGN 16 |
| 829 : | edgomez | 1174 | dequant_h263_inter_mmx: |
| 830 : | |||
| 831 : | edgomez | 1192 | mov edx, [esp+ 4] ; data |
| 832 : | mov ecx, [esp+ 8] ; coeff | ||
| 833 : | mov eax, [esp+12] ; quant | ||
| 834 : | movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 | ||
| 835 : | movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant | ||
| 836 : | mov eax, -16 | ||
| 837 : | edgomez | 1174 | |
| 838 : | edgomez | 1192 | ALIGN 16 |
| 839 : | edgomez | 1174 | .loop |
| 840 : | edgomez | 1192 | movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
| 841 : | movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] | ||
| 842 : | pxor mm1, mm1 | ||
| 843 : | pxor mm4, mm4 | ||
| 844 : | pcmpgtw mm1, mm0 ; sign(c) | ||
| 845 : | pcmpgtw mm4, mm3 ; sign(c') | ||
| 846 : | pxor mm2, mm2 | ||
| 847 : | pxor mm5, mm5 | ||
| 848 : | pcmpeqw mm2, mm0 ; c is zero | ||
| 849 : | pcmpeqw mm5, mm3 ; c' is zero | ||
| 850 : | pandn mm2, mm6 ; offset = isZero ? 0 : quant_add | ||
| 851 : | pandn mm5, mm6 | ||
| 852 : | pxor mm0, mm1 ; negate if negative | ||
| 853 : | pxor mm3, mm4 ; negate if negative | ||
| 854 : | psubw mm0, mm1 | ||
| 855 : | psubw mm3, mm4 | ||
| 856 : | pmullw mm0, mm7 ; *= 2Q | ||
| 857 : | pmullw mm3, mm7 ; *= 2Q | ||
| 858 : | paddw mm0, mm2 ; + offset | ||
| 859 : | paddw mm3, mm5 ; + offset | ||
| 860 : | paddw mm0, mm1 ; negate back | ||
| 861 : | paddw mm3, mm4 ; negate back | ||
| 862 : | ; saturates to +2047 | ||
| 863 : | movq mm2, [mmx_32767_minus_2047] | ||
| 864 : | add eax, 2 | ||
| 865 : | paddsw mm0, mm2 | ||
| 866 : | paddsw mm3, mm2 | ||
| 867 : | psubsw mm0, mm2 | ||
| 868 : | psubsw mm3, mm2 | ||
| 869 : | edgomez | 1174 | |
| 870 : | edgomez | 1192 | pxor mm0, mm1 |
| 871 : | pxor mm3, mm4 | ||
| 872 : | movq [edx + 8*eax + 8*16 - 2*8], mm0 | ||
| 873 : | movq [edx + 8*eax + 8*16+8 - 2*8], mm3 | ||
| 874 : | jnz near .loop | ||
| 875 : | edgomez | 1174 | |
| 876 : | edgomez | 1192 | xor eax, eax |
| 877 : | ret | ||
| 878 : | edgomez | 1174 | |
| 879 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 880 : | edgomez | 1174 | ; |
| 881 : | edgomez | 1176 | ; uint32_t dequant_h263_inter_xmm(int16_t * data, |
| 882 : | ; const int16_t * const coeff, | ||
| 883 : | edgomez | 1230 | ; const uint32_t quant, |
| 884 : | ; const uint16_t *mpeg_matrices); | ||
| 885 : | edgomez | 1174 | ; |
| 886 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 887 : | edgomez | 1174 | |
| 888 : | ; this is the same as dequant_inter_mmx, | ||
| 889 : | ; except that we're saturating using 'pminsw' (saves 2 cycles/loop) | ||
| 890 : | |||
| 891 : | edgomez | 1192 | ALIGN 16 |
| 892 : | edgomez | 1174 | dequant_h263_inter_xmm: |
| 893 : | |||
| 894 : | edgomez | 1192 | mov edx, [esp+ 4] ; data |
| 895 : | mov ecx, [esp+ 8] ; coeff | ||
| 896 : | mov eax, [esp+12] ; quant | ||
| 897 : | movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 | ||
| 898 : | movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant | ||
| 899 : | mov eax, -16 | ||
| 900 : | edgomez | 1174 | |
| 901 : | edgomez | 1192 | ALIGN 16 |
| 902 : | edgomez | 1174 | .loop |
| 903 : | edgomez | 1192 | movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
| 904 : | movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] | ||
| 905 : | pxor mm1, mm1 | ||
| 906 : | pxor mm4, mm4 | ||
| 907 : | pcmpgtw mm1, mm0 ; sign(c) | ||
| 908 : | pcmpgtw mm4, mm3 ; sign(c') | ||
| 909 : | pxor mm2, mm2 | ||
| 910 : | pxor mm5, mm5 | ||
| 911 : | pcmpeqw mm2, mm0 ; c is zero | ||
| 912 : | pcmpeqw mm5, mm3 ; c' is zero | ||
| 913 : | pandn mm2, mm6 ; offset = isZero ? 0 : quant_add | ||
