variance_sse2.c 10.7 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
2
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
John Koleszar's avatar
John Koleszar committed
3
 *
4
 *  Use of this source code is governed by a BSD-style license
5 6
 *  that can be found in the LICENSE file in the root of the source
 *  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.
John Koleszar's avatar
John Koleszar committed
9 10
 */

11
#include "./vp8_rtcd.h"
12
#include "vpx_config.h"
13
#include "vp8/common/variance.h"
John Koleszar's avatar
John Koleszar committed
14
#include "vpx_ports/mem.h"
Johann's avatar
Johann committed
15
#include "vp8/common/x86/filter_x86.h"
John Koleszar's avatar
John Koleszar committed
16

Johann's avatar
Johann committed
17 18 19 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);
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);
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);
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);
John Koleszar's avatar
John Koleszar committed
21 22 23

extern void vp8_filter_block2d_bil4x4_var_mmx
(
24
    const unsigned char *ref_ptr,
John Koleszar's avatar
John Koleszar committed
25
    int ref_pixels_per_line,
26
    const unsigned char *src_ptr,
John Koleszar's avatar
John Koleszar committed
27 28 29 30 31 32 33 34 35
    int src_pixels_per_line,
    const short *HFilter,
    const short *VFilter,
    int *sum,
    unsigned int *sumsquared
);

void vp8_filter_block2d_bil_var_sse2
(
36
    const unsigned char *ref_ptr,
John Koleszar's avatar
John Koleszar committed
37
    int ref_pixels_per_line,
38
    const unsigned char *src_ptr,
John Koleszar's avatar
John Koleszar committed
39 40
    int src_pixels_per_line,
    unsigned int Height,
41 42
    int  xoffset,
    int  yoffset,
John Koleszar's avatar
John Koleszar committed
43 44 45
    int *sum,
    unsigned int *sumsquared
);
46 47 48 49 50 51 52 53 54 55
void vp8_half_horiz_vert_variance8x_h_sse2
(
    const unsigned char *ref_ptr,
    int ref_pixels_per_line,
    const unsigned char *src_ptr,
    int src_pixels_per_line,
    unsigned int Height,
    int *sum,
    unsigned int *sumsquared
);
John Koleszar's avatar
John Koleszar committed
56 57
void vp8_half_horiz_vert_variance16x_h_sse2
(
58
    const unsigned char *ref_ptr,
John Koleszar's avatar
John Koleszar committed
59
    int ref_pixels_per_line,
60
    const unsigned char *src_ptr,
John Koleszar's avatar
John Koleszar committed
61 62 63 64 65
    int src_pixels_per_line,
    unsigned int Height,
    int *sum,
    unsigned int *sumsquared
);
66 67 68 69 70 71 72 73 74 75
void vp8_half_horiz_variance8x_h_sse2
(
    const unsigned char *ref_ptr,
    int ref_pixels_per_line,
    const unsigned char *src_ptr,
    int src_pixels_per_line,
    unsigned int Height,
    int *sum,
    unsigned int *sumsquared
);
John Koleszar's avatar
John Koleszar committed
76 77
void vp8_half_horiz_variance16x_h_sse2
(
78
    const unsigned char *ref_ptr,
John Koleszar's avatar
John Koleszar committed
79
    int ref_pixels_per_line,
80
    const unsigned char *src_ptr,
John Koleszar's avatar
John Koleszar committed
81 82 83 84 85
    int src_pixels_per_line,
    unsigned int Height,
    int *sum,
    unsigned int *sumsquared
);
86 87 88 89 90 91 92 93 94 95
void vp8_half_vert_variance8x_h_sse2
(
    const unsigned char *ref_ptr,
    int ref_pixels_per_line,
    const unsigned char *src_ptr,
    int src_pixels_per_line,
    unsigned int Height,
    int *sum,
    unsigned int *sumsquared
);
John Koleszar's avatar
John Koleszar committed
96 97
void vp8_half_vert_variance16x_h_sse2
(
98
    const unsigned char *ref_ptr,
John Koleszar's avatar
John Koleszar committed
99
    int ref_pixels_per_line,
100
    const unsigned char *src_ptr,
John Koleszar's avatar
John Koleszar committed
101 102 103 104 105 106 107 108
    int src_pixels_per_line,
    unsigned int Height,
    int *sum,
    unsigned int *sumsquared
);

unsigned int vp8_sub_pixel_variance4x4_wmt
(
109
    const unsigned char  *src_ptr,
John Koleszar's avatar
John Koleszar committed
110 111 112
    int  src_pixels_per_line,
    int  xoffset,
    int  yoffset,
113
    const unsigned char *dst_ptr,
John Koleszar's avatar
John Koleszar committed
114 115 116 117 118 119 120 121 122
    int dst_pixels_per_line,
    unsigned int *sse
)
{
    int xsum;
    unsigned int xxsum;
    vp8_filter_block2d_bil4x4_var_mmx(
        src_ptr, src_pixels_per_line,
        dst_ptr, dst_pixels_per_line,
Johann's avatar
Johann committed
123
        vp8_bilinear_filters_x86_4[xoffset], vp8_bilinear_filters_x86_4[yoffset],
John Koleszar's avatar
John Koleszar committed
124 125 126
        &xsum, &xxsum
    );
    *sse = xxsum;
127
    return (xxsum - (((unsigned int)xsum * xsum) >> 4));
John Koleszar's avatar
John Koleszar committed
128 129 130 131 132
}


unsigned int vp8_sub_pixel_variance8x8_wmt
(
133
    const unsigned char  *src_ptr,
John Koleszar's avatar
John Koleszar committed
134 135 136
    int  src_pixels_per_line,
    int  xoffset,
    int  yoffset,
137
    const unsigned char *dst_ptr,
John Koleszar's avatar
John Koleszar committed
138 139 140 141 142 143
    int dst_pixels_per_line,
    unsigned int *sse
)
{
    int xsum;
    unsigned int xxsum;
144 145 146

    if (xoffset == 4 && yoffset == 0)
    {
147
        vp8_half_horiz_variance8x_h_sse2(
148 149 150 151 152 153
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 8,
            &xsum, &xxsum);
    }
    else if (xoffset == 0 && yoffset == 4)
    {
154
        vp8_half_vert_variance8x_h_sse2(
155 156 157 158 159 160
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 8,
            &xsum, &xxsum);
    }
    else if (xoffset == 4 && yoffset == 4)
    {
161
        vp8_half_horiz_vert_variance8x_h_sse2(
162 163 164 165 166 167 168 169 170 171 172 173
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 8,
            &xsum, &xxsum);
    }
    else
    {
        vp8_filter_block2d_bil_var_sse2(
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 8,
            xoffset, yoffset,
            &xsum, &xxsum);
    }
John Koleszar's avatar
John Koleszar committed
174 175

    *sse = xxsum;
176
    return (xxsum - (((unsigned int)xsum * xsum) >> 6));
John Koleszar's avatar
John Koleszar committed
177 178 179 180
}

unsigned int vp8_sub_pixel_variance16x16_wmt
(
181
    const unsigned char  *src_ptr,
John Koleszar's avatar
John Koleszar committed
182 183 184
    int  src_pixels_per_line,
    int  xoffset,
    int  yoffset,
185
    const unsigned char *dst_ptr,
John Koleszar's avatar
John Koleszar committed
186 187 188 189 190 191 192 193
    int dst_pixels_per_line,
    unsigned int *sse
)
{
    int xsum0, xsum1;
    unsigned int xxsum0, xxsum1;


John Koleszar's avatar
John Koleszar committed
194 195 196
    /* note we could avoid these if statements if the calling function
     * just called the appropriate functions inside.
     */
John Koleszar's avatar
John Koleszar committed
197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222
    if (xoffset == 4 && yoffset == 0)
    {
        vp8_half_horiz_variance16x_h_sse2(
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 16,
            &xsum0, &xxsum0);
    }
    else if (xoffset == 0 && yoffset == 4)
    {
        vp8_half_vert_variance16x_h_sse2(
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 16,
            &xsum0, &xxsum0);
    }
    else if (xoffset == 4 && yoffset == 4)
    {
        vp8_half_horiz_vert_variance16x_h_sse2(
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 16,
            &xsum0, &xxsum0);
    }
    else
    {
        vp8_filter_block2d_bil_var_sse2(
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 16,
223
            xoffset, yoffset,
John Koleszar's avatar
John Koleszar committed
224 225 226 227 228 229
            &xsum0, &xxsum0
        );

        vp8_filter_block2d_bil_var_sse2(
            src_ptr + 8, src_pixels_per_line,
            dst_ptr + 8, dst_pixels_per_line, 16,
230
            xoffset, yoffset,
John Koleszar's avatar
John Koleszar committed
231 232
            &xsum1, &xxsum1
        );
233 234
        xsum0 += xsum1;
        xxsum0 += xxsum1;
John Koleszar's avatar
John Koleszar committed
235 236 237
    }

    *sse = xxsum0;
238
    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
John Koleszar's avatar
John Koleszar committed
239 240 241 242
}

unsigned int vp8_sub_pixel_variance16x8_wmt
(
243
    const unsigned char  *src_ptr,
John Koleszar's avatar
John Koleszar committed
244 245 246
    int  src_pixels_per_line,
    int  xoffset,
    int  yoffset,
247
    const unsigned char *dst_ptr,
John Koleszar's avatar
John Koleszar committed
248 249 250 251 252 253 254 255
    int dst_pixels_per_line,
    unsigned int *sse

)
{
    int xsum0, xsum1;
    unsigned int xxsum0, xxsum1;

256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
    if (xoffset == 4 && yoffset == 0)
    {
        vp8_half_horiz_variance16x_h_sse2(
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 8,
            &xsum0, &xxsum0);
    }
    else if (xoffset == 0 && yoffset == 4)
    {
        vp8_half_vert_variance16x_h_sse2(
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 8,
            &xsum0, &xxsum0);
    }
    else if (xoffset == 4 && yoffset == 4)
    {
        vp8_half_horiz_vert_variance16x_h_sse2(
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 8,
            &xsum0, &xxsum0);
    }
    else
    {
        vp8_filter_block2d_bil_var_sse2(
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 8,
            xoffset, yoffset,
            &xsum0, &xxsum0);
John Koleszar's avatar
John Koleszar committed
284

285 286 287 288 289
        vp8_filter_block2d_bil_var_sse2(
            src_ptr + 8, src_pixels_per_line,
            dst_ptr + 8, dst_pixels_per_line, 8,
            xoffset, yoffset,
            &xsum1, &xxsum1);
290 291
        xsum0 += xsum1;
        xxsum0 += xxsum1;
292
    }
John Koleszar's avatar
John Koleszar committed
293 294

    *sse = xxsum0;
295
    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 7));
John Koleszar's avatar
John Koleszar committed
296 297 298 299
}

unsigned int vp8_sub_pixel_variance8x16_wmt
(
300
    const unsigned char  *src_ptr,
John Koleszar's avatar
John Koleszar committed
301 302 303
    int  src_pixels_per_line,
    int  xoffset,
    int  yoffset,
304
    const unsigned char *dst_ptr,
John Koleszar's avatar
John Koleszar committed
305 306 307 308 309 310
    int dst_pixels_per_line,
    unsigned int *sse
)
{
    int xsum;
    unsigned int xxsum;
311 312 313

    if (xoffset == 4 && yoffset == 0)
    {
314
        vp8_half_horiz_variance8x_h_sse2(
315 316 317 318 319 320
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 16,
            &xsum, &xxsum);
    }
    else if (xoffset == 0 && yoffset == 4)
    {
321
        vp8_half_vert_variance8x_h_sse2(
322 323 324 325 326 327
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 16,
            &xsum, &xxsum);
    }
    else if (xoffset == 4 && yoffset == 4)
    {
328
        vp8_half_horiz_vert_variance8x_h_sse2(
329 330 331 332 333 334 335 336 337 338 339 340
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 16,
            &xsum, &xxsum);
    }
    else
    {
        vp8_filter_block2d_bil_var_sse2(
            src_ptr, src_pixels_per_line,
            dst_ptr, dst_pixels_per_line, 16,
            xoffset, yoffset,
            &xsum, &xxsum);
    }
John Koleszar's avatar
John Koleszar committed
341 342

    *sse = xxsum;
343
    return (xxsum - (((unsigned int)xsum * xsum) >> 7));
John Koleszar's avatar
John Koleszar committed
344 345
}

346 347 348 349 350 351 352 353

unsigned int vp8_variance_halfpixvar16x16_h_wmt(
    const unsigned char *src_ptr,
    int  src_pixels_per_line,
    const unsigned char *dst_ptr,
    int  dst_pixels_per_line,
    unsigned int *sse)
{
354 355
    int xsum0;
    unsigned int xxsum0;
356 357 358 359 360 361 362

    vp8_half_horiz_variance16x_h_sse2(
        src_ptr, src_pixels_per_line,
        dst_ptr, dst_pixels_per_line, 16,
        &xsum0, &xxsum0);

    *sse = xxsum0;
363
    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
364 365 366 367 368 369 370 371 372 373
}


unsigned int vp8_variance_halfpixvar16x16_v_wmt(
    const unsigned char *src_ptr,
    int  src_pixels_per_line,
    const unsigned char *dst_ptr,
    int  dst_pixels_per_line,
    unsigned int *sse)
{
374 375
    int xsum0;
    unsigned int xxsum0;
376 377 378 379 380 381
    vp8_half_vert_variance16x_h_sse2(
        src_ptr, src_pixels_per_line,
        dst_ptr, dst_pixels_per_line, 16,
        &xsum0, &xxsum0);

    *sse = xxsum0;
382
    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
383 384 385 386 387 388 389 390 391 392
}


unsigned int vp8_variance_halfpixvar16x16_hv_wmt(
    const unsigned char *src_ptr,
    int  src_pixels_per_line,
    const unsigned char *dst_ptr,
    int  dst_pixels_per_line,
    unsigned int *sse)
{
393 394
    int xsum0;
    unsigned int xxsum0;
395 396 397 398 399 400 401

    vp8_half_horiz_vert_variance16x_h_sse2(
        src_ptr, src_pixels_per_line,
        dst_ptr, dst_pixels_per_line, 16,
        &xsum0, &xxsum0);

    *sse = xxsum0;
402
    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
403
}