decode.c 129 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136
/*
 * 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 <errno.h>
#include <limits.h>
#include <string.h>
#include <stdio.h>
#include <inttypes.h>

#include "dav1d/data.h"

#include "common/intops.h"
#include "common/mem.h"

#include "src/decode.h"
#include "src/dequant_tables.h"
#include "src/env.h"
#include "src/qm.h"
#include "src/recon.h"
#include "src/ref.h"
#include "src/tables.h"
#include "src/thread_task.h"
#include "src/warpmv.h"

static void init_quant_tables(const Av1SequenceHeader *const seq_hdr,
                              const Av1FrameHeader *const frame_hdr,
                              const int qidx, uint16_t (*dq)[3][2])
{
    for (int i = 0; i < (frame_hdr->segmentation.enabled ? 8 : 1); i++) {
        const int yac = frame_hdr->segmentation.enabled ?
            iclip_u8(qidx + frame_hdr->segmentation.seg_data.d[i].delta_q) : qidx;
        const int ydc = iclip_u8(yac + frame_hdr->quant.ydc_delta);
        const int uac = iclip_u8(yac + frame_hdr->quant.uac_delta);
        const int udc = iclip_u8(yac + frame_hdr->quant.udc_delta);
        const int vac = iclip_u8(yac + frame_hdr->quant.vac_delta);
        const int vdc = iclip_u8(yac + frame_hdr->quant.vdc_delta);

        dq[i][0][0] = dav1d_dq_tbl[seq_hdr->bpc > 8][ydc][0];
        dq[i][0][1] = dav1d_dq_tbl[seq_hdr->bpc > 8][yac][1];
        dq[i][1][0] = dav1d_dq_tbl[seq_hdr->bpc > 8][udc][0];
        dq[i][1][1] = dav1d_dq_tbl[seq_hdr->bpc > 8][uac][1];
        dq[i][2][0] = dav1d_dq_tbl[seq_hdr->bpc > 8][vdc][0];
        dq[i][2][1] = dav1d_dq_tbl[seq_hdr->bpc > 8][vac][1];
    }
}

static int read_mv_component_diff(Dav1dTileContext *const t,
                                  CdfMvComponent *const mv_comp,
                                  const int have_fp)
{
    Dav1dTileState *const ts = t->ts;
    const Dav1dFrameContext *const f = t->f;
    const int have_hp = f->frame_hdr.hp;
    const int sign = msac_decode_bool_adapt(&ts->msac, mv_comp->sign);
    const int cl = msac_decode_symbol_adapt(&ts->msac, mv_comp->classes, 11);
    int up, fp, hp;

    if (!cl) {
        up = msac_decode_bool_adapt(&ts->msac, mv_comp->class0);
        if (have_fp) {
            fp = msac_decode_symbol_adapt(&ts->msac, mv_comp->class0_fp[up], 4);
            hp = have_hp ? msac_decode_bool_adapt(&ts->msac, mv_comp->class0_hp) : 1;
        } else {
            fp = 3;
            hp = 1;
        }
    } else {
        up = 1 << cl;
        for (int n = 0; n < cl; n++)
            up |= msac_decode_bool_adapt(&ts->msac, mv_comp->classN[n]) << n;
        if (have_fp) {
            fp = msac_decode_symbol_adapt(&ts->msac, mv_comp->classN_fp, 4);
            hp = have_hp ? msac_decode_bool_adapt(&ts->msac, mv_comp->classN_hp) : 1;
        } else {
            fp = 3;
            hp = 1;
        }
    }

    const int diff = ((up << 3) | (fp << 1) | hp) + 1;

    return sign ? -diff : diff;
}

static void read_mv_residual(Dav1dTileContext *const t, mv *const ref_mv,
                             CdfMvContext *const mv_cdf, const int have_fp)
{
    switch (msac_decode_symbol_adapt(&t->ts->msac, t->ts->cdf.mv.joint, N_MV_JOINTS)) {
    case MV_JOINT_HV:
        ref_mv->y += read_mv_component_diff(t, &mv_cdf->comp[0], have_fp);
        ref_mv->x += read_mv_component_diff(t, &mv_cdf->comp[1], have_fp);
        break;
    case MV_JOINT_H:
        ref_mv->x += read_mv_component_diff(t, &mv_cdf->comp[1], have_fp);
        break;
    case MV_JOINT_V:
        ref_mv->y += read_mv_component_diff(t, &mv_cdf->comp[0], have_fp);
        break;
    default:
        break;
    }
}

static void read_tx_tree(Dav1dTileContext *const t,
                         const enum RectTxfmSize from,
                         const int depth, uint16_t *const masks,
                         const int x_off, const int y_off)
{
    const Dav1dFrameContext *const f = t->f;
    const int bx4 = t->bx & 31, by4 = t->by & 31;
137
    const TxfmInfo *const t_dim = &dav1d_txfm_dimensions[from];
138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154
    const int txw = t_dim->lw, txh = t_dim->lh;
    int is_split;

    if (depth < 2 && from > (int) TX_4X4) {
        const int cat = 2 * (TX_64X64 - t_dim->max) - depth;
        const int a = t->a->tx[bx4] < txw;
        const int l = t->l.tx[by4] < txh;

        is_split = msac_decode_bool_adapt(&t->ts->msac, t->ts->cdf.m.txpart[cat][a + l]);
        if (is_split)
            masks[depth] |= 1 << (y_off * 4 + x_off);
    } else {
        is_split = 0;
    }

    if (is_split && t_dim->max > TX_8X8) {
        const enum RectTxfmSize sub = t_dim->sub;
155
        const TxfmInfo *const sub_t_dim = &dav1d_txfm_dimensions[sub];
156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178
        const int txsw = sub_t_dim->w, txsh = sub_t_dim->h;

        read_tx_tree(t, sub, depth + 1, masks, x_off * 2 + 0, y_off * 2 + 0);
        t->bx += txsw;
        if (txw >= txh && t->bx < f->bw)
            read_tx_tree(t, sub, depth + 1, masks, x_off * 2 + 1, y_off * 2 + 0);
        t->bx -= txsw;
        t->by += txsh;
        if (txh >= txw && t->by < f->bh) {
            read_tx_tree(t, sub, depth + 1, masks, x_off * 2 + 0, y_off * 2 + 1);
            t->bx += txsw;
            if (txw >= txh && t->bx < f->bw)
                read_tx_tree(t, sub, depth + 1, masks,
                             x_off * 2 + 1, y_off * 2 + 1);
            t->bx -= txsw;
        }
        t->by -= txsh;
    } else {
        memset(&t->a->tx[bx4], is_split ? TX_4X4 : txw, t_dim->w);
        memset(&t->l.tx[by4], is_split ? TX_4X4 : txh, t_dim->h);
    }
}

179
static int neg_deinterleave(int diff, int ref, int max) {
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
    if (!ref) return diff;
    if (ref >= (max - 1)) return max - diff - 1;
    if (2 * ref < max) {
        if (diff <= 2 * ref) {
            if (diff & 1)
                return ref + ((diff + 1) >> 1);
            else
                return ref - (diff >> 1);
        }
        return diff;
    } else {
        if (diff <= 2 * (max - ref - 1)) {
            if (diff & 1)
                return ref + ((diff + 1) >> 1);
            else
                return ref - (diff >> 1);
        }
        return max - (diff + 1);
    }
}

static void find_matching_ref(const Dav1dTileContext *const t,
                              const enum EdgeFlags intra_edge_flags,
                              const int bw4, const int bh4,
                              const int w4, const int h4,
                              const int have_left, const int have_top,
                              const int ref, uint64_t masks[2])
{
    const Dav1dFrameContext *const f = t->f;
    const ptrdiff_t b4_stride = f->b4_stride;
    const refmvs *const r = &f->mvs[t->by * b4_stride + t->bx];
    int count = 0;
    int have_topleft = have_top && have_left;
    int have_topright = imax(bw4, bh4) < 32 &&
                        have_top && t->bx + bw4 < t->ts->tiling.col_end &&
                        (intra_edge_flags & EDGE_I444_TOP_HAS_RIGHT);

217
#define bs(rp) dav1d_block_dimensions[sbtype_to_bs[(rp)->sb_type]]
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298
#define matches(rp) ((rp)->ref[0] == ref + 1 && (rp)->ref[1] == -1)

    if (have_top) {
        const refmvs *r2 = &r[-b4_stride];
        if (matches(r2)) {
            masks[0] |= 1;
            count = 1;
        }
        int aw4 = bs(r2)[0];
        if (aw4 >= bw4) {
            const int off = t->bx & (aw4 - 1);
            if (off) have_topleft = 0;
            if (aw4 - off > bw4) have_topright = 0;
        } else {
            unsigned mask = 1 << aw4;
            for (int x = aw4; x < w4; x += aw4) {
                r2 += aw4;
                if (matches(r2)) {
                    masks[0] |= mask;
                    if (++count >= 8) return;
                }
                aw4 = bs(r2)[0];
                mask <<= aw4;
            }
        }
    }
    if (have_left) {
        const refmvs *r2 = &r[-1];
        if (matches(r2)) {
            masks[1] |= 1;
            if (++count >= 8) return;
        }
        int lh4 = bs(r2)[1];
        if (lh4 >= bh4) {
            if (t->by & (lh4 - 1)) have_topleft = 0;
        } else {
            unsigned mask = 1 << lh4;
            for (int y = lh4; y < h4; y += lh4) {
                r2 += lh4 * b4_stride;
                if (matches(r2)) {
                    masks[1] |= mask;
                    if (++count >= 8) return;
                }
                lh4 = bs(r2)[1];
                mask <<= lh4;
            }
        }
    }
    if (have_topleft && matches(&r[-(1 + b4_stride)])) {
        masks[1] |= 1ULL << 32;
        if (++count >= 8) return;
    }
    if (have_topright && matches(&r[bw4 - b4_stride])) {
        masks[0] |= 1ULL << 32;
    }
#undef matches
}

static void derive_warpmv(const Dav1dTileContext *const t,
                          const int bw4, const int bh4,
                          const uint64_t masks[2], const struct mv mv,
                          WarpedMotionParams *const wmp)
{
    int pts[8][2 /* in, out */][2 /* x, y */], np = 0;
    const Dav1dFrameContext *const f = t->f;
    const ptrdiff_t b4_stride = f->b4_stride;
    const refmvs *const r = &f->mvs[t->by * b4_stride + t->bx];

#define add_sample(dx, dy, sx, sy, rp) do { \
    pts[np][0][0] = 16 * (2 * dx + sx * bs(rp)[0]) - 8; \
    pts[np][0][1] = 16 * (2 * dy + sy * bs(rp)[1]) - 8; \
    pts[np][1][0] = pts[np][0][0] + (rp)->mv[0].x; \
    pts[np][1][1] = pts[np][0][1] + (rp)->mv[0].y; \
    np++; \
} while (0)

    // use masks[] to find the projectable motion vectors in the edges
    if ((unsigned) masks[0] == 1 && !(masks[1] >> 32)) {
        const int off = t->bx & (bs(&r[-b4_stride])[0] - 1);
        add_sample(-off, 0, 1, -1, &r[-b4_stride]);
    } else for (unsigned off = 0, xmask = masks[0]; np < 8 && xmask;) { // top
299
        const int tz = ctz(xmask);
300 301 302 303 304 305 306 307 308
        off += tz;
        add_sample(off, 0, 1, -1, &r[off - b4_stride]);
        xmask >>= tz + 1;
        off += 1;
    }
    if (np < 8 && masks[1] == 1) {
        const int off = t->by & (bs(&r[-1])[1] - 1);
        add_sample(0, -off, -1, 1, &r[-1 - off * b4_stride]);
    } else for (unsigned off = 0, ymask = masks[1]; np < 8 && ymask;) { // left
309
        const int tz = ctz(ymask);
310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344
        off += tz;
        add_sample(0, off, -1, 1, &r[off * b4_stride - 1]);
        ymask >>= tz + 1;
        off += 1;
    }
    if (np < 8 && masks[1] >> 32) // top/left
        add_sample(0, 0, -1, -1, &r[-(1 + b4_stride)]);
    if (np < 8 && masks[0] >> 32) // top/right
        add_sample(bw4, 0, 1, -1, &r[bw4 - b4_stride]);
    assert(np > 0 && np <= 8);
#undef bs

