api.rs 31.8 KB
Newer Older
rzumer's avatar
rzumer committed
1 2 3 4 5 6 7 8 9
// Copyright (c) 2018, The rav1e contributors. All rights reserved
//
// This source code is subject to the terms of the BSD 2 Clause License and
// the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
// was not distributed with this source code in the LICENSE file, you can
// obtain it at www.aomedia.org/license/software. If the Alliance for Open
// Media Patent License 1.0 was not distributed with this source code in the
// PATENTS file, you can obtain it at www.aomedia.org/license/patent.

10
use arg_enum_proc_macro::ArgEnum;
Raphaël Zumer's avatar
Raphaël Zumer committed
11
use bitstream_io::*;
Raphaël Zumer's avatar
Raphaël Zumer committed
12 13 14
use crate::encoder::*;
use crate::metrics::calculate_frame_psnr;
use crate::partition::*;
15
use crate::rate::RCState;
16 17 18
use crate::rate::FRAME_NSUBTYPES;
use crate::rate::FRAME_SUBTYPE_I;
use crate::rate::FRAME_SUBTYPE_P;
Raphaël Zumer's avatar
Raphaël Zumer committed
19
use crate::scenechange::SceneChangeDetector;
20
use crate::util::Pixel;
Luca Barbato's avatar
Luca Barbato committed
21

Raphaël Zumer's avatar
Raphaël Zumer committed
22
use std::{cmp, fmt, io};
fbossen's avatar
fbossen committed
23
use std::collections::BTreeMap;
Luca Barbato's avatar
Luca Barbato committed
24
use std::sync::Arc;
Josh Holmer's avatar
Josh Holmer committed
25
use std::collections::BTreeSet;
26
use std::path::PathBuf;
Josh Holmer's avatar
Josh Holmer committed
27 28

const LOOKAHEAD_FRAMES: u64 = 10;
Luca Barbato's avatar
Luca Barbato committed
29 30 31

// TODO: use the num crate?
#[derive(Clone, Copy, Debug)]
32
#[repr(C)]
Luca Barbato's avatar
Luca Barbato committed
33 34 35
pub struct Rational {
  pub num: u64,
  pub den: u64
Luca Barbato's avatar
Luca Barbato committed
36 37
}

Luca Barbato's avatar
Luca Barbato committed
38 39 40
impl Rational {
  pub fn new(num: u64, den: u64) -> Self {
    Rational { num, den }
Luca Barbato's avatar
Luca Barbato committed
41 42 43
  }
}

44 45 46 47 48 49 50
#[derive(Clone, Copy, Debug)]
#[repr(C)]
pub struct Point {
  pub x: u16,
  pub y: u16
}

51 52

/// Encoder Settings impacting the bitstream produced
53
#[derive(Clone, Debug)]
54
pub struct EncoderConfig {
55
  // output size
56 57
  pub width: usize,
  pub height: usize,
58 59

  // data format and ancillary color information
60 61 62
  pub bit_depth: usize,
  pub chroma_sampling: ChromaSampling,
  pub chroma_sample_position: ChromaSamplePosition,
63 64 65 66 67 68
  pub pixel_range: PixelRange,
  pub color_description: Option<ColorDescription>,
  pub mastering_display: Option<MasteringDisplay>,
  pub content_light: Option<ContentLight>,

  // encoder configuration
69
  pub time_base: Rational,
Josh Holmer's avatar
Josh Holmer committed
70 71 72 73
  /// The *minimum* interval between two keyframes
  pub min_key_frame_interval: u64,
  /// The *maximum* interval between two keyframes
  pub max_key_frame_interval: u64,
74
  pub low_latency: bool,
75
  pub quantizer: usize,
76
  pub bitrate: i32,
77 78
  pub tune: Tune,
  pub speed_settings: SpeedSettings,
79 80
  /// `None` for one-pass encode. `Some(1)` or `Some(2)` for two-pass encoding.
  pub pass: Option<u8>,
81
  pub show_psnr: bool,
82
  pub stats_file: Option<PathBuf>,
83
  pub train_rdo: bool,
84 85 86 87
}

impl Default for EncoderConfig {
  fn default() -> Self {
88
    const DEFAULT_SPEED: usize = 5;
89
    Self::with_speed_preset(DEFAULT_SPEED)
90 91 92
  }
}

93 94 95
impl EncoderConfig {
  pub fn with_speed_preset(speed: usize) -> Self {
    EncoderConfig {
96 97
      width: 640,
      height: 480,
98

99 100 101
      bit_depth: 8,
      chroma_sampling: ChromaSampling::Cs420,
      chroma_sample_position: ChromaSamplePosition::Unknown,
102 103 104 105 106
      pixel_range: PixelRange::Unspecified,
      color_description: None,
      mastering_display: None,
      content_light: None,

107
      time_base: Rational { num: 30, den: 1 },
Josh Holmer's avatar
Josh Holmer committed
108 109
      min_key_frame_interval: 12,
      max_key_frame_interval: 240,
110
      low_latency: false,
111
      quantizer: 100,
112
      bitrate: 0,
113
      tune: Tune::default(),
114
      speed_settings: SpeedSettings::from_preset(speed),
115
      pass: None,
116
      show_psnr: false,
117
      stats_file: None,
118
      train_rdo: false
119 120 121 122 123 124 125 126 127 128 129
    }
  }
}

#[derive(Clone, Copy, Debug)]
pub struct SpeedSettings {
  pub min_block_size: BlockSize,
  pub multiref: bool,
  pub fast_deblock: bool,
  pub reduced_tx_set: bool,
  pub tx_domain_distortion: bool,
130
  pub tx_domain_rate: bool,
131 132 133 134
  pub encode_bottomup: bool,
  pub rdo_tx_decision: bool,
  pub prediction_modes: PredictionModesSetting,
  pub include_near_mvs: bool,
135
  pub no_scene_detection: bool,
136 137
  pub diamond_me: bool,
  pub cdef: bool
138 139 140 141 142
}

impl Default for SpeedSettings {
  fn default() -> Self {
    SpeedSettings {
143
      min_block_size: BlockSize::BLOCK_16X16,
144 145 146 147
      multiref: false,
      fast_deblock: false,
      reduced_tx_set: false,
      tx_domain_distortion: false,
148
      tx_domain_rate: false,
149 150 151 152 153
      encode_bottomup: false,
      rdo_tx_decision: false,
      prediction_modes: PredictionModesSetting::Simple,
      include_near_mvs: false,
      no_scene_detection: false,
154
      diamond_me: false,
155
      cdef: false,
156 157
    }
  }
158 159 160 161 162 163 164 165 166 167
}

impl SpeedSettings {
  pub fn from_preset(speed: usize) -> Self {
    SpeedSettings {
      min_block_size: Self::min_block_size_preset(speed),
      multiref: Self::multiref_preset(speed),
      fast_deblock: Self::fast_deblock_preset(speed),
      reduced_tx_set: Self::reduced_tx_set_preset(speed),
      tx_domain_distortion: Self::tx_domain_distortion_preset(speed),
168
      tx_domain_rate: Self::tx_domain_rate_preset(speed),
169 170 171 172
      encode_bottomup: Self::encode_bottomup_preset(speed),
      rdo_tx_decision: Self::rdo_tx_decision_preset(speed),
      prediction_modes: Self::prediction_modes_preset(speed),
      include_near_mvs: Self::include_near_mvs_preset(speed),
173
      no_scene_detection: Self::no_scene_detection_preset(speed),
174
      diamond_me: Self::diamond_me_preset(speed),
175
      cdef: Self::cdef_preset(speed),
176 177 178
    }
  }

179 180
  /// This preset is set this way because 8x8 with reduced TX set is faster but with equivalent
  /// or better quality compared to 16x16 or 32x32 (to which reduced TX set does not apply).
181
  fn min_block_size_preset(speed: usize) -> BlockSize {
182
    if speed == 0 {
183
      BlockSize::BLOCK_4X4
184
    } else if speed <= 8 {
185 186 187 188 189 190
      BlockSize::BLOCK_8X8
    } else {
      BlockSize::BLOCK_64X64
    }
  }

191 192 193 194
  /// Multiref is enabled automatically if low_latency is false,
  /// but if someone is setting low_latency to true manually,
  /// multiref has a large speed penalty with low quality gain.
  /// Because low_latency can be set manually, this setting is conservative.
195
  fn multiref_preset(speed: usize) -> bool {
196
    speed <= 1
197 198 199
  }

  fn fast_deblock_preset(speed: usize) -> bool {
200
    speed >= 8
201 202 203
  }

  fn reduced_tx_set_preset(speed: usize) -> bool {
204
    speed >= 5
205 206
  }

207 208 209
  /// TX domain distortion is always faster, with no significant quality change
  fn tx_domain_distortion_preset(_speed: usize) -> bool {
    true
210 211
  }

212 213 214 215
  fn tx_domain_rate_preset(_speed: usize) -> bool {
    false
  }

216 217 218 219 220 221 222 223 224
  fn encode_bottomup_preset(speed: usize) -> bool {
    speed == 0
  }

  fn rdo_tx_decision_preset(speed: usize) -> bool {
    speed <= 3
  }

  fn prediction_modes_preset(speed: usize) -> PredictionModesSetting {
225
    if speed <= 1 {
226
      PredictionModesSetting::ComplexAll
227
    } else if speed <= 5 {
228 229 230 231 232 233 234 235 236
      PredictionModesSetting::ComplexKeyframes
    } else {
      PredictionModesSetting::Simple
    }
  }

  fn include_near_mvs_preset(speed: usize) -> bool {
    speed <= 2
  }
237 238 239 240

  fn no_scene_detection_preset(speed: usize) -> bool {
    speed == 10
  }
241

242 243 244 245 246 247 248
  /// Currently Diamond ME gives better quality than full search on most videos,
  /// in addition to being faster.
  /// There are a few outliers, such as the Wikipedia test clip.
  ///
  /// TODO: Revisit this setting if full search quality improves in the future.
  fn diamond_me_preset(_speed: usize) -> bool {
    true
249
  }
250 251 252 253

  fn cdef_preset(_speed: usize) -> bool {
    true
  }
254 255
}

256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277
#[allow(dead_code, non_camel_case_types)]
#[derive(Debug, PartialEq, Clone, Copy, Serialize, Deserialize)]
#[repr(C)]
pub enum FrameType {
  KEY,
  INTER,
  INTRA_ONLY,
  SWITCH
}

impl fmt::Display for FrameType {
  fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
    use self::FrameType::*;
    match self {
      KEY => write!(f, "Key frame"),
      INTER => write!(f, "Inter frame"),
      INTRA_ONLY => write!(f, "Intra only frame"),
      SWITCH => write!(f, "Switching frame"),
    }
  }
}

278 279 280 281 282 283 284
#[derive(Clone, Copy, Debug, PartialOrd, PartialEq)]
pub enum PredictionModesSetting {
  Simple,
  ComplexKeyframes,
  ComplexAll,
}

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
#[derive(Copy, Clone, Debug, PartialEq)]
#[repr(C)]
pub enum ChromaSampling {
  Cs420,
  Cs422,
  Cs444,
  Cs400,
}

impl Default for ChromaSampling {
  fn default() -> Self {
    ChromaSampling::Cs420
  }
}

impl ChromaSampling {
  // Provides the sampling period in the horizontal and vertical axes.
  pub fn sampling_period(self) -> (usize, usize) {
    use self::ChromaSampling::*;
    match self {
      Cs420 => (2, 2),
      Cs422 => (2, 1),
      Cs444 => (1, 1),
      Cs400 => (2, 2),
    }
  }
}

#[derive(Copy, Clone, Debug, PartialEq)]
#[repr(C)]
pub enum ChromaSamplePosition {
  Unknown,
  Vertical,
  Colocated
}

impl Default for ChromaSamplePosition {
  fn default() -> Self {
    ChromaSamplePosition::Unknown
  }
}

327 328 329 330 331 332
#[derive(ArgEnum, Debug, Clone, Copy, PartialEq)]
#[repr(C)]
pub enum PixelRange {
    Unspecified = 0,
    Limited,
    Full,
333 334 335 336 337 338 339 340
}

impl Default for PixelRange {
    fn default() -> Self {
        PixelRange::Unspecified
    }
}

341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357
#[derive(ArgEnum, Debug, Clone, Copy, PartialEq)]
#[repr(C)]
pub enum MatrixCoefficients {
    Identity = 0,
    BT709,
    Unspecified,
    BT470M = 4,
    BT470BG,
    ST170M,
    ST240M,
    YCgCo,
    BT2020NonConstantLuminance,
    BT2020ConstantLuminance,
    ST2085,
    ChromaticityDerivedNonConstantLuminance,
    ChromaticityDerivedConstantLuminance,
    ICtCp,
358 359 360 361 362 363 364 365
}

impl Default for MatrixCoefficients {
    fn default() -> Self {
        MatrixCoefficients::Unspecified
    }
}

366 367 368 369 370 371 372 373 374 375 376 377 378 379 380
#[derive(ArgEnum, Debug, Clone, Copy, PartialEq)]
#[repr(C)]
pub enum ColorPrimaries {
    BT709 = 1,
    Unspecified,
    BT470M = 4,
    BT470BG,
    ST170M,
    ST240M,
    Film,
    BT2020,
    ST428,
    P3DCI,
    P3Display,
    Tech3213 = 22,
381 382 383 384 385 386 387 388
}

impl Default for ColorPrimaries {
    fn default() -> Self {
        ColorPrimaries::Unspecified
    }
}

389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408
#[derive(ArgEnum, Debug, Clone, Copy, PartialEq)]
#[repr(C)]
pub enum TransferCharacteristics {
    BT1886 = 1,
    Unspecified,
    BT470M = 4,
    BT470BG,
    ST170M,
    ST240M,
    Linear,
    Logarithmic100,
    Logarithmic316,
    XVYCC,
    BT1361E,
    SRGB,
    BT2020Ten,
    BT2020Twelve,
    PerceptualQuantizer,
    ST428,
    HybridLogGamma,
409 410 411 412 413 414 415 416
}

impl Default for TransferCharacteristics {
    fn default() -> Self {
        TransferCharacteristics::Unspecified
    }
}

417 418 419 420 421 422 423
#[derive(Copy, Clone, Debug)]
pub struct ColorDescription {
    pub color_primaries: ColorPrimaries,
    pub transfer_characteristics: TransferCharacteristics,
    pub matrix_coefficients: MatrixCoefficients
}

424 425 426 427 428 429 430 431 432 433 434 435 436 437
#[derive(Copy, Clone, Debug)]
pub struct MasteringDisplay {
    pub primaries: [Point; 3],
    pub white_point: Point,
    pub max_luminance: u32,
    pub min_luminance: u32,
}

#[derive(Copy, Clone, Debug)]
pub struct ContentLight {
    pub max_content_light_level: u16,
    pub max_frame_average_light_level: u16,
}

438
/// Contains all the encoder configuration
439
#[derive(Clone, Debug)]
440
pub struct Config {
Luca Barbato's avatar
Luca Barbato committed
441 442 443
  pub enc: EncoderConfig,
  /// The number of threads in the threadpool.
  pub threads: usize
Luca Barbato's avatar
Luca Barbato committed
444 445
}

446 447
const MAX_USABLE_THREADS: usize = 4;

Luca Barbato's avatar
Luca Barbato committed
448
impl Config {
449
  pub fn new_context<T: Pixel>(&self) -> Context<T> {
450 451
    assert!(8 * std::mem::size_of::<T>() >= self.enc.bit_depth, "The Pixel u{} does not match the Config bit_depth {}",
            8 * std::mem::size_of::<T>(), self.enc.bit_depth);
452 453 454
    // initialize with temporal delimiter
    let packet_data = TEMPORAL_DELIMITER.to_vec();

455 456 457 458 459 460
    let maybe_ac_qi_max = if self.enc.quantizer < 255 {
      Some(self.enc.quantizer as u8)
    } else {
      None
    };

461 462 463 464 465 466 467
    let threads = if self.threads == 0 {
      rayon::current_num_threads().min(MAX_USABLE_THREADS)
    } else {
      self.threads
    };

    let pool = rayon::ThreadPoolBuilder::new().num_threads(threads).build().unwrap();
Luca Barbato's avatar
Luca Barbato committed
468

Josh Holmer's avatar
Josh Holmer committed
469
    Context {
Luca Barbato's avatar
Luca Barbato committed
470
      inner: ContextInner {
Luca Barbato's avatar
Rustfmt  
Luca Barbato committed
471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494
        frame_count: 0,
        limit: 0,
        idx: 0,
        frames_processed: 0,
        frame_q: BTreeMap::new(),
        frame_invariants: BTreeMap::new(),
        keyframes: BTreeSet::new(),
        packet_data,
        segment_start_idx: 0,
        segment_start_frame: 0,
        keyframe_detector: SceneChangeDetector::new(self.enc.bit_depth),
        config: self.enc.clone(),
        rc_state: RCState::new(
          self.enc.width as i32,
          self.enc.height as i32,
          self.enc.time_base.num as i64,
          self.enc.time_base.den as i64,
          self.enc.bitrate,
          maybe_ac_qi_max,
          self.enc.max_key_frame_interval as i32
        ),
        maybe_prev_log_base_q: None,
        first_pass_data: FirstPassData { frames: Vec::new() },
        pool
495
      },
Luca Barbato's avatar
Rustfmt  
Luca Barbato committed
496
      config: self.enc.clone()
Josh Holmer's avatar
Josh Holmer committed
497
    }
Luca Barbato's avatar
Luca Barbato committed
498 499 500
  }
}

Luca Barbato's avatar
Luca Barbato committed
501
pub struct ContextInner<T: Pixel> {
fbossen's avatar
fbossen committed
502
  frame_count: u64,
503
  limit: u64,
504
  pub(crate) idx: u64,
Josh Holmer's avatar
Josh Holmer committed
505 506
  frames_processed: u64,
  /// Maps frame *number* to frames
507
  frame_q: BTreeMap<u64, Option<Arc<Frame<T>>>>, //    packet_q: VecDeque<Packet>
Josh Holmer's avatar
Josh Holmer committed
508
  /// Maps frame *idx* to frame data
Luca Barbato's avatar
Luca Barbato committed
509
  frame_invariants: BTreeMap<u64, FrameInvariants<T>>,
Josh Holmer's avatar
Josh Holmer committed
510
  /// A list of keyframe *numbers* in this encode. Needed so that we don't
Luca Barbato's avatar
Luca Barbato committed
511
  /// need to keep all of the frame_invariants in memory for the whole life of the encode.
Josh Holmer's avatar
Josh Holmer committed
512
  keyframes: BTreeSet<u64>,
513
  /// A storage space for reordered frames.
Josh Holmer's avatar
Josh Holmer committed
514 515 516
  packet_data: Vec<u8>,
  segment_start_idx: u64,
  segment_start_frame: u64,
517
  keyframe_detector: SceneChangeDetector<T>,
518
  pub(crate) config: EncoderConfig,
519
  rc_state: RCState,
520
  maybe_prev_log_base_q: Option<i64>,
521
  pub first_pass_data: FirstPassData,
Luca Barbato's avatar
Luca Barbato committed
522
  pool: rayon::ThreadPool,
Luca Barbato's avatar
Luca Barbato committed
523 524
}

Luca Barbato's avatar
Luca Barbato committed
525 526 527 528 529
pub struct Context<T: Pixel> {
  inner: ContextInner<T>,
  config: EncoderConfig,
}

Luca Barbato's avatar
Luca Barbato committed
530 531
#[derive(Clone, Copy, Debug)]
pub enum EncoderStatus {
Luca Barbato's avatar
Luca Barbato committed
532 533
  /// The encoder needs more data to produce an output Packet
  /// May be emitted by `Context::receive_packet`  when frame reordering is enabled.
Luca Barbato's avatar
Luca Barbato committed
534 535
  NeedMoreData,
  /// There are enough Frames queue
Luca Barbato's avatar
Luca Barbato committed
536
  /// May be emitted by `Context::send_frame` when the input queue is constrained
Luca Barbato's avatar
Luca Barbato committed
537
  EnoughData,
Luca Barbato's avatar
Luca Barbato committed
538 539 540 541 542
  /// The encoder already produced the number of frames requested
  /// May be emitted by `Context::receive_packet` after a flush request had been processed
  /// or the frame limit had been reached.
  LimitReached,
  /// Generic fatal error
Luca Barbato's avatar
Luca Barbato committed
543
  Failure,
Luca Barbato's avatar
Luca Barbato committed
544 545
}

546
pub struct Packet<T: Pixel> {
Luca Barbato's avatar
Luca Barbato committed
547
  pub data: Vec<u8>,
548
  pub rec: Option<Frame<T>>,
fbossen's avatar
fbossen committed
549
  pub number: u64,
550 551 552
  pub frame_type: FrameType,
  /// PSNR for Y, U, and V planes
  pub psnr: Option<(f64, f64, f64)>,
553 554
}

555
impl<T: Pixel> fmt::Display for Packet<T> {
556
  fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
Luca Barbato's avatar
Luca Barbato committed
557 558 559 560 561 562 563
    write!(
      f,
      "Frame {} - {} - {} bytes",
      self.number,
      self.frame_type,
      self.data.len()
    )
564
  }
Luca Barbato's avatar
Luca Barbato committed
565 566
}

567 568
impl<T: Pixel> Context<T> {
  pub fn new_frame(&self) -> Arc<Frame<T>> {
Josh Holmer's avatar
Josh Holmer committed
569
    Arc::new(Frame::new(
570 571 572
      self.config.width,
      self.config.height,
      self.config.chroma_sampling
Josh Holmer's avatar
Josh Holmer committed
573
    ))
Luca Barbato's avatar
Luca Barbato committed
574 575 576 577
  }

  pub fn send_frame<F>(&mut self, frame: F) -> Result<(), EncoderStatus>
  where
578 579
    F: Into<Option<Arc<Frame<T>>>>,
    T: Pixel,
Luca Barbato's avatar
Luca Barbato committed
580
  {
581 582 583 584 585 586
    let frame = frame.into();

    if frame.is_none() {
        self.inner.limit = self.inner.frame_count;
    }

Luca Barbato's avatar
Luca Barbato committed
587
    self.inner.send_frame(frame)
Luca Barbato's avatar
Luca Barbato committed
588 589
  }

Luca Barbato's avatar
Luca Barbato committed
590 591
  pub fn receive_packet(&mut self) -> Result<Packet<T>, EncoderStatus> {
    self.inner.receive_packet()
592 593
  }

Luca Barbato's avatar
Luca Barbato committed
594 595
  pub fn flush(&mut self) {
    self.send_frame(None).unwrap();
596 597
  }

598 599 600
  pub fn container_sequence_header(&mut self) -> Vec<u8> {
    fn sequence_header_inner(seq: &Sequence) -> io::Result<Vec<u8>> {
      let mut buf = Vec::new();
601

602
      {
603 604 605 606
        let mut bw = BitWriter::endian(&mut buf, BigEndian);
        bw.write_bit(true)?; // marker
        bw.write(7, 1)?; // version
        bw.write(3, seq.profile)?;
607
        bw.write(5, 31)?; // level
608 609 610 611 612 613 614 615 616 617 618 619
        bw.write_bit(false)?; // tier
        bw.write_bit(seq.bit_depth > 8)?; // high_bitdepth
        bw.write_bit(seq.bit_depth == 12)?; // twelve_bit
        bw.write_bit(seq.bit_depth == 1)?; // monochrome
        bw.write_bit(seq.bit_depth == 12)?; // twelve_bit
        bw.write_bit(seq.chroma_sampling != ChromaSampling::Cs444)?; // chroma_subsampling_x
        bw.write_bit(seq.chroma_sampling == ChromaSampling::Cs420)?; // chroma_subsampling_y
        bw.write(2, 0)?; // sample_position
        bw.write(3, 0)?; // reserved
        bw.write_bit(false)?; // initial_presentation_delay_present

        bw.write(4, 0)?; // reserved
620
      }
621

622 623 624
      Ok(buf)
    }

625
    let seq = Sequence::new(&self.config);
626 627

    sequence_header_inner(&seq).unwrap()
628 629
  }

Luca Barbato's avatar
Luca Barbato committed
630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646
  pub fn get_first_pass_data(&self) -> &FirstPassData {
    &self.inner.first_pass_data
  }

  pub fn set_limit(&mut self, limit: u64) {
    self.inner.set_limit(limit);
  }
}


impl<T: Pixel> ContextInner<T> {
  pub fn send_frame<F>(&mut self, frame: F) -> Result<(), EncoderStatus>
  where
    F: Into<Option<Arc<Frame<T>>>>,
    T: Pixel,
  {
    let idx = self.frame_count;
647 648 649 650 651
    let frame = frame.into();
    if frame.is_some() {
      self.frame_count += 1;
    }
    self.frame_q.insert(idx, frame);
Luca Barbato's avatar
Luca Barbato committed
652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675
    Ok(())
  }

  fn get_frame(&self, frame_number: u64) -> Arc<Frame<T>> {
    // Clones only the arc, so low cost overhead
    self.frame_q.get(&frame_number).as_ref().unwrap().as_ref().unwrap().clone()
  }

  pub fn get_frame_count(&self) -> u64 {
    self.frame_count
  }

  pub fn set_limit(&mut self, limit: u64) {
    self.limit = limit;
  }

  pub(crate) fn needs_more_lookahead(&self) -> bool {
    self.needs_more_frames(self.frame_count) && self.frames_processed + LOOKAHEAD_FRAMES > self.frame_q.keys().last().cloned().unwrap_or(0)
  }

  pub fn needs_more_frames(&self, frame_count: u64) -> bool {
    self.limit == 0 || frame_count < self.limit
  }

Josh Holmer's avatar
Josh Holmer committed
676
  fn next_keyframe(&self) -> u64 {
Luca Barbato's avatar
Luca Barbato committed
677
    let next_detected = self.frame_invariants.values()
Josh Holmer's avatar
Josh Holmer committed
678 679
      .find(|fi| fi.frame_type == FrameType::KEY && fi.number > self.segment_start_frame)
      .map(|fi| fi.number);
680
    let next_limit = self.segment_start_frame + self.config.max_key_frame_interval;
681
    let next_limit = if self.limit != 0 { next_limit.min(self.limit) } else { next_limit };
Josh Holmer's avatar
Josh Holmer committed
682 683 684 685 686
    if next_detected.is_none() {
      return next_limit;
    }
    cmp::min(next_detected.unwrap(), next_limit)
  }
fbossen's avatar
fbossen committed
687

688 689
  fn set_frame_properties(&mut self, idx: u64) -> Result<bool, EncoderStatus> {
    let (fi, end_of_subgop) = self.build_frame_properties(idx)?;
Luca Barbato's avatar
Luca Barbato committed
690
    self.frame_invariants.insert(idx, fi);
691

692
    Ok(end_of_subgop)
Josh Holmer's avatar
Josh Holmer committed
693 694
  }

695
  fn build_frame_properties(&mut self, idx: u64) -> Result<(FrameInvariants<T>, bool), EncoderStatus> {
Josh Holmer's avatar
Josh Holmer committed
696
    if idx == 0 {
697
      let seq = Sequence::new(&self.config);
698

Josh Holmer's avatar
Josh Holmer committed
699
      // The first frame will always be a key frame
Josh Holmer's avatar
Josh Holmer committed
700 701
      let fi = FrameInvariants::new_key_frame(
        &FrameInvariants::new(
702
          self.config.clone(),
703
          seq
Josh Holmer's avatar
Josh Holmer committed
704 705 706
        ),
        0
      );
707
      return Ok((fi, true));
Josh Holmer's avatar
Josh Holmer committed
708 709
    }

Luca Barbato's avatar
Luca Barbato committed
710
    let mut fi = self.frame_invariants[&(idx - 1)].clone();
Josh Holmer's avatar
Josh Holmer committed
711

712
    // FIXME: inter unsupported with 4:2:2 and 4:4:4 chroma sampling
713
    let chroma_sampling = self.config.chroma_sampling;
714 715 716
    let keyframe_only = chroma_sampling == ChromaSampling::Cs444 ||
      chroma_sampling == ChromaSampling::Cs422;

Josh Holmer's avatar
Josh Holmer committed
717 718 719 720 721 722
    // Initially set up the frame as an inter frame.
    // We need to determine what the frame number is before we can
    // look up the frame type. If reordering is enabled, the idx
    // may not match the frame number.
    let idx_in_segment = idx - self.segment_start_idx;
    if idx_in_segment > 0 {
723
      let next_keyframe = if keyframe_only { self.segment_start_frame + 1 } else { self.next_keyframe() };
724
      let (fi_temp, end_of_subgop) = FrameInvariants::new_inter_frame(
Josh Holmer's avatar
Josh Holmer committed
725
        &fi,
726 727 728 729
        self.segment_start_frame,
        idx_in_segment,
        next_keyframe
      );
Josh Holmer's avatar
Josh Holmer committed
730
      fi = fi_temp;
731
      if !end_of_subgop {
Josh Holmer's avatar
Josh Holmer committed
732 733 734
        if !fi.inter_cfg.unwrap().reorder
          || ((idx_in_segment - 1) % fi.inter_cfg.unwrap().group_len == 0
          && fi.number == (next_keyframe - 1))
735
        {
Josh Holmer's avatar
Josh Holmer committed
736
          self.segment_start_idx = idx;
Josh Holmer's avatar
Josh Holmer committed
737
          self.segment_start_frame = next_keyframe;
Josh Holmer's avatar
Josh Holmer committed
738
          fi.number = next_keyframe;
Josh Holmer's avatar
Josh Holmer committed
739
        } else {
740
          return Ok((fi, false));
Josh Holmer's avatar
Josh Holmer committed
741 742
        }
      }
fbossen's avatar
fbossen committed
743
    }
Luca Barbato's avatar
Luca Barbato committed
744

745 746
    match self.frame_q.get(&fi.number) {
      Some(Some(_)) => {},
747
      _ => { return Err(EncoderStatus::NeedMoreData); }
748 749
    }

Josh Holmer's avatar
Josh Holmer committed
750
    // Now that we know the frame number, look up the correct frame type
Josh Holmer's avatar
Josh Holmer committed
751 752 753 754 755 756 757
    let frame_type = self.determine_frame_type(fi.number);
    if frame_type == FrameType::KEY {
      self.segment_start_idx = idx;
      self.segment_start_frame = fi.number;
      self.keyframes.insert(fi.number);
    }
    fi.frame_type = frame_type;
Josh Holmer's avatar
Josh Holmer committed
758

Josh Holmer's avatar
Josh Holmer committed
759 760 761 762 763
    let idx_in_segment = idx - self.segment_start_idx;
    if idx_in_segment == 0 {
      fi = FrameInvariants::new_key_frame(&fi, self.segment_start_frame);
    } else {
      let next_keyframe = self.next_keyframe();
764
      let (fi_temp, end_of_subgop) = FrameInvariants::new_inter_frame(
Josh Holmer's avatar
Josh Holmer committed
765 766 767 768 769 770
        &fi,
        self.segment_start_frame,
        idx_in_segment,
        next_keyframe
      );
      fi = fi_temp;
771
      if !end_of_subgop {
772
        return Ok((fi, false));
Josh Holmer's avatar
Josh Holmer committed
773 774
      }
    }
775
    Ok((fi, true))
fbossen's avatar
fbossen committed
776
  }
Luca Barbato's avatar
Luca Barbato committed
777

778
  pub fn receive_packet(&mut self) -> Result<Packet<T>, EncoderStatus> {
779
    if self.limit != 0 && self.frames_processed == self.limit {
Luca Barbato's avatar
Luca Barbato committed
780
      return Err(EncoderStatus::LimitReached);
781 782
    }

783
    if self.needs_more_lookahead() {
Luca Barbato's avatar
Luca Barbato committed
784
      return Err(EncoderStatus::NeedMoreData);
785 786
    }

787 788
    let idx = {
      let mut idx = self.idx;
789
      while !self.set_frame_properties(idx)? {
790 791 792
        self.idx += 1;
        idx = self.idx;
      }
Luca Barbato's avatar
Luca Barbato committed
793

Luca Barbato's avatar
Luca Barbato committed
794
      if !self.needs_more_frames(self.frame_invariants[&idx].number) {
795
        self.idx += 1;
Luca Barbato's avatar
Luca Barbato committed
796
        return Err(EncoderStatus::LimitReached);
797 798 799
      }
      idx
    };
800

801
    let ret = {
Luca Barbato's avatar
Luca Barbato committed
802
      let fi = self.frame_invariants.get_mut(&idx).unwrap();
Luca Barbato's avatar
Luca Barbato committed
803 804
      if fi.show_existing_frame {
        self.idx += 1;
Luca Barbato's avatar
Luca Barbato committed
805

Luca Barbato's avatar
Luca Barbato committed
806
        let mut fs = FrameState::new(fi);
fbossen's avatar
fbossen committed
807

808 809
        // TODO: Record the bits spent here against the original frame for rate
        //  control purposes, or add a new frame subtype?
810
        let sef_data = encode_show_existing_frame(fi, &mut fs);
811
        self.packet_data.extend(sef_data);
Luca Barbato's avatar
Luca Barbato committed
812

Luca Barbato's avatar
Luca Barbato committed
813
        let rec = if fi.show_frame { Some(fs.rec) } else { None };
814
        let fi = fi.clone();
815
        self.finalize_packet(rec, &fi)
Luca Barbato's avatar
Luca Barbato committed
816 817 818
      } else {
        if let Some(f) = self.frame_q.get(&fi.number) {
          self.idx += 1;
fbossen's avatar
fbossen committed
819

820
          if let Some(frame) = f.clone() {
821
            let fti = fi.get_frame_subtype();
822 823
            let qps =
              self.rc_state.select_qi(self, fti, self.maybe_prev_log_base_q);
Luca Barbato's avatar
Luca Barbato committed
824
            let fi = self.frame_invariants.get_mut(&idx).unwrap();
825
            fi.set_quantizers(&qps);
Luca Barbato's avatar
Luca Barbato committed
826
            let mut fs = FrameState::new_with_frame(fi, frame.clone());
fbossen's avatar
fbossen committed
827

828
            // TODO: Trial encoding for first frame of each type.
Luca Barbato's avatar
Luca Barbato committed
829
            let data = self.pool.install(||encode_frame(fi, &mut fs));
830 831 832 833 834 835 836 837
            self.maybe_prev_log_base_q = Some(qps.log_base_q);
            // TODO: Add support for dropping frames.
            self.rc_state.update_state(
              (data.len() * 8) as i64,
              fti,
              qps.log_target_q,
              false
            );
Luca Barbato's avatar
Luca Barbato committed
838
            self.packet_data.extend(data);
fbossen's avatar
fbossen committed
839

Luca Barbato's avatar
Luca Barbato committed
840
            fs.rec.pad(fi.width, fi.height);
fbossen's avatar
fbossen committed
841

Luca Barbato's avatar
Luca Barbato committed
842 843
            // TODO avoid the clone by having rec Arc.
            let rec = if fi.show_frame { Some(fs.rec.clone()) } else { None };
fbossen's avatar
fbossen committed
844

Luca Barbato's avatar
Luca Barbato committed
845
            update_rec_buffer(fi, fs);
fbossen's avatar
fbossen committed
846

Luca Barbato's avatar
Luca Barbato committed
847
            if fi.show_frame {
848
              let fi = fi.clone();
849
              self.finalize_packet(rec, &fi)
Luca Barbato's avatar
Luca Barbato committed
850 851 852
            } else {
              Err(EncoderStatus::NeedMoreData)
            }
853 854 855
          } else {
            Err(EncoderStatus::NeedMoreData)
          }
fbossen's avatar
fbossen committed
856 857 858 859
        } else {
          Err(EncoderStatus::NeedMoreData)
        }
      }
860 861 862
    };

    if let Ok(ref pkt) = ret {
Luca Barbato's avatar
Luca Barbato committed
863
      self.garbage_collect(pkt.number);
864 865 866
    }

    ret
Luca Barbato's avatar
Luca Barbato committed
867 868
  }

869
  fn finalize_packet(&mut self, rec: Option<Frame<T>>, fi: &FrameInvariants<T>) -> Result<Packet<T>, EncoderStatus> {
870 871 872 873 874 875 876
    let data = self.packet_data.clone();
    self.packet_data.clear();
    if write_temporal_delimiter(&mut self.packet_data).is_err() {
      return Err(EncoderStatus::Failure);
    }

    let mut psnr = None;
877
    if self.config.show_psnr {
878
      if let Some(ref rec) = rec {
879
        let original_frame = self.get_frame(fi.number);
880 881 882 883 884 885 886 887
        psnr = Some(calculate_frame_psnr(
          &*original_frame,
          rec,
          fi.sequence.bit_depth
        ));
      }
    }

888
    if self.config.pass == Some(1) {
889 890 891
      self.first_pass_data.frames.push(FirstPassFrame::from(fi));
    }

892 893 894 895 896 897 898 899 900 901
    self.frames_processed += 1;
    Ok(Packet {
      data,
      rec,
      number: fi.number,
      frame_type: fi.frame_type,
      psnr
    })
  }

902
  fn garbage_collect(&mut self, cur_frame: u64) {
Josh Holmer's avatar
Josh Holmer committed
903 904 905 906 907 908 909 910 911 912
    if cur_frame == 0 {
      return;
    }
    for i in 0..cur_frame {
      self.frame_q.remove(&i);
    }
    if self.idx < 2 {
      return;
    }
    for i in 0..(self.idx - 1) {
Luca Barbato's avatar
Luca Barbato committed
913
      self.frame_invariants.remove(&i);
Josh Holmer's avatar
Josh Holmer committed
914 915 916 917 918
    }
  }

  fn determine_frame_type(&mut self, frame_number: u64) -> FrameType {
    if frame_number == 0 {
Josh Holmer's avatar
Josh Holmer committed
919 920
      return FrameType::KEY;
    }
921 922
    if self.config.speed_settings.no_scene_detection {
      if frame_number % self.config.max_key_frame_interval == 0 {
923 924 925 926 927
        return FrameType::KEY;
      } else {
        return FrameType::INTER;
      }
    }
Josh Holmer's avatar
Josh Holmer committed
928

Josh Holmer's avatar
Josh Holmer committed
929 930 931 932 933 934 935 936
    let prev_keyframe = self.keyframes.iter()
      .rfind(|&&keyframe| keyframe < frame_number)
      .cloned()
      .unwrap_or(0);
    let frame = match self.frame_q.get(&frame_number).cloned() {
      Some(frame) => frame,
      None => { return FrameType::KEY; }
    };
Josh Holmer's avatar
Josh Holmer committed
937
    if let Some(frame) = frame {
Josh Holmer's avatar
Josh Holmer committed
938
      let distance = frame_number - prev_keyframe;
939 940
      if distance < self.config.min_key_frame_interval {
        if distance + 1 == self.config.min_key_frame_interval {
941
          self.keyframe_detector.set_last_frame(frame, frame_number as usize);
Josh Holmer's avatar
Josh Holmer committed
942 943 944
        }
        return FrameType::INTER;
      }
945
      if distance >= self.config.max_key_frame_interval {
Josh Holmer's avatar
Josh Holmer committed
946 947
        return FrameType::KEY;
      }
Josh Holmer's avatar
Josh Holmer committed
948
      if self.keyframe_detector.detect_scene_change(frame, frame_number as usize) {
Josh Holmer's avatar
Josh Holmer committed
949 950 951 952 953
        return FrameType::KEY;
      }
    }
    FrameType::INTER
  }
954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984

  // Count the number of frames of each subtype in the next
  //  reservoir_frame_delay frames.
  // Returns the number of frames until the last keyframe in the next
  //  reservoir_frame_delay frames, or the end of the interval, whichever
  //  comes first.
  pub(crate) fn guess_frame_subtypes(
    &self, nframes: &mut [i32; FRAME_NSUBTYPES], reservoir_frame_delay: i32
  ) -> i32 {
    // TODO: Ideally this logic should be centralized, but the actual code used
    //  to determine a frame's subtype is spread over many places and
    //  intertwined with mutable state changes that occur when the frame is
    //  actually encoded.
    // So for now we just duplicate it here in stateless fashion.
    for fti in 0..FRAME_NSUBTYPES {
      nframes[fti] = 0;
    }
    let mut prev_keyframe = self.segment_start_idx;
    let mut acc: [i32; FRAME_NSUBTYPES] = [0; FRAME_NSUBTYPES];
    // Updates the frame counts with the accumulated values when we hit a
    //  keyframe.
    fn collect_counts(
      nframes: &mut [i32; FRAME_NSUBTYPES], acc: &mut [i32; FRAME_NSUBTYPES]
    ) {
      for fti in 0..FRAME_NSUBTYPES {
        nframes[fti] += acc[fti];
        acc[fti] = 0;
      }
      acc[FRAME_SUBTYPE_I] += 1;
    }
    for idx in self.idx..(self.idx + reservoir_frame_delay as u64) {
Luca Barbato's avatar
Luca Barbato committed
985
      if let Some(fd) = self.frame_invariants.get(&idx) {
986 987 988 989 990 991
        if fd.frame_type == FrameType::KEY {
          collect_counts(nframes, &mut acc);
          prev_keyframe = idx;
          continue;
        }
      } else if idx == 0
992
        || idx - prev_keyframe >= self.config.max_key_frame_interval
993 994 995 996 997 998 999
      {
        collect_counts(nframes, &mut acc);
        prev_keyframe = idx;
        continue;
      }
      // TODO: Implement golden P-frames.
      let mut fti = FRAME_SUBTYPE_P;
1000
      if !self.config.low_latency {
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026
        let pyramid_depth = 2;
        let group_src_len = 1 << pyramid_depth;
        let group_len = group_src_len + pyramid_depth;
        let idx_in_group = (idx - prev_keyframe - 1) % group_len;
        let lvl = if idx_in_group < pyramid_depth {
          idx_in_group
        } else {
          pos_to_lvl(idx_in_group - pyramid_depth + 1, pyramid_depth)
        };
        fti += lvl as usize;
      }
      acc[fti] += 1;
    }
    if prev_keyframe <= self.idx {
      // If there were no keyframes at all, or only the first frame was a
      //  keyframe, the accumulators never flushed and still contain counts for
      //  the entire buffer.
      // In both cases, we return these counts.
      collect_counts(nframes, &mut acc);
      reservoir_frame_delay
    } else {
      // Otherwise, we discard what remains in the accumulators as they contain
      //  the counts from and past the last keyframe.
      (prev_keyframe - self.idx) as i32
    }
  }
Luca Barbato's avatar
Luca Barbato committed
1027
}
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039

#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct FirstPassData {
  frames: Vec<FirstPassFrame>,
}

#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
pub struct FirstPassFrame {
  number: u64,
  frame_type: FrameType,
}

1040 1041
impl<T: Pixel> From<&FrameInvariants<T>> for FirstPassFrame {
  fn from(fi: &FrameInvariants<T>) -> FirstPassFrame {
1042 1043 1044 1045 1046
    FirstPassFrame {
      number: fi.number,
      frame_type: fi.frame_type,
    }
  }
1047
}
1048 1049 1050 1051 1052

#[cfg(test)]
mod test {
  use super::*;

Luca Barbato's avatar
Luca Barbato committed
1053 1054
  use interpolate_name::interpolate_test;

1055 1056 1057
  fn setup_encoder<T: Pixel>(
    w: usize, h: usize, speed: usize, quantizer: usize, bit_depth: usize,
    chroma_sampling: ChromaSampling, min_keyint: u64, max_keyint: u64,
Luca Barbato's avatar
Luca Barbato committed
1058 1059 1060
    bitrate: i32,
    low_latency: bool,
    no_scene_detection: bool,
1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072
  ) -> Context<T> {
    assert!(bit_depth == 8 || std::mem::size_of::<T>() > 1);
    let mut enc = EncoderConfig::with_speed_preset(speed);
    enc.quantizer = quantizer;
    enc.min_key_frame_interval = min_keyint;
    enc.max_key_frame_interval = max_keyint;
    enc.low_latency = low_latency;
    enc.width = w;
    enc.height = h;
    enc.bit_depth = bit_depth;
    enc.chroma_sampling = chroma_sampling;
    enc.bitrate = bitrate;
Luca Barbato's avatar
Luca Barbato committed
1073
    enc.speed_settings.no_scene_detection = no_scene_detection;
1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093

    let cfg = Config { enc, threads: 0 };

    cfg.new_context()
  }

  /*
  fn fill_frame<T: Pixel>(ra: &mut ChaChaRng, frame: &mut Frame<T>) {
    for plane in frame.planes.iter_mut() {
      let stride = plane.cfg.stride;
      for row in plane.data.chunks_mut(stride) {
        for pixel in row {
          let v: u8 = ra.gen();
          *pixel = T::cast_from(v);
        }
      }
    }
  }
  */

Luca Barbato's avatar
Luca Barbato committed
1094 1095 1096 1097 1098 1099

  #[interpolate_test(low_latency_no_scene_change, true, true)]
  #[interpolate_test(reorder_no_scene_change, false, true)]
  #[interpolate_test(low_latency_scene_change_detection, true, false)]
  #[interpolate_test(reorder_scene_change_detection, false, false)]
  fn flush(low_lantency: bool, no_scene_detection: bool) {
1100 1101
    let mut ctx = setup_encoder::<u8>(64, 80, 10, 100, 8, ChromaSampling::Cs420, 150, 200, 0, low_lantency, no_scene_detection);
    let limit = 41;
1102 1103 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 1130 1131 1132 1133 1134 1135

    ctx.set_limit(limit);

    for _ in  0..limit {
      let input = ctx.new_frame();
      let _ = ctx.send_frame(input);
    }

    ctx.flush();

    let mut count = 0;

    'out: for _ in 0..limit {
      loop {
        match ctx.receive_packet() {
          Ok(_) => {
            eprintln!("Packet Received {}/{}", count, limit);
            count += 1;
          },
          Err(EncoderStatus::EnoughData) => {
            eprintln!("{:?}", EncoderStatus::EnoughData);

            break 'out;
          }
          Err(e) => {
            eprintln!("{:?}", e);
            break;
          }
        }
      }
    }

    assert_eq!(limit, count);
  }
1136 1137 1138 1139 1140 1141 1142


  #[interpolate_test(low_latency_no_scene_change, true, true)]
  #[interpolate_test(reorder_no_scene_change, false, true)]
  #[interpolate_test(low_latency_scene_change_detection, true, false)]
  #[interpolate_test(reorder_scene_change_detection, false, false)]
  fn flush_unlimited(low_lantency: bool, no_scene_detection: bool) {
1143 1144
    let mut ctx = setup_encoder::<u8>(64, 80, 10, 100, 8, ChromaSampling::Cs420, 150, 200, 0, low_lantency, no_scene_detection);
    let limit = 41;
1145 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

    for _ in  0..limit {
      let input = ctx.new_frame();
      let _ = ctx.send_frame(input);
    }

    ctx.flush();

    let mut count = 0;

    'out: for _ in 0..limit {
      loop {
        match ctx.receive_packet() {
          Ok(_) => {
            eprintln!("Packet Received {}/{}", count, limit);
            count += 1;
          },
          Err(EncoderStatus::EnoughData) => {
            eprintln!("{:?}", EncoderStatus::EnoughData);

            break 'out;
          }
          Err(e) => {
            eprintln!("{:?}", e);
            break;
          }
        }
      }
    }

    assert_eq!(limit, count);
  }
1177
}