Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(948)

Issues Search
    My Issues | Starred     Open | Closed | All

Side by Side Diff: source/libvpx/vp9/common/x86/vp9_high_subpixel_bilinear_sse2.asm

Issue 668403002: libvpx: Pull from upstream (Closed) Base URL: svn://svn.chromium.org/chrome/trunk/deps/third_party/libvpx/
Patch Set: Created 6 years, 1 month ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch | Annotate | Revision Log
« no previous file with comments | « source/libvpx/vp9/common/x86/vp9_high_subpixel_8t_sse2.asm ('k') | source/libvpx/vp9/decoder/vp9_decodeframe.c » ('j') | no next file with comments »
Toggle Intra-line Diffs ('i') | Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
OLDNEW
1 ; 1 ;
2 ; Copyright (c) 2014 The WebM project authors. All Rights Reserved. 2 ; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
3 ; 3 ;
4 ; Use of this source code is governed by a BSD-style license 4 ; Use of this source code is governed by a BSD-style license
5 ; that can be found in the LICENSE file in the root of the source 5 ; that can be found in the LICENSE file in the root of the source
6 ; tree. An additional intellectual property rights grant can be found 6 ; tree. An additional intellectual property rights grant can be found
7 ; in the file PATENTS. All contributing project authors may 7 ; in the file PATENTS. All contributing project authors may
8 ; be found in the AUTHORS file in the root of the source tree. 8 ; be found in the AUTHORS file in the root of the source tree.
9 ; 9 ;
10 10
(...skipping 153 matching lines...) Expand 10 before | Expand all | Expand 10 after
164 %endif 164 %endif
165 movdqu [rdi], xmm0 ;store the result 165 movdqu [rdi], xmm0 ;store the result
166 movdqu [rdi + 16], xmm2 ;store the result 166 movdqu [rdi + 16], xmm2 ;store the result
167 167
168 lea rsi, [rsi + 2*rax] 168 lea rsi, [rsi + 2*rax]
169 lea rdi, [rdi + 2*rdx] 169 lea rdi, [rdi + 2*rdx]
170 dec rcx 170 dec rcx
171 %endm 171 %endm
172 %endif 172 %endif
173 173
174 global sym(vp9_high_filter_block1d4_v2_sse2) PRIVATE 174 global sym(vp9_highbd_filter_block1d4_v2_sse2) PRIVATE
175 sym(vp9_high_filter_block1d4_v2_sse2): 175 sym(vp9_highbd_filter_block1d4_v2_sse2):
176 push rbp 176 push rbp
177 mov rbp, rsp 177 mov rbp, rsp
178 SHADOW_ARGS_TO_STACK 7 178 SHADOW_ARGS_TO_STACK 7
179 push rsi 179 push rsi
180 push rdi 180 push rdi
181 ; end prolog 181 ; end prolog
182 182
183 HIGH_GET_PARAM_4 183 HIGH_GET_PARAM_4
184 .loop: 184 .loop:
185 movq xmm0, [rsi] ;load src 185 movq xmm0, [rsi] ;load src
186 movq xmm1, [rsi + 2*rax] 186 movq xmm1, [rsi + 2*rax]
187 187
188 HIGH_APPLY_FILTER_4 0 188 HIGH_APPLY_FILTER_4 0
189 jnz .loop 189 jnz .loop
190 190
191 ; begin epilog 191 ; begin epilog
192 pop rdi 192 pop rdi
193 pop rsi 193 pop rsi