    // select according to motion vector difference against a threshold
    int mvd[8], ret = 0;
    const int thresh = 4 * iclip(imax(bw4, bh4), 4, 28);
    for (int i = 0; i < np; i++) {
        mvd[i] = labs(pts[i][1][0] - pts[i][0][0] - mv.x) +
                 labs(pts[i][1][1] - pts[i][0][1] - mv.y);
        if (mvd[i] > thresh)
            mvd[i] = -1;
        else
            ret++;
    }
    if (!ret) {
        ret = 1;
    } else for (int i = 0, j = np - 1, k = 0; k < np - ret; k++, i++, j--) {
        while (mvd[i] != -1) i++;
        while (mvd[j] == -1) j--;
        assert(i != j);
        if (i > j) break;
        // replace the discarded samples;
        mvd[i] = mvd[j];
        memcpy(pts[i], pts[j], sizeof(*pts));
    }

345 346
    if (!dav1d_find_affine_int(pts, ret, bw4, bh4, mv, wmp, t->bx, t->by) &&
        !dav1d_get_shear_params(wmp))
347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428
    {
        wmp->type = WM_TYPE_AFFINE;
    } else
        wmp->type = WM_TYPE_IDENTITY;
}

static inline int findoddzero(const uint8_t *buf, int len) {
    for (int n = 0; n < len; n++)
        if (!buf[n * 2]) return 1;
    return 0;
}

static void read_pal_plane(Dav1dTileContext *const t, Av1Block *const b,
                           const int pl, const int sz_ctx,
                           const int bx4, const int by4)
{
    Dav1dTileState *const ts = t->ts;
    const Dav1dFrameContext *const f = t->f;
    const int pal_sz = b->pal_sz[pl] = 2 + msac_decode_symbol_adapt(&ts->msac,
                                                 ts->cdf.m.pal_sz[pl][sz_ctx], 7);
    uint16_t cache[16], used_cache[8];
    int l_cache = pl ? t->pal_sz_uv[1][by4] : t->l.pal_sz[by4];
    int n_cache = 0;
    // don't reuse above palette outside SB64 boundaries
    int a_cache = by4 & 15 ? pl ? t->pal_sz_uv[0][bx4] : t->a->pal_sz[bx4] : 0;
    const uint16_t *l = t->al_pal[1][by4][pl], *a = t->al_pal[0][bx4][pl];

    // fill/sort cache
    while (l_cache && a_cache) {
        if (*l < *a) {
            if (!n_cache || cache[n_cache - 1] != *l)
                cache[n_cache++] = *l;
            l++;
            l_cache--;
        } else {
            if (*a == *l) {
                l++;
                l_cache--;
            }
            if (!n_cache || cache[n_cache - 1] != *a)
                cache[n_cache++] = *a;
            a++;
            a_cache--;
        }
    }
    if (l_cache) {
        do {
            if (!n_cache || cache[n_cache - 1] != *l)
                cache[n_cache++] = *l;
            l++;
        } while (--l_cache > 0);
    } else if (a_cache) {
        do {
            if (!n_cache || cache[n_cache - 1] != *a)
                cache[n_cache++] = *a;
            a++;
        } while (--a_cache > 0);
    }

    // find reused cache entries
    int i = 0;
    for (int n = 0; n < n_cache && i < pal_sz; n++)
        if (msac_decode_bool(&ts->msac, 128 << 7))
            used_cache[i++] = cache[n];
    const int n_used_cache = i;

    // parse new entries
    uint16_t *const pal = f->frame_thread.pass ?
        f->frame_thread.pal[((t->by >> 1) + (t->bx & 1)) * (f->b4_stride >> 1) +
                            ((t->bx >> 1) + (t->by & 1))][pl] : t->pal[pl];
    if (i < pal_sz) {
        int prev = pal[i++] = msac_decode_bools(&ts->msac, f->cur.p.p.bpc);

        if (i < pal_sz) {
            int bits = f->cur.p.p.bpc - 3 + msac_decode_bools(&ts->msac, 2);
            const int max = (1 << f->cur.p.p.bpc) - 1;

            do {
                const int delta = msac_decode_bools(&ts->msac, bits);
                prev = pal[i++] = imin(prev + delta + !pl, max);
                if (prev + !pl >= max) {
                    for (; i < pal_sz; i++)
429
                        pal[i] = pal[i - 1] + !pl;
430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451
                    break;
                }
                bits = imin(bits, 1 + ulog2(max - prev - !pl));
            } while (i < pal_sz);
        }

        // merge cache+new entries
        int n = 0, m = n_used_cache;
        for (i = 0; i < pal_sz; i++) {
            if (n < n_used_cache && (m >= pal_sz || used_cache[n] <= pal[m])) {
                pal[i] = used_cache[n++];
            } else {
                assert(m < pal_sz);
                pal[i] = pal[m++];
            }
        }
    } else {
        memcpy(pal, used_cache, n_used_cache * sizeof(*used_cache));
    }

    if (DEBUG_BLOCK_INFO) {
        printf("Post-pal[pl=%d,sz=%d,cache_size=%d,used_cache=%d]: r=%d, cache=",
452
               pl, pal_sz, n_cache, n_used_cache, ts->msac.rng);
453 454 455
        for (int n = 0; n < n_cache; n++)
            printf("%c%02x", n ? ' ' : '[', cache[n]);
        printf("%s, pal=", n_cache ? "]" : "[]");
456
        for (int n = 0; n < pal_sz; n++)
457 458 459 460 461 462
            printf("%c%02x", n ? ' ' : '[', pal[n]);
        printf("]\n");
    }
}

static void read_pal_uv(Dav1dTileContext *const t, Av1Block *const b,
463
                        const int sz_ctx, const int bx4, const int by4)
464
{
465
    read_pal_plane(t, b, 1, sz_ctx, bx4, by4);
466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515

    // V pal coding
    Dav1dTileState *const ts = t->ts;
    const Dav1dFrameContext *const f = t->f;
    uint16_t *const pal = f->frame_thread.pass ?
        f->frame_thread.pal[((t->by >> 1) + (t->bx & 1)) * (f->b4_stride >> 1) +
                            ((t->bx >> 1) + (t->by & 1))][2] : t->pal[2];
    if (msac_decode_bool(&ts->msac, 128 << 7)) {
        const int bits = f->cur.p.p.bpc - 4 + msac_decode_bools(&ts->msac, 2);
        int prev = pal[0] = msac_decode_bools(&ts->msac, f->cur.p.p.bpc);
        const int max = (1 << f->cur.p.p.bpc) - 1;
        for (int i = 1; i < b->pal_sz[1]; i++) {
            int delta = msac_decode_bools(&ts->msac, bits);
            if (delta && msac_decode_bool(&ts->msac, 128 << 7)) delta = -delta;
            prev = pal[i] = (prev + delta) & max;
        }
    } else {
        for (int i = 0; i < b->pal_sz[1]; i++)
            pal[i] = msac_decode_bools(&ts->msac, f->cur.p.p.bpc);
    }
    if (DEBUG_BLOCK_INFO) {
        printf("Post-pal[pl=2]: r=%d ", ts->msac.rng);
        for (int n = 0; n < b->pal_sz[1]; n++)
            printf("%c%02x", n ? ' ' : '[', pal[n]);
        printf("]\n");
    }
}

// meant to be SIMD'able, so that theoretical complexity of this function
// times block size goes from w4*h4 to w4+h4-1
// a and b are previous two lines containing (a) top/left entries or (b)
// top/left entries, with a[0] being either the first top or first left entry,
// depending on top_offset being 1 or 0, and b being the first top/left entry
// for whichever has one. left_offset indicates whether the (len-1)th entry
// has a left neighbour.
// output is order[] and ctx for each member of this diagonal.
static void order_palette(const uint8_t *pal_idx, const ptrdiff_t stride,
                          const int i, const int first, const int last,
                          uint8_t (*const order)[8], uint8_t *const ctx)
{
    int have_top = i > first;

    pal_idx += first + (i - first) * stride;
    for (int j = first, n = 0; j >= last; have_top = 1, j--, n++, pal_idx += stride - 1) {
        const int have_left = j > 0;

        assert(have_left || have_top);

#define add(v_in) do { \
        const int v = v_in; \
516
        assert((unsigned)v < 8U); \
517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602
        order[n][o_idx++] = v; \
        mask |= 1 << v; \
    } while (0)

        unsigned mask = 0;
        int o_idx = 0;
        if (!have_left) {
            ctx[n] = 0;
            add(pal_idx[-stride]);
        } else if (!have_top) {
            ctx[n] = 0;
            add(pal_idx[-1]);
        } else {
            const int l = pal_idx[-1], t = pal_idx[-stride], tl = pal_idx[-(stride + 1)];
            const int same_t_l = t == l;
            const int same_t_tl = t == tl;
            const int same_l_tl = l == tl;
            const int same_all = same_t_l & same_t_tl & same_l_tl;

            if (same_all) {
                ctx[n] = 4;
                add(t);
            } else if (same_t_l) {
                ctx[n] = 3;
                add(t);
                add(tl);
            } else if (same_t_tl | same_l_tl) {
                ctx[n] = 2;
                add(tl);
                add(same_t_tl ? l : t);
            } else {
                ctx[n] = 1;
                add(imin(t, l));
                add(imax(t, l));
                add(tl);
            }
        }
        for (unsigned m = 1, bit = 0; m < 0x100; m <<= 1, bit++)
            if (!(mask & m))
                order[n][o_idx++] = bit;
        assert(o_idx == 8);
#undef add
    }
}

static void read_pal_indices(Dav1dTileContext *const t,
                             uint8_t *const pal_idx,
                             const Av1Block *const b, const int pl,
                             const int w4, const int h4,
                             const int bw4, const int bh4)
{
    Dav1dTileState *const ts = t->ts;
    const ptrdiff_t stride = bw4 * 4;
    pal_idx[0] = msac_decode_uniform(&ts->msac, b->pal_sz[pl]);
    uint16_t (*const color_map_cdf)[8 + 1] =
        ts->cdf.m.color_map[pl][b->pal_sz[pl] - 2];
    for (int i = 1; i < 4 * (w4 + h4) - 1; i++) {
        // top/left-to-bottom/right diagonals ("wave-front")
        uint8_t order[64][8], ctx[64];
        const int first = imin(i, w4 * 4 - 1);
        const int last = imax(0, i - h4 * 4 + 1);
        order_palette(pal_idx, stride, i, first, last, order, ctx);
        for (int j = first, m = 0; j >= last; j--, m++) {
            const int color_idx =
                msac_decode_symbol_adapt(&ts->msac, color_map_cdf[ctx[m]],
                                         b->pal_sz[pl]);
            pal_idx[(i - j) * stride + j] = order[m][color_idx];
        }
    }
    // fill invisible edges
    if (bw4 > w4)
        for (int y = 0; y < 4 * h4; y++)
            memset(&pal_idx[y * stride + 4 * w4],
                   pal_idx[y * stride + 4 * w4 - 1], 4 * (bw4 - w4));
    if (h4 < bh4) {
        const uint8_t *const src = &pal_idx[stride * (4 * h4 - 1)];
        for (int y = h4 * 4; y < bh4 * 4; y++)
            memcpy(&pal_idx[y * stride], src, bw4 * 4);
    }
}

static void read_vartx_tree(Dav1dTileContext *const t,
                            Av1Block *const b, const enum BlockSize bs,
                            const int bx4, const int by4)
{
    const Dav1dFrameContext *const f = t->f;
603
    const uint8_t *const b_dim = dav1d_block_dimensions[bs];
604 605 606 607
    const int bw4 = b_dim[0], bh4 = b_dim[1];

    // var-tx tree coding
    b->tx_split[0] = b->tx_split[1] = 0;
608
    b->max_ytx = dav1d_max_txfm_size_for_bs[bs][0];
609 610 611 612 613 614 615 616 617 618 619 620 621 622 623
    if (f->frame_hdr.segmentation.lossless[b->seg_id] ||
        b->max_ytx == TX_4X4)
    {
        b->max_ytx = b->uvtx = TX_4X4;
        if (f->frame_hdr.txfm_mode == TX_SWITCHABLE) {
            memset(&t->a->tx[bx4], TX_4X4, bw4);
            memset(&t->l.tx[by4], TX_4X4, bh4);
        }
    } else if (f->frame_hdr.txfm_mode != TX_SWITCHABLE || b->skip) {
        if (f->frame_hdr.txfm_mode == TX_SWITCHABLE) {
            memset(&t->a->tx[bx4], b_dim[2], bw4);
            memset(&t->l.tx[by4], b_dim[3], bh4);
        } else {
            assert(f->frame_hdr.txfm_mode == TX_LARGEST);
        }
624
        b->uvtx = dav1d_max_txfm_size_for_bs[bs][f->cur.p.p.layout];
625 626 627
    } else {
        assert(imin(bw4, bh4) <= 16 || b->max_ytx == TX_64X64);
        int y, x, y_off, x_off;
628
        const TxfmInfo *const ytx = &dav1d_txfm_dimensions[b->max_ytx];
629 630 631 632 633 634 635 636 637 638 639 640 641
        for (y = 0, y_off = 0; y < bh4; y += ytx->h, y_off++) {
            for (x = 0, x_off = 0; x < bw4; x += ytx->w, x_off++) {
                read_tx_tree(t, b->max_ytx, 0, b->tx_split, x_off, y_off);
                // contexts are updated inside read_tx_tree()
                t->bx += ytx->w;
            }
            t->bx -= x;
            t->by += ytx->h;
        }
        t->by -= y;
        if (DEBUG_BLOCK_INFO)
            printf("Post-vartxtree[%x/%x]: r=%d\n",
                   b->tx_split[0], b->tx_split[1], t->ts->msac.rng);
642
        b->uvtx = dav1d_max_txfm_size_for_bs[bs][f->cur.p.p.layout];
643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668
    }
}

static inline unsigned get_prev_frame_segid(const Dav1dFrameContext *const f,
                                            const int by, const int bx,
                                            const int w4, int h4,
                                            const uint8_t *ref_seg_map,
                                            const ptrdiff_t stride)
{
    unsigned seg_id = 8;

    assert(f->frame_hdr.primary_ref_frame != PRIMARY_REF_NONE);
    dav1d_thread_picture_wait(&f->refp[f->frame_hdr.primary_ref_frame],
                              (by + h4) * 4, PLANE_TYPE_BLOCK);

    ref_seg_map += by * stride + bx;
    do {
        for (int x = 0; x < w4; x++)
            seg_id = imin(seg_id, ref_seg_map[x]);
        ref_seg_map += stride;
    } while (--h4 > 0);
    assert(seg_id < 8);

    return seg_id;
}

669 670 671 672 673
static int decode_b(Dav1dTileContext *const t,
                    const enum BlockLevel bl,
                    const enum BlockSize bs,
                    const enum BlockPartition bp,
                    const enum EdgeFlags intra_edge_flags)
674 675 676 677 678
{
    Dav1dTileState *const ts = t->ts;
    const Dav1dFrameContext *const f = t->f;
    Av1Block b_mem, *const b = f->frame_thread.pass ?
        &f->frame_thread.b[t->by * f->b4_stride + t->bx] : &b_mem;
679
    const uint8_t *const b_dim = dav1d_block_dimensions[bs];
680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713
    const int bx4 = t->bx & 31, by4 = t->by & 31;
    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 cbx4 = bx4 >> ss_hor, cby4 = by4 >> ss_ver;
    const int bw4 = b_dim[0], bh4 = b_dim[1];
    const int w4 = imin(bw4, f->bw - t->bx), h4 = imin(bh4, f->bh - t->by);
    const int cbw4 = (bw4 + ss_hor) >> ss_hor, cbh4 = (bh4 + ss_ver) >> ss_ver;
    const int have_left = t->bx > ts->tiling.col_start;
    const int have_top = t->by > ts->tiling.row_start;
    const int has_chroma = f->seq_hdr.layout != DAV1D_PIXEL_LAYOUT_I400 &&
                           (bw4 > ss_hor || t->bx & 1) &&
                           (bh4 > ss_ver || t->by & 1);

    if (f->frame_thread.pass == 2) {
        if (b->intra) {
            f->bd_fn.recon_b_intra(t, bs, intra_edge_flags, b);

            if (has_chroma) {
                memset(&t->l.uvmode[cby4], b->uv_mode, cbh4);
                memset(&t->a->uvmode[cbx4], b->uv_mode, cbw4);
            }
            const enum IntraPredMode y_mode_nofilt =
                b->y_mode == FILTER_PRED ? DC_PRED : b->y_mode;
            memset(&t->l.mode[by4], y_mode_nofilt, bh4);
            memset(&t->a->mode[bx4], y_mode_nofilt, bw4);
        } else {
            if (b->comp_type == COMP_INTER_NONE && b->motion_mode == MM_WARP) {
                uint64_t mask[2] = { 0, 0 };
                find_matching_ref(t, intra_edge_flags, bw4, bh4, w4, h4,
                                  have_left, have_top, b->ref[0], mask);
                derive_warpmv(t, bw4, bh4, mask, b->mv[0], &t->warpmv);
            }
            f->bd_fn.recon_b_inter(t, bs, b);

714
            const uint8_t *const filter = dav1d_filter_dir[b->filter2d];
715 716 717 718 719 720 721 722 723 724 725
            memset(&t->l.filter[0][by4], filter[0], bh4);
            memset(&t->a->filter[0][bx4], filter[0], bw4);
            memset(&t->l.filter[1][by4], filter[1], bh4);
            memset(&t->a->filter[1][bx4], filter[1], bw4);
            if (has_chroma) {
                memset(&t->l.uvmode[cby4], DC_PRED, cbh4);
                memset(&t->a->uvmode[cbx4], DC_PRED, cbw4);
            }
        }
        memset(&t->l.intra[by4], b->intra, bh4);
        memset(&t->a->intra[bx4], b->intra, bw4);
726
        return 0;
727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772
    }

    const int cw4 = (w4 + ss_hor) >> ss_hor, ch4 = (h4 + ss_ver) >> ss_ver;

    b->bl = bl;
    b->bp = bp;
    b->bs = bs;

    // skip_mode
    if (f->frame_hdr.skip_mode_enabled && imin(bw4, bh4) > 1) {
        const int smctx = t->a->skip_mode[bx4] + t->l.skip_mode[by4];
        b->skip_mode = msac_decode_bool_adapt(&ts->msac,
                                              ts->cdf.m.skip_mode[smctx]);
        if (DEBUG_BLOCK_INFO)
            printf("Post-skipmode[%d]: r=%d\n", b->skip_mode, ts->msac.rng);
    } else {
        b->skip_mode = 0;
    }

    // segment_id (if seg_feature for skip/ref/gmv is enabled)
    int seg_pred = 0;
    if (f->frame_hdr.segmentation.enabled) {
        if (!f->frame_hdr.segmentation.update_map) {
            b->seg_id = f->prev_segmap ?
                        get_prev_frame_segid(f, t->by, t->bx, w4, h4,
                                             f->prev_segmap, f->b4_stride) : 0;
        } else if (f->frame_hdr.segmentation.seg_data.preskip) {
            if (f->frame_hdr.segmentation.temporal &&
                (seg_pred = msac_decode_bool_adapt(&ts->msac,
                                       ts->cdf.m.seg_pred[t->a->seg_pred[bx4] +
                                                          t->l.seg_pred[by4]])))
            {
                // temporal predicted seg_id
                b->seg_id = f->prev_segmap ?
                            get_prev_frame_segid(f, t->by, t->bx, w4, h4,
                                                 f->prev_segmap, f->b4_stride) : 0;
            } else {
                int seg_ctx;
                const unsigned pred_seg_id =
                    get_cur_frame_segid(t->by, t->bx, have_top, have_left,
                                        &seg_ctx, f->cur_segmap, f->b4_stride);
                const unsigned diff = msac_decode_symbol_adapt(&ts->msac,
                                                   ts->cdf.m.seg_id[seg_ctx],
                                                   NUM_SEGMENTS);
                const unsigned last_active_seg_id =
                    f->frame_hdr.segmentation.seg_data.last_active_segid;
773 774
                b->seg_id = neg_deinterleave(diff, pred_seg_id,
                                             last_active_seg_id + 1);
775
                if (b->seg_id > last_active_seg_id) b->seg_id = 0; // error?
776
                if (b->seg_id >= NUM_SEGMENTS) b->seg_id = 0; // error?
777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820
            }

            if (DEBUG_BLOCK_INFO)
                printf("Post-segid[preskip;%d]: r=%d\n",
                       b->seg_id, ts->msac.rng);
        }
    } else {
        b->seg_id = 0;
    }

    // skip
    const int sctx = t->a->skip[bx4] + t->l.skip[by4];
    b->skip = b->skip_mode ? 1 :
              msac_decode_bool_adapt(&ts->msac, ts->cdf.m.skip[sctx]);
    if (DEBUG_BLOCK_INFO)
        printf("Post-skip[%d]: r=%d\n", b->skip, ts->msac.rng);

    // segment_id
    if (f->frame_hdr.segmentation.enabled &&
        f->frame_hdr.segmentation.update_map &&
        !f->frame_hdr.segmentation.seg_data.preskip)
    {
        if (!b->skip && f->frame_hdr.segmentation.temporal &&
            (seg_pred = msac_decode_bool_adapt(&ts->msac,
                                   ts->cdf.m.seg_pred[t->a->seg_pred[bx4] +
                                                      t->l.seg_pred[by4]])))
        {
            // temporal predicted seg_id
            b->seg_id = f->prev_segmap ?
                        get_prev_frame_segid(f, t->by, t->bx, w4, h4,
                                             f->prev_segmap, f->b4_stride) : 0;
        } else {
            int seg_ctx;
            const unsigned pred_seg_id =
                get_cur_frame_segid(t->by, t->bx, have_top, have_left,
                                    &seg_ctx, f->cur_segmap, f->b4_stride);
            if (b->skip) {
                b->seg_id = pred_seg_id;
            } else {
                const unsigned diff = msac_decode_symbol_adapt(&ts->msac,
                                                   ts->cdf.m.seg_id[seg_ctx],
                                                   NUM_SEGMENTS);
                const unsigned last_active_seg_id =
                    f->frame_hdr.segmentation.seg_data.last_active_segid;
821 822
                b->seg_id = neg_deinterleave(diff, pred_seg_id,
                                             last_active_seg_id + 1);
823 824
                if (b->seg_id > last_active_seg_id) b->seg_id = 0; // error?
            }
825
            if (b->seg_id >= NUM_SEGMENTS) b->seg_id = 0; // error?
826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854
        }

        if (DEBUG_BLOCK_INFO)
            printf("Post-segid[postskip;%d]: r=%d\n",
                   b->seg_id, ts->msac.rng);
    }

