highbd_subpel_variance_impl_sse2.asm 30.8 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
;
;  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
;
;  Use of this source code is governed by a BSD-style license
;  that can be found in the LICENSE file in the root of the source
;  tree. An additional intellectual property rights grant can be found
;  in the file PATENTS.  All contributing project authors may
;  be found in the AUTHORS file in the root of the source tree.
;

%include "third_party/x86inc/x86inc.asm"

SECTION_RODATA
pw_8: times  8 dw  8
bilin_filter_m_sse2: times  8 dw 16
                     times  8 dw  0
                     times  8 dw 14
                     times  8 dw  2
                     times  8 dw 12
                     times  8 dw  4
                     times  8 dw 10
                     times  8 dw  6
                     times 16 dw  8
                     times  8 dw  6
                     times  8 dw 10
                     times  8 dw  4
                     times  8 dw 12
                     times  8 dw  2
                     times  8 dw 14

SECTION .text

Johann's avatar
Johann committed
33
; int vpx_sub_pixel_varianceNxh(const uint8_t *src, ptrdiff_t src_stride,
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81
;                               int x_offset, int y_offset,
;                               const uint8_t *dst, ptrdiff_t dst_stride,
;                               int height, unsigned int *sse);
;
; This function returns the SE and stores SSE in the given pointer.

%macro SUM_SSE 6 ; src1, dst1, src2, dst2, sum, sse
  psubw                %3, %4
  psubw                %1, %2
  mova                 %4, %3       ; make copies to manipulate to calc sum
  mova                 %2, %1       ; use originals for calc sse
  pmaddwd              %3, %3
  paddw                %4, %2
  pmaddwd              %1, %1
  movhlps              %2, %4
  paddd                %6, %3
  paddw                %4, %2
  pxor                 %2, %2
  pcmpgtw              %2, %4       ; mask for 0 > %4 (sum)
  punpcklwd            %4, %2       ; sign-extend word to dword
  paddd                %6, %1
  paddd                %5, %4

%endmacro

%macro STORE_AND_RET 0
%if mmsize == 16
  ; if H=64 and W=16, we have 8 words of each 2(1bit)x64(6bit)x9bit=16bit
  ; in m6, i.e. it _exactly_ fits in a signed word per word in the xmm reg.
  ; We have to sign-extend it before adding the words within the register
  ; and outputing to a dword.
  movhlps              m3, m7
  movhlps              m4, m6
  paddd                m7, m3
  paddd                m6, m4
  pshufd               m3, m7, 0x1
  pshufd               m4, m6, 0x1
  paddd                m7, m3
  paddd                m6, m4
  mov                  r1, ssem         ; r1 = unsigned int *sse
  movd               [r1], m7           ; store sse
  movd                rax, m6           ; store sum as return value
%endif
  RET
%endmacro

%macro INC_SRC_BY_SRC_STRIDE  0
%if ARCH_X86=1 && CONFIG_PIC=1
82 83
  add                srcq, src_stridemp
  add                srcq, src_stridemp
84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104
%else
  lea                srcq, [srcq + src_strideq*2]
%endif
%endmacro

%macro SUBPEL_VARIANCE 1-2 0 ; W
%define bilin_filter_m bilin_filter_m_sse2
%define filter_idx_shift 5


%ifdef PIC    ; 64bit PIC
  %if %2 == 1 ; avg
    cglobal highbd_sub_pixel_avg_variance%1xh, 9, 10, 13, src, src_stride, \
                                      x_offset, y_offset, \
                                      dst, dst_stride, \
                                      sec, sec_stride, height, sse
    %define sec_str sec_strideq
  %else
    cglobal highbd_sub_pixel_variance%1xh, 7, 8, 13, src, src_stride, x_offset, \
                                  y_offset, dst, dst_stride, height, sse
  %endif
105
  %define block_height heightd
106 107 108 109 110 111 112 113 114
  %define bilin_filter sseq
%else
  %if ARCH_X86=1 && CONFIG_PIC=1
    %if %2 == 1 ; avg
      cglobal highbd_sub_pixel_avg_variance%1xh, 7, 7, 13, src, src_stride, \
                                  x_offset, y_offset, \
                                  dst, dst_stride, \
                                  sec, sec_stride, \
                                  height, sse, g_bilin_filter, g_pw_8
115
      %define block_height dword heightm
116 117 118
      %define sec_str sec_stridemp