| 914 : | pandn mm5, mm6 | ||
| 915 : | pxor mm0, mm1 ; negate if negative | ||
| 916 : | pxor mm3, mm4 ; negate if negative | ||
| 917 : | psubw mm0, mm1 | ||
| 918 : | psubw mm3, mm4 | ||
| 919 : | pmullw mm0, mm7 ; *= 2Q | ||
| 920 : | pmullw mm3, mm7 ; *= 2Q | ||
| 921 : | paddw mm0, mm2 ; + offset | ||
| 922 : | paddw mm3, mm5 ; + offset | ||
| 923 : | paddw mm0, mm1 ; start restoring sign | ||
| 924 : | paddw mm3, mm4 ; start restoring sign | ||
| 925 : | ; saturates to +2047 | ||
| 926 : | movq mm2, [mmx_2047] | ||
| 927 : | pminsw mm0, mm2 | ||
| 928 : | add eax, 2 | ||
| 929 : | pminsw mm3, mm2 | ||
| 930 : | edgomez | 1174 | |
| 931 : | edgomez | 1192 | pxor mm0, mm1 ; finish restoring sign |
| 932 : | pxor mm3, mm4 ; finish restoring sign | ||
| 933 : | movq [edx + 8*eax + 8*16 - 2*8], mm0 | ||
| 934 : | movq [edx + 8*eax + 8*16+8 - 2*8], mm3 | ||
| 935 : | jnz near .loop | ||
| 936 : | edgomez | 1174 | |
| 937 : | edgomez | 1192 | xor eax, eax |
| 938 : | ret | ||
| 939 : | edgomez | 1174 | |
| 940 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 941 : | edgomez | 1174 | ; |
| 942 : | edgomez | 1176 | ; uint32_t dequant_h263_inter_sse2(int16_t * data, |
| 943 : | ; const int16_t * const coeff, | ||
| 944 : | edgomez | 1230 | ; const uint32_t quant, |
| 945 : | ; const uint16_t *mpeg_matrices); | ||
| 946 : | edgomez | 1174 | ; |
| 947 : | edgomez | 1192 | ;----------------------------------------------------------------------------- |
| 948 : | |||
| 949 : | ALIGN 16 | ||
| 950 : | edgomez | 1174 | dequant_h263_inter_sse2: |
| 951 : | edgomez | 1192 | mov edx, [esp + 4] ; data |
| 952 : | mov ecx, [esp + 8] ; coeff | ||
| 953 : | mov eax, [esp + 12] ; quant | ||
| 954 : | movq mm6, [mmx_add + eax * 8 - 8] | ||
| 955 : | movq mm7, [mmx_mul + eax * 8 - 8] | ||
| 956 : | movq2dq xmm6, mm6 | ||
| 957 : | movq2dq xmm7, mm7 | ||
| 958 : | movlhps xmm6, xmm6 | ||
| 959 : | movlhps xmm7, xmm7 | ||
| 960 : | mov eax, -16 | ||
| 961 : | edgomez | 1174 | |
| 962 : | edgomez | 1192 | ALIGN 16 |
| 963 : | edgomez | 1174 | .loop |
| 964 : | edgomez | 1192 | movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
| 965 : | movdqa xmm3, [ecx + 8*16 + 8*eax + 16] | ||
| 966 : | edgomez | 1174 | |
| 967 : | edgomez | 1192 | pxor xmm1, xmm1 |
| 968 : | pxor xmm4, xmm4 | ||
| 969 : | pcmpgtw xmm1, xmm0 ; sign(c) | ||
| 970 : | pcmpgtw xmm4, xmm3 | ||
| 971 : | pxor xmm2, xmm2 | ||
| 972 : | pxor xmm5, xmm5 | ||
| 973 : | pcmpeqw xmm2, xmm0 ; c is zero | ||
| 974 : | pcmpeqw xmm5, xmm3 | ||
| 975 : | pandn xmm2, xmm6 | ||
| 976 : | pandn xmm5, xmm6 | ||
| 977 : | pxor xmm0, xmm1 ; negate if negative | ||
| 978 : | pxor xmm3, xmm4 | ||
| 979 : | psubw xmm0, xmm1 | ||
| 980 : | psubw xmm3, xmm4 | ||
| 981 : | pmullw xmm0, xmm7 ; *= 2Q | ||
| 982 : | pmullw xmm3, xmm7 | ||
| 983 : | paddw xmm0, xmm2 ; + offset | ||
| 984 : | paddw xmm3, xmm5 | ||
| 985 : | edgomez | 1174 | |
| 986 : | edgomez | 1192 | paddw xmm0, xmm1 ; start restoring sign |
| 987 : | paddw xmm3, xmm4 | ||
| 988 : | edgomez | 1174 | |
| 989 : | edgomez | 1192 | ; saturates to +2047 |
| 990 : | movdqa xmm2, [sse2_2047] | ||
| 991 : | pminsw xmm0, xmm2 | ||
| 992 : | add eax, 4 | ||
| 993 : | pminsw xmm3, xmm2 | ||
| 994 : | edgomez | 1174 | |
| 995 : | edgomez | 1192 | pxor xmm0, xmm1 ; finish restoring sign |
| 996 : | pxor xmm3, xmm4 | ||
| 997 : | movdqa [edx + 8*16 - 8*4 + 8*eax], xmm0 | ||
| 998 : | movdqa [edx + 8*16 - 8*4 + 8*eax + 16], xmm3 | ||
| 999 : | jnz near .loop | ||
| 1000 : | edgomez | 1174 | |
| 1001 : | edgomez | 1192 | xor eax, eax |
| 1002 : | ret |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |