Parent Directory
| 
Revision Log
Revision 1793 - (view) (download)
| 1 : | edgomez | 1586 | ;/***************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * - MPEG4 Quantization H263 implementation / MMX optimized - | ||
| 5 : | ; * | ||
| 6 : | ; * Copyright(C) 2001-2003 Peter Ross <pross@xvid.org> | ||
| 7 : | ; * 2002-2003 Pascal Massimino <skal@planet-d.net> | ||
| 8 : | ; * 2004 Andre Werthmann <wertmann@aei.mpg.de> | ||
| 9 : | ; * | ||
| 10 : | ; * This program is free software ; you can redistribute it and/or modify | ||
| 11 : | ; * it under the terms of the GNU General Public License as published by | ||
| 12 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
| 13 : | ; * (at your option) any later version. | ||
| 14 : | ; * | ||
| 15 : | ; * This program is distributed in the hope that it will be useful, | ||
| 16 : | ; * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
| 17 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 18 : | ; * GNU General Public License for more details. | ||
| 19 : | ; * | ||
| 20 : | ; * You should have received a copy of the GNU General Public License | ||
| 21 : | ; * along with this program ; if not, write to the Free Software | ||
| 22 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 23 : | ; * | ||
| 24 : | Isibaar | 1793 | ; * $Id: quantize_h263_mmx.asm,v 1.3 2008-11-11 20:46:24 Isibaar Exp $ |
| 25 : | edgomez | 1586 | ; * |
| 26 : | ; ****************************************************************************/ | ||
| 27 : | |||
| 28 : | ; enable dequant saturate [-2048,2047], test purposes only. | ||
| 29 : | %define SATURATE | ||
| 30 : | |||
| 31 : | BITS 64 | ||
| 32 : | |||
| 33 : | %macro cglobal 1 | ||
| 34 : | %ifdef PREFIX | ||
| 35 : | %ifdef MARK_FUNCS | ||
| 36 : | global _%1:function %1.endfunc-%1 | ||
| 37 : | %define %1 _%1:function %1.endfunc-%1 | ||
| 38 : | Isibaar | 1793 | %define ENDFUNC .endfunc |
| 39 : | edgomez | 1586 | %else |
| 40 : | global _%1 | ||
| 41 : | %define %1 _%1 | ||
| 42 : | Isibaar | 1793 | %define ENDFUNC |
| 43 : | edgomez | 1586 | %endif |
| 44 : | %else | ||
| 45 : | %ifdef MARK_FUNCS | ||
| 46 : | global %1:function %1.endfunc-%1 | ||
| 47 : | Isibaar | 1793 | %define ENDFUNC .endfunc |
| 48 : | edgomez | 1586 | %else |
| 49 : | global %1 | ||
| 50 : | Isibaar | 1793 | %define ENDFUNC |
| 51 : | edgomez | 1586 | %endif |
| 52 : | %endif | ||
| 53 : | %endmacro | ||
| 54 : | |||
| 55 : | ;============================================================================= | ||
| 56 : | ; Read only Local data | ||
| 57 : | ;============================================================================= | ||
| 58 : | |||
| 59 : | %ifdef FORMAT_COFF | ||
| 60 : | SECTION .rodata | ||
| 61 : | %else | ||
| 62 : | SECTION .rodata align=16 | ||
| 63 : | %endif | ||
| 64 : | |||
| 65 : | ALIGN 16 | ||
| 66 : | plus_one: | ||
| 67 : | times 8 dw 1 | ||
| 68 : | |||
| 69 : | ;----------------------------------------------------------------------------- | ||
| 70 : | ; | ||
| 71 : | ; subtract by Q/2 table | ||
| 72 : | ; | ||
| 73 : | ;----------------------------------------------------------------------------- | ||
| 74 : | |||
| 75 : | ALIGN 16 | ||
| 76 : | mmx_sub: | ||
| 77 : | %assign quant 1 | ||
| 78 : | %rep 31 | ||
| 79 : | times 4 dw quant / 2 | ||
| 80 : | %assign quant quant+1 | ||
| 81 : | %endrep | ||
| 82 : | |||
| 83 : | ;----------------------------------------------------------------------------- | ||
| 84 : | ; | ||
| 85 : | ; divide by 2Q table | ||
| 86 : | ; | ||
| 87 : | ; use a shift of 16 to take full advantage of _pmulhw_ | ||
| 88 : | ; for q=1, _pmulhw_ will overflow so it is treated seperately | ||
| 89 : | ; (3dnow2 provides _pmulhuw_ which wont cause overflow) | ||
| 90 : | ; | ||
| 91 : | ;----------------------------------------------------------------------------- | ||
| 92 : | |||
| 93 : | ALIGN 16 | ||
| 94 : | mmx_div: | ||
| 95 : | %assign quant 1 | ||
| 96 : | %rep 31 | ||
| 97 : | times 4 dw (1<<16) / (quant*2) + 1 | ||
| 98 : | %assign quant quant+1 | ||
| 99 : | %endrep | ||
| 100 : | |||
| 101 : | ;----------------------------------------------------------------------------- | ||
| 102 : | ; | ||
| 103 : | ; add by (odd(Q) ? Q : Q - 1) table | ||
| 104 : | ; | ||
| 105 : | ;----------------------------------------------------------------------------- | ||
| 106 : | |||
| 107 : | ALIGN 16 | ||
| 108 : | mmx_add: | ||
| 109 : | %assign quant 1 | ||
| 110 : | %rep 31 | ||
| 111 : | %if quant % 2 != 0 | ||
| 112 : | times 4 dw quant | ||
| 113 : | %else | ||
| 114 : | times 4 dw quant - 1 | ||
| 115 : | %endif | ||
| 116 : | %assign quant quant+1 | ||
| 117 : | %endrep | ||
| 118 : | |||
| 119 : | ;----------------------------------------------------------------------------- | ||
| 120 : | ; | ||
| 121 : | ; multiple by 2Q table | ||
| 122 : | ; | ||
| 123 : | ;----------------------------------------------------------------------------- | ||
| 124 : | |||
| 125 : | ALIGN 16 | ||
| 126 : | mmx_mul: | ||
| 127 : | %assign quant 1 | ||
| 128 : | %rep 31 | ||
| 129 : | times 4 dw quant*2 | ||
| 130 : | %assign quant quant+1 | ||
| 131 : | %endrep | ||
| 132 : | |||
| 133 : | ;----------------------------------------------------------------------------- | ||
| 134 : | ; | ||
| 135 : | ; saturation limits | ||
| 136 : | ; | ||
| 137 : | ;----------------------------------------------------------------------------- | ||
| 138 : | |||
| 139 : | ALIGN 16 | ||
| 140 : | sse2_2047: | ||
| 141 : | times 8 dw 2047 | ||
| 142 : | |||
| 143 : | ALIGN 16 | ||