      ; Store bilin_filter and pw_8 location in stack
119 120 121 122
      %if GET_GOT_DEFINED == 1
        GET_GOT eax
        add esp, 4                ; restore esp
      %endif
123 124 125 126 127 128 129 130 131 132 133 134

      lea ecx, [GLOBAL(bilin_filter_m)]
      mov g_bilin_filterm, ecx

      lea ecx, [GLOBAL(pw_8)]
      mov g_pw_8m, ecx

      LOAD_IF_USED 0, 1         ; load eax, ecx back
    %else
      cglobal highbd_sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, \
                                x_offset, y_offset, dst, dst_stride, height, \
                                sse, g_bilin_filter, g_pw_8
135
      %define block_height heightd
136 137

      ; Store bilin_filter and pw_8 location in stack
138 139 140 141
      %if GET_GOT_DEFINED == 1
        GET_GOT eax
        add esp, 4                ; restore esp
      %endif
142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159

      lea ecx, [GLOBAL(bilin_filter_m)]
      mov g_bilin_filterm, ecx

      lea ecx, [GLOBAL(pw_8)]
      mov g_pw_8m, ecx

      LOAD_IF_USED 0, 1         ; load eax, ecx back
    %endif
  %else
    %if %2 == 1 ; avg
      cglobal highbd_sub_pixel_avg_variance%1xh, 7 + 2 * ARCH_X86_64, \
                        7 + 2 * ARCH_X86_64, 13, src, src_stride, \
                                             x_offset, y_offset, \
                                             dst, dst_stride, \
                                             sec, sec_stride, \
                                             height, sse
      %if ARCH_X86_64
160
      %define block_height heightd
161 162
      %define sec_str sec_strideq
      %else
163
      %define block_height dword heightm
164 165 166 167 168
      %define sec_str sec_stridemp
      %endif
    %else
      cglobal highbd_sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, \
                              x_offset, y_offset, dst, dst_stride, height, sse
169
      %define block_height heightd
170 171 172 173 174 175 176 177 178 179 180
    %endif

    %define bilin_filter bilin_filter_m
  %endif
%endif

  ASSERT               %1 <= 16         ; m6 overflows if w > 16
  pxor                 m6, m6           ; sum
  pxor                 m7, m7           ; sse

%if %1 < 16
181
  sar                   block_height, 1
182
%endif
183 184 185
%if %2 == 1 ; avg
  shl             sec_str, 1
%endif
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209

  ; FIXME(rbultje) replace by jumptable?
  test          x_offsetd, x_offsetd
  jnz .x_nonzero
  ; x_offset == 0
  test          y_offsetd, y_offsetd
  jnz .x_zero_y_nonzero

  ; x_offset == 0 && y_offset == 0
.x_zero_y_zero_loop:
%if %1 == 16
  movu                 m0, [srcq]
  movu                 m2, [srcq + 16]
  mova                 m1, [dstq]
  mova                 m3, [dstq + 16]
%if %2 == 1 ; avg
  pavgw                m0, [secq]
  pavgw                m2, [secq+16]
%endif
  SUM_SSE              m0, m1, m2, m3, m6, m7

  lea                srcq, [srcq + src_strideq*2]
  lea                dstq, [dstq + dst_strideq*2]
%if %2 == 1 ; avg
210
  add                secq, sec_str
211 212 213 214 215 216 217 218
%endif
%else ; %1 < 16
  movu                 m0, [srcq]
  movu                 m2, [srcq + src_strideq*2]
  mova                 m1, [dstq]
  mova                 m3, [dstq + dst_strideq*2]
%if %2 == 1 ; avg
  pavgw                m0, [secq]
219 220
  add                secq, sec_str
  pavgw                m2, [secq]
221 222 223 224 225 226
%endif
  SUM_SSE              m0, m1, m2, m3, m6, m7

  lea                srcq, [srcq + src_strideq*4]
  lea                dstq, [dstq + dst_strideq*4]
%if %2 == 1 ; avg
227
  add                secq, sec_str
228 229
%endif
%endif
230
  dec                   block_height
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
  jg .x_zero_y_zero_loop
  STORE_AND_RET

.x_zero_y_nonzero:
  cmp           y_offsetd, 8
  jne .x_zero_y_nonhalf

  ; x_offset == 0 && y_offset == 0.5
.x_zero_y_half_loop:
%if %1 == 16
  movu                 m0, [srcq]
  movu                 m1, [srcq+16]
  movu                 m4, [srcq+src_strideq*2]
  movu                 m5, [srcq+src_strideq*2+16]
  mova                 m2, [dstq]
  mova                 m3, [dstq+16]
  pavgw                m0, m4
  pavgw                m1, m5
%if %2 == 1 ; avg
  pavgw                m0, [secq]
  pavgw                m1, [secq+16]
%endif
  SUM_SSE              m0, m2, m1, m3, m6, m7

  lea                srcq, [srcq + src_strideq*2]
  lea                dstq, [dstq + dst_strideq*2]
%if %2 == 1 ; avg
258
  add                secq, sec_str
259 260 261 262 263 264 265 266 267 268 269
%endif
%else ; %1 < 16
  movu                 m0, [srcq]
  movu                 m1, [srcq+src_strideq*2]
  movu                 m5, [srcq+src_strideq*4]
  mova                 m2, [dstq]
  mova                 m3, [dstq+dst_strideq*2]
  pavgw                m0, m1
  pavgw                m1, m5
%if %2 == 1 ; avg
  pavgw                m0, [secq]
270 271
  add                secq, sec_str
  pavgw                m1, [secq]
272 273 274 275 276 277
%endif
  SUM_SSE              m0, m2, m1, m3, m6, m7

  lea                srcq, [srcq + src_strideq*4]
  lea                dstq, [dstq + dst_strideq*4]
%if %2 == 1 ; avg
278
  add                secq, sec_str
279 280
%endif
%endif
281
  dec                   block_height
282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346
  jg .x_zero_y_half_loop
  STORE_AND_RET

.x_zero_y_nonhalf:
  ; x_offset == 0 && y_offset == bilin interpolation
%ifdef PIC
  lea        bilin_filter, [bilin_filter_m]
%endif
  shl           y_offsetd, filter_idx_shift
%if ARCH_X86_64 && mmsize == 16
  mova                 m8, [bilin_filter+y_offsetq]
  mova                 m9, [bilin_filter+y_offsetq+16]
  mova                m10, [pw_8]
%define filter_y_a m8
%define filter_y_b m9
%define filter_rnd m10
%else ; x86-32 or mmx
%if ARCH_X86=1 && CONFIG_PIC=1
; x_offset == 0, reuse x_offset reg
%define tempq x_offsetq
  add y_offsetq, g_bilin_filterm
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
  mov tempq, g_pw_8m
%define filter_rnd [tempq]
%else
  add           y_offsetq, bilin_filter
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
%define filter_rnd [pw_8]
%endif
%endif

.x_zero_y_other_loop:
%if %1 == 16
  movu                 m0, [srcq]
  movu                 m1, [srcq + 16]
  movu                 m4, [srcq+src_strideq*2]
  movu                 m5, [srcq+src_strideq*2+16]
  mova                 m2, [dstq]
  mova                 m3, [dstq+16]
  ; FIXME(rbultje) instead of out=((num-x)*in1+x*in2+rnd)>>log2(num), we can
  ; also do out=in1+(((num-x)*(in2-in1)+rnd)>>log2(num)). Total number of
  ; instructions is the same (5), but it is 1 mul instead of 2, so might be
  ; slightly faster because of pmullw latency. It would also cut our rodata
  ; tables in half for this function, and save 1-2 registers on x86-64.
  pmullw               m1, filter_y_a
  pmullw               m5, filter_y_b
  paddw                m1, filter_rnd
  pmullw               m0, filter_y_a
  pmullw               m4, filter_y_b
  paddw                m0, filter_rnd
  paddw                m1, m5
  paddw                m0, m4
  psrlw                m1, 4
  psrlw                m0, 4
%if %2 == 1 ; avg
  pavgw                m0, [secq]
  pavgw                m1, [secq+16]
%endif
  SUM_SSE              m0, m2, m1, m3, m6, m7

  lea                srcq, [srcq + src_strideq*2]
  lea                dstq, [dstq + dst_strideq*2]
%if %2 == 1 ; avg
347
  add                secq, sec_str
348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
%endif
%else ; %1 < 16
  movu                 m0, [srcq]
  movu                 m1, [srcq+src_strideq*2]
  movu                 m5, [srcq+src_strideq*4]
  mova                 m4, m1
  mova                 m2, [dstq]
  mova                 m3, [dstq+dst_strideq*2]
  pmullw               m1, filter_y_a
  pmullw               m5, filter_y_b
  paddw                m1, filter_rnd
  pmullw               m0, filter_y_a
  pmullw               m4, filter_y_b
  paddw                m0, filter_rnd
  paddw                m1, m5
  paddw                m0, m4
  psrlw                m1, 4
  psrlw                m0, 4
%if %2 == 1 ; avg
  pavgw                m0, [secq]
368 369
  add                secq, sec_str
  pavgw                m1, [secq]
370 371 372 373 374 375
%endif
  SUM_SSE              m0, m2, m1, m3, m6, m7

  lea                srcq, [srcq + src_strideq*4]
  lea                dstq, [dstq + dst_strideq*4]
%if %2 == 1 ; avg
376
  add                secq, sec_str
377 378
%endif
%endif
379
  dec                   block_height
380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412
  jg .x_zero_y_other_loop
%undef filter_y_a
%undef filter_y_b
%undef filter_rnd
  STORE_AND_RET

.x_nonzero:
  cmp           x_offsetd, 8
  jne .x_nonhalf
  ; x_offset == 0.5
  test          y_offsetd, y_offsetd
  jnz .x_half_y_nonzero

  ; x_offset == 0.5 && y_offset == 0
.x_half_y_zero_loop:
%if %1 == 16
  movu                 m0, [srcq]
  movu                 m1, [srcq + 16]
  movu                 m4, [srcq + 2]
  movu                 m5, [srcq + 18]
  mova                 m2, [dstq]
  mova                 m3, [dstq + 16]
  pavgw                m0, m4
  pavgw                m1, m5
%if %2 == 1 ; avg
  pavgw                m0, [secq]
  pavgw                m1, [secq+16]
%endif
  SUM_SSE              m0, m2, m1, m3, m6, m7

  lea                srcq, [srcq + src_strideq*2]
  lea                dstq, [dstq + dst_strideq*2]
%if %2 == 1 ; avg
413
  add                secq, sec_str
414 415 416 417 418 419 420 421 422 423 424 425
%endif
%else ; %1 < 16
  movu                 m0, [srcq]
  movu                 m1, [srcq + src_strideq*2]
  movu                 m4, [srcq + 2]
  movu                 m5, [srcq + src_strideq*2 + 2]
  mova                 m2, [dstq]
  mova                 m3, [dstq + dst_strideq*2]
  pavgw                m0, m4
  pavgw                m1, m5
%if %2 == 1 ; avg
  pavgw                m0, [secq]
426 427
  add                secq, sec_str
  pavgw                m1, [secq]
428 429 430 431 432 433
%endif
  SUM_SSE              m0, m2, m1, m3, m6, m7

  lea                srcq, [srcq + src_strideq*4]
  lea                dstq, [dstq + dst_strideq*4]
%if %2 == 1 ; avg
434
  add                secq, sec_str
435 436
%endif
%endif
437
  dec                   block_height
438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475
  jg .x_half_y_zero_loop
  STORE_AND_RET

.x_half_y_nonzero:
  cmp           y_offsetd, 8
  jne .x_half_y_nonhalf

  ; x_offset == 0.5 && y_offset == 0.5
%if %1 == 16
  movu                 m0, [srcq]
  movu                 m1, [srcq+16]
  movu                 m2, [srcq+2]
  movu                 m3, [srcq+18]
  lea                srcq, [srcq + src_strideq*2]
  pavgw                m0, m2
  pavgw                m1, m3
.x_half_y_half_loop:
  movu                 m2, [srcq]
  movu                 m3, [srcq + 16]
  movu                 m4, [srcq + 2]
  movu                 m5, [srcq + 18]
  pavgw                m2, m4
  pavgw                m3, m5
  pavgw                m0, m2
  pavgw                m1, m3
  mova                 m4, [dstq]
  mova                 m5, [dstq + 16]
%if %2 == 1 ; avg
  pavgw                m0, [secq]
  pavgw                m1, [secq+16]
%endif
  SUM_SSE              m0, m4, m1, m5, m6, m7
  mova                 m0, m2
  mova                 m1, m3

  lea                srcq, [srcq + src_strideq*2]
  lea                dstq, [dstq + dst_strideq*2]
%if %2 == 1 ; avg
476
  add                secq, sec_str
477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495
%endif
%else ; %1 < 16
  movu                 m0, [srcq]
  movu                 m2, [srcq+2]
  lea                srcq, [srcq + src_strideq*2]
  pavgw                m0, m2
.x_half_y_half_loop:
  movu                 m2, [srcq]
  movu                 m3, [srcq + src_strideq*2]
  movu                 m4, [srcq + 2]
  movu                 m5, [srcq + src_strideq*2 + 2]
  pavgw                m2, m4
  pavgw                m3, m5
  pavgw                m0, m2
  pavgw                m2, m3
  mova                 m4, [dstq]
  mova                 m5, [dstq + dst_strideq*2]
%if %2 == 1 ; avg
  pavgw                m0, [secq]
496 497
  add                secq, sec_str
  pavgw                m2, [secq]
498 499 500 501 502 503 504
%endif
  SUM_SSE              m0, m4, m2, m5, m6, m7
  mova                 m0, m3

  lea                srcq, [srcq + src_strideq*4]
  lea                dstq, [dstq + dst_strideq*4]
%if %2 == 1 ; avg
505
  add                secq, sec_str
506 507
%endif
%endif
508
  dec                   block_height
509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581
  jg .x_half_y_half_loop
  STORE_AND_RET

.x_half_y_nonhalf:
  ; x_offset == 0.5 && y_offset == bilin interpolation
%ifdef PIC
  lea        bilin_filter, [bilin_filter_m]
%endif
  shl           y_offsetd, filter_idx_shift
%if ARCH_X86_64 && mmsize == 16
  mova                 m8, [bilin_filter+y_offsetq]
  mova                 m9, [bilin_filter+y_offsetq+16]
  mova                m10, [pw_8]
%define filter_y_a m8
%define filter_y_b m9
%define filter_rnd m10
%else  ; x86_32
%if ARCH_X86=1 && CONFIG_PIC=1
; x_offset == 0.5. We can reuse x_offset reg
%define tempq x_offsetq
  add y_offsetq, g_bilin_filterm
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
  mov tempq, g_pw_8m
%define filter_rnd [tempq]
%else
  add           y_offsetq, bilin_filter
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
%define filter_rnd [pw_8]
%endif
%endif

%if %1 == 16
  movu                 m0, [srcq]
  movu                 m1, [srcq+16]
  movu                 m2, [srcq+2]
  movu                 m3, [srcq+18]
  lea                srcq, [srcq + src_strideq*2]
  pavgw                m0, m2
  pavgw                m1, m3
.x_half_y_other_loop:
  movu                 m2, [srcq]
  movu                 m3, [srcq+16]
  movu                 m4, [srcq+2]
  movu                 m5, [srcq+18]
  pavgw                m2, m4
  pavgw                m3, m5
  mova                 m4, m2
  mova                 m5, m3
  pmullw               m1, filter_y_a
  pmullw               m3, filter_y_b
  paddw                m1, filter_rnd
  paddw                m1, m3
  pmullw               m0, filter_y_a
  pmullw               m2, filter_y_b
  paddw                m0, filter_rnd
  psrlw                m1, 4
  paddw                m0, m2
  mova                 m2, [dstq]
  psrlw                m0, 4
  mova                 m3, [dstq+16]
%if %2 == 1 ; avg
  pavgw                m0, [secq]
  pavgw                m1, [secq+16]
%endif
  SUM_SSE              m0, m2, m1, m3, m6, m7
  mova                 m0, m4
  mova                 m1, m5

  lea                srcq, [srcq + src_strideq*2]
  lea                dstq, [dstq + dst_strideq*2]
%if %2 == 1 ; avg
582
  add                secq, sec_str
583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611
%endif
%else ; %1 < 16
  movu                 m0, [srcq]
  movu                 m2, [srcq+2]
  lea                srcq, [srcq + src_strideq*2]
  pavgw                m0, m2
.x_half_y_other_loop:
  movu                 m2, [srcq]
  movu                 m3, [srcq+src_strideq*2]
  movu                 m4, [srcq+2]
  movu                 m5, [srcq+src_strideq*2+2]
  pavgw                m2, m4
  pavgw                m3, m5
  mova                 m4, m2
  mova                 m5, m3
  pmullw               m4, filter_y_a
  pmullw               m3, filter_y_b
  paddw                m4, filter_rnd
  paddw                m4, m3
  pmullw               m0, filter_y_a
  pmullw               m2, filter_y_b
  paddw                m0, filter_rnd
  psrlw                m4, 4
  paddw                m0, m2
  mova                 m2, [dstq]
  psrlw                m0, 4
  mova                 m3, [dstq+dst_strideq*2]
%if %2 == 1 ; avg
  pavgw                m0, [secq]
612 613
  add                secq, sec_str
  pavgw                m4, [secq]
614 615 616 617 618 619 620
%endif
  SUM_SSE              m0, m2, m4, m3, m6, m7
  mova                 m0, m5

  lea                srcq, [srcq + src_strideq*4]
  lea                dstq, [dstq + dst_strideq*4]
%if %2 == 1 ; avg
621
  add                secq, sec_str
622 623
%endif
%endif
624
  dec                   block_height
625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690
  jg .x_half_y_other_loop
%undef filter_y_a
%undef filter_y_b
%undef filter_rnd
  STORE_AND_RET

.x_nonhalf:
  test          y_offsetd, y_offsetd
  jnz .x_nonhalf_y_nonzero

  ; x_offset == bilin interpolation && y_offset == 0
%ifdef PIC
  lea        bilin_filter, [bilin_filter_m]
%endif
  shl           x_offsetd, filter_idx_shift
%if ARCH_X86_64 && mmsize == 16
  mova                 m8, [bilin_filter+x_offsetq]
  mova                 m9, [bilin_filter+x_offsetq+16]
  mova                m10, [pw_8]
%define filter_x_a m8
%define filter_x_b m9
%define filter_rnd m10
%else    ; x86-32
%if ARCH_X86=1 && CONFIG_PIC=1
; y_offset == 0. We can reuse y_offset reg.
%define tempq y_offsetq
  add x_offsetq, g_bilin_filterm
%define filter_x_a [x_offsetq]
%define filter_x_b [x_offsetq+16]
  mov tempq, g_pw_8m
%define filter_rnd [tempq]
%else
  add           x_offsetq, bilin_filter
%define filter_x_a [x_offsetq]
%define filter_x_b [x_offsetq+16]
%define filter_rnd [pw_8]
%endif
%endif

.x_other_y_zero_loop:
%if %1 == 16
  movu                 m0, [srcq]
  movu                 m1, [srcq+16]
  movu                 m2, [srcq+2]
  movu                 m3, [srcq+18]
  mova                 m4, [dstq]
  mova                 m5, [dstq+16]
  pmullw               m1, filter_x_a
  pmullw               m3, filter_x_b
  paddw                m1, filter_rnd
  pmullw               m0, filter_x_a
  pmullw               m2, filter_x_b
  paddw                m0, filter_rnd
  paddw                m1, m3
  paddw                m0, m2
  psrlw                m1, 4
  psrlw                m0, 4
%if %2 == 1 ; avg
  pavgw                m0, [secq]
  pavgw                m1, [secq+16]
%endif
  SUM_SSE              m0, m4, m1, m5, m6, m7

  lea                srcq, [srcq+src_strideq*2]
  lea                dstq, [dstq+dst_strideq*2]
%if %2 == 1 ; avg
691
  add                secq, sec_str
692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711
%endif
%else ; %1 < 16
  movu                 m0, [srcq]
  movu                 m1, [srcq+src_strideq*2]
  movu                 m2, [srcq+2]
  movu                 m3, [srcq+src_strideq*2+2]
  mova                 m4, [dstq]
  mova                 m5, [dstq+dst_strideq*2]
  pmullw               m1, filter_x_a
  pmullw               m3, filter_x_b
  paddw                m1, filter_rnd
  pmullw               m0, filter_x_a
  pmullw               m2, filter_x_b
  paddw                m0, filter_rnd
  paddw                m1, m3
  paddw                m0, m2
  psrlw                m1, 4
  psrlw                m0, 4
%if %2 == 1 ; avg
  pavgw                m0, [secq]
712 713
  add                secq, sec_str
  pavgw                m1, [secq]
714 715 716 717 718 719
%endif
  SUM_SSE              m0, m4, m1, m5, m6, m7

  lea                srcq, [srcq+src_strideq*4]
  lea                dstq, [dstq+dst_strideq*4]
%if %2 == 1 ; avg
720
  add                secq, sec_str
721 722
%endif
%endif
723
  dec                   block_height
724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808
  jg .x_other_y_zero_loop
%undef filter_x_a
%undef filter_x_b
%undef filter_rnd
  STORE_AND_RET

.x_nonhalf_y_nonzero:
  cmp           y_offsetd, 8
  jne .x_nonhalf_y_nonhalf

  ; x_offset == bilin interpolation && y_offset == 0.5
%ifdef PIC
  lea        bilin_filter, [bilin_filter_m]
%endif
  shl           x_offsetd, filter_idx_shift
%if ARCH_X86_64 && mmsize == 16
  mova                 m8, [bilin_filter+x_offsetq]
  mova                 m9, [bilin_filter+x_offsetq+16]
  mova                m10, [pw_8]
%define filter_x_a m8
%define filter_x_b m9
%define filter_rnd m10
%else    ; x86-32
%if ARCH_X86=1 && CONFIG_PIC=1
; y_offset == 0.5. We can reuse y_offset reg.
%define tempq y_offsetq
  add x_offsetq, g_bilin_filterm
%define filter_x_a [x_offsetq]
%define filter_x_b [x_offsetq+16]
  mov tempq, g_pw_8m
%define filter_rnd [tempq]
%else
  add           x_offsetq, bilin_filter
%define filter_x_a [x_offsetq]
%define filter_x_b [x_offsetq+16]
%define filter_rnd [pw_8]
%endif
%endif

