picklpf.c 14.7 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
 */


John Koleszar's avatar
John Koleszar committed
12
#include "vp8/common/onyxc_int.h"
John Koleszar's avatar
John Koleszar committed
13
14
15
16
17
#include "onyx_int.h"
#include "quantize.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx_scale/yv12extend.h"
#include "vpx_scale/vpxscale.h"
John Koleszar's avatar
John Koleszar committed
18
#include "vp8/common/alloccommon.h"
19
20
21
#if ARCH_ARM
#include "vpx_ports/arm.h"
#endif
John Koleszar's avatar
John Koleszar committed
22
23
24
25
26
27
28
29
30
31
32
33
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
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143

extern void vp8_loop_filter_frame(VP8_COMMON *cm,    MACROBLOCKD *mbd,  int filt_val);
extern void vp8_loop_filter_frame_yonly(VP8_COMMON *cm,    MACROBLOCKD *mbd,  int filt_val, int sharpness_lvl);
extern int vp8_calc_ss_err(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *dest, const vp8_variance_rtcd_vtable_t *rtcd);
#if HAVE_ARMV7
extern void vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc);
#endif

#if CONFIG_RUNTIME_CPU_DETECT
#define IF_RTCD(x) (x)
#else
#define IF_RTCD(x) NULL
#endif

extern void
(*vp8_yv12_copy_partial_frame_ptr)(YV12_BUFFER_CONFIG *src_ybc,
                                   YV12_BUFFER_CONFIG *dst_ybc,
                                   int Fraction);
void
vp8_yv12_copy_partial_frame(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc, int Fraction)
{
    unsigned char *src_y, *dst_y;
    int yheight;
    int ystride;
    int border;
    int yoffset;
    int linestocopy;

    border   = src_ybc->border;
    yheight  = src_ybc->y_height;
    ystride  = src_ybc->y_stride;

    linestocopy = (yheight >> (Fraction + 4));

    if (linestocopy < 1)
        linestocopy = 1;

    linestocopy <<= 4;

    yoffset  = ystride * ((yheight >> 5) * 16 - 8);
    src_y = src_ybc->y_buffer + yoffset;
    dst_y = dst_ybc->y_buffer + yoffset;

    vpx_memcpy(dst_y, src_y, ystride *(linestocopy + 16));
}

static int vp8_calc_partial_ssl_err(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *dest, int Fraction, const vp8_variance_rtcd_vtable_t *rtcd)
{
    int i, j;
    int Total = 0;
    int srcoffset, dstoffset;
    unsigned char *src = source->y_buffer;
    unsigned char *dst = dest->y_buffer;

    int linestocopy = (source->y_height >> (Fraction + 4));
    (void)rtcd;

    if (linestocopy < 1)
        linestocopy = 1;

    linestocopy <<= 4;


    srcoffset = source->y_stride   * (dest->y_height >> 5) * 16;
    dstoffset = dest->y_stride     * (dest->y_height >> 5) * 16;

    src += srcoffset;
    dst += dstoffset;

    // Loop through the Y plane raw and reconstruction data summing (square differences)
    for (i = 0; i < linestocopy; i += 16)
    {
        for (j = 0; j < source->y_width; j += 16)
        {
            unsigned int sse;
            Total += VARIANCE_INVOKE(rtcd, mse16x16)(src + j, source->y_stride, dst + j, dest->y_stride, &sse);
        }

        src += 16 * source->y_stride;
        dst += 16 * dest->y_stride;
    }

    return Total;
}

extern void vp8_loop_filter_partial_frame
(
    VP8_COMMON *cm,
    MACROBLOCKD *mbd,
    int default_filt_lvl,
    int sharpness_lvl,
    int Fraction
);

// Enforce a minimum filter level based upon baseline Q
static int get_min_filter_level(VP8_COMP *cpi, int base_qindex)
{
    int min_filter_level;

    if (cpi->source_alt_ref_active && cpi->common.refresh_golden_frame && !cpi->common.refresh_alt_ref_frame)
        min_filter_level = 0;
    else
    {
        if (base_qindex <= 6)
            min_filter_level = 0;
        else if (base_qindex <= 16)
            min_filter_level = 1;
        else
            min_filter_level = (base_qindex / 8);
    }

    return min_filter_level;
}

// Enforce a maximum filter level based upon baseline Q
static int get_max_filter_level(VP8_COMP *cpi, int base_qindex)
{
    // PGW August 2006: Highest filter values almost always a bad idea

    // jbb chg: 20100118 - not so any more with this overquant stuff allow high values
    // with lots of intra coming in.
    int max_filter_level = MAX_LOOP_FILTER ;//* 3 / 4;
Johann's avatar
Johann committed
144
    (void)base_qindex;
John Koleszar's avatar
John Koleszar committed
145

146
    if (cpi->twopass.section_intra_rating > 8)
John Koleszar's avatar
John Koleszar committed
147
148
149
150
151
152
153
154
155
156
157
        max_filter_level = MAX_LOOP_FILTER * 3 / 4;

    return max_filter_level;
}

void vp8cx_pick_filter_level_fast(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi)
{
    VP8_COMMON *cm = &cpi->common;

    int best_err = 0;
    int filt_err = 0;
Johann's avatar
Johann committed
158
159
    int min_filter_level = get_min_filter_level(cpi, cm->base_qindex);
    int max_filter_level = get_max_filter_level(cpi, cm->base_qindex);
John Koleszar's avatar
John Koleszar committed
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
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
223
224
225
226
227
228
229
230
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
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
284
285
286
287
288
289
290
    int filt_val;
    int best_filt_val = cm->filter_level;

    //  Make a copy of the unfiltered / processed recon buffer
    //vp8_yv12_copy_frame_ptr( cm->frame_to_show, &cpi->last_frame_uf  );
    vp8_yv12_copy_partial_frame_ptr(cm->frame_to_show, &cpi->last_frame_uf, 3);

    if (cm->frame_type == KEY_FRAME)
        cm->sharpness_level = 0;
    else
        cm->sharpness_level = cpi->oxcf.Sharpness;

    // Start the search at the previous frame filter level unless it is now out of range.
    if (cm->filter_level < min_filter_level)
        cm->filter_level = min_filter_level;
    else if (cm->filter_level > max_filter_level)
        cm->filter_level = max_filter_level;

    filt_val = cm->filter_level;
    best_filt_val = filt_val;

    // Set up alternate filter values

    // Get the err using the previous frame's filter value.
    vp8_loop_filter_partial_frame(cm, &cpi->mb.e_mbd, filt_val, 0  , 3);
    cm->last_frame_type = cm->frame_type;
    cm->last_filter_type = cm->filter_type;
    cm->last_sharpness_level = cm->sharpness_level;

    best_err = vp8_calc_partial_ssl_err(sd, cm->frame_to_show, 3, IF_RTCD(&cpi->rtcd.variance));

    //  Re-instate the unfiltered frame
    vp8_yv12_copy_partial_frame_ptr(&cpi->last_frame_uf, cm->frame_to_show, 3);

    filt_val -= (1 + ((filt_val > 10) ? 1 : 0));

    // Search lower filter levels
    while (filt_val >= min_filter_level)
    {
        // Apply the loop filter
        vp8_loop_filter_partial_frame(cm, &cpi->mb.e_mbd, filt_val, 0, 3);
        cm->last_frame_type = cm->frame_type;
        cm->last_filter_type = cm->filter_type;
        cm->last_sharpness_level = cm->sharpness_level;

        // Get the err for filtered frame
        filt_err = vp8_calc_partial_ssl_err(sd, cm->frame_to_show, 3, IF_RTCD(&cpi->rtcd.variance));


        //  Re-instate the unfiltered frame
        vp8_yv12_copy_partial_frame_ptr(&cpi->last_frame_uf, cm->frame_to_show, 3);


        // Update the best case record or exit loop.
        if (filt_err < best_err)
        {
            best_err = filt_err;
            best_filt_val = filt_val;
        }
        else
            break;

        // Adjust filter level
        filt_val -= (1 + ((filt_val > 10) ? 1 : 0));
    }

    // Search up (note that we have already done filt_val = cm->filter_level)
    filt_val = cm->filter_level + (1 + ((filt_val > 10) ? 1 : 0));

    if (best_filt_val == cm->filter_level)
    {
        // Resist raising filter level for very small gains
        best_err -= (best_err >> 10);

        while (filt_val < max_filter_level)
        {
            // Apply the loop filter
            vp8_loop_filter_partial_frame(cm, &cpi->mb.e_mbd, filt_val, 0, 3);
            cm->last_frame_type = cm->frame_type;
            cm->last_filter_type = cm->filter_type;
            cm->last_sharpness_level = cm->sharpness_level;

            // Get the err for filtered frame
            filt_err = vp8_calc_partial_ssl_err(sd, cm->frame_to_show, 3, IF_RTCD(&cpi->rtcd.variance));

            //  Re-instate the unfiltered frame
            vp8_yv12_copy_partial_frame_ptr(&cpi->last_frame_uf, cm->frame_to_show, 3);

            // Update the best case record or exit loop.
            if (filt_err < best_err)
            {
                // Do not raise filter level if improvement is < 1 part in 4096
                best_err = filt_err - (filt_err >> 10);

                best_filt_val = filt_val;
            }
            else
                break;

            // Adjust filter level
            filt_val += (1 + ((filt_val > 10) ? 1 : 0));
        }
    }

    cm->filter_level = best_filt_val;

    if (cm->filter_level < min_filter_level)
        cm->filter_level = min_filter_level;

    if (cm->filter_level > max_filter_level)
        cm->filter_level = max_filter_level;
}

// Stub function for now Alt LF not used
void vp8cx_set_alt_lf_level(VP8_COMP *cpi, int filt_val)
{
    MACROBLOCKD *mbd = &cpi->mb.e_mbd;
    (void) filt_val;

    mbd->segment_feature_data[MB_LVL_ALT_LF][0] = cpi->segment_feature_data[MB_LVL_ALT_LF][0];
    mbd->segment_feature_data[MB_LVL_ALT_LF][1] = cpi->segment_feature_data[MB_LVL_ALT_LF][1];
    mbd->segment_feature_data[MB_LVL_ALT_LF][2] = cpi->segment_feature_data[MB_LVL_ALT_LF][2];
    mbd->segment_feature_data[MB_LVL_ALT_LF][3] = cpi->segment_feature_data[MB_LVL_ALT_LF][3];
}

void vp8cx_pick_filter_level(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi)
{
    VP8_COMMON *cm = &cpi->common;

    int best_err = 0;
    int filt_err = 0;
Johann's avatar
Johann committed
291
292
    int min_filter_level = get_min_filter_level(cpi, cm->base_qindex);
    int max_filter_level = get_max_filter_level(cpi, cm->base_qindex);
John Koleszar's avatar
John Koleszar committed
293
294
295
296
297
298
299
300
301
302
303
304

    int filter_step;
    int filt_high = 0;
    int filt_mid = cm->filter_level;      // Start search at previous frame filter level
    int filt_low = 0;
    int filt_best;
    int filt_direction = 0;

    int Bias = 0;                       // Bias against raising loop filter and in favour of lowering it

    //  Make a copy of the unfiltered / processed recon buffer
#if HAVE_ARMV7
305
306
307
308
309
310
311
312
313
314
315
316
317
318
#if CONFIG_RUNTIME_CPU_DETECT
    if (cm->rtcd.flags & HAS_NEON)
#endif
    {
        vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(cm->frame_to_show, &cpi->last_frame_uf);
    }
#if CONFIG_RUNTIME_CPU_DETECT
    else
#endif
#endif
#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT
    {
        vp8_yv12_copy_frame_ptr(cm->frame_to_show, &cpi->last_frame_uf);
    }
John Koleszar's avatar
John Koleszar committed
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
347
348
#endif

    if (cm->frame_type == KEY_FRAME)
        cm->sharpness_level = 0;
    else
        cm->sharpness_level = cpi->oxcf.Sharpness;

    // Start the search at the previous frame filter level unless it is now out of range.
    filt_mid = cm->filter_level;

    if (filt_mid < min_filter_level)
        filt_mid = min_filter_level;
    else if (filt_mid > max_filter_level)
        filt_mid = max_filter_level;

    // Define the initial step size
    filter_step = (filt_mid < 16) ? 4 : filt_mid / 4;

    // Get baseline error score
    vp8cx_set_alt_lf_level(cpi, filt_mid);
    vp8_loop_filter_frame_yonly(cm, &cpi->mb.e_mbd, filt_mid, 0);
    cm->last_frame_type = cm->frame_type;
    cm->last_filter_type = cm->filter_type;
    cm->last_sharpness_level = cm->sharpness_level;

    best_err = vp8_calc_ss_err(sd, cm->frame_to_show, IF_RTCD(&cpi->rtcd.variance));
    filt_best = filt_mid;

    //  Re-instate the unfiltered frame
#if HAVE_ARMV7
349
350
351
352
353
354
355
356
357
358
359
360
361
362
#if CONFIG_RUNTIME_CPU_DETECT
    if (cm->rtcd.flags & HAS_NEON)
#endif
    {
        vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(&cpi->last_frame_uf, cm->frame_to_show);
    }
#if CONFIG_RUNTIME_CPU_DETECT
    else
#endif
#endif
#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT
    {
        vp8_yv12_copy_frame_yonly_ptr(&cpi->last_frame_uf, cm->frame_to_show);
    }
John Koleszar's avatar
John Koleszar committed
363
364
365
366
367
368
369
#endif

    while (filter_step > 0)
    {
        Bias = (best_err >> (15 - (filt_mid / 8))) * filter_step; //PGW change 12/12/06 for small images

        // jbb chg: 20100118 - in sections with lots of new material coming in don't bias as much to a low filter value
370
371
        if (cpi->twopass.section_intra_rating < 20)
            Bias = Bias * cpi->twopass.section_intra_rating / 20;
John Koleszar's avatar
John Koleszar committed
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388

        filt_high = ((filt_mid + filter_step) > max_filter_level) ? max_filter_level : (filt_mid + filter_step);
        filt_low = ((filt_mid - filter_step) < min_filter_level) ? min_filter_level : (filt_mid - filter_step);

        if ((filt_direction <= 0) && (filt_low != filt_mid))
        {
            // Get Low filter error score
            vp8cx_set_alt_lf_level(cpi, filt_low);
            vp8_loop_filter_frame_yonly(cm, &cpi->mb.e_mbd, filt_low, 0);
            cm->last_frame_type = cm->frame_type;
            cm->last_filter_type = cm->filter_type;
            cm->last_sharpness_level = cm->sharpness_level;

            filt_err = vp8_calc_ss_err(sd, cm->frame_to_show, IF_RTCD(&cpi->rtcd.variance));

            //  Re-instate the unfiltered frame
#if HAVE_ARMV7
389
390
391
392
393
394
395
396
397
398
399
400
401
402
#if CONFIG_RUNTIME_CPU_DETECT
            if (cm->rtcd.flags & HAS_NEON)
#endif
            {
                vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(&cpi->last_frame_uf, cm->frame_to_show);
            }
#if CONFIG_RUNTIME_CPU_DETECT
            else
#endif
#endif
#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT
            {
                vp8_yv12_copy_frame_yonly_ptr(&cpi->last_frame_uf, cm->frame_to_show);
            }
John Koleszar's avatar
John Koleszar committed
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
#endif

            // If value is close to the best so far then bias towards a lower loop filter value.
            if ((filt_err - Bias) < best_err)
            {
                // Was it actually better than the previous best?
                if (filt_err < best_err)
                    best_err = filt_err;

                filt_best = filt_low;
            }
        }

        // Now look at filt_high
        if ((filt_direction >= 0) && (filt_high != filt_mid))
        {
            vp8cx_set_alt_lf_level(cpi, filt_high);
            vp8_loop_filter_frame_yonly(cm, &cpi->mb.e_mbd, filt_high, 0);
            cm->last_frame_type = cm->frame_type;
            cm->last_filter_type = cm->filter_type;
            cm->last_sharpness_level = cm->sharpness_level;

            filt_err = vp8_calc_ss_err(sd, cm->frame_to_show, IF_RTCD(&cpi->rtcd.variance));

            //  Re-instate the unfiltered frame
#if HAVE_ARMV7
429
430
431
432
433
434
435
436
437
438
439
440
441
442
#if CONFIG_RUNTIME_CPU_DETECT
            if (cm->rtcd.flags & HAS_NEON)
#endif
            {
                vp8_yv12_copy_frame_yonly_no_extend_frame_borders_neon(&cpi->last_frame_uf, cm->frame_to_show);
            }
#if CONFIG_RUNTIME_CPU_DETECT
            else
#endif
#endif
#if !HAVE_ARMV7 || CONFIG_RUNTIME_CPU_DETECT
            {
                vp8_yv12_copy_frame_yonly_ptr(&cpi->last_frame_uf, cm->frame_to_show);
            }
John Koleszar's avatar
John Koleszar committed
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
#endif

            // Was it better than the previous best?
            if (filt_err < (best_err - Bias))
            {
                best_err = filt_err;
                filt_best = filt_high;
            }
        }

        // Half the step distance if the best filter value was the same as last time
        if (filt_best == filt_mid)
        {
            filter_step = filter_step / 2;
            filt_direction = 0;
        }
        else
        {
            filt_direction = (filt_best < filt_mid) ? -1 : 1;
            filt_mid = filt_best;
        }
    }

    cm->filter_level = filt_best;
}