| 144 : | mmx_2047: | ||
| 145 : | times 4 dw 2047 | ||
| 146 : | |||
| 147 : | ALIGN 8 | ||
| 148 : | mmx_32768_minus_2048: | ||
| 149 : | times 4 dw (32768-2048) | ||
| 150 : | |||
| 151 : | mmx_32767_minus_2047: | ||
| 152 : | times 4 dw (32767-2047) | ||
| 153 : | |||
| 154 : | |||
| 155 : | ;============================================================================= | ||
| 156 : | ; Code | ||
| 157 : | ;============================================================================= | ||
| 158 : | |||
| 159 : | SECTION .text align=16 | ||
| 160 : | |||
| 161 : | cglobal quant_h263_intra_x86_64 | ||
| 162 : | cglobal quant_h263_inter_x86_64 | ||
| 163 : | cglobal dequant_h263_intra_x86_64 | ||
| 164 : | cglobal dequant_h263_inter_x86_64 | ||
| 165 : | |||
| 166 : | ;----------------------------------------------------------------------------- | ||
| 167 : | ; | ||
| 168 : | ; uint32_t quant_h263_intra_x86_64(int16_t * coeff, | ||
| 169 : | ; const int16_t const * data, | ||
| 170 : | ; const uint32_t quant, | ||
| 171 : | ; const uint32_t dcscalar, | ||
| 172 : | ; const uint16_t *mpeg_matrices); | ||
| 173 : | ; Port of the 32bit mmx cousin | ||
| 174 : | ;----------------------------------------------------------------------------- | ||
| 175 : | |||
| 176 : | ALIGN 16 | ||
| 177 : | quant_h263_intra_x86_64: | ||
| 178 : | mov rax, rdx ; quant | ||
| 179 : | ; rsi is data | ||
| 180 : | ; rdi is coeff | ||
| 181 : | mov r8, rcx ; save dscalar | ||
| 182 : | |||
| 183 : | xor rcx, rcx | ||
| 184 : | cmp rax, 1 | ||
| 185 : | jz .q1loop | ||
| 186 : | |||
| 187 : | lea r9, [mmx_div wrt rip] | ||
| 188 : | movq mm7, [r9 + rax * 8 - 8] | ||
| 189 : | |||
| 190 : | ALIGN 16 | ||
| 191 : | Isibaar | 1793 | .loop: |
| 192 : | edgomez | 1586 | movq mm0, [rsi + 8*rcx] ; mm0 = [1st] |
| 193 : | movq mm3, [rsi + 8*rcx + 8] | ||
| 194 : | pxor mm1, mm1 ; mm1 = 0 | ||
| 195 : | pxor mm4, mm4 ; | ||
| 196 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 197 : | pcmpgtw mm4, mm3 ; | ||
| 198 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 199 : | pxor mm3, mm4 ; | ||
| 200 : | psubw mm0, mm1 ; displace | ||
| 201 : | psubw mm3, mm4 ; | ||
| 202 : | pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 | ||
| 203 : | pmulhw mm3, mm7 ; | ||
| 204 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 205 : | pxor mm3, mm4 ; | ||
| 206 : | psubw mm0, mm1 ; undisplace | ||
| 207 : | psubw mm3, mm4 ; | ||
| 208 : | movq [rdi + 8*rcx], mm0 | ||
| 209 : | movq [rdi + 8*rcx + 8], mm3 | ||
| 210 : | |||
| 211 : | add rcx, 2 | ||
| 212 : | cmp rcx, 16 | ||
| 213 : | jnz .loop | ||
| 214 : | |||
| 215 : | Isibaar | 1793 | .done: |
| 216 : | edgomez | 1586 | |
| 217 : | ; caclulate data[0] // (int32_t)dcscalar) | ||
| 218 : | mov rcx, r8 ; dscalar | ||