%if %1 == 16
  movu                 m0, [srcq]
  movu                 m1, [srcq+16]
  movu                 m2, [srcq+2]
  movu                 m3, [srcq+18]
  pmullw               m0, filter_x_a
  pmullw               m2, filter_x_b
  paddw                m0, filter_rnd
  pmullw               m1, filter_x_a
  pmullw               m3, filter_x_b
  paddw                m1, filter_rnd
  paddw                m0, m2
  paddw                m1, m3
  psrlw                m0, 4
  psrlw                m1, 4
  lea                srcq, [srcq+src_strideq*2]
.x_other_y_half_loop:
  movu                 m2, [srcq]
  movu                 m3, [srcq+16]
  movu                 m4, [srcq+2]
  movu                 m5, [srcq+18]
  pmullw               m2, filter_x_a
  pmullw               m4, filter_x_b
  paddw                m2, filter_rnd
  pmullw               m3, filter_x_a
  pmullw               m5, filter_x_b
  paddw                m3, filter_rnd
  paddw                m2, m4
  paddw                m3, m5
  mova                 m4, [dstq]
  mova                 m5, [dstq+16]
  psrlw                m2, 4
  psrlw                m3, 4
  pavgw                m0, m2
  pavgw                m1, m3
%if %2 == 1 ; avg
  pavgw                m0, [secq]
  pavgw                m1, [secq+16]
%endif
  SUM_SSE              m0, m4, m1, m5, m6, m7
  mova                 m0, m2
  mova                 m1, m3

  lea                srcq, [srcq+src_strideq*2]
  lea                dstq, [dstq+dst_strideq*2]
%if %2 == 1 ; avg
809
  add                secq, sec_str
810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840
%endif
%else ; %1 < 16
  movu                 m0, [srcq]
  movu                 m2, [srcq+2]
  pmullw               m0, filter_x_a
  pmullw               m2, filter_x_b
  paddw                m0, filter_rnd
  paddw                m0, m2
  psrlw                m0, 4
  lea                srcq, [srcq+src_strideq*2]
.x_other_y_half_loop:
  movu                 m2, [srcq]
  movu                 m3, [srcq+src_strideq*2]
  movu                 m4, [srcq+2]
  movu                 m5, [srcq+src_strideq*2+2]
  pmullw               m2, filter_x_a
  pmullw               m4, filter_x_b
  paddw                m2, filter_rnd
  pmullw               m3, filter_x_a
  pmullw               m5, filter_x_b
  paddw                m3, filter_rnd
  paddw                m2, m4
  paddw                m3, m5
  mova                 m4, [dstq]
  mova                 m5, [dstq+dst_strideq*2]
  psrlw                m2, 4
  psrlw                m3, 4
  pavgw                m0, m2
  pavgw                m2, m3
%if %2 == 1 ; avg
  pavgw                m0, [secq]
841 842
  add                secq, sec_str
  pavgw                m2, [secq]
843 844 845 846 847 848 849
%endif
  SUM_SSE              m0, m4, m2, m5, m6, m7
  mova                 m0, m3

  lea                srcq, [srcq+src_strideq*4]
  lea                dstq, [dstq+dst_strideq*4]
%if %2 == 1 ; avg
850
  add                secq, sec_str
851 852
%endif
%endif
853
  dec                   block_height
854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963
  jg .x_other_y_half_loop
%undef filter_x_a
%undef filter_x_b
%undef filter_rnd
  STORE_AND_RET

.x_nonhalf_y_nonhalf:
; loading filter - this is same as in 8-bit depth
%ifdef PIC
  lea        bilin_filter, [bilin_filter_m]
%endif
  shl           x_offsetd, filter_idx_shift ; filter_idx_shift = 5
  shl           y_offsetd, filter_idx_shift
%if ARCH_X86_64 && mmsize == 16
  mova                 m8, [bilin_filter+x_offsetq]
  mova                 m9, [bilin_filter+x_offsetq+16]
  mova                m10, [bilin_filter+y_offsetq]
  mova                m11, [bilin_filter+y_offsetq+16]
  mova                m12, [pw_8]
%define filter_x_a m8
%define filter_x_b m9
%define filter_y_a m10
%define filter_y_b m11
%define filter_rnd m12
%else   ; x86-32
%if ARCH_X86=1 && CONFIG_PIC=1
; In this case, there is NO unused register. Used src_stride register. Later,
; src_stride has to be loaded from stack when it is needed.
%define tempq src_strideq
  mov tempq, g_bilin_filterm
  add           x_offsetq, tempq
  add           y_offsetq, tempq
%define filter_x_a [x_offsetq]
%define filter_x_b [x_offsetq+16]
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]

  mov tempq, g_pw_8m
%define filter_rnd [tempq]
%else
  add           x_offsetq, bilin_filter
  add           y_offsetq, bilin_filter
%define filter_x_a [x_offsetq]
%define filter_x_b [x_offsetq+16]
%define filter_y_a [y_offsetq]
%define filter_y_b [y_offsetq+16]
%define filter_rnd [pw_8]
%endif
%endif
; end of load filter

  ; x_offset == bilin interpolation && y_offset == bilin interpolation
%if %1 == 16
  movu                 m0, [srcq]
  movu                 m2, [srcq+2]
  movu                 m1, [srcq+16]
  movu                 m3, [srcq+18]
  pmullw               m0, filter_x_a
  pmullw               m2, filter_x_b
  paddw                m0, filter_rnd
  pmullw               m1, filter_x_a
  pmullw               m3, filter_x_b
  paddw                m1, filter_rnd
  paddw                m0, m2
  paddw                m1, m3
  psrlw                m0, 4
  psrlw                m1, 4

  INC_SRC_BY_SRC_STRIDE

.x_other_y_other_loop:
  movu                 m2, [srcq]
  movu                 m4, [srcq+2]
  movu                 m3, [srcq+16]
  movu                 m5, [srcq+18]
  pmullw               m2, filter_x_a
  pmullw               m4, filter_x_b
  paddw                m2, filter_rnd
  pmullw               m3, filter_x_a
  pmullw               m5, filter_x_b
  paddw                m3, filter_rnd
  paddw                m2, m4
  paddw                m3, m5
  psrlw                m2, 4
  psrlw                m3, 4
  mova                 m4, m2
  mova                 m5, m3
  pmullw               m0, filter_y_a
  pmullw               m2, filter_y_b
  paddw                m0, filter_rnd
  pmullw               m1, filter_y_a
  pmullw               m3, filter_y_b
  paddw                m0, m2
  paddw                m1, filter_rnd
  mova                 m2, [dstq]
  paddw                m1, m3
  psrlw                m0, 4
  psrlw                m1, 4
  mova                 m3, [dstq+16]
%if %2 == 1 ; avg
  pavgw                m0, [secq]
  pavgw                m1, [secq+16]
%endif
  SUM_SSE              m0, m2, m1, m3, m6, m7
  mova                 m0, m4
  mova                 m1, m5

  INC_SRC_BY_SRC_STRIDE
  lea                dstq, [dstq + dst_strideq * 2]
%if %2 == 1 ; avg
964
  add                secq, sec_str
965 966 967 968 969 970 971 972 973 974 975 976 977 978 979
%endif
%else ; %1 < 16
  movu                 m0, [srcq]
  movu                 m2, [srcq+2]
  pmullw               m0, filter_x_a
  pmullw               m2, filter_x_b
  paddw                m0, filter_rnd
  paddw                m0, m2
  psrlw                m0, 4

  INC_SRC_BY_SRC_STRIDE

.x_other_y_other_loop:
  movu                 m2, [srcq]
  movu                 m4, [srcq+2]
980 981 982
  INC_SRC_BY_SRC_STRIDE
  movu                 m3, [srcq]
  movu                 m5, [srcq+2]
983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008
  pmullw               m2, filter_x_a
  pmullw               m4, filter_x_b
  paddw                m2, filter_rnd
  pmullw               m3, filter_x_a
  pmullw               m5, filter_x_b
  paddw                m3, filter_rnd
  paddw                m2, m4
  paddw                m3, m5
  psrlw                m2, 4
  psrlw                m3, 4
  mova                 m4, m2
  mova                 m5, m3
  pmullw               m0, filter_y_a
  pmullw               m2, filter_y_b
  paddw                m0, filter_rnd
  pmullw               m4, filter_y_a
  pmullw               m3, filter_y_b
  paddw                m0, m2
  paddw                m4, filter_rnd
  mova                 m2, [dstq]
  paddw                m4, m3
  psrlw                m0, 4
  psrlw                m4, 4
  mova                 m3, [dstq+dst_strideq*2]
%if %2 == 1 ; avg
  pavgw                m0, [secq]
1009 1010
  add                secq, sec_str
  pavgw                m4, [secq]
1011 1012 1013 1014
%endif
  SUM_SSE              m0, m2, m4, m3, m6, m7
  mova                 m0, m5

1015
  INC_SRC_BY_SRC_STRIDE
1016 1017
  lea                dstq, [dstq + dst_strideq * 4]
%if %2 == 1 ; avg
1018
  add                secq, sec_str
1019 1020
%endif
%endif
1021
  dec                   block_height
1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037
  jg .x_other_y_other_loop
%undef filter_x_a
%undef filter_x_b
%undef filter_y_a
%undef filter_y_b
%undef filter_rnd
  STORE_AND_RET
%endmacro

INIT_XMM sse2
SUBPEL_VARIANCE  8
SUBPEL_VARIANCE 16

INIT_XMM sse2
SUBPEL_VARIANCE  8, 1
SUBPEL_VARIANCE 16, 1