vp8/encoder/x86/variance_sse2.c

   1 /*
   2  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
   3  *
   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
   6  *  tree. An additional intellectual property rights grant can be found
   7  *  in the file PATENTS.  All contributing project authors may
   8  *  be found in the AUTHORS file in the root of the source tree.
   9  */
  10
  11
  12 #include "vp8/encoder/variance.h"
  13 #include "vp8/common/pragmas.h"
  14 #include "vpx_ports/mem.h"
  15
  16 extern void filter_block1d_h6_mmx(const unsigned char *src_ptr, unsigned short *output_ptr, unsigned int src_pixels_per_line, unsigned int pixel_step, unsigned int output_height, unsigned int output_width, short *vp7_filter);
  17 extern void filter_block1d_v6_mmx(const short *src_ptr, unsigned char *output_ptr, unsigned int pixels_per_line, unsigned int pixel_step, unsigned int output_height, unsigned int output_width, short *vp7_filter);
  18 extern void filter_block1d8_h6_sse2(const unsigned char *src_ptr, unsigned short *output_ptr, unsigned int src_pixels_per_line, unsigned int pixel_step, unsigned int output_height, unsigned int output_width, short *vp7_filter);
  19 extern void filter_block1d8_v6_sse2(const short *src_ptr, unsigned char *output_ptr, unsigned int pixels_per_line, unsigned int pixel_step, unsigned int output_height, unsigned int output_width, short *vp7_filter);
  20
  21 extern void vp8_filter_block2d_bil4x4_var_mmx
  22 (
  23     const unsigned char *ref_ptr,
  24     int ref_pixels_per_line,
  25     const unsigned char *src_ptr,
  26     int src_pixels_per_line,
  27     const short *HFilter,
  28     const short *VFilter,
  29     int *sum,
  30     unsigned int *sumsquared
  31 );
  32
  33 extern unsigned int vp8_get4x4var_mmx
  34 (
  35     const unsigned char *src_ptr,
  36     int  source_stride,
  37     const unsigned char *ref_ptr,
  38     int  recon_stride,
  39     unsigned int *SSE,
  40     int *Sum
  41 );
  42
  43 unsigned int vp8_get_mb_ss_sse2
  44 (
  45     const short *src_ptr
  46 );
  47 unsigned int vp8_get16x16var_sse2
  48 (
  49     const unsigned char *src_ptr,
  50     int source_stride,
  51     const unsigned char *ref_ptr,
  52     int recon_stride,
  53     unsigned int *SSE,
  54     int *Sum
  55 );
  56 unsigned int vp8_get16x16pred_error_sse2
  57 (
  58     const unsigned char *src_ptr,
  59     int src_stride,
  60     const unsigned char *ref_ptr,
  61     int ref_stride
  62 );
  63 unsigned int vp8_get8x8var_sse2
  64 (
  65     const unsigned char *src_ptr,
  66     int source_stride,
  67     const unsigned char *ref_ptr,
  68     int recon_stride,
  69     unsigned int *SSE,
  70     int *Sum
  71 );
  72 void vp8_filter_block2d_bil_var_sse2
  73 (
  74     const unsigned char *ref_ptr,
  75     int ref_pixels_per_line,
  76     const unsigned char *src_ptr,
  77     int src_pixels_per_line,
  78     unsigned int Height,
  79     int  xoffset,
  80     int  yoffset,
  81     int *sum,
  82     unsigned int *sumsquared
  83 );
  84 void vp8_half_horiz_vert_variance16x_h_sse2
  85 (
  86     const unsigned char *ref_ptr,
  87     int ref_pixels_per_line,
  88     const unsigned char *src_ptr,
  89     int src_pixels_per_line,
  90     unsigned int Height,
  91     int *sum,
  92     unsigned int *sumsquared
  93 );
  94 void vp8_half_horiz_variance16x_h_sse2