| 219 : | mov rdx, rcx | ||
| 220 : | movsx eax, word [rsi] ; data[0] with sign extend | ||
| 221 : | shr rdx, 1 ; edx = dcscalar /2 | ||
| 222 : | cmp eax, 0 | ||
| 223 : | jg .gtzero | ||
| 224 : | |||
| 225 : | sub rax, rdx | ||
| 226 : | jmp short .mul | ||
| 227 : | |||
| 228 : | Isibaar | 1793 | .gtzero: |
| 229 : | edgomez | 1586 | add rax, rdx |
| 230 : | Isibaar | 1793 | .mul: |
| 231 : | edgomez | 1586 | cdq ; expand eax -> edx:eax |
| 232 : | idiv ecx ; eax = edx:eax / dcscalar | ||
| 233 : | mov [rdi], ax ; coeff[0] = ax | ||
| 234 : | |||
| 235 : | xor rax, rax ; return(0); | ||
| 236 : | |||
| 237 : | ret | ||
| 238 : | |||
| 239 : | ALIGN 16 | ||
| 240 : | Isibaar | 1793 | .q1loop: |
| 241 : | edgomez | 1586 | movq mm0, [rsi + 8*rcx] ; mm0 = [1st] |
| 242 : | movq mm3, [rsi + 8*rcx + 8] | ||
| 243 : | pxor mm1, mm1 ; mm1 = 0 | ||
| 244 : | pxor mm4, mm4 ; | ||
| 245 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 246 : | pcmpgtw mm4, mm3 ; | ||
| 247 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 248 : | pxor mm3, mm4 ; | ||
| 249 : | psubw mm0, mm1 ; displace | ||
| 250 : | psubw mm3, mm4 ; | ||
| 251 : | psrlw mm0, 1 ; mm0 >>= 1 (/2) | ||
| 252 : | psrlw mm3, 1 ; | ||
| 253 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 254 : | pxor mm3, mm4 | ||
| 255 : | psubw mm0, mm1 ; undisplace | ||
| 256 : | psubw mm3, mm4 ; | ||
| 257 : | movq [rdi + 8*rcx], mm0 | ||
| 258 : | movq [rdi + 8*rcx + 8], mm3 | ||
| 259 : | |||
| 260 : | add rcx, 2 | ||
| 261 : | cmp rcx, 16 | ||
| 262 : | jnz .q1loop | ||
| 263 : | |||
| 264 : | jmp .done | ||
| 265 : | Isibaar | 1793 | ENDFUNC |
| 266 : | edgomez | 1586 | |
| 267 : | |||
| 268 : | ;----------------------------------------------------------------------------- | ||
| 269 : | ; | ||
| 270 : | ; uint32_t quant_h263_inter_x86_64(int16_t * coeff, | ||
| 271 : | ; const int16_t const * data, | ||
| 272 : | ; const uint32_t quant, | ||
| 273 : | ; const uint16_t *mpeg_matrices); | ||
| 274 : | ; Port of the 32bit mmx cousin | ||
| 275 : | ;----------------------------------------------------------------------------- | ||
| 276 : | |||
| 277 : | ALIGN 16 | ||
| 278 : | quant_h263_inter_x86_64: | ||
| 279 : | mov rax, rdx ; quant | ||
| 280 : | ; rsi is data | ||
| 281 : | ; rdi is coeff | ||
| 282 : | |||
| 283 : | xor rcx, rcx | ||
| 284 : | |||
| 285 : | pxor mm5, mm5 ; sum | ||
| 286 : | lea r9, [mmx_sub wrt rip] | ||
| 287 : | movq mm6, [r9 + rax * 8 - 8] ; sub | ||
| 288 : | |||
| 289 : | cmp rax, 1 | ||
| 290 : | jz .q1loop | ||
| 291 : | |||
| 292 : | lea r9, [mmx_div wrt rip] | ||
| 293 : | movq mm7, [r9 + rax * 8 - 8] ; divider | ||
| 294 : | |||
| 295 : | ALIGN 8 | ||
| 296 : | Isibaar | 1793 | .loop: |
| 297 : | edgomez | 1586 | movq mm0, [rsi + 8*rcx] ; mm0 = [1st] |
