1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/common/x86/variance_sse2.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,558 @@
1.4 +/*
1.5 + * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license
1.8 + * that can be found in the LICENSE file in the root of the source
1.9 + * tree. An additional intellectual property rights grant can be found
1.10 + * in the file PATENTS. All contributing project authors may
1.11 + * be found in the AUTHORS file in the root of the source tree.
1.12 + */
1.13 +
1.14 +#include "vpx_config.h"
1.15 +#include "vp8/common/variance.h"
1.16 +#include "vp8/common/pragmas.h"
1.17 +#include "vpx_ports/mem.h"
1.18 +#include "vp8/common/x86/filter_x86.h"
1.19 +
1.20 +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 *filter);
1.21 +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 *filter);
1.22 +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 *filter);
1.23 +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 *filter);
1.24 +
1.25 +extern void vp8_filter_block2d_bil4x4_var_mmx
1.26 +(
1.27 + const unsigned char *ref_ptr,
1.28 + int ref_pixels_per_line,
1.29 + const unsigned char *src_ptr,
1.30 + int src_pixels_per_line,
1.31 + const short *HFilter,
1.32 + const short *VFilter,
1.33 + int *sum,
1.34 + unsigned int *sumsquared
1.35 +);
1.36 +
1.37 +extern unsigned int vp8_get4x4var_mmx
1.38 +(
1.39 + const unsigned char *src_ptr,
1.40 + int source_stride,
1.41 + const unsigned char *ref_ptr,
1.42 + int recon_stride,
1.43 + unsigned int *SSE,
1.44 + int *Sum
1.45 +);
1.46 +
1.47 +unsigned int vp8_get_mb_ss_sse2
1.48 +(
1.49 + const short *src_ptr
1.50 +);
1.51 +unsigned int vp8_get16x16var_sse2
1.52 +(
1.53 + const unsigned char *src_ptr,
1.54 + int source_stride,
1.55 + const unsigned char *ref_ptr,
1.56 + int recon_stride,
1.57 + unsigned int *SSE,
1.58 + int *Sum
1.59 +);
1.60 +unsigned int vp8_get8x8var_sse2
1.61 +(
1.62 + const unsigned char *src_ptr,
1.63 + int source_stride,
1.64 + const unsigned char *ref_ptr,
1.65 + int recon_stride,
1.66 + unsigned int *SSE,
1.67 + int *Sum
1.68 +);
1.69 +void vp8_filter_block2d_bil_var_sse2
1.70 +(
1.71 + const unsigned char *ref_ptr,
1.72 + int ref_pixels_per_line,
1.73 + const unsigned char *src_ptr,
1.74 + int src_pixels_per_line,
1.75 + unsigned int Height,
1.76 + int xoffset,
1.77 + int yoffset,
1.78 + int *sum,
1.79 + unsigned int *sumsquared
1.80 +);
1.81 +void vp8_half_horiz_vert_variance8x_h_sse2
1.82 +(
1.83 + const unsigned char *ref_ptr,
1.84 + int ref_pixels_per_line,
1.85 + const unsigned char *src_ptr,
1.86 + int src_pixels_per_line,
1.87 + unsigned int Height,
1.88 + int *sum,
1.89 + unsigned int *sumsquared
1.90 +);
1.91 +void vp8_half_horiz_vert_variance16x_h_sse2
1.92 +(
1.93 + const unsigned char *ref_ptr,
1.94 + int ref_pixels_per_line,
1.95 + const unsigned char *src_ptr,
1.96 + int src_pixels_per_line,
1.97 + unsigned int Height,
1.98 + int *sum,
1.99 + unsigned int *sumsquared
1.100 +);
1.101 +void vp8_half_horiz_variance8x_h_sse2
1.102 +(
1.103 + const unsigned char *ref_ptr,
1.104 + int ref_pixels_per_line,
1.105 + const unsigned char *src_ptr,
1.106 + int src_pixels_per_line,
1.107 + unsigned int Height,
1.108 + int *sum,
1.109 + unsigned int *sumsquared
1.110 +);
1.111 +void vp8_half_horiz_variance16x_h_sse2
1.112 +(
1.113 + const unsigned char *ref_ptr,
1.114 + int ref_pixels_per_line,
1.115 + const unsigned char *src_ptr,
1.116 + int src_pixels_per_line,
1.117 + unsigned int Height,
1.118 + int *sum,
1.119 + unsigned int *sumsquared
1.120 +);
1.121 +void vp8_half_vert_variance8x_h_sse2
1.122 +(
1.123 + const unsigned char *ref_ptr,
1.124 + int ref_pixels_per_line,
1.125 + const unsigned char *src_ptr,
1.126 + int src_pixels_per_line,
1.127 + unsigned int Height,
1.128 + int *sum,
1.129 + unsigned int *sumsquared
1.130 +);
1.131 +void vp8_half_vert_variance16x_h_sse2
1.132 +(
1.133 + const unsigned char *ref_ptr,
1.134 + int ref_pixels_per_line,
1.135 + const unsigned char *src_ptr,
1.136 + int src_pixels_per_line,
1.137 + unsigned int Height,
1.138 + int *sum,
1.139 + unsigned int *sumsquared
1.140 +);
1.141 +
1.142 +unsigned int vp8_variance4x4_wmt(
1.143 + const unsigned char *src_ptr,
1.144 + int source_stride,
1.145 + const unsigned char *ref_ptr,
1.146 + int recon_stride,
1.147 + unsigned int *sse)
1.148 +{
1.149 + unsigned int var;
1.150 + int avg;
1.151 +
1.152 + vp8_get4x4var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg) ;
1.153 + *sse = var;
1.154 + return (var - (((unsigned int)avg * avg) >> 4));
1.155 +
1.156 +}
1.157 +
1.158 +unsigned int vp8_variance8x8_wmt
1.159 +(
1.160 + const unsigned char *src_ptr,
1.161 + int source_stride,
1.162 + const unsigned char *ref_ptr,
1.163 + int recon_stride,
1.164 + unsigned int *sse)
1.165 +{
1.166 + unsigned int var;
1.167 + int avg;
1.168 +
1.169 + vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg) ;
1.170 + *sse = var;
1.171 + return (var - (((unsigned int)avg * avg) >> 6));
1.172 +
1.173 +}
1.174 +
1.175 +
1.176 +unsigned int vp8_variance16x16_wmt
1.177 +(
1.178 + const unsigned char *src_ptr,
1.179 + int source_stride,
1.180 + const unsigned char *ref_ptr,
1.181 + int recon_stride,
1.182 + unsigned int *sse)
1.183 +{
1.184 + unsigned int sse0;
1.185 + int sum0;
1.186 +
1.187 +
1.188 + vp8_get16x16var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
1.189 + *sse = sse0;
1.190 + return (sse0 - (((unsigned int)sum0 * sum0) >> 8));
1.191 +}
1.192 +unsigned int vp8_mse16x16_wmt(
1.193 + const unsigned char *src_ptr,
1.194 + int source_stride,
1.195 + const unsigned char *ref_ptr,
1.196 + int recon_stride,
1.197 + unsigned int *sse)
1.198 +{
1.199 +
1.200 + unsigned int sse0;
1.201 + int sum0;
1.202 + vp8_get16x16var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
1.203 + *sse = sse0;
1.204 + return sse0;
1.205 +
1.206 +}
1.207 +
1.208 +
1.209 +unsigned int vp8_variance16x8_wmt
1.210 +(
1.211 + const unsigned char *src_ptr,
1.212 + int source_stride,
1.213 + const unsigned char *ref_ptr,
1.214 + int recon_stride,
1.215 + unsigned int *sse)
1.216 +{
1.217 + unsigned int sse0, sse1, var;
1.218 + int sum0, sum1, avg;
1.219 +
1.220 + vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
1.221 + vp8_get8x8var_sse2(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
1.222 +
1.223 + var = sse0 + sse1;
1.224 + avg = sum0 + sum1;
1.225 + *sse = var;
1.226 + return (var - (((unsigned int)avg * avg) >> 7));
1.227 +
1.228 +}
1.229 +
1.230 +unsigned int vp8_variance8x16_wmt
1.231 +(
1.232 + const unsigned char *src_ptr,
1.233 + int source_stride,
1.234 + const unsigned char *ref_ptr,
1.235 + int recon_stride,
1.236 + unsigned int *sse)
1.237 +{
1.238 + unsigned int sse0, sse1, var;
1.239 + int sum0, sum1, avg;
1.240 +
1.241 + vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
1.242 + vp8_get8x8var_sse2(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse1, &sum1) ;
1.243 +
1.244 + var = sse0 + sse1;
1.245 + avg = sum0 + sum1;
1.246 + *sse = var;
1.247 + return (var - (((unsigned int)avg * avg) >> 7));
1.248 +
1.249 +}
1.250 +
1.251 +unsigned int vp8_sub_pixel_variance4x4_wmt
1.252 +(
1.253 + const unsigned char *src_ptr,
1.254 + int src_pixels_per_line,
1.255 + int xoffset,
1.256 + int yoffset,
1.257 + const unsigned char *dst_ptr,
1.258 + int dst_pixels_per_line,
1.259 + unsigned int *sse
1.260 +)
1.261 +{
1.262 + int xsum;
1.263 + unsigned int xxsum;
1.264 + vp8_filter_block2d_bil4x4_var_mmx(
1.265 + src_ptr, src_pixels_per_line,
1.266 + dst_ptr, dst_pixels_per_line,
1.267 + vp8_bilinear_filters_x86_4[xoffset], vp8_bilinear_filters_x86_4[yoffset],
1.268 + &xsum, &xxsum
1.269 + );
1.270 + *sse = xxsum;
1.271 + return (xxsum - (((unsigned int)xsum * xsum) >> 4));
1.272 +}
1.273 +
1.274 +
1.275 +unsigned int vp8_sub_pixel_variance8x8_wmt
1.276 +(
1.277 + const unsigned char *src_ptr,
1.278 + int src_pixels_per_line,
1.279 + int xoffset,
1.280 + int yoffset,
1.281 + const unsigned char *dst_ptr,
1.282 + int dst_pixels_per_line,
1.283 + unsigned int *sse
1.284 +)
1.285 +{
1.286 + int xsum;
1.287 + unsigned int xxsum;
1.288 +
1.289 + if (xoffset == 4 && yoffset == 0)
1.290 + {
1.291 + vp8_half_horiz_variance8x_h_sse2(
1.292 + src_ptr, src_pixels_per_line,
1.293 + dst_ptr, dst_pixels_per_line, 8,
1.294 + &xsum, &xxsum);
1.295 + }
1.296 + else if (xoffset == 0 && yoffset == 4)
1.297 + {
1.298 + vp8_half_vert_variance8x_h_sse2(
1.299 + src_ptr, src_pixels_per_line,
1.300 + dst_ptr, dst_pixels_per_line, 8,
1.301 + &xsum, &xxsum);
1.302 + }
1.303 + else if (xoffset == 4 && yoffset == 4)
1.304 + {
1.305 + vp8_half_horiz_vert_variance8x_h_sse2(
1.306 + src_ptr, src_pixels_per_line,
1.307 + dst_ptr, dst_pixels_per_line, 8,
1.308 + &xsum, &xxsum);
1.309 + }
1.310 + else
1.311 + {
1.312 + vp8_filter_block2d_bil_var_sse2(
1.313 + src_ptr, src_pixels_per_line,
1.314 + dst_ptr, dst_pixels_per_line, 8,
1.315 + xoffset, yoffset,
1.316 + &xsum, &xxsum);
1.317 + }
1.318 +
1.319 + *sse = xxsum;
1.320 + return (xxsum - (((unsigned int)xsum * xsum) >> 6));
1.321 +}
1.322 +
1.323 +unsigned int vp8_sub_pixel_variance16x16_wmt
1.324 +(
1.325 + const unsigned char *src_ptr,
1.326 + int src_pixels_per_line,
1.327 + int xoffset,
1.328 + int yoffset,
1.329 + const unsigned char *dst_ptr,
1.330 + int dst_pixels_per_line,
1.331 + unsigned int *sse
1.332 +)
1.333 +{
1.334 + int xsum0, xsum1;
1.335 + unsigned int xxsum0, xxsum1;
1.336 +
1.337 +
1.338 + /* note we could avoid these if statements if the calling function
1.339 + * just called the appropriate functions inside.
1.340 + */
1.341 + if (xoffset == 4 && yoffset == 0)
1.342 + {
1.343 + vp8_half_horiz_variance16x_h_sse2(
1.344 + src_ptr, src_pixels_per_line,
1.345 + dst_ptr, dst_pixels_per_line, 16,
1.346 + &xsum0, &xxsum0);
1.347 + }
1.348 + else if (xoffset == 0 && yoffset == 4)
1.349 + {
1.350 + vp8_half_vert_variance16x_h_sse2(
1.351 + src_ptr, src_pixels_per_line,
1.352 + dst_ptr, dst_pixels_per_line, 16,
1.353 + &xsum0, &xxsum0);
1.354 + }
1.355 + else if (xoffset == 4 && yoffset == 4)
1.356 + {
1.357 + vp8_half_horiz_vert_variance16x_h_sse2(
1.358 + src_ptr, src_pixels_per_line,
1.359 + dst_ptr, dst_pixels_per_line, 16,
1.360 + &xsum0, &xxsum0);
1.361 + }
1.362 + else
1.363 + {
1.364 + vp8_filter_block2d_bil_var_sse2(
1.365 + src_ptr, src_pixels_per_line,
1.366 + dst_ptr, dst_pixels_per_line, 16,
1.367 + xoffset, yoffset,
1.368 + &xsum0, &xxsum0
1.369 + );
1.370 +
1.371 + vp8_filter_block2d_bil_var_sse2(
1.372 + src_ptr + 8, src_pixels_per_line,
1.373 + dst_ptr + 8, dst_pixels_per_line, 16,
1.374 + xoffset, yoffset,
1.375 + &xsum1, &xxsum1
1.376 + );
1.377 + xsum0 += xsum1;
1.378 + xxsum0 += xxsum1;
1.379 + }
1.380 +
1.381 + *sse = xxsum0;
1.382 + return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
1.383 +}
1.384 +
1.385 +unsigned int vp8_sub_pixel_mse16x16_wmt(
1.386 + const unsigned char *src_ptr,
1.387 + int src_pixels_per_line,
1.388 + int xoffset,
1.389 + int yoffset,
1.390 + const unsigned char *dst_ptr,
1.391 + int dst_pixels_per_line,
1.392 + unsigned int *sse
1.393 +)
1.394 +{
1.395 + vp8_sub_pixel_variance16x16_wmt(src_ptr, src_pixels_per_line, xoffset, yoffset, dst_ptr, dst_pixels_per_line, sse);
1.396 + return *sse;
1.397 +}
1.398 +
1.399 +unsigned int vp8_sub_pixel_variance16x8_wmt
1.400 +(
1.401 + const unsigned char *src_ptr,
1.402 + int src_pixels_per_line,
1.403 + int xoffset,
1.404 + int yoffset,
1.405 + const unsigned char *dst_ptr,
1.406 + int dst_pixels_per_line,
1.407 + unsigned int *sse
1.408 +
1.409 +)
1.410 +{
1.411 + int xsum0, xsum1;
1.412 + unsigned int xxsum0, xxsum1;
1.413 +
1.414 + if (xoffset == 4 && yoffset == 0)
1.415 + {
1.416 + vp8_half_horiz_variance16x_h_sse2(
1.417 + src_ptr, src_pixels_per_line,
1.418 + dst_ptr, dst_pixels_per_line, 8,
1.419 + &xsum0, &xxsum0);
1.420 + }
1.421 + else if (xoffset == 0 && yoffset == 4)
1.422 + {
1.423 + vp8_half_vert_variance16x_h_sse2(
1.424 + src_ptr, src_pixels_per_line,
1.425 + dst_ptr, dst_pixels_per_line, 8,
1.426 + &xsum0, &xxsum0);
1.427 + }
1.428 + else if (xoffset == 4 && yoffset == 4)
1.429 + {
1.430 + vp8_half_horiz_vert_variance16x_h_sse2(
1.431 + src_ptr, src_pixels_per_line,
1.432 + dst_ptr, dst_pixels_per_line, 8,
1.433 + &xsum0, &xxsum0);
1.434 + }
1.435 + else
1.436 + {
1.437 + vp8_filter_block2d_bil_var_sse2(
1.438 + src_ptr, src_pixels_per_line,
1.439 + dst_ptr, dst_pixels_per_line, 8,
1.440 + xoffset, yoffset,
1.441 + &xsum0, &xxsum0);
1.442 +
1.443 + vp8_filter_block2d_bil_var_sse2(
1.444 + src_ptr + 8, src_pixels_per_line,
1.445 + dst_ptr + 8, dst_pixels_per_line, 8,
1.446 + xoffset, yoffset,
1.447 + &xsum1, &xxsum1);
1.448 + xsum0 += xsum1;
1.449 + xxsum0 += xxsum1;
1.450 + }
1.451 +
1.452 + *sse = xxsum0;
1.453 + return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 7));
1.454 +}
1.455 +
1.456 +unsigned int vp8_sub_pixel_variance8x16_wmt
1.457 +(
1.458 + const unsigned char *src_ptr,
1.459 + int src_pixels_per_line,
1.460 + int xoffset,
1.461 + int yoffset,
1.462 + const unsigned char *dst_ptr,
1.463 + int dst_pixels_per_line,
1.464 + unsigned int *sse
1.465 +)
1.466 +{
1.467 + int xsum;
1.468 + unsigned int xxsum;
1.469 +
1.470 + if (xoffset == 4 && yoffset == 0)
1.471 + {
1.472 + vp8_half_horiz_variance8x_h_sse2(
1.473 + src_ptr, src_pixels_per_line,
1.474 + dst_ptr, dst_pixels_per_line, 16,
1.475 + &xsum, &xxsum);
1.476 + }
1.477 + else if (xoffset == 0 && yoffset == 4)
1.478 + {
1.479 + vp8_half_vert_variance8x_h_sse2(
1.480 + src_ptr, src_pixels_per_line,
1.481 + dst_ptr, dst_pixels_per_line, 16,
1.482 + &xsum, &xxsum);
1.483 + }
1.484 + else if (xoffset == 4 && yoffset == 4)
1.485 + {
1.486 + vp8_half_horiz_vert_variance8x_h_sse2(
1.487 + src_ptr, src_pixels_per_line,
1.488 + dst_ptr, dst_pixels_per_line, 16,
1.489 + &xsum, &xxsum);
1.490 + }
1.491 + else
1.492 + {
1.493 + vp8_filter_block2d_bil_var_sse2(
1.494 + src_ptr, src_pixels_per_line,
1.495 + dst_ptr, dst_pixels_per_line, 16,
1.496 + xoffset, yoffset,
1.497 + &xsum, &xxsum);
1.498 + }
1.499 +
1.500 + *sse = xxsum;
1.501 + return (xxsum - (((unsigned int)xsum * xsum) >> 7));
1.502 +}
1.503 +
1.504 +
1.505 +unsigned int vp8_variance_halfpixvar16x16_h_wmt(
1.506 + const unsigned char *src_ptr,
1.507 + int src_pixels_per_line,
1.508 + const unsigned char *dst_ptr,
1.509 + int dst_pixels_per_line,
1.510 + unsigned int *sse)
1.511 +{
1.512 + int xsum0;
1.513 + unsigned int xxsum0;
1.514 +
1.515 + vp8_half_horiz_variance16x_h_sse2(
1.516 + src_ptr, src_pixels_per_line,
1.517 + dst_ptr, dst_pixels_per_line, 16,
1.518 + &xsum0, &xxsum0);
1.519 +
1.520 + *sse = xxsum0;
1.521 + return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
1.522 +}
1.523 +
1.524 +
1.525 +unsigned int vp8_variance_halfpixvar16x16_v_wmt(
1.526 + const unsigned char *src_ptr,
1.527 + int src_pixels_per_line,
1.528 + const unsigned char *dst_ptr,
1.529 + int dst_pixels_per_line,
1.530 + unsigned int *sse)
1.531 +{
1.532 + int xsum0;
1.533 + unsigned int xxsum0;
1.534 + vp8_half_vert_variance16x_h_sse2(
1.535 + src_ptr, src_pixels_per_line,
1.536 + dst_ptr, dst_pixels_per_line, 16,
1.537 + &xsum0, &xxsum0);
1.538 +
1.539 + *sse = xxsum0;
1.540 + return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
1.541 +}
1.542 +
1.543 +
1.544 +unsigned int vp8_variance_halfpixvar16x16_hv_wmt(
1.545 + const unsigned char *src_ptr,
1.546 + int src_pixels_per_line,
1.547 + const unsigned char *dst_ptr,
1.548 + int dst_pixels_per_line,
1.549 + unsigned int *sse)
1.550 +{
1.551 + int xsum0;
1.552 + unsigned int xxsum0;
1.553 +
1.554 + vp8_half_horiz_vert_variance16x_h_sse2(
1.555 + src_ptr, src_pixels_per_line,
1.556 + dst_ptr, dst_pixels_per_line, 16,
1.557 + &xsum0, &xxsum0);
1.558 +
1.559 + *sse = xxsum0;
1.560 + return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
1.561 +}
