still265 0.6.1

Pure-Rust HEVC/H.265 still-image encoder with BPG and HEIC support
Documentation
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
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
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
345
346
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
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
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
//! 8x8 PartNxN search path.

use crate::internal::bpg_hevc_decode::hevc::slice::IntraPredMode;

use crate::cabac::CabacEstimator;
use crate::encoder::Encoder;
use crate::primitives::intra_angs;
use crate::primitives::{satd_u8, satd_u16};

use super::canvas::CanvasSaved;
use super::depth::StillSearchDepth;
use super::emit;
use super::eval::{BlockTrial, ResidualPricingMode};
use super::ledger::{StillSearchLedger, WorkBucket};
use super::plan::{CuPlan, NxnInfoPlan, ParentChromaPlan, TtPlan};
use super::price::luma_mode_bits;
use super::rough::build_luma_shortlist;
use super::source::CtuSourceCache;
use crate::effort::{AngularFamily, ModeClass, ModeCost};
use crate::encoder::types::{CHROMA_DM_IDX, chroma_pred_mode};

/// Rough-mode cache for the four PartNxN PUs. Uses a fixed-size array
/// large enough for any shortlist policy max_modes + MPMs.
const NXN_MAX_SHORTLIST: usize = 16;

pub(super) struct NxnRoughSet {
    shortlist: [[u8; NXN_MAX_SHORTLIST]; 4],
    len: [usize; 4],
}

impl NxnRoughSet {
    fn new() -> Self {
        Self {
            shortlist: [[0; NXN_MAX_SHORTLIST]; 4],
            len: [0; 4],
        }
    }

    fn set(&mut self, pu: usize, modes: &[u8]) {
        self.len[pu] = modes.len();
        self.shortlist[pu][..modes.len()].copy_from_slice(modes);
    }

    fn modes(&self, pu: usize) -> &[u8] {
        &self.shortlist[pu][..self.len[pu]]
    }
}

impl<S> StillSearchDepth<S>
where
    S: CtuSourceCache,
{
    /// Build and price an 8x8 `PartNxN` candidate. Luma is coded as four 4x4
    /// PUs/TUs in decoder z-order; each PU's selected recon is pushed before the
    /// next PU so legal intra references see prior siblings through the overlay.
    /// For Phase 7 this supports 4:2:0 parent chroma, 4:4:4 per-PU chroma,
    /// and grayscale. 4:2:2 stacked parent chroma is still deferred.
    pub(super) fn decide_cu_part_nxn(
        &mut self,
        state: &mut Encoder<'_>,
        x0: u32,
        y0: u32,
        ct_depth: u8,
        lambda: f64,
    ) -> (CuPlan, f64) {
        state.stats.partnxn_cu_trials += 1;
        state.set_ct_depth(x0, y0, 3, ct_depth);
        if state.aq.active {
            state.aq_set_cu_qp(x0, y0);
        }

        let pu_xy = [(x0, y0), (x0 + 4, y0), (x0, y0 + 4), (x0 + 4, y0 + 4)];
        let mut luma_modes = [IntraPredMode::Dc.as_u8(); 4];
        let mut mpms = [[IntraPredMode::Dc; 3]; 4];
        let mut kids = Vec::with_capacity(4);
        let mut cost = 0.0;
        let mut rough_set = NxnRoughSet::new();

        let evolve = super::env::ctx_evolve_search();
        for (pu, &(px, py)) in pu_xy.iter().enumerate() {
            let mpm = super::intra_mpm(state, px, py);
            mpms[pu] = mpm;
            self.fill_nxn_rough_set(state, px, py, pu, mpm, lambda, &mut rough_set);
            let (mode, luma, recon, pu_cost) =
                self.eval_nxn4_batch(state, px, py, pu, mpm, lambda, &rough_set);
            self.overlay.reattach(recon);
            luma_modes[pu] = mode;
            state.store_mode(px, py, 2, mode);
            cost += pu_cost;
            if evolve {
                // Commit the winning PU's luma syntax (mode trials never
                // commit): CBF bin + residual context evolution, so later
                // PUs price from post-PU contexts like x265's coder does.
                self.workspace.price_cur.models[crate::contexts::ctx::CBF_LUMA]
                    .update(luma.cbf as u8);
                if luma.cbf {
                    if let Some(id) = luma.coeff {
                        let scan = crate::residual::get_scan_order(2, mode, 0, state.cat);
                        let _ = crate::residual::estimate_residual_bits(
                            &mut self.workspace.price_cur,
                            self.workspace.coeffs.get(id),
                            2,
                            0,
                            scan,
                            state.sign_data_hiding,
                        );
                    }
                }
            }

            let mut cb = emit::empty_block();
            let mut cr = emit::empty_block();
            let chroma_log2 = if state.cat == 3 { 2 } else { 0 };
            if state.cat == 3 {
                let prev_commit = self.workspace.commit_ctx;
                let prev_skip = self.workspace.commit_ctx_skip_chroma;
                self.workspace.commit_ctx = evolve;
                self.workspace.commit_ctx_skip_chroma = false;
                let chroma_mode = IntraPredMode::from_u8(mode).unwrap_or(IntraPredMode::Dc);
                let cb_e = self.eval_component(
                    state,
                    px,
                    py,
                    2,
                    1,
                    chroma_mode,
                    state.cur_qp_c,
                    1,
                    lambda,
                    super::eval::search_trial_quant(state.effort_template.luma.exact.quant),
                    ResidualPricingMode::Exact,
                    true,
                );
                cost += cb_e.cost;
                cb = cb_e.into_plan_block();
                let cr_e = self.eval_component(
                    state,
                    px,
                    py,
                    2,
                    2,
                    chroma_mode,
                    state.cur_qp_c,
                    1,
                    lambda,
                    super::eval::search_trial_quant(state.effort_template.luma.exact.quant),
                    ResidualPricingMode::Exact,
                    true,
                );
                cost += cr_e.cost;
                cr = cr_e.into_plan_block();
                self.workspace.commit_ctx = prev_commit;
                self.workspace.commit_ctx_skip_chroma = prev_skip;
            }

            let tt = emit::leaf_tu(
                2,
                chroma_log2,
                1,
                mode,
                mode,
                luma.into_plan_block(),
                cb,
                cr,
                emit::empty_block(),
                emit::empty_block(),
            );
            kids.push(tt);
        }

        let mut parent_chroma = None;
        if matches!(state.cat, 1 | 2) {
            let prev_commit = self.workspace.commit_ctx;
            let prev_skip = self.workspace.commit_ctx_skip_chroma;
            self.workspace.commit_ctx = evolve;
            self.workspace.commit_ctx_skip_chroma = false;
            let (pc, pc_cost) =
                self.eval_part_nxn_parent_chroma_subsampled(state, x0, y0, luma_modes[0], lambda);
            self.workspace.commit_ctx = prev_commit;
            self.workspace.commit_ctx_skip_chroma = prev_skip;
            cost += pc_cost;
            parent_chroma = Some(pc);
        }

        let (cbf_cb, cbf_cb1, cbf_cr, cbf_cr1) = if state.cat == 3 {
            (
                kids.iter().any(TtPlan::cbf_cb),
                false,
                kids.iter().any(TtPlan::cbf_cr),
                false,
            )
        } else {
            (
                parent_chroma.as_ref().is_some_and(|c| c.cb.cbf),
                parent_chroma.as_ref().is_some_and(|c| c.cb1.cbf),
                parent_chroma.as_ref().is_some_and(|c| c.cr.cbf),
                parent_chroma.as_ref().is_some_and(|c| c.cr1.cbf),
            )
        };
        let tt = TtPlan::Split {
            log2_size: 3,
            trafo_depth: 0,
            cbf_cb,
            cbf_cr,
            cbf_cb1,
            cbf_cr1,
            parent_chroma,
            kids,
        };

        if evolve {
            // NxN CU header syntax: part_mode=NxN, four prev_intra flags,
            // then the chroma mode(s). Header models are disjoint from
            // content models, so applying them after content preserves the
            // writer's per-model bin sequences.
            let ctxs = &mut self.workspace.price_cur;
            ctxs.models[crate::contexts::ctx::PART_MODE].update(0);
            for pu in 0..4 {
                let in_mpm = mpms[pu].iter().any(|m| m.as_u8() == luma_modes[pu]);
                ctxs.models[crate::contexts::ctx::PREV_INTRA_LUMA_PRED_FLAG].update(in_mpm as u8);
            }
            if state.cat != 0 {
                let n = if state.cat == 3 { 4 } else { 1 };
                for _ in 0..n {
                    ctxs.models[crate::contexts::ctx::INTRA_CHROMA_PRED_MODE].update(0);
                }
            }
        }

        let mut leaf = emit::cu_leaf(mpms[0], luma_modes[0], tt);
        leaf.chroma_mode_idx = CHROMA_DM_IDX;
        leaf.nxn = Some(NxnInfoPlan {
            luma_modes,
            mpms,
            chroma_mode_idx: [CHROMA_DM_IDX; 4],
        });
        (CuPlan::Leaf(leaf), cost)
    }

    fn eval_part_nxn_parent_chroma_subsampled(
        &mut self,
        state: &Encoder<'_>,
        x0: u32,
        y0: u32,
        chroma_mode: u8,
        lambda: f64,
    ) -> (ParentChromaPlan, f64) {
        debug_assert!(matches!(state.cat, 1 | 2));
        let cx = x0 / 2;
        let cy = if state.cat == 1 { y0 / 2 } else { y0 };
        let clog2 = 2;
        let step = 1u32 << clog2;
        let pred_mode = IntraPredMode::from_u8(chroma_pred_mode(state.cat, chroma_mode))
            .unwrap_or(IntraPredMode::Dc);

        let cb0 = self.eval_component(
            state,
            cx,
            cy,
            clog2,
            1,
            pred_mode,
            state.cur_qp_c,
            0,
            lambda,
            super::eval::search_trial_quant(state.effort_template.luma.exact.quant),
            ResidualPricingMode::Exact,
            true,
        );
        let mut cost = cb0.cost;
        let mut cb1 = None;
        if state.cat == 2 {
            let cb = self.eval_component(
                state,
                cx,
                cy + step,
                clog2,
                1,
                pred_mode,
                state.cur_qp_c,
                0,
                lambda,
                super::eval::search_trial_quant(state.effort_template.luma.exact.quant),
                ResidualPricingMode::Exact,
                true,
            );
            cost += cb.cost;
            cb1 = Some(cb);
        }

        let cr0 = self.eval_component(
            state,
            cx,
            cy,
            clog2,
            2,
            pred_mode,
            state.cur_qp_c,
            0,
            lambda,
            super::eval::search_trial_quant(state.effort_template.luma.exact.quant),
            ResidualPricingMode::Exact,
            true,
        );
        cost += cr0.cost;
        let mut cr1 = None;
        if state.cat == 2 {
            let cr = self.eval_component(
                state,
                cx,
                cy + step,
                clog2,
                2,
                pred_mode,
                state.cur_qp_c,
                0,
                lambda,
                super::eval::search_trial_quant(state.effort_template.luma.exact.quant),
                ResidualPricingMode::Exact,
                true,
            );
            cost += cr.cost;
            cr1 = Some(cr);
        }

        let parent = ParentChromaPlan {
            log2_size: clog2,
            chroma_mode,
            cb: cb0.into_plan_block(),
            cb1: cb1
                .map(BlockTrial::into_plan_block)
                .unwrap_or_else(emit::empty_block),
            cr: cr0.into_plan_block(),
            cr1: cr1
                .map(BlockTrial::into_plan_block)
                .unwrap_or_else(emit::empty_block),
        };
        (parent, cost)
    }

    fn fill_nxn_rough_set(
        &mut self,
        state: &Encoder<'_>,
        x0: u32,
        y0: u32,
        pu: usize,
        mpm: [IntraPredMode; 3],
        lambda: f64,
        set: &mut NxnRoughSet,
    ) {
        let timer = StillSearchLedger::start_timer();
        let mpm_u8 = [mpm[0].as_u8(), mpm[1].as_u8(), mpm[2].as_u8()];
        let scale = CabacEstimator::SCALE as f64;
        let size = 4usize;
        let lambda_sad = lambda.sqrt();
        let mut rough = [(0.0f64, 0u8); 35];
        // Build the 4x4 angular predictions (modes 2..=34) once from a single
        // overlay-aware reference border, mirroring the CU-level rough batch.
        // Planar/DC (0/1) stay on the per-mode predictor.
        let n = size * size;
        if state.bit_depth == 8 {
            // Predict directly to u8 — one batch narrow, no per-mode narrow loop.
            let mut batch_u8 = [0u8; intra_angs::ANGULAR_MODES * 16];
            self.predict_all_angular_into_u8(
                state,
                x0,
                y0,
                2,
                &mut batch_u8[..intra_angs::ANGULAR_MODES * n],
            );
            let mut src = [0u8; 16];
            self.source.sample_block_u8(0, x0, y0, size, &mut src);
            let mut pred = [0u8; 16];
            let mut tmp_u16 = Vec::with_capacity(16);
            for (idx, m) in (0u8..=1).enumerate() {
                let mode = IntraPredMode::from_u8(m).expect("0..=1 are valid intra modes");
                self.predict_into_u8(state, x0, y0, 2, 0, mode, &mut pred, &mut tmp_u16);
                let satd = satd_u8(&src, size, &pred, size, size);
                let mbits = luma_mode_bits(&self.workspace.price_cur, mpm_u8, m);
                rough[idx] = (satd as f64 + lambda_sad * mbits as f64 / scale, m);
            }
            for m in 2u8..=34 {
                let off = intra_angs::slot_offset(m, 2);
                let satd = satd_u8(&src, size, &batch_u8[off..off + n], size, size);
                let mbits = luma_mode_bits(&self.workspace.price_cur, mpm_u8, m);
                rough[m as usize] = (satd as f64 + lambda_sad * mbits as f64 / scale, m);
            }
        } else {
            let mut batch = [0u16; intra_angs::ANGULAR_MODES * 16];
            self.predict_all_angular_into_u16(
                state,
                x0,
                y0,
                2,
                &mut batch[..intra_angs::ANGULAR_MODES * n],
            );
            let mut src = [0u16; 16];
            for y in 0..size {
                for x in 0..size {
                    src[y * size + x] = self.source.sample(0, x0 + x as u32, y0 + y as u32);
                }
            }
            let mut pred = [0u16; 16];
            for (idx, m) in (0u8..=1).enumerate() {
                let mode = IntraPredMode::from_u8(m).expect("0..=1 are valid intra modes");
                self.predict_into(state, x0, y0, 2, 0, mode, &mut pred);
                let satd = satd_u16(&src, size, &pred, size, size);
                let mbits = luma_mode_bits(&self.workspace.price_cur, mpm_u8, m);
                rough[idx] = (satd as f64 + lambda_sad * mbits as f64 / scale, m);
            }
            for m in 2u8..=34 {
                let off = intra_angs::slot_offset(m, 2);
                let slot = &batch[off..off + n];
                let satd = satd_u16(&src, size, slot, size, size);
                let mbits = luma_mode_bits(&self.workspace.price_cur, mpm_u8, m);
                rough[m as usize] = (satd as f64 + lambda_sad * mbits as f64 / scale, m);
            }
        }
        rough.sort_by(|a, b| {
            a.0.partial_cmp(&b.0)
                .unwrap_or(std::cmp::Ordering::Equal)
                .then_with(|| a.1.cmp(&b.1))
        });
        self.workspace.ledger.bump(WorkBucket::NxnRough);
        self.workspace
            .ledger
            .finish_timer(WorkBucket::NxnRough, timer);

        let sl_policy = &state.effort_template.luma.shortlist;
        fn classify_nxn(m: u8) -> (ModeClass, Option<AngularFamily>) {
            if m == 0 {
                (ModeClass::Planar, None)
            } else if m == 1 {
                (ModeClass::Dc, None)
            } else {
                (ModeClass::Angular, AngularFamily::classify(m))
            }
        }
        let rough_costs: Vec<ModeCost> = rough
            .iter()
            .map(|&(cost, m)| {
                let (class, family) = classify_nxn(m);
                ModeCost {
                    mode: m,
                    cost,
                    satd: 0,
                    class,
                    family,
                }
            })
            .collect();
        let modes = build_luma_shortlist(&rough_costs, mpm_u8, sl_policy);
        set.set(pu, &modes);
    }

    /// Exact-evaluate the selected PU's rough set. The name is intentionally the
    /// planned batch boundary; this first version still evaluates one PU at a
    /// time while preserving the later function shape.
    fn eval_nxn4_batch(
        &mut self,
        state: &mut Encoder<'_>,
        x0: u32,
        y0: u32,
        pu: usize,
        mpm: [IntraPredMode; 3],
        lambda: f64,
        rough_set: &NxnRoughSet,
    ) -> (u8, BlockTrial, CanvasSaved, f64) {
        let mpm_u8 = [mpm[0].as_u8(), mpm[1].as_u8(), mpm[2].as_u8()];
        let scale = CabacEstimator::SCALE as f64;
        let mut best: Option<(u8, BlockTrial, CanvasSaved, f64)> = None;
        for &mode in rough_set.modes(pu) {
            let timer = StillSearchLedger::start_timer();
            let mark = self.overlay.mark();
            let pred_mode = IntraPredMode::from_u8(mode).unwrap_or(IntraPredMode::Dc);
            let trial = self.eval_component(
                state,
                x0,
                y0,
                2,
                0,
                pred_mode,
                state.cur_qp_y,
                1,
                lambda,
                super::eval::search_trial_quant(state.effort_template.luma.exact.quant),
                ResidualPricingMode::Exact,
                true,
            );
            let mbits = luma_mode_bits(&self.workspace.price_cur, mpm_u8, mode);
            let total = trial.cost + lambda * mbits as f64 / scale;
            let recon = self.overlay.detach_from(mark);
            match &best {
                Some((_, _, _, best_total)) if *best_total <= total => {}
                _ => best = Some((mode, trial, recon, total)),
            }
            self.workspace.ledger.bump(WorkBucket::NxnBatch);
            self.workspace
                .ledger
                .finish_timer(WorkBucket::NxnBatch, timer);
        }

        let (mode, trial, recon, total) = best.expect("PartNxN shortlist is never empty");
        state.stats.partnxn_code_block_calls += 1;
        (mode, trial, recon, total)
    }
}