lr_apply.c 12.3 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
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
/*
 * Copyright © 2018, VideoLAN and dav1d authors
 * Copyright © 2018, Two Orioles, LLC
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *
 * 1. Redistributions of source code must retain the above copyright notice, this
 *    list of conditions and the following disclaimer.
 *
 * 2. Redistributions in binary form must reproduce the above copyright notice,
 *    this list of conditions and the following disclaimer in the documentation
 *    and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#include "config.h"

#include <stdio.h>

#include "common/intops.h"

#include "src/lr_apply.h"


enum LrRestorePlanes {
    LR_RESTORE_Y = 1 << 0,
    LR_RESTORE_U = 1 << 1,
    LR_RESTORE_V = 1 << 2,
};

static void backup_lpf(pixel *dst, ptrdiff_t dst_stride,
                       const pixel *src, ptrdiff_t src_stride,
                       const int first_stripe_h, const int next_stripe_h,
                       int row, const int row_h, const int w, const int h)
{
    src_stride = PXSTRIDE(src_stride);
    dst_stride = PXSTRIDE(dst_stride);
    if (row) {
        // Copy the top part of the stored loop filtered pixels from the
        // previous sb row needed above the first stripe of this sb row.
        pixel_copy(&dst[dst_stride *  0], &dst[dst_stride *  8], w);
        pixel_copy(&dst[dst_stride *  1], &dst[dst_stride *  9], w);
        pixel_copy(&dst[dst_stride *  2], &dst[dst_stride * 10], w);
        pixel_copy(&dst[dst_stride *  3], &dst[dst_stride * 11], w);
    }

    int stripe_h = first_stripe_h;
    dst += 4 * dst_stride;
    src += (stripe_h - 2) * src_stride;
    for (; row + stripe_h <= row_h; row += stripe_h) {
        for (int i = 0; i < 4; i++) {
            pixel_copy(dst, src, w);
            dst += dst_stride;
            src += src_stride;
        }
        stripe_h = next_stripe_h;
        src += (stripe_h - 4) * src_stride;
    }
}

void bytefn(dav1d_lr_copy_lpf)(Dav1dFrameContext *const f,
                               /*const*/ pixel *const src[3], const int sby)
{
    const int stripe_h = 64 - (8 * !sby);
    const ptrdiff_t offset = 8 * !!sby;
    const ptrdiff_t *const src_stride = f->cur.p.stride;

    // TODO Also check block level restore type to reduce copying.
    const int restore_planes =
        ((f->frame_hdr.restoration.type[0] != RESTORATION_NONE) << 0) +
        ((f->frame_hdr.restoration.type[1] != RESTORATION_NONE) << 1) +
        ((f->frame_hdr.restoration.type[2] != RESTORATION_NONE) << 2);

    if (restore_planes & LR_RESTORE_Y) {
        const int h = f->bh << 2;
        const int w = f->bw << 2;
        const int row_h = imin((sby + 1) << (6 + f->seq_hdr.sb128), h);
        const int y_stripe = (sby << (6 + f->seq_hdr.sb128)) - offset;
        backup_lpf(f->lf.lr_lpf_line_ptr[0], sizeof(pixel) * f->b4_stride * 4,
                   src[0] - offset * PXSTRIDE(src_stride[0]),
                   src_stride[0], stripe_h, 64, y_stripe, row_h, w, h);
    }
    if (restore_planes & (LR_RESTORE_U | LR_RESTORE_V)) {
        const int ss_ver = f->cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420;
        const int ss_hor = f->cur.p.p.layout != DAV1D_PIXEL_LAYOUT_I444;
        const int h = f->bh << (2 - ss_ver);
        const int w = f->bw << (2 - ss_hor);
        const int row_h = imin((sby + 1) << ((6 - ss_ver) + f->seq_hdr.sb128), h);
        const int stripe_h_uv = stripe_h >> ss_ver;
        const ptrdiff_t offset_uv = offset >> ss_ver;
        const int y_stripe =
            (sby << ((6 - ss_ver) + f->seq_hdr.sb128)) - offset_uv;

        if (restore_planes & LR_RESTORE_U) {
            backup_lpf(f->lf.lr_lpf_line_ptr[1], sizeof(pixel) * f->b4_stride * 4,
                       src[1] - offset_uv * PXSTRIDE(src_stride[1]),
                       src_stride[1], stripe_h_uv, 32, y_stripe,
                       row_h, w, h);
        }
        if (restore_planes & LR_RESTORE_V) {
            backup_lpf(f->lf.lr_lpf_line_ptr[2], sizeof(pixel) * f->b4_stride * 4,
                       src[2] - offset_uv * PXSTRIDE(src_stride[1]),
                       src_stride[1], stripe_h_uv, 32, y_stripe,
                       row_h, w, h);
        }
    }
}


static void lr_stripe(const Dav1dFrameContext *const f, pixel *p, int x, int y,
123
124
                      const int plane, const int unit_w, const int row_h,
                      const Av1RestorationUnit *const lr, enum LrEdgeFlags edges)
125
126
{
    const Dav1dDSPContext *const dsp = f->dsp;
127
    const int ss_ver = !!plane * f->cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420;
128
129
130
131
132
    const int sbrow_has_bottom = (edges & LR_HAVE_BOTTOM);
    const pixel *lpf = f->lf.lr_lpf_line_ptr[plane] + x;
    const ptrdiff_t p_stride = f->cur.p.stride[!!plane];
    const ptrdiff_t lpf_stride = sizeof(pixel) * f->b4_stride * 4;

133
134
    // The first stripe of the frame is shorter by 8 luma pixel rows.
    int stripe_h = imin((64 - 8 * !y) >> ss_ver, row_h - y);
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168

    // FIXME [8] might be easier for SIMD
    int16_t filterh[7], filterv[7];
    if (lr->type == RESTORATION_WIENER) {
        filterh[0] = filterh[6] = lr->filter_h[0];
        filterh[1] = filterh[5] = lr->filter_h[1];
        filterh[2] = filterh[4] = lr->filter_h[2];
        filterh[3] = -((filterh[0] + filterh[1] + filterh[2]) * 2);

        filterv[0] = filterv[6] = lr->filter_v[0];
        filterv[1] = filterv[5] = lr->filter_v[1];
        filterv[2] = filterv[4] = lr->filter_v[2];
        filterv[3] = -((filterv[0] + filterv[1] + filterv[2]) * 2);
    }

    while (y + stripe_h <= row_h) {
        // TODO Look into getting rid of the this if
        if (y + stripe_h == row_h) {
            edges &= ~LR_HAVE_BOTTOM;
        } else {
            edges |= LR_HAVE_BOTTOM;
        }
        if (lr->type == RESTORATION_WIENER) {
            dsp->lr.wiener(p, p_stride, lpf, lpf_stride, unit_w, stripe_h,
                           filterh, filterv, edges);
        } else {
            assert(lr->type == RESTORATION_SGRPROJ);
            dsp->lr.selfguided(p, p_stride, lpf, lpf_stride, unit_w, stripe_h,
                               lr->sgr_idx, lr->sgr_weights, edges);
        }

        y += stripe_h;
        if (y + stripe_h > row_h && sbrow_has_bottom) break;
        p += stripe_h * PXSTRIDE(p_stride);
169
170
        edges |= LR_HAVE_TOP;
        stripe_h = imin(64 >> ss_ver, row_h - y);
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
        if (stripe_h == 0) break;
        lpf += 4 * PXSTRIDE(lpf_stride);
    }
}

static void backup3xU(pixel *dst, const pixel *src, const ptrdiff_t src_stride,
                      int u)
{
    for (; u > 0; u--, dst += 3, src += PXSTRIDE(src_stride))
        pixel_copy(dst, src, 3);
}

static void restore3xU(pixel *dst, const ptrdiff_t dst_stride, const pixel *src,
                       int u)
{
    for (; u > 0; u--, dst += PXSTRIDE(dst_stride), src += 3)
        pixel_copy(dst, src, 3);
}

static void lr_sbrow(const Dav1dFrameContext *const f, pixel *p, const int y,
                     const int w, const int h, const int row_h, const int plane)
{
    const int ss_ver = !!plane * f->cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420;
    const ptrdiff_t p_stride = f->cur.p.stride[!!plane];

    const int unit_size_log2 = f->frame_hdr.restoration.unit_size[!!plane];
    const int unit_size = 1 << unit_size_log2;
    const int half_unit_size = unit_size >> 1;
    const int max_unit_size = unit_size + half_unit_size;

201
    // Y coordinate of the sbrow (y is 8 luma pixel rows above row_y)
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
    const int row_y = y + ((8 >> ss_ver) * !!y);

    // FIXME This is an ugly hack to lookup the proper AV1Filter unit for
    // chroma planes. Question: For Multithreaded decoding, is it better
    // to store the chroma LR information with collocated Luma information?
    // In other words. For a chroma restoration unit locate at 128,128 and
    // with a 4:2:0 chroma subsampling, do we store the filter information at
    // the AV1Filter unit located at (128,128) or (256,256)
    // TODO Support chroma subsampling.
    const int shift = plane ? 6 : 7;

    int ruy = (row_y >> unit_size_log2);
    // Merge last restoration unit if its height is < half_unit_size
    if (ruy > 0) ruy -= (ruy << unit_size_log2) + half_unit_size > h;

217
218
219
    // The first stripe of the frame is shorter by 8 luma pixel rows.
    const int filter_h =
        imin(((1 << (6 + f->seq_hdr.sb128)) - 8 * !y) >> ss_ver, h - y);
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

    pixel pre_lr_border[filter_h * 3];
    pixel post_lr_border[filter_h * 3];

    int unit_w = unit_size;

    enum LrEdgeFlags edges = (y > 0 ? LR_HAVE_TOP : 0) |
                             (row_h < h ? LR_HAVE_BOTTOM : 0);

    for (int x = 0, rux = 0; x < w; x+= unit_w, rux++, edges |= LR_HAVE_LEFT) {
        // TODO Clean up this if statement.
        if (x + max_unit_size > w) {
            unit_w = w - x;
            edges &= ~LR_HAVE_RIGHT;
        } else {
            edges |= LR_HAVE_RIGHT;
        }

        // Based on the position of the restoration unit, find the corresponding
        // AV1Filter unit.
        const int unit_idx = ((ruy & 16) >> 3) + ((rux & 16) >> 4);
        const Av1RestorationUnit *const lr =
            &f->lf.mask[(((ruy << unit_size_log2) >> shift) * f->sb128w) +
                        (x >> shift)].lr[plane][unit_idx];

        if (edges & LR_HAVE_LEFT) {
            restore3xU(p - 3, p_stride, pre_lr_border, filter_h);
        }
        // FIXME Don't backup if the next restoration unit is RESTORE_NONE
        // This also requires not restoring in the same conditions.
        if (edges & LR_HAVE_RIGHT) {
            backup3xU(pre_lr_border, p + unit_w - 3, p_stride, filter_h);
        }
        if (lr->type != RESTORATION_NONE) {
254
            lr_stripe(f, p, x, y, plane, unit_w, row_h, lr, edges);
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
291
292
293
294
295
296
297
298
299
300
301
302
        }
        if (edges & LR_HAVE_LEFT) {
            restore3xU(p - 3, p_stride, post_lr_border, filter_h);
        }
        if (edges & LR_HAVE_RIGHT) {
            backup3xU(post_lr_border, p + unit_w - 3, p_stride, filter_h);
        }
        p += unit_w;
    }
}

void bytefn(dav1d_lr_sbrow)(Dav1dFrameContext *const f, pixel *const dst[3],
                            const int sby)
{
    const ptrdiff_t offset_y = 8 * !!sby;
    const ptrdiff_t *const dst_stride = f->cur.p.stride;

    const int restore_planes =
        ((f->frame_hdr.restoration.type[0] != RESTORATION_NONE) << 0) +
        ((f->frame_hdr.restoration.type[1] != RESTORATION_NONE) << 1) +
        ((f->frame_hdr.restoration.type[2] != RESTORATION_NONE) << 2);

    if (restore_planes & LR_RESTORE_Y) {
        const int h = f->bh << 2;
        const int w = f->bw << 2;
        const int row_h = imin((sby + 1) << (6 + f->seq_hdr.sb128), h);
        const int y_stripe = (sby << (6 + f->seq_hdr.sb128)) - offset_y;
        lr_sbrow(f, dst[0] - offset_y * PXSTRIDE(dst_stride[0]), y_stripe, w,
                 h, row_h, 0);
    }
    if (restore_planes & (LR_RESTORE_U | LR_RESTORE_V)) {
        const int ss_ver = f->cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420;
        const int ss_hor = f->cur.p.p.layout != DAV1D_PIXEL_LAYOUT_I444;
        const int h = f->bh << (2 - ss_ver);
        const int w = f->bw << (2 - ss_hor);
        const int row_h = imin((sby + 1) << ((6 - ss_ver) + f->seq_hdr.sb128), h);
        const ptrdiff_t offset_uv = offset_y >> ss_ver;
        const int y_stripe =
            (sby << ((6 - ss_ver) + f->seq_hdr.sb128)) - offset_uv;
        if (restore_planes & LR_RESTORE_U)
            lr_sbrow(f, dst[1] - offset_uv * PXSTRIDE(dst_stride[1]), y_stripe,
                     w, h, row_h, 1);

        if (restore_planes & LR_RESTORE_V)
            lr_sbrow(f, dst[2] - offset_uv * PXSTRIDE(dst_stride[1]), y_stripe,
                     w, h, row_h, 2);
    }
}