  95 (
  96     const unsigned char *ref_ptr,
  97     int ref_pixels_per_line,
  98     const unsigned char *src_ptr,
  99     int src_pixels_per_line,
 100     unsigned int Height,
 101     int *sum,
 102     unsigned int *sumsquared
 103 );
 104 void vp8_half_vert_variance16x_h_sse2
 105 (
 106     const unsigned char *ref_ptr,
 107     int ref_pixels_per_line,
 108     const unsigned char *src_ptr,
 109     int src_pixels_per_line,
 110     unsigned int Height,
 111     int *sum,
 112     unsigned int *sumsquared
 113 );
 114
 115 DECLARE_ALIGNED(16, extern short, vp8_vp7_bilinear_filters_mmx[8][8]);
 116
 117 unsigned int vp8_variance4x4_wmt(
 118     const unsigned char *src_ptr,
 119     int  source_stride,
 120     const unsigned char *ref_ptr,
 121     int  recon_stride)
 122 {
 123     unsigned int var;
 124     int avg;
 125
 126     vp8_get4x4var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg) ;
 127     return (var - ((avg * avg) >> 4));
 128
 129 }
 130
 131
 132
 133 unsigned int vp8_variance8x8_wmt
 134 (
 135     const unsigned char *src_ptr,
 136     int  source_stride,
 137     const unsigned char *ref_ptr,
 138     int  recon_stride)
 139 {
 140     unsigned int var;
 141     int avg;
 142
 143     vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg) ;
 144
 145     return (var - ((avg * avg) >> 6));
 146
 147 }
 148
 149
 150 unsigned int vp8_variance16x16_wmt
 151 (
 152     const unsigned char *src_ptr,
 153     int  source_stride,
 154     const unsigned char *ref_ptr,
 155     int  recon_stride,
 156     unsigned int *sse)
 157 {
 158     unsigned int sse0;
 159     int sum0;
 160
 161
 162     vp8_get16x16var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
 163     *sse = sse0;
 164     return (sse0 - ((sum0 * sum0) >> 8));
 165 }
 166 unsigned int vp8_mse16x16_wmt(
 167     const unsigned char *src_ptr,
 168     int  source_stride,
 169     const unsigned char *ref_ptr,
 170     int  recon_stride,
 171     unsigned int *sse)
 172 {
 173
 174     unsigned int sse0;
 175     int sum0;
 176     vp8_get16x16var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
 177     *sse = sse0;
 178     return sse0;
 179
 180 }
 181
 182
 183 unsigned int vp8_variance16x8_wmt
 184 (
 185     const unsigned char *src_ptr,
 186     int  source_stride,
 187     const unsigned char *ref_ptr,
 188     int  recon_stride,
 189     unsigned int *sse)
 190 {
 191     unsigned int sse0, sse1, var;
 192     int sum0, sum1, avg;
 193
 194     vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
 195     vp8_get8x8var_sse2(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
 196
 197     var = sse0 + sse1;
 198     avg = sum0 + sum1;
 199     *sse = var;
 200     return (var - ((avg * avg) >> 7));
 201
 202 }
 203
 204 unsigned int vp8_variance8x16_wmt
 205 (
 206     const unsigned char *src_ptr,
 207     int  source_stride,
 208     const unsigned char *ref_ptr,
 209     int  recon_stride,
 210     unsigned int *sse)
 211 {
 212     unsigned int sse0, sse1, var;
 213     int sum0, sum1, avg;
 214
 215     vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
 216     vp8_get8x8var_sse2(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse1, &sum1) ;
 217
 218     var = sse0 + sse1;
 219     avg = sum0 + sum1;
 220     *sse = var;
 221     return (var - ((avg * avg) >> 7));
 222
 223 }
 224
 225 unsigned int vp8_sub_pixel_variance4x4_wmt