| 298 : | movq mm3, [rsi + 8*rcx + 8] | ||
| 299 : | pxor mm1, mm1 ; mm1 = 0 | ||
| 300 : | pxor mm4, mm4 ; | ||
| 301 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 302 : | pcmpgtw mm4, mm3 ; | ||
| 303 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 304 : | pxor mm3, mm4 ; | ||
| 305 : | psubw mm0, mm1 ; displace | ||
| 306 : | psubw mm3, mm4 ; | ||
| 307 : | psubusw mm0, mm6 ; mm0 -= sub (unsigned, dont go < 0) | ||
| 308 : | psubusw mm3, mm6 ; | ||
| 309 : | pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 | ||
| 310 : | pmulhw mm3, mm7 ; | ||
| 311 : | paddw mm5, mm0 ; sum += mm0 | ||
| 312 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 313 : | paddw mm5, mm3 ; | ||
| 314 : | pxor mm3, mm4 ; | ||
| 315 : | psubw mm0, mm1 ; undisplace | ||
| 316 : | psubw mm3, mm4 | ||
| 317 : | movq [rdi + 8*rcx], mm0 | ||
| 318 : | movq [rdi + 8*rcx + 8], mm3 | ||
| 319 : | |||
| 320 : | add rcx, 2 | ||
| 321 : | cmp rcx, 16 | ||
| 322 : | jnz .loop | ||
| 323 : | |||
| 324 : | Isibaar | 1793 | .done: |
| 325 : | edgomez | 1586 | pmaddwd mm5, [plus_one wrt rip] |
| 326 : | movq mm0, mm5 | ||
| 327 : | psrlq mm5, 32 | ||
| 328 : | paddd mm0, mm5 | ||
| 329 : | |||
| 330 : | movd rax, mm0 ; return sum | ||
| 331 : | |||
| 332 : | ret | ||
| 333 : | |||
| 334 : | ALIGN 8 | ||
| 335 : | Isibaar | 1793 | .q1loop: |
| 336 : | edgomez | 1586 | movq mm0, [rsi + 8*rcx] ; mm0 = [1st] |
| 337 : | movq mm3, [rsi + 8*rcx+ 8] ; | ||
| 338 : | pxor mm1, mm1 ; mm1 = 0 | ||
| 339 : | pxor mm4, mm4 ; | ||
| 340 : | pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) | ||
| 341 : | pcmpgtw mm4, mm3 ; | ||
| 342 : | pxor mm0, mm1 ; mm0 = |mm0| | ||
| 343 : | pxor mm3, mm4 ; | ||
| 344 : | psubw mm0, mm1 ; displace | ||
| 345 : | psubw mm3, mm4 ; | ||
| 346 : | psubusw mm0, mm6 ; mm0 -= sub (unsigned, dont go < 0) | ||
| 347 : | psubusw mm3, mm6 ; | ||
| 348 : | psrlw mm0, 1 ; mm0 >>= 1 (/2) | ||
| 349 : | psrlw mm3, 1 ; | ||
| 350 : | paddw mm5, mm0 ; sum += mm0 | ||
| 351 : | pxor mm0, mm1 ; mm0 *= sign(mm0) | ||
| 352 : | paddw mm5, mm3 ; | ||
| 353 : | pxor mm3, mm4 ; | ||
| 354 : | psubw mm0, mm1 ; undisplace | ||
| 355 : | psubw mm3, mm4 | ||
| 356 : | movq [rdi + 8*rcx], mm0 | ||
| 357 : | movq [rdi + 8*rcx + 8], mm3 | ||
| 358 : | |||
| 359 : | add rcx, 2 | ||
| 360 : | cmp rcx, 16 | ||
| 361 : | jnz .q1loop | ||
| 362 : | |||
| 363 : | jmp .done | ||
| 364 : | Isibaar | 1793 | ENDFUNC |
| 365 : | edgomez | 1586 | |
| 366 : | |||
| 367 : | ;----------------------------------------------------------------------------- | ||
| 368 : | ; | ||
| 369 : | ; uint32_t dequant_h263_intra_x86_64(int16_t *data, | ||
| 370 : | ; const int16_t const *coeff, | ||
| 371 : | ; const uint32_t quant, | ||
| 372 : | ; const uint32_t dcscalar, | ||