194 UNSHADOW_ARGS 194 UNSHADOW_ARGS
195 pop rbp 195 pop rbp
196 ret 196 ret
197 197
198 %if ARCH_X86_64 198 %if ARCH_X86_64
199 global sym(vp9_high_filter_block1d8_v2_sse2) PRIVATE 199 global sym(vp9_highbd_filter_block1d8_v2_sse2) PRIVATE
200 sym(vp9_high_filter_block1d8_v2_sse2): 200 sym(vp9_highbd_filter_block1d8_v2_sse2):
201 push rbp 201 push rbp
202 mov rbp, rsp 202 mov rbp, rsp
203 SHADOW_ARGS_TO_STACK 7 203 SHADOW_ARGS_TO_STACK 7
204 SAVE_XMM 8 204 SAVE_XMM 8
205 push rsi 205 push rsi
206 push rdi 206 push rdi
207 ; end prolog 207 ; end prolog
208 208
209 HIGH_GET_PARAM 209 HIGH_GET_PARAM
210 .loop: 210 .loop:
211 movdqu xmm0, [rsi] ;0 211 movdqu xmm0, [rsi] ;0
212 movdqu xmm1, [rsi + 2*rax] ;1 212 movdqu xmm1, [rsi + 2*rax] ;1
213 213
214 HIGH_APPLY_FILTER_8 0 214 HIGH_APPLY_FILTER_8 0
215 jnz .loop 215 jnz .loop
216 216
217 ; begin epilog 217 ; begin epilog
218 pop rdi 218 pop rdi
219 pop rsi 219 pop rsi
220 RESTORE_XMM 220 RESTORE_XMM
221 UNSHADOW_ARGS 221 UNSHADOW_ARGS
222 pop rbp 222 pop rbp
223 ret 223 ret
224 224
225 global sym(vp9_high_filter_block1d16_v2_sse2) PRIVATE 225 global sym(vp9_highbd_filter_block1d16_v2_sse2) PRIVATE
226 sym(vp9_high_filter_block1d16_v2_sse2): 226 sym(vp9_highbd_filter_block1d16_v2_sse2):
227 push rbp 227 push rbp
228 mov rbp, rsp 228 mov rbp, rsp
229 SHADOW_ARGS_TO_STACK 7 229 SHADOW_ARGS_TO_STACK 7
230 SAVE_XMM 9 230 SAVE_XMM 9
231 push rsi 231 push rsi
232 push rdi 232 push rdi
233 ; end prolog 233 ; end prolog
234 234
235 HIGH_GET_PARAM 235 HIGH_GET_PARAM
236 .loop: 236 .loop:
237 movdqu xmm0, [rsi] ;0 237 movdqu xmm0, [rsi] ;0
238 movdqu xmm2, [rsi + 16] 238 movdqu xmm2, [rsi + 16]
239 movdqu xmm1, [rsi + 2*rax] ;1 239 movdqu xmm1, [rsi + 2*rax] ;1
240 movdqu xmm3, [rsi + 2*rax + 16] 240 movdqu xmm3, [rsi + 2*rax + 16]
241 241
242 HIGH_APPLY_FILTER_16 0 242 HIGH_APPLY_FILTER_16 0
243 jnz .loop 243 jnz .loop
244 244
245 ; begin epilog 245 ; begin epilog
246 pop rdi 246 pop rdi
247 pop rsi 247 pop rsi
248 RESTORE_XMM 248 RESTORE_XMM
249 UNSHADOW_ARGS 249 UNSHADOW_ARGS
250 pop rbp 250 pop rbp
251 ret 251 ret
252 %endif 252 %endif
253 253
254 global sym(vp9_high_filter_block1d4_v2_avg_sse2) PRIVATE 254 global sym(vp9_highbd_filter_block1d4_v2_avg_sse2) PRIVATE
255 sym(vp9_high_filter_block1d4_v2_avg_sse2): 255 sym(vp9_highbd_filter_block1d4_v2_avg_sse2):
256 push rbp 256 push rbp
257 mov rbp, rsp 257 mov rbp, rsp
258 SHADOW_ARGS_TO_STACK 7 258 SHADOW_ARGS_TO_STACK 7
259 push rsi 259 push rsi
260 push rdi 260 push rdi
261 ; end prolog 261 ; end prolog
262 262
263 HIGH_GET_PARAM_4 263 HIGH_GET_PARAM_4
264 .loop: 264 .loop:
265 movq xmm0, [rsi] ;load src 265 movq xmm0, [rsi] ;load src
266 movq xmm1, [rsi + 2*rax] 266 movq xmm1, [rsi + 2*rax]
267 267
268 HIGH_APPLY_FILTER_4 1 268 HIGH_APPLY_FILTER_4 1
269 jnz .loop 269 jnz .loop
270 270
271 ; begin epilog 271 ; begin epilog
272 pop rdi 272 pop rdi
273 pop rsi 273 pop rsi
274 UNSHADOW_ARGS 274 UNSHADOW_ARGS
275 pop rbp 275 pop rbp
276 ret 276 ret
277 277
278 %if ARCH_X86_64 278 %if ARCH_X86_64
279 global sym(vp9_high_filter_block1d8_v2_avg_sse2) PRIVATE 279 global sym(vp9_highbd_filter_block1d8_v2_avg_sse2) PRIVATE
280 sym(vp9_high_filter_block1d8_v2_avg_sse2): 280 sym(vp9_highbd_filter_block1d8_v2_avg_sse2):
281 push rbp 281 push rbp
282 mov rbp, rsp 282 mov rbp, rsp
283 SHADOW_ARGS_TO_STACK 7 283 SHADOW_ARGS_TO_STACK 7
284 SAVE_XMM 8 284 SAVE_XMM 8
285 push rsi 285 push rsi
286 push rdi 286 push rdi
287 ; end prolog 287 ; end prolog
288 288
289 HIGH_GET_PARAM 289 HIGH_GET_PARAM
290 .loop: 290 .loop:
291 movdqu xmm0, [rsi] ;0 291 movdqu xmm0, [rsi] ;0
292 movdqu xmm1, [rsi + 2*rax] ;1 292 movdqu xmm1, [rsi + 2*rax] ;1
293 293
294 HIGH_APPLY_FILTER_8 1 294 HIGH_APPLY_FILTER_8 1
295 jnz .loop 295 jnz .loop
296 296
297 ; begin epilog 297 ; begin epilog
298 pop rdi 298 pop rdi
299 pop rsi 299 pop rsi
300 RESTORE_XMM 300 RESTORE_XMM
301 UNSHADOW_ARGS 301 UNSHADOW_ARGS
302 pop rbp 302 pop rbp
303 ret 303 ret
304 304
305 global sym(vp9_high_filter_block1d16_v2_avg_sse2) PRIVATE 305 global sym(vp9_highbd_filter_block1d16_v2_avg_sse2) PRIVATE
306 sym(vp9_high_filter_block1d16_v2_avg_sse2): 306 sym(vp9_highbd_filter_block1d16_v2_avg_sse2):
307 push rbp 307 push rbp
308 mov rbp, rsp 308 mov rbp, rsp
309 SHADOW_ARGS_TO_STACK 7 309 SHADOW_ARGS_TO_STACK 7
310 SAVE_XMM 9 310 SAVE_XMM 9
311 push rsi 311 push rsi
312 push rdi 312 push rdi
313 ; end prolog 313 ; end prolog
314 314
315 HIGH_GET_PARAM 315 HIGH_GET_PARAM
316 .loop: 316 .loop:
317 movdqu xmm0, [rsi] ;0 317 movdqu xmm0, [rsi] ;0
318 movdqu xmm1, [rsi + 2*rax] ;1 318 movdqu xmm1, [rsi + 2*rax] ;1
319 movdqu xmm2, [rsi + 16] 319 movdqu xmm2, [rsi + 16]
320 movdqu xmm3, [rsi + 2*rax + 16] 320 movdqu xmm3, [rsi + 2*rax + 16]
321 321
322 HIGH_APPLY_FILTER_16 1 322 HIGH_APPLY_FILTER_16 1
323 jnz .loop 323 jnz .loop
324 324
325 ; begin epilog 325 ; begin epilog
326 pop rdi 326 pop rdi
327 pop rsi 327 pop rsi
328 RESTORE_XMM 328 RESTORE_XMM
329 UNSHADOW_ARGS 329 UNSHADOW_ARGS
330 pop rbp 330 pop rbp
331 ret 331 ret
332 %endif 332 %endif
333 333
334 global sym(vp9_high_filter_block1d4_h2_sse2) PRIVATE 334 global sym(vp9_highbd_filter_block1d4_h2_sse2) PRIVATE
335 sym(vp9_high_filter_block1d4_h2_sse2): 335 sym(vp9_highbd_filter_block1d4_h2_sse2):
336 push rbp 336 push rbp
337 mov rbp, rsp 337 mov rbp, rsp
338 SHADOW_ARGS_TO_STACK 7 338 SHADOW_ARGS_TO_STACK 7
339 push rsi 339 push rsi
340 push rdi 340 push rdi
341 ; end prolog 341 ; end prolog
342 342
343 HIGH_GET_PARAM_4 343 HIGH_GET_PARAM_4
344 .loop: 344 .loop:
345 movdqu xmm0, [rsi] ;load src 345 movdqu xmm0, [rsi] ;load src
346 movdqa xmm1, xmm0 346 movdqa xmm1, xmm0
347 psrldq xmm1, 2 347 psrldq xmm1, 2
348 348
349 HIGH_APPLY_FILTER_4 0 349 HIGH_APPLY_FILTER_4 0
350 jnz .loop 350 jnz .loop
351 351
352 ; begin epilog 352 ; begin epilog
353 pop rdi 353 pop rdi
354 pop rsi 354 pop rsi
355 UNSHADOW_ARGS 355 UNSHADOW_ARGS
356 pop rbp 356 pop rbp
357 ret 357 ret
358 358
359 %if ARCH_X86_64 359 %if ARCH_X86_64
360 global sym(vp9_high_filter_block1d8_h2_sse2) PRIVATE 360 global sym(vp9_highbd_filter_block1d8_h2_sse2) PRIVATE
361 sym(vp9_high_filter_block1d8_h2_sse2): 361 sym(vp9_highbd_filter_block1d8_h2_sse2):
362 push rbp 362 push rbp
363 mov rbp, rsp 363 mov rbp, rsp
364 SHADOW_ARGS_TO_STACK 7 364 SHADOW_ARGS_TO_STACK 7
365 SAVE_XMM 8 365 SAVE_XMM 8
366 push rsi 366 push rsi
367 push rdi 367 push rdi
368 ; end prolog 368 ; end prolog
369 369
370 HIGH_GET_PARAM 370 HIGH_GET_PARAM
371 .loop: 371 .loop:
372 movdqu xmm0, [rsi] ;load src 372 movdqu xmm0, [rsi] ;load src
373 movdqu xmm1, [rsi + 2] 373 movdqu xmm1, [rsi + 2]
374 374
375 HIGH_APPLY_FILTER_8 0 375 HIGH_APPLY_FILTER_8 0
376 jnz .loop 376 jnz .loop
377 377
378 ; begin epilog 378 ; begin epilog
379 pop rdi 379 pop rdi
380 pop rsi 380 pop rsi
381 RESTORE_XMM 381 RESTORE_XMM
382 UNSHADOW_ARGS 382 UNSHADOW_ARGS
383 pop rbp 383 pop rbp
384 ret 384 ret
385 385
386 global sym(vp9_high_filter_block1d16_h2_sse2) PRIVATE 386 global sym(vp9_highbd_filter_block1d16_h2_sse2) PRIVATE
387 sym(vp9_high_filter_block1d16_h2_sse2): 387 sym(vp9_highbd_filter_block1d16_h2_sse2):
388 push rbp 388 push rbp
389 mov rbp, rsp 389 mov rbp, rsp
390 SHADOW_ARGS_TO_STACK 7 390 SHADOW_ARGS_TO_STACK 7
391 SAVE_XMM 9 391 SAVE_XMM 9
392 push rsi 392 push rsi
393 push rdi 393 push rdi
394 ; end prolog 394 ; end prolog
395 395
396 HIGH_GET_PARAM 396 HIGH_GET_PARAM
397 .loop: 397 .loop:
398 movdqu xmm0, [rsi] ;load src 398 movdqu xmm0, [rsi] ;load src
399 movdqu xmm1, [rsi + 2] 399 movdqu xmm1, [rsi + 2]
400 movdqu xmm2, [rsi + 16] 400 movdqu xmm2, [rsi + 16]
401 movdqu xmm3, [rsi + 18] 401 movdqu xmm3, [rsi + 18]
402 402
403 HIGH_APPLY_FILTER_16 0 403 HIGH_APPLY_FILTER_16 0
404 jnz .loop 404 jnz .loop
405 405
406 ; begin epilog 406 ; begin epilog
407 pop rdi 407 pop rdi
408 pop rsi 408 pop rsi
409 RESTORE_XMM 409 RESTORE_XMM
410 UNSHADOW_ARGS 410 UNSHADOW_ARGS
411 pop rbp 411 pop rbp
412 ret 412 ret
413 %endif 413 %endif
414 414
415 global sym(vp9_high_filter_block1d4_h2_avg_sse2) PRIVATE 415 global sym(vp9_highbd_filter_block1d4_h2_avg_sse2) PRIVATE
416 sym(vp9_high_filter_block1d4_h2_avg_sse2): 416 sym(vp9_highbd_filter_block1d4_h2_avg_sse2):
417 push rbp 417 push rbp
418 mov rbp, rsp 418 mov rbp, rsp
419 SHADOW_ARGS_TO_STACK 7 419 SHADOW_ARGS_TO_STACK 7
420 push rsi 420 push rsi
421 push rdi 421 push rdi
422 ; end prolog 422 ; end prolog
423 423
424 HIGH_GET_PARAM_4 424 HIGH_GET_PARAM_4
425 .loop: 425 .loop:
426 movdqu xmm0, [rsi] ;load src 426 movdqu xmm0, [rsi] ;load src
427 movdqa xmm1, xmm0 427 movdqa xmm1, xmm0
428 psrldq xmm1, 2 428 psrldq xmm1, 2
429 429
430 HIGH_APPLY_FILTER_4 1 430 HIGH_APPLY_FILTER_4 1
431 jnz .loop 431 jnz .loop
432 432
433 ; begin epilog 433 ; begin epilog
434 pop rdi 434 pop rdi
435 pop rsi 435 pop rsi
436 UNSHADOW_ARGS 436 UNSHADOW_ARGS
437 pop rbp 437 pop rbp
438 ret 438 ret
439 439
440 %if ARCH_X86_64 440 %if ARCH_X86_64
441 global sym(vp9_high_filter_block1d8_h2_avg_sse2) PRIVATE 441 global sym(vp9_highbd_filter_block1d8_h2_avg_sse2) PRIVATE
442 sym(vp9_high_filter_block1d8_h2_avg_sse2): 442 sym(vp9_highbd_filter_block1d8_h2_avg_sse2):
443 push rbp 443 push rbp
444 mov rbp, rsp 444 mov rbp, rsp
445 SHADOW_ARGS_TO_STACK 7 445 SHADOW_ARGS_TO_STACK 7
446 SAVE_XMM 8 446 SAVE_XMM 8
447 push rsi 447 push rsi
448 push rdi 448 push rdi
449 ; end prolog 449 ; end prolog
450 450
451 HIGH_GET_PARAM 451 HIGH_GET_PARAM
452 .loop: 452 .loop:
453 movdqu xmm0, [rsi] ;load src 453 movdqu xmm0, [rsi] ;load src
454 movdqu xmm1, [rsi + 2] 454 movdqu xmm1, [rsi + 2]
455 455
456 HIGH_APPLY_FILTER_8 1 456 HIGH_APPLY_FILTER_8 1
457 jnz .loop 457 jnz .loop
458 458
459 ; begin epilog 459 ; begin epilog
460 pop rdi 460 pop rdi
461 pop rsi 461 pop rsi
462 RESTORE_XMM 462 RESTORE_XMM
463 UNSHADOW_ARGS 463 UNSHADOW_ARGS
464 pop rbp 464 pop rbp
465 ret 465 ret
466 466
467 global sym(vp9_high_filter_block1d16_h2_avg_sse2) PRIVATE 467 global sym(vp9_highbd_filter_block1d16_h2_avg_sse2) PRIVATE
468 sym(vp9_high_filter_block1d16_h2_avg_sse2): 468 sym(vp9_highbd_filter_block1d16_h2_avg_sse2):
469 push rbp 469 push rbp
470 mov rbp, rsp 470 mov rbp, rsp
471 SHADOW_ARGS_TO_STACK 7 471 SHADOW_ARGS_TO_STACK 7
472 SAVE_XMM 9 472 SAVE_XMM 9
473 push rsi 473 push rsi
474 push rdi 474 push rdi
475 ; end prolog 475 ; end prolog
476 476
477 HIGH_GET_PARAM 477 HIGH_GET_PARAM
478 .loop: 478 .loop:
479 movdqu xmm0, [rsi] ;load src 479 movdqu xmm0, [rsi] ;load src
480 movdqu xmm1, [rsi + 2] 480 movdqu xmm1, [rsi + 2]
481 movdqu xmm2, [rsi + 16] 481 movdqu xmm2, [rsi + 16]
482 movdqu xmm3, [rsi + 18] 482 movdqu xmm3, [rsi + 18]
483 483
484 HIGH_APPLY_FILTER_16 1 484 HIGH_APPLY_FILTER_16 1
485 jnz .loop 485 jnz .loop
486 486
487 ; begin epilog 487 ; begin epilog
488 pop rdi 488 pop rdi
489 pop rsi 489 pop rsi
490 RESTORE_XMM 490 RESTORE_XMM
491 UNSHADOW_ARGS 491 UNSHADOW_ARGS
492 pop rbp 492 pop rbp
493 ret 493 ret
494 %endif 494 %endif
OLDNEW
« no previous file with comments | « source/libvpx/vp9/common/x86/vp9_high_subpixel_8t_sse2.asm ('k') | source/libvpx/vp9/decoder/vp9_decodeframe.c » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698