 226 (
 227     const unsigned char  *src_ptr,
 228     int  src_pixels_per_line,
 229     int  xoffset,
 230     int  yoffset,
 231     const unsigned char *dst_ptr,
 232     int dst_pixels_per_line,
 233     unsigned int *sse
 234 )
 235 {
 236     int xsum;
 237     unsigned int xxsum;
 238     vp8_filter_block2d_bil4x4_var_mmx(
 239         src_ptr, src_pixels_per_line,
 240         dst_ptr, dst_pixels_per_line,
 241         vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset],
 242         &xsum, &xxsum
 243     );
 244     *sse = xxsum;
 245     return (xxsum - ((xsum * xsum) >> 4));
 246 }
 247
 248
 249 unsigned int vp8_sub_pixel_variance8x8_wmt
 250 (
 251     const unsigned char  *src_ptr,
 252     int  src_pixels_per_line,
 253     int  xoffset,
 254     int  yoffset,
 255     const unsigned char *dst_ptr,
 256     int dst_pixels_per_line,
 257     unsigned int *sse
 258 )
 259 {
 260     int xsum;
 261     unsigned int xxsum;
 262
 263     if (xoffset == 4 && yoffset == 0)
 264     {
 265         vp8_half_horiz_variance16x_h_sse2(
 266             src_ptr, src_pixels_per_line,
 267             dst_ptr, dst_pixels_per_line, 8,
 268             &xsum, &xxsum);
 269     }
 270     else if (xoffset == 0 && yoffset == 4)
 271     {
 272         vp8_half_vert_variance16x_h_sse2(
 273             src_ptr, src_pixels_per_line,
 274             dst_ptr, dst_pixels_per_line, 8,
 275             &xsum, &xxsum);
 276     }
 277     else if (xoffset == 4 && yoffset == 4)
 278     {
 279         vp8_half_horiz_vert_variance16x_h_sse2(
 280             src_ptr, src_pixels_per_line,
 281             dst_ptr, dst_pixels_per_line, 8,
 282             &xsum, &xxsum);
 283     }
 284     else
 285     {
 286         vp8_filter_block2d_bil_var_sse2(
 287             src_ptr, src_pixels_per_line,
 288             dst_ptr, dst_pixels_per_line, 8,
 289             xoffset, yoffset,
 290             &xsum, &xxsum);
 291     }
 292
 293     *sse = xxsum;
 294     return (xxsum - ((xsum * xsum) >> 6));
 295 }
 296
 297 unsigned int vp8_sub_pixel_variance16x16_wmt
 298 (
 299     const unsigned char  *src_ptr,
 300     int  src_pixels_per_line,
 301     int  xoffset,
 302     int  yoffset,
 303     const unsigned char *dst_ptr,
 304     int dst_pixels_per_line,
 305     unsigned int *sse
 306 )
 307 {
 308     int xsum0, xsum1;
 309     unsigned int xxsum0, xxsum1;
 310
 311
 312     // note we could avoid these if statements if the calling function
 313     // just called the appropriate functions inside.
 314     if (xoffset == 4 && yoffset == 0)
 315     {
 316         vp8_half_horiz_variance16x_h_sse2(
 317             src_ptr, src_pixels_per_line,
 318             dst_ptr, dst_pixels_per_line, 16,
 319             &xsum0, &xxsum0);
 320
 321         vp8_half_horiz_variance16x_h_sse2(
 322             src_ptr + 8, src_pixels_per_line,
 323             dst_ptr + 8, dst_pixels_per_line, 16,
 324             &xsum1, &xxsum1);
 325     }
 326     else if (xoffset == 0 && yoffset == 4)
 327     {
 328         vp8_half_vert_variance16x_h_sse2(
 329             src_ptr, src_pixels_per_line,
 330             dst_ptr, dst_pixels_per_line, 16,
 331             &xsum0, &xxsum0);
 332
 333         vp8_half_vert_variance16x_h_sse2(
 334             src_ptr + 8, src_pixels_per_line,
 335             dst_ptr + 8, dst_pixels_per_line, 16,
 336             &xsum1, &xxsum1);
 337     }
 338     else if (xoffset == 4 && yoffset == 4)
 339     {
 340         vp8_half_horiz_vert_variance16x_h_sse2(
 341             src_ptr, src_pixels_per_line,
 342             dst_ptr, dst_pixels_per_line, 16,
 343             &xsum0, &xxsum0);
 344
 345         vp8_half_horiz_vert_variance16x_h_sse2(
 346             src_ptr + 8, src_pixels_per_line,
 347             dst_ptr + 8, dst_pixels_per_line, 16,
 348             &xsum1, &xxsum1);
 349     }
 350     else
 351     {
 352         vp8_filter_block2d_bil_var_sse2(
 353             src_ptr, src_pixels_per_line,
 354             dst_ptr, dst_pixels_per_line, 16,
 355             xoffset, yoffset,
 356             &xsum0, &xxsum0
 357         );
 358
 359
 360         vp8_filter_block2d_bil_var_sse2(
 361             src_ptr + 8, src_pixels_per_line,
 362             dst_ptr + 8, dst_pixels_per_line, 16,
 363             xoffset, yoffset,
 364             &xsum1, &xxsum1
 365         );
 366     }
 367
 368     xsum0 += xsum1;
 369     xxsum0 += xxsum1;
 370     *sse = xxsum0;
 371     return (xxsum0 - ((xsum0 * xsum0) >> 8));
 372 }
 373
 374 unsigned int vp8_sub_pixel_mse16x16_wmt(
 375     const unsigned char  *src_ptr,
 376     int  src_pixels_per_line,
 377     int  xoffset,
 378     int  yoffset,
 379     const unsigned char *dst_ptr,
 380     int dst_pixels_per_line,
 381     unsigned int *sse
 382 )
 383 {
 384     vp8_sub_pixel_variance16x16_wmt(src_ptr, src_pixels_per_line, xoffset, yoffset, dst_ptr, dst_pixels_per_line, sse);
 385     return *sse;
 386 }
 387
 388 unsigned int vp8_sub_pixel_variance16x8_wmt
 389 (
 390     const unsigned char  *src_ptr,
 391     int  src_pixels_per_line,
 392     int  xoffset,
 393     int  yoffset,
 394     const unsigned char *dst_ptr,
 395     int dst_pixels_per_line,
 396     unsigned int *sse
 397
 398 )
 399 {
 400     int xsum0, xsum1;
 401     unsigned int xxsum0, xxsum1;
 402
 403     if (xoffset == 4 && yoffset == 0)
 404     {
 405         vp8_half_horiz_variance16x_h_sse2(
 406             src_ptr, src_pixels_per_line,
 407             dst_ptr, dst_pixels_per_line, 8,
 408             &xsum0, &xxsum0);
 409
 410         vp8_half_horiz_variance16x_h_sse2(
 411             src_ptr + 8, src_pixels_per_line,
 412             dst_ptr + 8, dst_pixels_per_line, 8,
 413             &xsum1, &xxsum1);
 414     }
 415     else if (xoffset == 0 && yoffset == 4)
 416     {
 417         vp8_half_vert_variance16x_h_sse2(
 418             src_ptr, src_pixels_per_line,
 419             dst_ptr, dst_pixels_per_line, 8,
 420             &xsum0, &xxsum0);
 421
 422         vp8_half_vert_variance16x_h_sse2(
 423             src_ptr + 8, src_pixels_per_line,
 424             dst_ptr + 8, dst_pixels_per_line, 8,
 425             &xsum1, &xxsum1);
 426     }
 427     else if (xoffset == 4 && yoffset == 4)
 428     {
 429         vp8_half_horiz_vert_variance16x_h_sse2(
 430             src_ptr, src_pixels_per_line,
 431             dst_ptr, dst_pixels_per_line, 8,
 432             &xsum0, &xxsum0);
 433
 434         vp8_half_horiz_vert_variance16x_h_sse2(
 435             src_ptr + 8, src_pixels_per_line,
 436             dst_ptr + 8, dst_pixels_per_line, 8,
 437             &xsum1, &xxsum1);
 438     }
 439     else
 440     {
 441         vp8_filter_block2d_bil_var_sse2(
 442             src_ptr, src_pixels_per_line,
 443             dst_ptr, dst_pixels_per_line, 8,
 444             xoffset, yoffset,
 445             &xsum0, &xxsum0);
 446
 447         vp8_filter_block2d_bil_var_sse2(
 448             src_ptr + 8, src_pixels_per_line,
 449             dst_ptr + 8, dst_pixels_per_line, 8,
 450             xoffset, yoffset,
 451             &xsum1, &xxsum1);
 452     }
 453
 454     xsum0 += xsum1;
 455     xxsum0 += xxsum1;
 456
 457     *sse = xxsum0;
 458     return (xxsum0 - ((xsum0 * xsum0) >> 7));
 459 }
 460
 461 unsigned int vp8_sub_pixel_variance8x16_wmt
 462 (
 463     const unsigned char  *src_ptr,
 464     int  src_pixels_per_line,
 465     int  xoffset,
 466     int  yoffset,
 467     const unsigned char *dst_ptr,
 468     int dst_pixels_per_line,
 469     unsigned int *sse
 470 )
 471 {
 472     int xsum;
 473     unsigned int xxsum;
 474
 475     if (xoffset == 4 && yoffset == 0)
 476     {
 477         vp8_half_horiz_variance16x_h_sse2(
 478             src_ptr, src_pixels_per_line,
 479             dst_ptr, dst_pixels_per_line, 16,
 480             &xsum, &xxsum);
 481     }
 482     else if (xoffset == 0 && yoffset == 4)
 483     {
 484         vp8_half_vert_variance16x_h_sse2(
 485             src_ptr, src_pixels_per_line,
 486             dst_ptr, dst_pixels_per_line, 16,
 487             &xsum, &xxsum);
 488     }
 489     else if (xoffset == 4 && yoffset == 4)
 490     {
 491         vp8_half_horiz_vert_variance16x_h_sse2(
 492             src_ptr, src_pixels_per_line,
 493             dst_ptr, dst_pixels_per_line, 16,
 494             &xsum, &xxsum);
 495     }
 496     else
 497     {
 498         vp8_filter_block2d_bil_var_sse2(
 499             src_ptr, src_pixels_per_line,
 500             dst_ptr, dst_pixels_per_line, 16,
 501             xoffset, yoffset,
 502             &xsum, &xxsum);
 503     }
 504
 505     *sse = xxsum;
 506     return (xxsum - ((xsum * xsum) >> 7));
 507 }
 508
 509 unsigned int vp8_i_variance16x16_wmt(
 510     const unsigned char *src_ptr,
 511     int  source_stride,
 512     const unsigned char *ref_ptr,
 513     int  recon_stride,
 514     unsigned int *sse)
 515 {
 516     unsigned int sse0, sse1, sse2, sse3, var;
 517     int sum0, sum1, sum2, sum3, avg;
 518
 519
 520     vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
 521     vp8_get8x8var_sse2(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
 522     vp8_get8x8var_sse2(src_ptr + (source_stride >> 1), source_stride, ref_ptr + (recon_stride >> 1), recon_stride, &sse2, &sum2) ;
 523     vp8_get8x8var_sse2(src_ptr + (source_stride >> 1) + 8, source_stride, ref_ptr + (recon_stride >> 1) + 8, recon_stride, &sse3, &sum3);
 524
 525     var = sse0 + sse1 + sse2 + sse3;
 526     avg = sum0 + sum1 + sum2 + sum3;
 527
 528     *sse = var;
 529     return (var - ((avg * avg) >> 8));
 530
 531 }
 532
 533 unsigned int vp8_i_variance8x16_wmt(
 534     const unsigned char *src_ptr,
 535     int  source_stride,
 536     const unsigned char *ref_ptr,
 537     int  recon_stride,
 538     unsigned int *sse)
 539 {
 540     unsigned int sse0, sse1, var;
 541     int sum0, sum1, avg;
 542     vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
 543     vp8_get8x8var_sse2(src_ptr + (source_stride >> 1), source_stride, ref_ptr + (recon_stride >> 1), recon_stride, &sse1, &sum1) ;
 544
 545     var = sse0 + sse1;
 546     avg = sum0 + sum1;
 547