| 373 : | ; const uint16_t *mpeg_matrices); | ||
| 374 : | ; port of the 32bit xmm cousin | ||
| 375 : | ;----------------------------------------------------------------------------- | ||
| 376 : | |||
| 377 : | ; this is the same as dequant_inter_mmx, except that we're | ||
| 378 : | ; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) | ||
| 379 : | |||
| 380 : | ALIGN 16 | ||
| 381 : | dequant_h263_intra_x86_64: | ||
| 382 : | |||
| 383 : | mov rax, rdx ; quant | ||
| 384 : | mov [rsp-8], rcx ; save dscalar | ||
| 385 : | mov rcx, rsi ; coeff | ||
| 386 : | mov rdx, rdi ; data | ||
| 387 : | |||
| 388 : | lea r9, [mmx_add wrt rip] | ||
| 389 : | movq mm6, [r9 + rax*8 - 8] ; quant or quant-1 | ||
| 390 : | lea r9, [mmx_mul wrt rip] | ||
| 391 : | movq mm7, [r9 + rax*8 - 8] ; 2*quant | ||
| 392 : | mov rax, -16 | ||
| 393 : | |||
| 394 : | ALIGN 16 | ||
| 395 : | Isibaar | 1793 | .loop: |
| 396 : | edgomez | 1586 | movq mm0, [rcx+8*rax+8*16] ; c = coeff[i] |
| 397 : | movq mm3, [rcx+8*rax+8*16 + 8] ; c' = coeff[i+1] | ||
| 398 : | pxor mm1, mm1 | ||
| 399 : | pxor mm4, mm4 | ||
| 400 : | pcmpgtw mm1, mm0 ; sign(c) | ||
| 401 : | pcmpgtw mm4, mm3 ; sign(c') | ||
| 402 : | pxor mm2, mm2 | ||
| 403 : | pxor mm5, mm5 | ||
| 404 : | pcmpeqw mm2, mm0 ; c is zero | ||
| 405 : | pcmpeqw mm5, mm3 ; c' is zero | ||
| 406 : | pandn mm2, mm6 ; offset = isZero ? 0 : quant_add | ||
| 407 : | pandn mm5, mm6 | ||
| 408 : | pxor mm0, mm1 ; negate if negative | ||
| 409 : | pxor mm3, mm4 ; negate if negative | ||
| 410 : | psubw mm0, mm1 | ||
| 411 : | psubw mm3, mm4 | ||
| 412 : | pmullw mm0, mm7 ; *= 2Q | ||
| 413 : | pmullw mm3, mm7 ; *= 2Q | ||
| 414 : | paddw mm0, mm2 ; + offset | ||
| 415 : | paddw mm3, mm5 ; + offset | ||
| 416 : | paddw mm0, mm1 ; negate back | ||
| 417 : | paddw mm3, mm4 ; negate back | ||
| 418 : | |||
| 419 : | ; saturates to +2047 | ||
| 420 : | movq mm2, [mmx_2047 wrt rip] | ||
| 421 : | pminsw mm0, mm2 | ||
| 422 : | add rax, 2 | ||
| 423 : | pminsw mm3, mm2 | ||
| 424 : | |||
| 425 : | pxor mm0, mm1 | ||
| 426 : | pxor mm3, mm4 | ||
| 427 : | movq [rdx + 8*rax + 8*16 - 2*8], mm0 | ||
| 428 : | movq [rdx + 8*rax + 8*16+8 - 2*8], mm3 | ||
| 429 : | jnz near .loop | ||
| 430 : | |||
| 431 : | ; deal with DC | ||
| 432 : | movd mm0, [rcx] | ||
| 433 : | pmullw mm0, [rsp-8] ; dscalar | ||
| 434 : | movq mm2, [mmx_32767_minus_2047 wrt rip] | ||
| 435 : | paddsw mm0, mm2 | ||
| 436 : | psubsw mm0, mm2 | ||
| 437 : | movq mm2, [mmx_32768_minus_2048 wrt rip] | ||
| 438 : | psubsw mm0, mm2 | ||
| 439 : | paddsw mm0, mm2 | ||
| 440 : | movd rax, mm0 | ||
| 441 : | mov [rdx], ax | ||
| 442 : | |||
| 443 : | xor rax, rax | ||
| 444 : | ret | ||
| 445 : | Isibaar | 1793 | ENDFUNC |
| 446 : | edgomez | 1586 | |
| 447 : | |||