    // cdef index
    if (!b->skip) {
        const int idx = f->seq_hdr.sb128 ? ((t->bx & 16) >> 4) +
                                           ((t->by & 16) >> 3) : 0;
        if (t->cur_sb_cdef_idx_ptr[idx] == -1) {
            const int v = msac_decode_bools(&ts->msac, f->frame_hdr.cdef.n_bits);
            t->cur_sb_cdef_idx_ptr[idx] = v;
            if (bw4 > 16) t->cur_sb_cdef_idx_ptr[idx + 1] = v;
            if (bh4 > 16) t->cur_sb_cdef_idx_ptr[idx + 2] = v;
            if (bw4 == 32 && bh4 == 32) t->cur_sb_cdef_idx_ptr[idx + 3] = v;

            if (DEBUG_BLOCK_INFO)
                printf("Post-cdef_idx[%d]: r=%d\n",
                        *t->cur_sb_cdef_idx_ptr, ts->msac.rng);
        }
    }

    // delta-q/lf
    if (!(t->bx & (31 >> !f->seq_hdr.sb128)) &&
        !(t->by & (31 >> !f->seq_hdr.sb128)))
    {
        const int prev_qidx = ts->last_qidx;
855
        const int have_delta_q = f->frame_hdr.delta.q.present &&
856
            (bs != (f->seq_hdr.sb128 ? BS_128x128 : BS_64x64) || !b->skip);
857 858 859 860

        int8_t prev_delta_lf[4];
        memcpy(prev_delta_lf, ts->last_delta_lf, 4);

861 862 863 864 865 866 867 868
        if (have_delta_q) {
            int delta_q = msac_decode_symbol_adapt(&ts->msac, ts->cdf.m.delta_q, 4);
            if (delta_q == 3) {
                const int n_bits = 1 + msac_decode_bools(&ts->msac, 3);
                delta_q = msac_decode_bools(&ts->msac, n_bits) + 1 + (1 << n_bits);
            }
            if (delta_q) {
                if (msac_decode_bool(&ts->msac, 128 << 7)) delta_q = -delta_q;
869
                delta_q *= 1 << f->frame_hdr.delta.q.res_log2;
870 871 872 873 874
            }
            ts->last_qidx = iclip(ts->last_qidx + delta_q, 1, 255);
            if (have_delta_q && DEBUG_BLOCK_INFO)
                printf("Post-delta_q[%d->%d]: r=%d\n",
                       delta_q, ts->last_qidx, ts->msac.rng);
875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900

            if (f->frame_hdr.delta.lf.present) {
                const int n_lfs = f->frame_hdr.delta.lf.multi ?
                    f->seq_hdr.layout != DAV1D_PIXEL_LAYOUT_I400 ? 4 : 2 : 1;

                for (int i = 0; i < n_lfs; i++) {
                    int delta_lf =
                        msac_decode_symbol_adapt(&ts->msac,
                        ts->cdf.m.delta_lf[i + f->frame_hdr.delta.lf.multi], 4);
                    if (delta_lf == 3) {
                        const int n_bits = 1 + msac_decode_bools(&ts->msac, 3);
                        delta_lf = msac_decode_bools(&ts->msac, n_bits) +
                                   1 + (1 << n_bits);
                    }
                    if (delta_lf) {
                        if (msac_decode_bool(&ts->msac, 128 << 7))
                            delta_lf = -delta_lf;
                        delta_lf *= 1 << f->frame_hdr.delta.lf.res_log2;
                    }
                    ts->last_delta_lf[i] =
                        iclip(ts->last_delta_lf[i] + delta_lf, -63, 63);
                    if (have_delta_q && DEBUG_BLOCK_INFO)
                        printf("Post-delta_lf[%d:%d]: r=%d\n", i, delta_lf,
                               ts->msac.rng);
                }
            }
901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938
        }
        if (ts->last_qidx == f->frame_hdr.quant.yac) {
            // assign frame-wide q values to this sb
            ts->dq = f->dq;
        } else if (ts->last_qidx != prev_qidx) {
            // find sb-specific quant parameters
            init_quant_tables(&f->seq_hdr, &f->frame_hdr, ts->last_qidx, ts->dqmem);
            ts->dq = ts->dqmem;
        }
        if (!memcmp(ts->last_delta_lf, (int8_t[4]) { 0, 0, 0, 0 }, 4)) {
            // assign frame-wide lf values to this sb
            ts->lflvl = f->lf.lvl;
        } else if (memcmp(ts->last_delta_lf, prev_delta_lf, 4)) {
            // find sb-specific lf lvl parameters
            dav1d_calc_lf_values(ts->lflvlmem, &f->frame_hdr, ts->last_delta_lf);
            ts->lflvl = ts->lflvlmem;
        }
    }

    if (b->skip_mode) {
        b->intra = 0;
    } else if (f->frame_hdr.frame_type & 1) {
        const int ictx = get_intra_ctx(t->a, &t->l, by4, bx4,
                                       have_top, have_left);
        b->intra = !msac_decode_bool_adapt(&ts->msac, ts->cdf.m.intra[ictx]);
        if (DEBUG_BLOCK_INFO)
            printf("Post-intra[%d]: r=%d\n", b->intra, ts->msac.rng);
    } else if (f->frame_hdr.allow_intrabc) {
        b->intra = !msac_decode_bool_adapt(&ts->msac, ts->cdf.m.intrabc);
        if (DEBUG_BLOCK_INFO)
            printf("Post-intrabcflag[%d]: r=%d\n", b->intra, ts->msac.rng);
    } else {
        b->intra = 1;
    }

    // intra/inter-specific stuff
    if (b->intra) {
        uint16_t *const ymode_cdf = f->frame_hdr.frame_type & 1 ?
939 940 941
            ts->cdf.m.y_mode[dav1d_ymode_size_context[bs]] :
            ts->cdf.kfym[dav1d_intra_mode_context[t->a->mode[bx4]]]
                        [dav1d_intra_mode_context[t->l.mode[by4]]];
942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958
        b->y_mode = msac_decode_symbol_adapt(&ts->msac, ymode_cdf,
                                              N_INTRA_PRED_MODES);
        if (DEBUG_BLOCK_INFO)
            printf("Post-ymode[%d]: r=%d\n", b->y_mode, ts->msac.rng);

        // angle delta
        if (b_dim[2] + b_dim[3] >= 2 && b->y_mode >= VERT_PRED &&
            b->y_mode <= VERT_LEFT_PRED)
        {
            uint16_t *const acdf = ts->cdf.m.angle_delta[b->y_mode - VERT_PRED];
            const int angle = msac_decode_symbol_adapt(&ts->msac, acdf, 7);
            b->y_angle = angle - 3;
        } else {
            b->y_angle = 0;
        }

        if (has_chroma) {
959 960
            const int cfl_allowed = f->frame_hdr.segmentation.lossless[b->seg_id] ?
                cbw4 == 1 && cbh4 == 1 : !!(cfl_allowed_mask & (1 << bs));
961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024
            uint16_t *const uvmode_cdf = ts->cdf.m.uv_mode[cfl_allowed][b->y_mode];
            b->uv_mode = msac_decode_symbol_adapt(&ts->msac, uvmode_cdf,
                                         N_UV_INTRA_PRED_MODES - !cfl_allowed);
            if (DEBUG_BLOCK_INFO)
                printf("Post-uvmode[%d]: r=%d\n", b->uv_mode, ts->msac.rng);

            if (b->uv_mode == CFL_PRED) {
#define SIGN(a) (!!(a) + ((a) > 0))
                const int sign =
                    msac_decode_symbol_adapt(&ts->msac, ts->cdf.m.cfl_sign, 8) + 1;
                const int sign_u = sign * 0x56 >> 8, sign_v = sign - sign_u * 3;
                assert(sign_u == sign / 3);
                if (sign_u) {
                    const int ctx = (sign_u == 2) * 3 + sign_v;
                    b->cfl_alpha[0] = msac_decode_symbol_adapt(&ts->msac,
                                            ts->cdf.m.cfl_alpha[ctx], 16) + 1;
                    if (sign_u == 1) b->cfl_alpha[0] = -b->cfl_alpha[0];
                } else {
                    b->cfl_alpha[0] = 0;
                }
                if (sign_v) {
                    const int ctx = (sign_v == 2) * 3 + sign_u;
                    b->cfl_alpha[1] = msac_decode_symbol_adapt(&ts->msac,
                                            ts->cdf.m.cfl_alpha[ctx], 16) + 1;
                    if (sign_v == 1) b->cfl_alpha[1] = -b->cfl_alpha[1];
                } else {
                    b->cfl_alpha[1] = 0;
                }
#undef SIGN
                if (DEBUG_BLOCK_INFO)
                    printf("Post-uvalphas[%d/%d]: r=%d\n",
                           b->cfl_alpha[0], b->cfl_alpha[1], ts->msac.rng);
            } else if (b_dim[2] + b_dim[3] >= 2 && b->uv_mode >= VERT_PRED &&
                       b->uv_mode <= VERT_LEFT_PRED)
            {
                uint16_t *const acdf = ts->cdf.m.angle_delta[b->uv_mode - VERT_PRED];
                const int angle = msac_decode_symbol_adapt(&ts->msac, acdf, 7);
                b->uv_angle = angle - 3;
            } else {
                b->uv_angle = 0;
            }
        }

        b->pal_sz[0] = b->pal_sz[1] = 0;
        if (f->frame_hdr.allow_screen_content_tools &&
            imax(bw4, bh4) <= 16 && bw4 + bh4 >= 4)
        {
            const int sz_ctx = b_dim[2] + b_dim[3] - 2;
            if (b->y_mode == DC_PRED) {
                const int pal_ctx = (t->a->pal_sz[bx4] > 0) + (t->l.pal_sz[by4] > 0);
                const int use_y_pal =
                    msac_decode_bool_adapt(&ts->msac, ts->cdf.m.pal_y[sz_ctx][pal_ctx]);
                if (DEBUG_BLOCK_INFO)
                    printf("Post-y_pal[%d]: r=%d\n", use_y_pal, ts->msac.rng);
                if (use_y_pal)
                    read_pal_plane(t, b, 0, sz_ctx, bx4, by4);
            }

            if (has_chroma && b->uv_mode == DC_PRED) {
                const int pal_ctx = b->pal_sz[0] > 0;
                const int use_uv_pal =
                    msac_decode_bool_adapt(&ts->msac, ts->cdf.m.pal_uv[pal_ctx]);
                if (DEBUG_BLOCK_INFO)
                    printf("Post-uv_pal[%d]: r=%d\n", use_uv_pal, ts->msac.rng);
1025 1026
                if (use_uv_pal) // see aomedia bug 2183 for why we use luma coordinates
                    read_pal_uv(t, b, sz_ctx, bx4, by4);
1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071
            }
        }

        if (b->y_mode == DC_PRED && !b->pal_sz[0] &&
            imax(b_dim[2], b_dim[3]) <= 3 && f->seq_hdr.filter_intra)
        {
            const int is_filter = msac_decode_bool_adapt(&ts->msac,
                                            ts->cdf.m.use_filter_intra[bs]);
            if (is_filter) {
                b->y_mode = FILTER_PRED;
                b->y_angle = msac_decode_symbol_adapt(&ts->msac,
                                                  ts->cdf.m.filter_intra, 5);
            }
            if (DEBUG_BLOCK_INFO)
                printf("Post-filterintramode[%d/%d]: r=%d\n",
                       b->y_mode, b->y_angle, ts->msac.rng);
        }

        if (b->pal_sz[0]) {
            uint8_t *pal_idx;
            if (f->frame_thread.pass) {
                pal_idx = ts->frame_thread.pal_idx;
                ts->frame_thread.pal_idx += bw4 * bh4 * 16;
            } else
                pal_idx = t->scratch.pal_idx;
            read_pal_indices(t, pal_idx, b, 0, w4, h4, bw4, bh4);
            if (DEBUG_BLOCK_INFO)
                printf("Post-y-pal-indices: r=%d\n", ts->msac.rng);
        }

        if (has_chroma && b->pal_sz[1]) {
            uint8_t *pal_idx;
            if (f->frame_thread.pass) {
                pal_idx = ts->frame_thread.pal_idx;
                ts->frame_thread.pal_idx += cbw4 * cbh4 * 16;
            } else
                pal_idx = &t->scratch.pal_idx[bw4 * bh4 * 16];
            read_pal_indices(t, pal_idx, b, 1, cw4, ch4, cbw4, cbh4);
            if (DEBUG_BLOCK_INFO)
                printf("Post-uv-pal-indices: r=%d\n", ts->msac.rng);
        }

        const TxfmInfo *t_dim;
        if (f->frame_hdr.segmentation.lossless[b->seg_id]) {
            b->tx = b->uvtx = (int) TX_4X4;
1072
            t_dim = &dav1d_txfm_dimensions[TX_4X4];
1073
        } else {
1074 1075 1076
            b->tx = dav1d_max_txfm_size_for_bs[bs][0];
            b->uvtx = dav1d_max_txfm_size_for_bs[bs][f->cur.p.p.layout];
            t_dim = &dav1d_txfm_dimensions[b->tx];
1077
            if (f->frame_hdr.txfm_mode == TX_SWITCHABLE && t_dim->max > TX_4X4) {
1078
                const int tctx = get_tx_ctx(t->a, &t->l, t_dim, by4, bx4);
1079 1080 1081 1082 1083 1084
                uint16_t *const tx_cdf = ts->cdf.m.txsz[t_dim->max - 1][tctx];
                int depth = msac_decode_symbol_adapt(&ts->msac, tx_cdf,
                                                     imin(t_dim->max + 1, 3));

                while (depth--) {
                    b->tx = t_dim->sub;
1085
                    t_dim = &dav1d_txfm_dimensions[b->tx];
1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101
                }
            }
            if (DEBUG_BLOCK_INFO)
                printf("Post-tx[%d]: r=%d\n", b->tx, ts->msac.rng);
        }

        // reconstruction
        if (f->frame_thread.pass == 1) {
            f->bd_fn.read_coef_blocks(t, bs, b);
        } else {
            f->bd_fn.recon_b_intra(t, bs, intra_edge_flags, b);
        }

        dav1d_create_lf_mask_intra(t->lf_mask, f->lf.level, f->b4_stride,
                                   &f->frame_hdr, (const uint8_t (*)[8][2])
                                   &ts->lflvl[b->seg_id][0][0][0],
1102 1103
                                   t->bx, t->by, (f->cur.p.p.w + 3) >> 2,
                                   (f->cur.p.p.h + 3) >> 2, bs,
1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129
                                   b->tx, b->uvtx, f->cur.p.p.layout,
                                   &t->a->tx_lpf_y[bx4], &t->l.tx_lpf_y[by4],
                                   has_chroma ? &t->a->tx_lpf_uv[cbx4] : NULL,
                                   has_chroma ? &t->l.tx_lpf_uv[cby4] : NULL);

        // update contexts
        memset(&t->a->tx_intra[bx4], t_dim->lw, bw4);
        memset(&t->l.tx_intra[by4], t_dim->lh, bh4);
        const enum IntraPredMode y_mode_nofilt =
            b->y_mode == FILTER_PRED ? DC_PRED : b->y_mode;
        memset(&t->l.mode[by4], y_mode_nofilt, bh4);
        memset(&t->a->mode[bx4], y_mode_nofilt, bw4);
        memset(&t->l.pal_sz[by4], b->pal_sz[0], bh4);
        memset(&t->a->pal_sz[bx4], b->pal_sz[0], bw4);
        if (b->pal_sz[0]) {
            uint16_t *const pal = f->frame_thread.pass ?
                f->frame_thread.pal[((t->by >> 1) + (t->bx & 1)) * (f->b4_stride >> 1) +
                                    ((t->bx >> 1) + (t->by & 1))][0] : t->pal[0];
            for (int x = 0; x < bw4; x++)
                memcpy(t->al_pal[0][bx4 + x][0], pal, 16);
            for (int y = 0; y < bh4; y++)
                memcpy(t->al_pal[1][by4 + y][0], pal, 16);
        }
        if (has_chroma) {
            memset(&t->l.uvmode[cby4], b->uv_mode, cbh4);
            memset(&t->a->uvmode[cbx4], b->uv_mode, cbw4);
1130 1131 1132
            // see aomedia bug 2183 for why we use luma coordinates here
            memset(&t->pal_sz_uv[1][by4], b->pal_sz[1], bh4);
            memset(&t->pal_sz_uv[0][bx4], b->pal_sz[1], bw4);
1133 1134 1135 1136
            if (b->pal_sz[1]) for (int pl = 1; pl < 3; pl++) {
                uint16_t *const pal = f->frame_thread.pass ?
                    f->frame_thread.pal[((t->by >> 1) + (t->bx & 1)) * (f->b4_stride >> 1) +
                                        ((t->bx >> 1) + (t->by & 1))][pl] : t->pal[pl];
1137 1138 1139 1140 1141
                // see aomedia bug 2183 for why we use luma coordinates here
                for (int x = 0; x < bw4; x++)
                    memcpy(t->al_pal[0][bx4 + x][pl], pal, 16);
                for (int y = 0; y < bh4; y++)
                    memcpy(t->al_pal[1][by4 + y][pl], pal, 16);
1142
            }
1143 1144 1145
        } else { // see aomedia bug 2183 for why we reset this
            memset(&t->pal_sz_uv[1][by4], 0, bh4);
            memset(&t->pal_sz_uv[0][bx4], 0, bw4);
1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191
        }
        if ((f->frame_hdr.frame_type & 1) || f->frame_hdr.allow_intrabc) {
            memset(&t->a->tx[bx4], t_dim->lw, bw4);
            memset(&t->l.tx[by4], t_dim->lh, bh4);
            splat_intraref(f->mvs, f->b4_stride, t->by, t->bx, bs,
                           y_mode_nofilt);
        }
        if (f->frame_hdr.frame_type & 1) {
            memset(&t->l.comp_type[by4], COMP_INTER_NONE, bh4);
            memset(&t->a->comp_type[bx4], COMP_INTER_NONE, bw4);
            memset(&t->l.ref[0][by4], -1, bh4);
            memset(&t->a->ref[0][bx4], -1, bw4);
            memset(&t->l.ref[1][by4], -1, bh4);
            memset(&t->a->ref[1][bx4], -1, bw4);
            memset(&t->l.filter[0][by4], N_SWITCHABLE_FILTERS, bh4);
            memset(&t->a->filter[0][bx4], N_SWITCHABLE_FILTERS, bw4);
            memset(&t->l.filter[1][by4], N_SWITCHABLE_FILTERS, bh4);
            memset(&t->a->filter[1][bx4], N_SWITCHABLE_FILTERS, bw4);
        }
    } else if (!(f->frame_hdr.frame_type & 1)) {
        // intra block copy
        candidate_mv mvstack[8];
        int n_mvs;
        mv mvlist[2][2];
        av1_find_ref_mvs(mvstack, &n_mvs, mvlist, NULL,
                         (int[2]) { -1, -1 }, f->bw, f->bh,
                         bs, bp, t->by, t->bx, ts->tiling.col_start,
                         ts->tiling.col_end, ts->tiling.row_start,
                         ts->tiling.row_end, f->libaom_cm);

        if (mvlist[0][0].y | mvlist[0][0].x)
            b->mv[0] = mvlist[0][0];
        else if (mvlist[0][1].y | mvlist[0][1].x)
            b->mv[0] = mvlist[0][1];
        else {
            if (t->by - (16 << f->seq_hdr.sb128) < ts->tiling.row_start) {
                b->mv[0].y = 0;
                b->mv[0].x = -(512 << f->seq_hdr.sb128) - 2048;
            } else {
                b->mv[0].y = -(512 << f->seq_hdr.sb128);
                b->mv[0].x = 0;
            }
        }

        const struct mv ref = b->mv[0];
        read_mv_residual(t, &b->mv[0], &ts->cdf.dmv, 0);
1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247

        // clip intrabc motion vector to decoded parts of current tile
        int border_left = ts->tiling.col_start * 4;
        int border_top  = ts->tiling.row_start * 4;
        if (has_chroma) {
            if (bw4 < 2 &&  ss_hor)
                border_left += 4;
            if (bh4 < 2 &&  ss_ver)
                border_top  += 4;
        }
        int src_left   = t->bx * 4 + (b->mv[0].x >> 3);
        int src_top    = t->by * 4 + (b->mv[0].y >> 3);
        int src_right  = src_left + bw4 * 4;
        int src_bottom = src_top  + bh4 * 4;

        // check against left or right tile boundary and adjust if necessary
        if (src_left < border_left) {
            src_right += border_left - src_left;
            src_left  += border_left - src_left;
        } else if (src_right > ts->tiling.col_end * 4) {
            src_left  -= src_right - ts->tiling.col_end * 4;
            src_right -= src_right - ts->tiling.col_end * 4;
        }
        // check against top tile boundary and adjust if necessary
        if (src_top < border_top) {
            src_bottom += border_top - src_top;
            src_top    += border_top - src_top;
        }

        const int sbx = (t->bx >> (4 + f->seq_hdr.sb128)) << (6 + f->seq_hdr.sb128);
        const int sby = (t->by >> (4 + f->seq_hdr.sb128)) << (6 + f->seq_hdr.sb128);
        const int sb_size = 1 << (6 + f->seq_hdr.sb128);
        // check for overlap with current superblock
        if (src_bottom > sby && src_right > sbx) {
            if (src_top - border_top >= src_bottom - sby) {
                // if possible move src up into the previous suberblock row
                src_top    -= src_bottom - sby;
                src_bottom -= src_bottom - sby;
            } else if (src_left - border_left >= src_right - sbx) {
                // if possible move src left into the previous suberblock
                src_left  -= src_right - sbx;
                src_right -= src_right - sbx;
            }
        }
        // move src up if it is below current superblock row
        if (src_bottom > sby + sb_size) {
            src_top    -= src_bottom - (sby + sb_size);
            src_bottom -= src_bottom - (sby + sb_size);
        }
        // error out if mv still overlaps with the current superblock
        if (src_bottom > sby && src_right > sbx)
            return -1;

        b->mv[0].x = (src_left - t->bx * 4) * 8;
        b->mv[0].y = (src_top  - t->by * 4) * 8;

1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
        if (DEBUG_BLOCK_INFO)
            printf("Post-dmv[%d/%d,ref=%d/%d|%d/%d]: r=%d\n",
                   b->mv[0].y, b->mv[0].x, ref.y, ref.x,
                   mvlist[0][0].y, mvlist[0][0].x, ts->msac.rng);
        read_vartx_tree(t, b, bs, bx4, by4);

        // reconstruction
        if (f->frame_thread.pass == 1) {
            f->bd_fn.read_coef_blocks(t, bs, b);
        } else {
            f->bd_fn.recon_b_inter(t, bs, b);
        }

        splat_intrabc_mv(f->mvs, f->b4_stride, t->by, t->bx, bs, b->mv[0]);

        memset(&t->a->tx_intra[bx4], b_dim[2], bw4);
        memset(&t->l.tx_intra[by4], b_dim[3], bh4);
        memset(&t->l.mode[by4], DC_PRED, bh4);
        memset(&t->a->mode[bx4], DC_PRED, bw4);
        memset(&t->l.pal_sz[by4], 0, bh4);
        memset(&t->a->pal_sz[bx4], 0, bw4);
1269 1270 1271
        // see aomedia bug 2183 for why this is outside if (has_chroma)
        memset(&t->pal_sz_uv[1][by4], 0, bh4);
        memset(&t->pal_sz_uv[0][bx4], 0, bw4);
1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310
        if (has_chroma) {
            memset(&t->l.uvmode[cby4], DC_PRED, cbh4);
            memset(&t->a->uvmode[cbx4], DC_PRED, cbw4);
        }
    } else {
        // inter-specific mode/mv coding
        int is_comp, has_subpel_filter;

        if (b->skip_mode) {
            is_comp = 1;
        } else if (f->frame_hdr.switchable_comp_refs && imin(bw4, bh4) > 1) {
            const int ctx = get_comp_ctx(t->a, &t->l, by4, bx4,
                                         have_top, have_left);
            is_comp = msac_decode_bool_adapt(&ts->msac, ts->cdf.m.comp[ctx]);
            if (DEBUG_BLOCK_INFO)
                printf("Post-compflag[%d]: r=%d\n", is_comp, ts->msac.rng);
        } else {
            is_comp = 0;
        }

        if (b->skip_mode) {
            b->ref[0] = f->frame_hdr.skip_mode_refs[0];
            b->ref[1] = f->frame_hdr.skip_mode_refs[1];
            b->comp_type = COMP_INTER_AVG;
            b->inter_mode = NEARESTMV_NEARESTMV;
            b->drl_idx = 0;
            has_subpel_filter = 0;

            candidate_mv mvstack[8];
            int n_mvs, ctx;
            mv mvlist[2][2];
            av1_find_ref_mvs(mvstack, &n_mvs, mvlist, &ctx,
                             (int[2]) { b->ref[0], b->ref[1] }, f->bw, f->bh,
                             bs, bp, t->by, t->bx, ts->tiling.col_start,
                             ts->tiling.col_end, ts->tiling.row_start,
                             ts->tiling.row_end, f->libaom_cm);

            b->mv[0] = mvstack[0].this_mv;
            b->mv[1] = mvstack[0].comp_mv;
1311 1312
            fix_mv_precision(&f->frame_hdr, &b->mv[0]);
            fix_mv_precision(&f->frame_hdr, &b->mv[1]);
1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393
            if (DEBUG_BLOCK_INFO)
                printf("Post-skipmodeblock[mv=1:y=%d,x=%d,2:y=%d,x=%d,refs=%d+%d\n",
                       b->mv[0].y, b->mv[0].x, b->mv[1].y, b->mv[1].x,
                       b->ref[0], b->ref[1]);
        } else if (is_comp) {
            const int dir_ctx = get_comp_dir_ctx(t->a, &t->l, by4, bx4,
                                                 have_top, have_left);
            if (msac_decode_bool_adapt(&ts->msac, ts->cdf.m.comp_dir[dir_ctx])) {
                // bidir - first reference (fw)
                const int ctx1 = av1_get_fwd_ref_ctx(t->a, &t->l, by4, bx4,
                                                     have_top, have_left);
                if (msac_decode_bool_adapt(&ts->msac,
                                           ts->cdf.m.comp_fwd_ref[0][ctx1]))
                {
                    const int ctx2 = av1_get_fwd_ref_2_ctx(t->a, &t->l, by4, bx4,
                                                           have_top, have_left);
                    b->ref[0] = 2 + msac_decode_bool_adapt(&ts->msac,
                                            ts->cdf.m.comp_fwd_ref[2][ctx2]);
                } else {
                    const int ctx2 = av1_get_fwd_ref_1_ctx(t->a, &t->l, by4, bx4,
                                                           have_top, have_left);
                    b->ref[0] = msac_decode_bool_adapt(&ts->msac,
                                            ts->cdf.m.comp_fwd_ref[1][ctx2]);
                }

                // second reference (bw)
                const int ctx3 = av1_get_bwd_ref_ctx(t->a, &t->l, by4, bx4,
                                                     have_top, have_left);
                if (msac_decode_bool_adapt(&ts->msac,
                                           ts->cdf.m.comp_bwd_ref[0][ctx3]))
                {
                    b->ref[1] = 6;
                } else {
                    const int ctx4 = av1_get_bwd_ref_1_ctx(t->a, &t->l, by4, bx4,
                                                           have_top, have_left);
                    b->ref[1] = 4 + msac_decode_bool_adapt(&ts->msac,
                                           ts->cdf.m.comp_bwd_ref[1][ctx4]);
                }
            } else {
                // unidir
                const int uctx_p = av1_get_uni_p_ctx(t->a, &t->l, by4, bx4,
                                                     have_top, have_left);
                if (msac_decode_bool_adapt(&ts->msac,
                                           ts->cdf.m.comp_uni_ref[0][uctx_p]))
                {
                    b->ref[0] = 4;
                    b->ref[1] = 6;
                } else {
                    const int uctx_p1 = av1_get_uni_p1_ctx(t->a, &t->l, by4, bx4,
                                                           have_top, have_left);
                    b->ref[0] = 0;
                    b->ref[1] = 1 + msac_decode_bool_adapt(&ts->msac,
                                           ts->cdf.m.comp_uni_ref[1][uctx_p1]);
                    if (b->ref[1] == 2) {
                        const int uctx_p2 = av1_get_uni_p2_ctx(t->a, &t->l, by4, bx4,
                                                               have_top, have_left);
                        b->ref[1] += msac_decode_bool_adapt(&ts->msac,
                                           ts->cdf.m.comp_uni_ref[2][uctx_p2]);
                    }
                }
            }
            if (DEBUG_BLOCK_INFO)
                printf("Post-refs[%d/%d]: r=%d\n",
                       b->ref[0], b->ref[1], ts->msac.rng);

            candidate_mv mvstack[8];
            int n_mvs, ctx;
            mv mvlist[2][2];
            av1_find_ref_mvs(mvstack, &n_mvs, mvlist, &ctx,
                             (int[2]) { b->ref[0], b->ref[1] }, f->bw, f->bh,
                             bs, bp, t->by, t->bx, ts->tiling.col_start,
                             ts->tiling.col_end, ts->tiling.row_start,
                             ts->tiling.row_end, f->libaom_cm);

            b->inter_mode = msac_decode_symbol_adapt(&ts->msac,
                                             ts->cdf.m.comp_inter_mode[ctx],
                                             N_COMP_INTER_PRED_MODES);
            if (DEBUG_BLOCK_INFO)
                printf("Post-compintermode[%d,ctx=%d,n_mvs=%d]: r=%d\n",
                       b->inter_mode, ctx, n_mvs, ts->msac.rng);

1394
            const uint8_t *const im = dav1d_comp_inter_pred_modes[b->inter_mode];
1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431
            b->drl_idx = 0;
            if (b->inter_mode == NEWMV_NEWMV) {
                if (n_mvs > 1) {
                    const int drl_ctx_v1 = get_drl_context(mvstack, 0);
                    b->drl_idx += msac_decode_bool_adapt(&ts->msac,
                                             ts->cdf.m.drl_bit[drl_ctx_v1]);
                    if (b->drl_idx == 1 && n_mvs > 2) {
                        const int drl_ctx_v2 = get_drl_context(mvstack, 1);
                        b->drl_idx += msac_decode_bool_adapt(&ts->msac,
                                             ts->cdf.m.drl_bit[drl_ctx_v2]);
                    }
                    if (DEBUG_BLOCK_INFO)
                        printf("Post-drlidx[%d,n_mvs=%d]: r=%d\n",
                               b->drl_idx, n_mvs, ts->msac.rng);
                }
            } else if (im[0] == NEARMV || im[1] == NEARMV) {
                b->drl_idx = 1;
                if (n_mvs > 2) {
                    const int drl_ctx_v2 = get_drl_context(mvstack, 1);
                    b->drl_idx += msac_decode_bool_adapt(&ts->msac,
                                             ts->cdf.m.drl_bit[drl_ctx_v2]);
                    if (b->drl_idx == 2 && n_mvs > 3) {
                        const int drl_ctx_v3 = get_drl_context(mvstack, 2);
                        b->drl_idx += msac_decode_bool_adapt(&ts->msac,
                                             ts->cdf.m.drl_bit[drl_ctx_v3]);
                    }
                    if (DEBUG_BLOCK_INFO)
                        printf("Post-drlidx[%d,n_mvs=%d]: r=%d\n",
                               b->drl_idx, n_mvs, ts->msac.rng);
                }
            }

#define assign_comp_mv(idx, pfx) \
            switch (im[idx]) { \
            case NEARMV: \
            case NEARESTMV: \
                b->mv[idx] = mvstack[b->drl_idx].pfx##_mv; \
1432
                fix_mv_precision(&f->frame_hdr, &b->mv[idx]); \
1433 1434 1435 1436 1437 1438
                break; \
            case GLOBALMV: \
                has_subpel_filter |= \
                    f->frame_hdr.gmv[b->ref[idx]].type == WM_TYPE_TRANSLATION; \
                b->mv[idx] = get_gmv_2d(&f->frame_hdr.gmv[b->ref[idx]], \
                                        t->bx, t->by, bw4, bh4, &f->frame_hdr); \
1439
                fix_mv_precision(&f->frame_hdr, &b->mv[idx]); \
1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489
                break; \
            case NEWMV: \
                b->mv[idx] = mvstack[b->drl_idx].pfx##_mv; \
                read_mv_residual(t, &b->mv[idx], &ts->cdf.mv, \
                                 !f->frame_hdr.force_integer_mv); \
                break; \
            }
            has_subpel_filter = imin(bw4, bh4) == 1 ||
                                b->inter_mode != GLOBALMV_GLOBALMV;
            assign_comp_mv(0, this);
            assign_comp_mv(1, comp);
#undef assign_comp_mv
            if (DEBUG_BLOCK_INFO)
                printf("Post-residual_mv[1:y=%d,x=%d,2:y=%d,x=%d]: r=%d\n",
                       b->mv[0].y, b->mv[0].x, b->mv[1].y, b->mv[1].x,
                       ts->msac.rng);

            // jnt_comp vs. seg vs. wedge
            int is_segwedge = 0;
            if (f->seq_hdr.masked_compound) {
                const int mask_ctx = get_mask_comp_ctx(t->a, &t->l, by4, bx4);

                is_segwedge = msac_decode_bool_adapt(&ts->msac,
                                                 ts->cdf.m.mask_comp[mask_ctx]);
                if (DEBUG_BLOCK_INFO)
                    printf("Post-segwedge_vs_jntavg[%d,ctx=%d]: r=%d\n",
                           is_segwedge, mask_ctx, ts->msac.rng);
            }

            if (!is_segwedge) {
                if (f->seq_hdr.jnt_comp) {
                    const int jnt_ctx =
                        get_jnt_comp_ctx(f->seq_hdr.order_hint_n_bits,
                                         f->cur.p.poc, f->refp[b->ref[0]].p.poc,
                                         f->refp[b->ref[1]].p.poc, t->a, &t->l,
                                         by4, bx4);
                    b->comp_type = COMP_INTER_WEIGHTED_AVG +
                        msac_decode_bool_adapt(&ts->msac,
                                               ts->cdf.m.jnt_comp[jnt_ctx]);
                    if (DEBUG_BLOCK_INFO)
                        printf("Post-jnt_comp[%d,ctx=%d[ac:%d,ar:%d,lc:%d,lr:%d]]: r=%d\n",
                               b->comp_type == COMP_INTER_AVG,
                               jnt_ctx, t->a->comp_type[bx4], t->a->ref[0][bx4],
                               t->l.comp_type[by4], t->l.ref[0][by4],
                               ts->msac.rng);
                } else {
                    b->comp_type = COMP_INTER_AVG;
                }
            } else {
                if (wedge_allowed_mask & (1 << bs)) {
1490
                    const int ctx = dav1d_wedge_ctx_lut[bs];
1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558
                    b->comp_type = COMP_INTER_WEDGE -
                        msac_decode_bool_adapt(&ts->msac,
                                               ts->cdf.m.wedge_comp[ctx]);
                    if (b->comp_type == COMP_INTER_WEDGE)
                        b->wedge_idx = msac_decode_symbol_adapt(&ts->msac,
                                                ts->cdf.m.wedge_idx[ctx], 16);
                } else {
                    b->comp_type = COMP_INTER_SEG;
                }
                b->mask_sign = msac_decode_bool(&ts->msac, 128 << 7);
                if (DEBUG_BLOCK_INFO)
                    printf("Post-seg/wedge[%d,wedge_idx=%d,sign=%d]: r=%d\n",
                           b->comp_type == COMP_INTER_WEDGE,
                           b->wedge_idx, b->mask_sign, ts->msac.rng);
            }
        } else {
            b->comp_type = COMP_INTER_NONE;

            // ref
            const int ctx1 = av1_get_ref_ctx(t->a, &t->l, by4, bx4,
                                             have_top, have_left);
            if (msac_decode_bool_adapt(&ts->msac, ts->cdf.m.ref[0][ctx1])) {
                const int ctx2 = av1_get_ref_2_ctx(t->a, &t->l, by4, bx4,
                                                   have_top, have_left);
                if (msac_decode_bool_adapt(&ts->msac, ts->cdf.m.ref[1][ctx2])) {
                    b->ref[0] = 6;
                } else {
                    const int ctx3 = av1_get_ref_6_ctx(t->a, &t->l, by4, bx4,
                                                       have_top, have_left);
                    b->ref[0] = 4 + msac_decode_bool_adapt(&ts->msac,
                                                       ts->cdf.m.ref[5][ctx3]);
                }
            } else {
                const int ctx2 = av1_get_ref_3_ctx(t->a, &t->l, by4, bx4,
                                                   have_top, have_left);
                if (msac_decode_bool_adapt(&ts->msac, ts->cdf.m.ref[2][ctx2])) {
                    const int ctx3 = av1_get_ref_5_ctx(t->a, &t->l, by4, bx4,
                                                       have_top, have_left);
                    b->ref[0] = 2 + msac_decode_bool_adapt(&ts->msac,
                                                       ts->cdf.m.ref[4][ctx3]);
                } else {
                    const int ctx3 = av1_get_ref_4_ctx(t->a, &t->l, by4, bx4,
                                                       have_top, have_left);
                    b->ref[0] = msac_decode_bool_adapt(&ts->msac,
                                                       ts->cdf.m.ref[3][ctx3]);
                }
            }
            b->ref[1] = -1;
            if (DEBUG_BLOCK_INFO)
                printf("Post-ref[%d]: r=%d\n", b->ref[0], ts->msac.rng);

            candidate_mv mvstack[8];
            int n_mvs, ctx;
            mv mvlist[2][2];
            av1_find_ref_mvs(mvstack, &n_mvs, mvlist, &ctx,
                             (int[2]) { b->ref[0], -1 }, f->bw, f->bh, bs, bp,
                             t->by, t->bx, ts->tiling.col_start,
                             ts->tiling.col_end, ts->tiling.row_start,
                             ts->tiling.row_end, f->libaom_cm);

            // mode parsing and mv derivation from ref_mvs
            if (msac_decode_bool_adapt(&ts->msac, ts->cdf.m.newmv_mode[ctx & 7])) {
                if (!msac_decode_bool_adapt(&ts->msac,
                                        ts->cdf.m.globalmv_mode[(ctx >> 3) & 1]))
                {
                    b->inter_mode = GLOBALMV;
                    b->mv[0] = get_gmv_2d(&f->frame_hdr.gmv[b->ref[0]],
                                          t->bx, t->by, bw4, bh4, &f->frame_hdr);
1559
                    fix_mv_precision(&f->frame_hdr, &b->mv[0]);
1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587
                    has_subpel_filter = imin(bw4, bh4) == 1 ||
                        f->frame_hdr.gmv[b->ref[0]].type == WM_TYPE_TRANSLATION;
                } else {
                    has_subpel_filter = 1;
                    if (msac_decode_bool_adapt(&ts->msac,
                                       ts->cdf.m.refmv_mode[(ctx >> 4) & 15]))
                    {
                        b->inter_mode = NEARMV;
                        b->drl_idx = 1;
                        if (n_mvs > 2) {
                            const int drl_ctx_v2 = get_drl_context(mvstack, 1);
                            b->drl_idx += msac_decode_bool_adapt(&ts->msac,
                                                 ts->cdf.m.drl_bit[drl_ctx_v2]);
                            if (b->drl_idx == 2 && n_mvs > 3) {
                                const int drl_ctx_v3 =
                                    get_drl_context(mvstack, 2);
                                b->drl_idx += msac_decode_bool_adapt(&ts->msac,
                                                 ts->cdf.m.drl_bit[drl_ctx_v3]);
                            }
                        }
                    } else {
                        b->inter_mode = NEARESTMV;
                        b->drl_idx = 0;
                    }
                    if (b->drl_idx >= 2) {
                        b->mv[0] = mvstack[b->drl_idx].this_mv;
                    } else {
                        b->mv[0] = mvlist[0][b->drl_idx];
1588
                        fix_mv_precision(&f->frame_hdr, &b->mv[0]);
1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613
                    }
                }

                if (DEBUG_BLOCK_INFO)
                    printf("Post-intermode[%d,drl=%d,mv=y:%d,x:%d,n_mvs=%d]: r=%d\n",
                           b->inter_mode, b->drl_idx, b->mv[0].y, b->mv[0].x, n_mvs,
                           ts->msac.rng);
            } else {
                has_subpel_filter = 1;
                b->inter_mode = NEWMV;
                b->drl_idx = 0;
                if (n_mvs > 1) {
                    const int drl_ctx_v1 = get_drl_context(mvstack, 0);
                    b->drl_idx += msac_decode_bool_adapt(&ts->msac,
                                                 ts->cdf.m.drl_bit[drl_ctx_v1]);
                    if (b->drl_idx == 1 && n_mvs > 2) {
                        const int drl_ctx_v2 = get_drl_context(mvstack, 1);
                        b->drl_idx += msac_decode_bool_adapt(&ts->msac,
                                                 ts->cdf.m.drl_bit[drl_ctx_v2]);
                    }
                }
                if (n_mvs > 1) {
                    b->mv[0] = mvstack[b->drl_idx].this_mv;
                } else {
                    b->mv[0] = mvlist[0][0];
1614
                    fix_mv_precision(&f->frame_hdr, &b->mv[0]);
1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626
                }
                if (DEBUG_BLOCK_INFO)
                    printf("Post-intermode[%d,drl=%d]: r=%d\n",
                           b->inter_mode, b->drl_idx, ts->msac.rng);
                read_mv_residual(t, &b->mv[0], &ts->cdf.mv,
                                 !f->frame_hdr.force_integer_mv);
                if (DEBUG_BLOCK_INFO)
                    printf("Post-residualmv[mv=y:%d,x:%d]: r=%d\n",
                           b->mv[0].y, b->mv[0].x, ts->msac.rng);
            }

            // interintra flags
1627
            const int ii_sz_grp = dav1d_ymode_size_context[bs];
1628 1629 1630 1631 1632 1633 1634
            if (f->seq_hdr.inter_intra &&
                interintra_allowed_mask & (1 << bs) &&
                msac_decode_bool_adapt(&ts->msac, ts->cdf.m.interintra[ii_sz_grp]))
            {
                b->interintra_mode = msac_decode_symbol_adapt(&ts->msac,
                                          ts->cdf.m.interintra_mode[ii_sz_grp],
                                          N_INTER_INTRA_PRED_MODES);
1635
                const int wedge_ctx = dav1d_wedge_ctx_lut[bs];
1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734
                b->interintra_type = INTER_INTRA_BLEND +
                    msac_decode_bool_adapt(&ts->msac,
                                           ts->cdf.m.interintra_wedge[wedge_ctx]);
                if (b->interintra_type == INTER_INTRA_WEDGE)
                    b->wedge_idx = msac_decode_symbol_adapt(&ts->msac,
                                            ts->cdf.m.wedge_idx[wedge_ctx], 16);
            } else {
                b->interintra_type = INTER_INTRA_NONE;
            }
            if (DEBUG_BLOCK_INFO && f->seq_hdr.inter_intra &&
                interintra_allowed_mask & (1 << bs))
            {
                printf("Post-interintra[t=%d,m=%d,w=%d]: r=%d\n",
                       b->interintra_type, b->interintra_mode,
                       b->wedge_idx, ts->msac.rng);
            }

            // motion variation
            if (f->frame_hdr.switchable_motion_mode &&
                b->interintra_type == INTER_INTRA_NONE && imin(bw4, bh4) >= 2 &&
                // is not warped global motion
                !(!f->frame_hdr.force_integer_mv && b->inter_mode == GLOBALMV &&
                  f->frame_hdr.gmv[b->ref[0]].type > WM_TYPE_TRANSLATION) &&
                // has overlappable neighbours
                ((have_left && findoddzero(&t->l.intra[by4 + 1], h4 >> 1)) ||
                 (have_top && findoddzero(&t->a->intra[bx4 + 1], w4 >> 1))))
            {
                // reaching here means the block allows obmc - check warp by
                // finding matching-ref blocks in top/left edges
                uint64_t mask[2] = { 0, 0 };
                find_matching_ref(t, intra_edge_flags, bw4, bh4, w4, h4,
                                  have_left, have_top, b->ref[0], mask);
                const int allow_warp = !f->frame_hdr.force_integer_mv &&
                    f->frame_hdr.warp_motion && (mask[0] | mask[1]);

                b->motion_mode = allow_warp ?
                    msac_decode_symbol_adapt(&ts->msac, ts->cdf.m.motion_mode[bs], 3) :
                    msac_decode_bool_adapt(&ts->msac, ts->cdf.m.obmc[bs]);
                if (b->motion_mode == MM_WARP) {
                    has_subpel_filter = 0;
                    derive_warpmv(t, bw4, bh4, mask, b->mv[0], &t->warpmv);
#define signabs(v) v < 0 ? '-' : ' ', abs(v)
                    if (DEBUG_BLOCK_INFO)
                        printf("[ %c%x %c%x %c%x\n  %c%x %c%x %c%x ]\n"
                               "alpha=%c%x, beta=%c%x, gamma=%c%x, delta=%c%x\n",
                               signabs(t->warpmv.matrix[0]),
                               signabs(t->warpmv.matrix[1]),
                               signabs(t->warpmv.matrix[2]),
                               signabs(t->warpmv.matrix[3]),
                               signabs(t->warpmv.matrix[4]),
                               signabs(t->warpmv.matrix[5]),
                               signabs(t->warpmv.alpha),
                               signabs(t->warpmv.beta),
                               signabs(t->warpmv.gamma),
                               signabs(t->warpmv.delta));
#undef signabs
                }

                if (DEBUG_BLOCK_INFO)
                    printf("Post-motionmode[%d]: r=%d [mask: 0x%" PRIu64 "x/0x%"
                           PRIu64 "x]\n", b->motion_mode, ts->msac.rng, mask[0],
                            mask[1]);
            } else {
                b->motion_mode = MM_TRANSLATION;
            }
        }

        // subpel filter
        enum FilterMode filter[2];
        if (f->frame_hdr.subpel_filter_mode == FILTER_SWITCHABLE) {
            if (has_subpel_filter) {
                const int comp = b->comp_type != COMP_INTER_NONE;
                const int ctx1 = get_filter_ctx(t->a, &t->l, comp, 0, b->ref[0],
                                                by4, bx4);
                filter[0] = msac_decode_symbol_adapt(&ts->msac,
                    ts->cdf.m.filter[0][ctx1], N_SWITCHABLE_FILTERS);
                if (f->seq_hdr.dual_filter) {
                    const int ctx2 = get_filter_ctx(t->a, &t->l, comp, 1,
                                                    b->ref[0], by4, bx4);
                    if (DEBUG_BLOCK_INFO)
                        printf("Post-subpel_filter1[%d,ctx=%d]: r=%d\n",
                               filter[0], ctx1, ts->msac.rng);
                    filter[1] = msac_decode_symbol_adapt(&ts->msac,
                        ts->cdf.m.filter[1][ctx2], N_SWITCHABLE_FILTERS);
                    if (DEBUG_BLOCK_INFO)
                        printf("Post-subpel_filter2[%d,ctx=%d]: r=%d\n",
                               filter[1], ctx2, ts->msac.rng);
                } else {
                    filter[1] = filter[0];
                    if (DEBUG_BLOCK_INFO)
                        printf("Post-subpel_filter[%d,ctx=%d]: r=%d\n",
                               filter[0], ctx1, ts->msac.rng);
                }
            } else {
                filter[0] = filter[1] = FILTER_8TAP_REGULAR;
            }
        } else {
            filter[0] = filter[1] = f->frame_hdr.subpel_filter_mode;
        }
1735
        b->filter2d = dav1d_filter_2d[filter[1]][filter[0]];
1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751

        read_vartx_tree(t, b, bs, bx4, by4);

        // reconstruction
        if (f->frame_thread.pass == 1) {
            f->bd_fn.read_coef_blocks(t, bs, b);
        } else {
            f->bd_fn.recon_b_inter(t, bs, b);
        }

        const int is_globalmv =
            b->inter_mode == (is_comp ? GLOBALMV_GLOBALMV : GLOBALMV);
        const uint8_t (*const lf_lvls)[8][2] = (const uint8_t (*)[8][2])
            &ts->lflvl[b->seg_id][0][b->ref[0] + 1][!is_globalmv];
        dav1d_create_lf_mask_inter(t->lf_mask, f->lf.level, f->b4_stride,
                                   &f->frame_hdr, lf_lvls, t->bx, t->by,
1752 1753 1754
                                   (f->cur.p.p.w + 3) >> 2,
                                   (f->cur.p.p.h + 3) >> 2,
                                   b->skip, bs, b->tx_split,
1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771
                                   b->uvtx, f->cur.p.p.layout,
                                   &t->a->tx_lpf_y[bx4], &t->l.tx_lpf_y[by4],
                                   has_chroma ? &t->a->tx_lpf_uv[cbx4] : NULL,
                                   has_chroma ? &t->l.tx_lpf_uv[cby4] : NULL);

        // context updates
        if (is_comp) {
            splat_tworef_mv(f->mvs, f->b4_stride, t->by, t->bx, bs,
                            b->inter_mode, b->ref[0], b->ref[1],
                            b->mv[0], b->mv[1]);
        } else {
            splat_oneref_mv(f->mvs, f->b4_stride, t->by, t->bx, bs,
                            b->inter_mode, b->ref[0], b->mv[0],
                            b->interintra_type);
        }
        memset(&t->l.pal_sz[by4], 0, bh4);
        memset(&t->a->pal_sz[bx4], 0, bw4);
1772 1773 1774
        // see aomedia bug 2183 for why this is outside if (has_chroma)
        memset(&t->pal_sz_uv[1][by4], 0, bh4);
        memset(&t->pal_sz_uv[0][bx4], 0, bw4);
1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818
        if (has_chroma) {
            memset(&t->l.uvmode[cby4], DC_PRED, cbh4);
            memset(&t->a->uvmode[cbx4], DC_PRED, cbw4);
        }
        memset(&t->a->tx_intra[bx4], b_dim[2], bw4);
        memset(&t->l.tx_intra[by4], b_dim[3], bh4);
        memset(&t->l.comp_type[by4], b->comp_type, bh4);
        memset(&t->a->comp_type[bx4], b->comp_type, bw4);
        memset(&t->l.filter[0][by4], filter[0], bh4);
        memset(&t->a->filter[0][bx4], filter[0], bw4);
        memset(&t->l.filter[1][by4], filter[1], bh4);
        memset(&t->a->filter[1][bx4], filter[1], bw4);
        memset(&t->l.mode[by4], b->inter_mode, bh4);
        memset(&t->a->mode[bx4], b->inter_mode, bw4);
        memset(&t->l.ref[0][by4], b->ref[0], bh4);
        memset(&t->a->ref[0][bx4], b->ref[0], bw4);
        memset(&t->l.ref[1][by4], b->ref[1], bh4);
        memset(&t->a->ref[1][bx4], b->ref[1], bw4);
    }

    // update contexts
    if (f->frame_hdr.segmentation.enabled &&
        f->frame_hdr.segmentation.update_map)
    {
        uint8_t *seg_ptr = &f->cur_segmap[t->by * f->b4_stride + t->bx];
        for (int y = 0; y < bh4; y++) {
            memset(seg_ptr, b->seg_id, bw4);
            seg_ptr += f->b4_stride;
        }
    }
    memset(&t->l.seg_pred[by4], seg_pred, bh4);
    memset(&t->a->seg_pred[bx4], seg_pred, bw4);
    memset(&t->l.skip_mode[by4], b->skip_mode, bh4);
    memset(&t->a->skip_mode[bx4], b->skip_mode, bw4);
    memset(&t->l.intra[by4], b->intra, bh4);
    memset(&t->a->intra[bx4], b->intra, bw4);
    memset(&t->l.skip[by4], b->skip, bh4);
    memset(&t->a->skip[bx4], b->skip, bw4);
    if (!b->skip) {
        uint32_t *noskip_mask = &t->lf_mask->noskip_mask[by4];
        const unsigned mask = ((1ULL << bw4) - 1) << bx4;
        for (int y = 0; y < bh4; y++)
            *noskip_mask++ |= mask;
    }
1819 1820

    return 0;
1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867
}

static int decode_sb(Dav1dTileContext *const t, const enum BlockLevel bl,
                     const EdgeNode *const node)
{
    const Dav1dFrameContext *const f = t->f;
    const int hsz = 16 >> bl;
    const int have_h_split = f->bw > t->bx + hsz;
    const int have_v_split = f->bh > t->by + hsz;

    if (!have_h_split && !have_v_split) {
        assert(bl < BL_8X8);
        return decode_sb(t, bl + 1, ((const EdgeBranch *) node)->split[0]);
    }

    uint16_t *pc;
    enum BlockPartition bp;
    int ctx, bx8, by8;
    if (f->frame_thread.pass != 2) {
        if (0 && bl == BL_64X64)
            printf("poc=%d,y=%d,x=%d,bl=%d,r=%d\n",
                   f->frame_hdr.frame_offset, t->by, t->bx, bl, t->ts->msac.rng);
        bx8 = (t->bx & 31) >> 1;
        by8 = (t->by & 31) >> 1;
        ctx = get_partition_ctx(t->a, &t->l, bl, by8, bx8);
        pc = t->ts->cdf.m.partition[bl][ctx];
    }

    if (have_h_split && have_v_split) {
        if (f->frame_thread.pass == 2) {
            const Av1Block *const b = &f->frame_thread.b[t->by * f->b4_stride + t->bx];
            bp = b->bl == bl ? b->bp : PARTITION_SPLIT;
        } else {
            const unsigned n_part = bl == BL_8X8 ? N_SUB8X8_PARTITIONS :
                bl == BL_128X128 ? N_PARTITIONS - 2 : N_PARTITIONS;
            bp = msac_decode_symbol_adapt(&t->ts->msac, pc, n_part);
            if (f->cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I422 &&
                (bp == PARTITION_V || bp == PARTITION_V4 ||
                 bp == PARTITION_T_LEFT_SPLIT || bp == PARTITION_T_RIGHT_SPLIT))
            {
                return 1;
            }
            if (DEBUG_BLOCK_INFO)
                printf("poc=%d,y=%d,x=%d,bl=%d,ctx=%d,bp=%d: r=%d\n",
                       f->frame_hdr.frame_offset, t->by, t->bx, bl, ctx, bp,
                       t->ts->msac.rng);
        }
1868
        const uint8_t *const b = dav1d_block_sizes[bl][bp];
1869 1870 1871

        switch (bp) {
        case PARTITION_NONE:
1872 1873
            if (decode_b(t, bl, b[0], PARTITION_NONE, node->o))
                return -1;
1874 1875
            break;
        case PARTITION_H:
1876 1877
            if (decode_b(t, bl, b[0], PARTITION_H, node->h[0]))
                return -1;
1878
            t->by += hsz;
1879 1880
            if (decode_b(t, bl, b[0], PARTITION_H, node->h[1]))
                return -1;
1881 1882 1883
            t->by -= hsz;
            break;
        case PARTITION_V:
1884 1885
            if (decode_b(t, bl, b[0], PARTITION_V, node->v[0]))
                return -1;
1886
            t->bx += hsz;
1887 1888
            if (decode_b(t, bl, b[0], PARTITION_V, node->v[1]))
                return -1;
1889 1890 1891 1892 1893 1894
            t->bx -= hsz;
            break;
        case PARTITION_SPLIT:
            if (bl == BL_8X8) {
                const EdgeTip *const tip = (const EdgeTip *) node;
                assert(hsz == 1);
1895 1896
                if (decode_b(t, bl, BS_4x4, PARTITION_SPLIT, tip->split[0]))
                    return -1;
1897 1898
                const enum Filter2d tl_filter = t->tl_4x4_filter;
                t->bx++;
1899 1900
                if (decode_b(t, bl, BS_4x4, PARTITION_SPLIT, tip->split[1]))
                    return -1;
1901 1902
                t->bx--;
                t->by++;
1903 1904
                if (decode_b(t, bl, BS_4x4, PARTITION_SPLIT, tip->split[2]))
                    return -1;
1905 1906
                t->bx++;
                t->tl_4x4_filter = tl_filter;
1907 1908
                if (decode_b(t, bl, BS_4x4, PARTITION_SPLIT, tip->split[3]))
                    return -1;
1909 1910 1911 1912
                t->bx--;
                t->by--;
            } else {
                const EdgeBranch *const branch = (const EdgeBranch *) node;
1913 1914
                if (decode_sb(t, bl + 1, branch->split[0]))
                    return 1;
1915
                t->bx += hsz;
1916 1917
                if (decode_sb(t, bl + 1, branch->split[1]))
                    return 1;
1918 1919
                t->bx -= hsz;
                t->by += hsz;
1920 1921
                if (decode_sb(t, bl + 1, branch->split[2]))
                    return 1;
1922
                t->bx += hsz;
1923 1924
                if (decode_sb(t, bl + 1, branch->split[3]))
                    return 1;
1925 1926 1927 1928 1929 1930
                t->bx -= hsz;
                t->by -= hsz;
            }
            break;
        case PARTITION_T_TOP_SPLIT: {
            const EdgeBranch *const branch = (const EdgeBranch *) node;
1931 1932
            if (decode_b(t, bl, b[0], PARTITION_T_TOP_SPLIT, branch->tts[0]))
                return -1;
1933
            t->bx += hsz;
1934 1935
            if (decode_b(t, bl, b[0], PARTITION_T_TOP_SPLIT, branch->tts[1]))
                return -1;
1936 1937
            t->bx -= hsz;
            t->by += hsz;
1938 1939
            if (decode_b(t, bl, b[1], PARTITION_T_TOP_SPLIT, branch->tts[2]))
                return -1;