 548     *sse = var;
 549     return (var - ((avg * avg) >> 7));
 550
 551 }
 552
 553
 554 unsigned int vp8_i_sub_pixel_variance16x16_wmt
 555 (
 556     const unsigned char  *src_ptr,
 557     int  src_pixels_per_line,
 558     int  xoffset,
 559     int  yoffset,
 560     const unsigned char *dst_ptr,
 561     int dst_pixels_per_line,
 562     unsigned int *sse
 563 )
 564 {
 565     return vp8_sub_pixel_variance16x16_wmt(src_ptr, (src_pixels_per_line >> 1), xoffset, yoffset, dst_ptr, (dst_pixels_per_line >> 1), sse);
 566 }
 567
 568
 569 unsigned int vp8_i_sub_pixel_variance8x16_wmt
 570 (
 571     const unsigned char  *src_ptr,
 572     int  src_pixels_per_line,
 573     int  xoffset,
 574     int  yoffset,
 575     const unsigned char *dst_ptr,
 576     int dst_pixels_per_line,
 577     unsigned int *sse
 578 )
 579 {
 580
 581     return vp8_sub_pixel_variance8x16_wmt(src_ptr, (src_pixels_per_line >> 1), xoffset, yoffset, dst_ptr, (dst_pixels_per_line >> 1), sse);
 582 }
 583
 584
 585 unsigned int vp8_variance_halfpixvar16x16_h_wmt(
 586     const unsigned char *src_ptr,
 587     int  src_pixels_per_line,
 588     const unsigned char *dst_ptr,
 589     int  dst_pixels_per_line,
 590     unsigned int *sse)
 591 {
 592     int xsum0, xsum1;
 593     unsigned int xxsum0, xxsum1;
 594
 595     vp8_half_horiz_variance16x_h_sse2(
 596         src_ptr, src_pixels_per_line,
 597         dst_ptr, dst_pixels_per_line, 16,
 598         &xsum0, &xxsum0);
 599
 600     vp8_half_horiz_variance16x_h_sse2(
 601         src_ptr + 8, src_pixels_per_line,
 602         dst_ptr + 8, dst_pixels_per_line, 16,
 603         &xsum1, &xxsum1);
 604
 605     xsum0 += xsum1;
 606     xxsum0 += xxsum1;
 607     *sse = xxsum0;
 608     return (xxsum0 - ((xsum0 * xsum0) >> 8));
 609 }
 610
 611
 612 unsigned int vp8_variance_halfpixvar16x16_v_wmt(
 613     const unsigned char *src_ptr,
 614     int  src_pixels_per_line,
 615     const unsigned char *dst_ptr,
 616     int  dst_pixels_per_line,
 617     unsigned int *sse)
 618 {
 619     int xsum0, xsum1;
 620     unsigned int xxsum0, xxsum1;
 621
 622     vp8_half_vert_variance16x_h_sse2(
 623         src_ptr, src_pixels_per_line,
 624         dst_ptr, dst_pixels_per_line, 16,
 625         &xsum0, &xxsum0);
 626
 627     vp8_half_vert_variance16x_h_sse2(
 628         src_ptr + 8, src_pixels_per_line,
 629         dst_ptr + 8, dst_pixels_per_line, 16,
 630         &xsum1, &xxsum1);
 631
 632     xsum0 += xsum1;
 633     xxsum0 += xxsum1;
 634     *sse = xxsum0;
 635     return (xxsum0 - ((xsum0 * xsum0) >> 8));
 636 }
 637
 638
 639 unsigned int vp8_variance_halfpixvar16x16_hv_wmt(
 640     const unsigned char *src_ptr,
 641     int  src_pixels_per_line,
 642     const unsigned char *dst_ptr,
 643     int  dst_pixels_per_line,
 644     unsigned int *sse)
 645 {
 646     int xsum0, xsum1;
 647     unsigned int xxsum0, xxsum1;
 648
 649     vp8_half_horiz_vert_variance16x_h_sse2(
 650         src_ptr, src_pixels_per_line,
 651         dst_ptr, dst_pixels_per_line, 16,
 652         &xsum0, &xxsum0);
 653
 654     vp8_half_horiz_vert_variance16x_h_sse2(
 655         src_ptr + 8, src_pixels_per_line,
 656         dst_ptr + 8, dst_pixels_per_line, 16,
 657         &xsum1, &xxsum1);
 658
 659     xsum0 += xsum1;
 660     xxsum0 += xxsum1;
 661     *sse = xxsum0;
 662     return (xxsum0 - ((xsum0 * xsum0) >> 8));
 663 }