| 448 : | ;----------------------------------------------------------------------------- | ||
| 449 : | ; | ||
| 450 : | ; uint32_t dequant_h263_inter_x86_64(int16_t * data, | ||
| 451 : | ; const int16_t * const coeff, | ||
| 452 : | ; const uint32_t quant, | ||
| 453 : | ; const uint16_t *mpeg_matrices); | ||
| 454 : | ; Port of the 32bit xmm cousin | ||
| 455 : | ;----------------------------------------------------------------------------- | ||
| 456 : | |||
| 457 : | ; this is the same as dequant_inter_mmx, | ||
| 458 : | ; except that we're saturating using 'pminsw' (saves 2 cycles/loop) | ||
| 459 : | |||
| 460 : | ALIGN 16 | ||
| 461 : | dequant_h263_inter_x86_64: | ||
| 462 : | |||
| 463 : | mov rax, rdx ; quant | ||
| 464 : | mov rcx, rsi ; coeff | ||
| 465 : | mov rdx, rdi ; data | ||
| 466 : | |||
| 467 : | lea r9, [mmx_add wrt rip] | ||
| 468 : | movq mm6, [r9 + rax*8 - 8] ; quant or quant-1 | ||
| 469 : | lea r9, [mmx_mul wrt rip] | ||
| 470 : | movq mm7, [r9 + rax*8 - 8] ; 2*quant | ||
| 471 : | mov rax, -16 | ||
| 472 : | |||
| 473 : | ALIGN 16 | ||
| 474 : | Isibaar | 1793 | .loop: |
| 475 : | edgomez | 1586 | movq mm0, [rcx+8*rax+8*16] ; c = coeff[i] |
| 476 : | movq mm3, [rcx+8*rax+8*16 + 8] ; c' = coeff[i+1] | ||
| 477 : | pxor mm1, mm1 | ||
| 478 : | pxor mm4, mm4 | ||
| 479 : | pcmpgtw mm1, mm0 ; sign(c) | ||
| 480 : | pcmpgtw mm4, mm3 ; sign(c') | ||
| 481 : | pxor mm2, mm2 | ||
| 482 : | pxor mm5, mm5 | ||
| 483 : | pcmpeqw mm2, mm0 ; c is zero | ||
| 484 : | pcmpeqw mm5, mm3 ; c' is zero | ||
| 485 : | pandn mm2, mm6 ; offset = isZero ? 0 : quant_add | ||
| 486 : | pandn mm5, mm6 | ||
| 487 : | pxor mm0, mm1 ; negate if negative | ||
| 488 : | pxor mm3, mm4 ; negate if negative | ||
| 489 : | psubw mm0, mm1 | ||
| 490 : | psubw mm3, mm4 | ||
| 491 : | pmullw mm0, mm7 ; *= 2Q | ||
| 492 : | pmullw mm3, mm7 ; *= 2Q | ||
| 493 : | paddw mm0, mm2 ; + offset | ||
| 494 : | paddw mm3, mm5 ; + offset | ||
| 495 : | paddw mm0, mm1 ; start restoring sign | ||
| 496 : | paddw mm3, mm4 ; start restoring sign | ||
| 497 : | ; saturates to +2047 | ||
| 498 : | movq mm2, [mmx_2047 wrt rip] | ||
| 499 : | pminsw mm0, mm2 | ||
| 500 : | add rax, 2 | ||
| 501 : | pminsw mm3, mm2 | ||
| 502 : | |||
| 503 : | pxor mm0, mm1 ; finish restoring sign | ||
| 504 : | pxor mm3, mm4 ; finish restoring sign | ||
| 505 : | movq [rdx + 8*rax + 8*16 - 2*8], mm0 | ||
| 506 : | movq [rdx + 8*rax + 8*16+8 - 2*8], mm3 | ||
| 507 : | jnz near .loop | ||
| 508 : | |||
| 509 : | xor rax, rax | ||
| 510 : | ret | ||
| 511 : | Isibaar | 1793 | ENDFUNC |
| 512 : | Isibaar | 1790 | |
| 513 : | %ifidn __OUTPUT_FORMAT__,elf | ||
| 514 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
| 515 : | %endif | ||
| 516 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |