api.rs 25.8 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.

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

Raphaël Zumer's avatar
Raphaël Zumer committed
21
use std::{cmp, fmt, io};
fbossen's avatar
fbossen committed
22
use std::collections::BTreeMap;
Luca Barbato's avatar
Luca Barbato committed
23
use std::sync::Arc;
Josh Holmer's avatar
Josh Holmer committed
24
use std::collections::BTreeSet;
25
use std::path::PathBuf;
Josh Holmer's avatar
Josh Holmer committed
26 27

const LOOKAHEAD_FRAMES: u64 = 10;
Luca Barbato's avatar
Luca Barbato committed
28 29 30

// TODO: use the num crate?
#[derive(Clone, Copy, Debug)]
31
#[repr(C)]
Luca Barbato's avatar
Luca Barbato committed
32 33 34
pub struct Rational {
  pub num: u64,
  pub den: u64
Luca Barbato's avatar
Luca Barbato committed
35 36
}

Luca Barbato's avatar
Luca Barbato committed
37 38 39
impl Rational {
  pub fn new(num: u64, den: u64) -> Self {
    Rational { num, den }
Luca Barbato's avatar
Luca Barbato committed
40 41 42
  }
}

43 44 45 46 47 48 49
#[derive(Clone, Copy, Debug)]
#[repr(C)]
pub struct Point {
  pub x: u16,
  pub y: u16
}

50
#[derive(Clone, Debug)]
51
pub struct EncoderConfig {
52
  // output size
53 54
  pub width: usize,
  pub height: usize,
55 56

  // data format and ancillary color information
57 58 59
  pub bit_depth: usize,
  pub chroma_sampling: ChromaSampling,
  pub chroma_sample_position: ChromaSamplePosition,
60 61 62 63 64 65
  pub pixel_range: PixelRange,
  pub color_description: Option<ColorDescription>,
  pub mastering_display: Option<MasteringDisplay>,
  pub content_light: Option<ContentLight>,

  // encoder configuration
66
  pub time_base: Rational,
Josh Holmer's avatar
Josh Holmer committed
67 68 69 70
  /// The *minimum* interval between two keyframes
  pub min_key_frame_interval: u64,
  /// The *maximum* interval between two keyframes
  pub max_key_frame_interval: u64,
71
  pub low_latency: bool,
72
  pub quantizer: usize,
73
  pub bitrate: i32,
74 75
  pub tune: Tune,
  pub speed_settings: SpeedSettings,
76 77
  /// `None` for one-pass encode. `Some(1)` or `Some(2)` for two-pass encoding.
  pub pass: Option<u8>,
78
  pub show_psnr: bool,
79
  pub stats_file: Option<PathBuf>,
80 81 82 83
}

impl Default for EncoderConfig {
  fn default() -> Self {
84 85
    const DEFAULT_SPEED: usize = 3;
    Self::with_speed_preset(DEFAULT_SPEED)
86 87 88
  }
}

89 90 91
impl EncoderConfig {
  pub fn with_speed_preset(speed: usize) -> Self {
    EncoderConfig {
92 93
      width: 640,
      height: 480,
94

95 96 97
      bit_depth: 8,
      chroma_sampling: ChromaSampling::Cs420,
      chroma_sample_position: ChromaSamplePosition::Unknown,
98 99 100 101 102
      pixel_range: PixelRange::Unspecified,
      color_description: None,
      mastering_display: None,
      content_light: None,

103
      time_base: Rational { num: 30, den: 1 },
Josh Holmer's avatar
Josh Holmer committed
104 105
      min_key_frame_interval: 12,
      max_key_frame_interval: 240,
106
      low_latency: false,
107
      quantizer: 100,
108
      bitrate: 0,
109
      tune: Tune::Psnr,
110
      speed_settings: SpeedSettings::from_preset(speed),
111
      pass: None,
112
      show_psnr: false,
113
      stats_file: None,
114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
    }
  }
}

#[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,
  pub encode_bottomup: bool,
  pub rdo_tx_decision: bool,
  pub prediction_modes: PredictionModesSetting,
  pub include_near_mvs: bool,
129 130 131 132 133 134
  pub no_scene_detection: bool,
}

impl Default for SpeedSettings {
  fn default() -> Self {
    SpeedSettings {
135
      min_block_size: BlockSize::BLOCK_16X16,
136 137 138 139 140 141 142 143 144 145 146
      multiref: false,
      fast_deblock: false,
      reduced_tx_set: false,
      tx_domain_distortion: false,
      encode_bottomup: false,
      rdo_tx_decision: false,
      prediction_modes: PredictionModesSetting::Simple,
      include_near_mvs: false,
      no_scene_detection: false,
    }
  }
147 148 149 150 151 152 153 154 155 156 157 158 159 160
}

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),
      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),
161
      no_scene_detection: Self::no_scene_detection_preset(speed),
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
    }
  }

  fn min_block_size_preset(speed: usize) -> BlockSize {
    if speed <= 1 {
      BlockSize::BLOCK_4X4
    } else if speed <= 2 {
      BlockSize::BLOCK_8X8
    } else if speed <= 3 {
      BlockSize::BLOCK_16X16
    } else if speed <= 4 {
      BlockSize::BLOCK_32X32
    } else {
      BlockSize::BLOCK_64X64
    }
  }

  fn multiref_preset(speed: usize) -> bool {
    speed <= 2
  }

  fn fast_deblock_preset(speed: usize) -> bool {
    speed >= 4
  }

  fn reduced_tx_set_preset(speed: usize) -> bool {
    speed >= 2
  }

  fn tx_domain_distortion_preset(speed: usize) -> bool {
192
    speed >= 4
193 194 195 196 197 198 199 200 201 202 203
  }

  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 {
204
    if speed <= 1 {
205
      PredictionModesSetting::ComplexAll
206
    } else if speed <= 3 {
207 208 209 210 211 212 213 214 215
      PredictionModesSetting::ComplexKeyframes
    } else {
      PredictionModesSetting::Simple
    }
  }

  fn include_near_mvs_preset(speed: usize) -> bool {
    speed <= 2
  }
216 217 218 219

  fn no_scene_detection_preset(speed: usize) -> bool {
    speed == 10
  }
220 221
}

222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243
#[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"),
    }
  }
}

244 245 246 247 248 249 250
#[derive(Clone, Copy, Debug, PartialOrd, PartialEq)]
pub enum PredictionModesSetting {
  Simple,
  ComplexKeyframes,
  ComplexAll,
}

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
#[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
  }
}

293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308
arg_enum!{
  #[derive(Debug, Clone, Copy, PartialEq)]
  #[repr(C)]
  pub enum PixelRange {
      Unspecified = 0,
      Limited,
      Full,
  }
}

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

309 310
arg_enum!{
  #[derive(Debug, Clone, Copy, PartialEq)]
311
  #[repr(C)]
312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327
  pub enum MatrixCoefficients {
      Identity = 0,
      BT709,
      Unspecified,
      BT470M = 4,
      BT470BG,
      ST170M,
      ST240M,
      YCgCo,
      BT2020NonConstantLuminance,
      BT2020ConstantLuminance,
      ST2085,
      ChromaticityDerivedNonConstantLuminance,
      ChromaticityDerivedConstantLuminance,
      ICtCp,
  }
328 329 330 331 332 333 334 335
}

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

336
arg_enum!{
337
  #[derive(Debug, Clone, Copy, PartialEq)]
338
  #[repr(C)]
339 340 341 342 343 344 345 346 347 348 349 350 351 352
  pub enum ColorPrimaries {
      BT709 = 1,
      Unspecified,
      BT470M = 4,
      BT470BG,
      ST170M,
      ST240M,
      Film,
      BT2020,
      ST428,
      P3DCI,
      P3Display,
      Tech3213 = 22,
  }
353 354 355 356 357 358 359 360
}

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

361
arg_enum!{
362
  #[derive(Debug, Clone, Copy, PartialEq)]
363
  #[repr(C)]
364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382
  pub enum TransferCharacteristics {
      BT1886 = 1,
      Unspecified,
      BT470M = 4,
      BT470BG,
      ST170M,
      ST240M,
      Linear,
      Logarithmic100,
      Logarithmic316,
      XVYCC,
      BT1361E,
      SRGB,
      BT2020Ten,
      BT2020Twelve,
      PerceptualQuantizer,
      ST428,
      HybridLogGamma,
  }
383 384 385 386 387 388 389 390
}

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

391 392 393 394 395 396 397
#[derive(Copy, Clone, Debug)]
pub struct ColorDescription {
    pub color_primaries: ColorPrimaries,
    pub transfer_characteristics: TransferCharacteristics,
    pub matrix_coefficients: MatrixCoefficients
}

398 399 400 401 402 403 404 405 406 407 408 409 410 411
#[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,
}

412
/// Contain all the encoder configuration
413
#[derive(Clone, Debug)]
414
pub struct Config {
Luca Barbato's avatar
Luca Barbato committed
415 416 417 418
  pub enc: EncoderConfig
}

impl Config {
Luca Barbato's avatar
Luca Barbato committed
419 420
  pub fn parse(&mut self, key: &str, value: &str) -> Result<(), EncoderStatus> {
    match key {
421
      "low_latency" => self.enc.low_latency = value.parse().map_err(|_e| ParseError)?,
Josh Holmer's avatar
Josh Holmer committed
422 423
      "min_key_frame_interval" => self.enc.min_key_frame_interval = value.parse().map_err(|_e| ParseError)?,
      "key_frame_interval" => self.enc.max_key_frame_interval = value.parse().map_err(|_e| ParseError)?,
424
      "quantizer" => self.enc.quantizer = value.parse().map_err(|_e| ParseError)?,
425
      "speed" => self.enc.speed_settings = SpeedSettings::from_preset(value.parse().map_err(|_e| ParseError)?),
426 427
      "tune" => self.enc.tune = value.parse().map_err(|_e| ParseError)?,
      _ => return Err(InvalidKey)
Luca Barbato's avatar
Luca Barbato committed
428 429 430 431 432
    }

    Ok(())
  }

Luca Barbato's avatar
Luca Barbato committed
433
  pub fn new_context(&self) -> Context {
434
    #[cfg(feature = "aom")]
Luca Barbato's avatar
Luca Barbato committed
435
    unsafe {
Luca Barbato's avatar
Luca Barbato committed
436 437
      av1_rtcd();
      aom_dsp_rtcd();
Luca Barbato's avatar
Luca Barbato committed
438 439
    }

440 441 442
    // initialize with temporal delimiter
    let packet_data = TEMPORAL_DELIMITER.to_vec();

443 444 445 446 447 448
    let maybe_ac_qi_max = if self.enc.quantizer < 255 {
      Some(self.enc.quantizer as u8)
    } else {
      None
    };

Josh Holmer's avatar
Josh Holmer committed
449 450
    Context {
      frame_count: 0,
451
      limit: 0,
Josh Holmer's avatar
Josh Holmer committed
452
      idx: 0,
Josh Holmer's avatar
Josh Holmer committed
453
      frames_processed: 0,
Josh Holmer's avatar
Josh Holmer committed
454
      frame_q: BTreeMap::new(),
Josh Holmer's avatar
Josh Holmer committed
455 456
      frame_data: BTreeMap::new(),
      keyframes: BTreeSet::new(),
457
      packet_data,
Josh Holmer's avatar
Josh Holmer committed
458 459
      segment_start_idx: 0,
      segment_start_frame: 0,
460
      keyframe_detector: SceneChangeDetector::new(self.enc.bit_depth),
461
      config: self.clone(),
462 463 464 465 466 467 468 469 470 471
      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,
472 473 474
      first_pass_data: FirstPassData {
        frames: Vec::new(),
      },
Josh Holmer's avatar
Josh Holmer committed
475
    }
Luca Barbato's avatar
Luca Barbato committed
476 477 478 479
  }
}

pub struct Context {
Luca Barbato's avatar
Luca Barbato committed
480
  //    timebase: Rational,
fbossen's avatar
fbossen committed
481
  frame_count: u64,
482
  limit: u64,
483
  pub(crate) idx: u64,
Josh Holmer's avatar
Josh Holmer committed
484 485
  frames_processed: u64,
  /// Maps frame *number* to frames
486
  frame_q: BTreeMap<u64, Option<Arc<Frame>>>, //    packet_q: VecDeque<Packet>
Josh Holmer's avatar
Josh Holmer committed
487 488 489 490 491
  /// Maps frame *idx* to frame data
  frame_data: BTreeMap<u64, FrameInvariants>,
  /// A list of keyframe *numbers* in this encode. Needed so that we don't
  /// need to keep all of the frame_data in memory for the whole life of the encode.
  keyframes: BTreeSet<u64>,
492
  /// A storage space for reordered frames.
Josh Holmer's avatar
Josh Holmer committed
493 494 495 496
  packet_data: Vec<u8>,
  segment_start_idx: u64,
  segment_start_frame: u64,
  keyframe_detector: SceneChangeDetector,
497
  pub config: Config,
498
  rc_state: RCState,
499
  maybe_prev_log_base_q: Option<i64>,
500
  pub first_pass_data: FirstPassData,
Luca Barbato's avatar
Luca Barbato committed
501 502 503 504 505 506 507 508 509
}

#[derive(Clone, Copy, Debug)]
pub enum EncoderStatus {
  /// The encoder needs more Frames to produce an output Packet
  NeedMoreData,
  /// There are enough Frames queue
  EnoughData,
  ///
Luca Barbato's avatar
Luca Barbato committed
510 511 512
  Failure,
  InvalidKey,
  ParseError
Luca Barbato's avatar
Luca Barbato committed
513 514 515 516
}

pub struct Packet {
  pub data: Vec<u8>,
fbossen's avatar
fbossen committed
517 518
  pub rec: Option<Frame>,
  pub number: u64,
519 520 521
  pub frame_type: FrameType,
  /// PSNR for Y, U, and V planes
  pub psnr: Option<(f64, f64, f64)>,
522 523 524 525
}

impl fmt::Display for Packet {
  fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
Luca Barbato's avatar
Luca Barbato committed
526 527 528 529 530 531 532
    write!(
      f,
      "Frame {} - {} - {} bytes",
      self.number,
      self.frame_type,
      self.data.len()
    )
533
  }
Luca Barbato's avatar
Luca Barbato committed
534 535 536 537
}

impl Context {
  pub fn new_frame(&self) -> Arc<Frame> {
Josh Holmer's avatar
Josh Holmer committed
538
    Arc::new(Frame::new(
539 540
      self.config.enc.width,
      self.config.enc.height,
541
      self.config.enc.chroma_sampling
Josh Holmer's avatar
Josh Holmer committed
542
    ))
Luca Barbato's avatar
Luca Barbato committed
543 544 545 546 547 548
  }

  pub fn send_frame<F>(&mut self, frame: F) -> Result<(), EncoderStatus>
  where
    F: Into<Option<Arc<Frame>>>
  {
Josh Holmer's avatar
Josh Holmer committed
549 550
    let idx = self.frame_count;
    self.frame_q.insert(idx, frame.into());
551
    self.frame_count += 1;
Luca Barbato's avatar
Luca Barbato committed
552 553 554
    Ok(())
  }

555 556 557 558
  pub fn get_frame_count(&self) -> u64 {
    self.frame_count
  }

559 560
  pub fn set_limit(&mut self, limit: u64) {
    self.limit = limit;
561 562
  }

Josh Holmer's avatar
Josh Holmer committed
563 564 565 566
  pub 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)
  }

567
  pub fn needs_more_frames(&self, frame_count: u64) -> bool {
568
    self.limit == 0 || frame_count < self.limit
569 570
  }

571 572 573
  pub fn container_sequence_header(&mut self) -> Vec<u8> {
    fn sequence_header_inner(seq: &Sequence) -> io::Result<Vec<u8>> {
      let mut buf = Vec::new();
574

575
      {
576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592
        let mut bw = BitWriter::endian(&mut buf, BigEndian);
        bw.write_bit(true)?; // marker
        bw.write(7, 1)?; // version
        bw.write(3, seq.profile)?;
        bw.write(5, 32)?; // level
        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
593
      }
594

595 596 597
      Ok(buf)
    }

Josh Holmer's avatar
Josh Holmer committed
598
    sequence_header_inner(&self.frame_data[&0].sequence).unwrap()
599 600
  }

Josh Holmer's avatar
Josh Holmer committed
601
  fn next_keyframe(&self) -> u64 {
Josh Holmer's avatar
Josh Holmer committed
602 603 604 605
    let next_detected = self.frame_data.values()
      .find(|fi| fi.frame_type == FrameType::KEY && fi.number > self.segment_start_frame)
      .map(|fi| fi.number);
    let next_limit = self.segment_start_frame + self.config.enc.max_key_frame_interval;
Josh Holmer's avatar
Josh Holmer committed
606 607 608 609 610
    if next_detected.is_none() {
      return next_limit;
    }
    cmp::min(next_detected.unwrap(), next_limit)
  }
fbossen's avatar
fbossen committed
611

612
  fn set_frame_properties(&mut self, idx: u64) -> bool {
613
    let (fi, end_of_subgop) = self.build_frame_properties(idx);
614 615 616
    self.frame_data.insert(idx, fi);

    end_of_subgop
Josh Holmer's avatar
Josh Holmer committed
617 618
  }

619
  fn build_frame_properties(&mut self, idx: u64) -> (FrameInvariants, bool) {
Josh Holmer's avatar
Josh Holmer committed
620
    if idx == 0 {
621
      let seq = Sequence::new(&self.config.enc);
622

Josh Holmer's avatar
Josh Holmer committed
623
      // The first frame will always be a key frame
Josh Holmer's avatar
Josh Holmer committed
624 625
      let fi = FrameInvariants::new_key_frame(
        &FrameInvariants::new(
626
          self.config.enc.clone(),
627
          seq
Josh Holmer's avatar
Josh Holmer committed
628 629 630
        ),
        0
      );
631
      return (fi, true);
Josh Holmer's avatar
Josh Holmer committed
632 633
    }

Josh Holmer's avatar
Josh Holmer committed
634 635
    let mut fi = self.frame_data[&(idx - 1)].clone();

636 637 638 639 640
    // FIXME: inter unsupported with 4:2:2 and 4:4:4 chroma sampling
    let chroma_sampling = self.config.enc.chroma_sampling;
    let keyframe_only = chroma_sampling == ChromaSampling::Cs444 ||
      chroma_sampling == ChromaSampling::Cs422;

Josh Holmer's avatar
Josh Holmer committed
641 642 643 644 645 646
    // 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 {
647
      let next_keyframe = if keyframe_only { self.segment_start_frame + 1 } else { self.next_keyframe() };
648
      let (fi_temp, end_of_subgop) = FrameInvariants::new_inter_frame(
Josh Holmer's avatar
Josh Holmer committed
649
        &fi,
650 651 652 653
        self.segment_start_frame,
        idx_in_segment,
        next_keyframe
      );
Josh Holmer's avatar
Josh Holmer committed
654
      fi = fi_temp;
655
      if !end_of_subgop {
Josh Holmer's avatar
Josh Holmer committed
656 657 658
        if !fi.inter_cfg.unwrap().reorder
          || ((idx_in_segment - 1) % fi.inter_cfg.unwrap().group_len == 0
          && fi.number == (next_keyframe - 1))
659
        {
Josh Holmer's avatar
Josh Holmer committed
660
          self.segment_start_idx = idx;
Josh Holmer's avatar
Josh Holmer committed
661
          self.segment_start_frame = next_keyframe;
Josh Holmer's avatar
Josh Holmer committed
662
          fi.number = next_keyframe;
Josh Holmer's avatar
Josh Holmer committed
663
        } else {
664
          return (fi, false);
Josh Holmer's avatar
Josh Holmer committed
665 666
        }
      }
fbossen's avatar
fbossen committed
667
    }
Luca Barbato's avatar
Luca Barbato committed
668

669 670
    match self.frame_q.get(&fi.number) {
      Some(Some(_)) => {},
Josh Holmer's avatar
Josh Holmer committed
671
      _ => { return (fi, false); }
672 673
    }

Josh Holmer's avatar
Josh Holmer committed
674
    // Now that we know the frame number, look up the correct frame type
Josh Holmer's avatar
Josh Holmer committed
675 676 677 678 679 680 681
    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
682

Josh Holmer's avatar
Josh Holmer committed
683 684 685 686 687
    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();
688
      let (fi_temp, end_of_subgop) = FrameInvariants::new_inter_frame(
Josh Holmer's avatar
Josh Holmer committed
689 690 691 692 693 694
        &fi,
        self.segment_start_frame,
        idx_in_segment,
        next_keyframe
      );
      fi = fi_temp;
695 696
      if !end_of_subgop {
        return (fi, false);
Josh Holmer's avatar
Josh Holmer committed
697 698
      }
    }
699
    (fi, true)
fbossen's avatar
fbossen committed
700
  }
Luca Barbato's avatar
Luca Barbato committed
701

fbossen's avatar
fbossen committed
702
  pub fn receive_packet(&mut self) -> Result<Packet, EncoderStatus> {
703 704 705 706 707 708
    let idx = {
      let mut idx = self.idx;
      while !self.set_frame_properties(idx) {
        self.idx += 1;
        idx = self.idx;
      }
Luca Barbato's avatar
Luca Barbato committed
709

710 711 712 713 714 715
      if !self.needs_more_frames(self.frame_data.get(&idx).unwrap().number) {
        self.idx += 1;
        return Err(EncoderStatus::EnoughData);
      }
      idx
    };
716

717
    let ret = {
Luca Barbato's avatar
Luca Barbato committed
718 719 720
      let fi = self.frame_data.get_mut(&idx).unwrap();
      if fi.show_existing_frame {
        self.idx += 1;
Luca Barbato's avatar
Luca Barbato committed
721

Luca Barbato's avatar
Luca Barbato committed
722
        let mut fs = FrameState::new(fi);
fbossen's avatar
fbossen committed
723

724 725
        // TODO: Record the bits spent here against the original frame for rate
        //  control purposes, or add a new frame subtype?
726 727
        let sef_data = encode_frame(fi, &mut fs);
        self.packet_data.extend(sef_data);
Luca Barbato's avatar
Luca Barbato committed
728

Luca Barbato's avatar
Luca Barbato committed
729
        let rec = if fi.show_frame { Some(fs.rec) } else { None };
730 731
        let fi = fi.clone();
        self.finalize_packet(&*fs.input, rec, &fi)
Luca Barbato's avatar
Luca Barbato committed
732 733 734
      } else {
        if let Some(f) = self.frame_q.get(&fi.number) {
          self.idx += 1;
fbossen's avatar
fbossen committed
735

736
          if let Some(frame) = f.clone() {
737
            let fti = fi.get_frame_subtype();
738 739 740
            let qps =
              self.rc_state.select_qi(self, fti, self.maybe_prev_log_base_q);
            let fi = self.frame_data.get_mut(&idx).unwrap();
741
            fi.set_quantizers(&qps);
Luca Barbato's avatar
Luca Barbato committed
742
            let mut fs = FrameState::new_with_frame(fi, frame.clone());
fbossen's avatar
fbossen committed
743

744
            // TODO: Trial encoding for first frame of each type.
Luca Barbato's avatar
Luca Barbato committed
745
            let data = encode_frame(fi, &mut fs);
746 747 748 749 750 751 752 753
            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
754
            self.packet_data.extend(data);
fbossen's avatar
fbossen committed
755

Luca Barbato's avatar
Luca Barbato committed
756
            fs.rec.pad(fi.width, fi.height);
fbossen's avatar
fbossen committed
757

Luca Barbato's avatar
Luca Barbato committed
758 759
            // 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
760

Luca Barbato's avatar
Luca Barbato committed
761
            update_rec_buffer(fi, fs);
fbossen's avatar
fbossen committed
762

Luca Barbato's avatar
Luca Barbato committed
763
            if fi.show_frame {
764 765
              let fi = fi.clone();
              self.finalize_packet(&*frame, rec, &fi)
Luca Barbato's avatar
Luca Barbato committed
766 767 768
            } else {
              Err(EncoderStatus::NeedMoreData)
            }
769 770 771
          } else {
            Err(EncoderStatus::NeedMoreData)
          }
fbossen's avatar
fbossen committed
772 773 774 775
        } else {
          Err(EncoderStatus::NeedMoreData)
        }
      }
776 777 778
    };

    if let Ok(ref pkt) = ret {
Luca Barbato's avatar
Luca Barbato committed
779
      self.garbage_collect(pkt.number);
780 781 782
    }

    ret
Luca Barbato's avatar
Luca Barbato committed
783 784
  }

785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802
  fn finalize_packet(&mut self, original_frame: &Frame, rec: Option<Frame>, fi: &FrameInvariants) -> Result<Packet, EncoderStatus> {
    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;
    if self.config.enc.show_psnr {
      if let Some(ref rec) = rec {
        psnr = Some(calculate_frame_psnr(
          &*original_frame,
          rec,
          fi.sequence.bit_depth
        ));
      }
    }

803 804 805 806
    if self.config.enc.pass == Some(1) {
      self.first_pass_data.frames.push(FirstPassFrame::from(fi));
    }

807 808 809 810 811 812 813 814 815 816
    self.frames_processed += 1;
    Ok(Packet {
      data,
      rec,
      number: fi.number,
      frame_type: fi.frame_type,
      psnr
    })
  }

817
  fn garbage_collect(&mut self, cur_frame: u64) {
Josh Holmer's avatar
Josh Holmer committed
818 819 820 821 822 823 824 825 826 827 828 829 830 831
    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) {
      self.frame_data.remove(&i);
    }
  }

Luca Barbato's avatar
Luca Barbato committed
832
  pub fn flush(&mut self) {
fbossen's avatar
fbossen committed
833
    self.frame_q.insert(self.frame_count, None);
834
    self.frame_count += 1;
Luca Barbato's avatar
Luca Barbato committed
835
  }
Josh Holmer's avatar
Josh Holmer committed
836

Josh Holmer's avatar
Josh Holmer committed
837 838
  fn determine_frame_type(&mut self, frame_number: u64) -> FrameType {
    if frame_number == 0 {
Josh Holmer's avatar
Josh Holmer committed
839 840
      return FrameType::KEY;
    }
841 842 843 844 845 846 847
    if self.config.enc.speed_settings.no_scene_detection {
      if frame_number % self.config.enc.max_key_frame_interval == 0 {
        return FrameType::KEY;
      } else {
        return FrameType::INTER;
      }
    }
Josh Holmer's avatar
Josh Holmer committed
848

Josh Holmer's avatar
Josh Holmer committed
849 850 851 852 853 854 855 856
    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
857
    if let Some(frame) = frame {
Josh Holmer's avatar
Josh Holmer committed
858 859 860
      let distance = frame_number - prev_keyframe;
      if distance < self.config.enc.min_key_frame_interval {
        if distance + 1 == self.config.enc.min_key_frame_interval {
Josh Holmer's avatar
Josh Holmer committed
861 862
          // Run the detector for the current frame, so that it will contain this frame's information
          // to compare against the next frame. We can ignore the results for this frame.
Josh Holmer's avatar
Josh Holmer committed
863
          self.keyframe_detector.detect_scene_change(frame, frame_number as usize);
Josh Holmer's avatar
Josh Holmer committed
864 865 866
        }
        return FrameType::INTER;
      }
Josh Holmer's avatar
Josh Holmer committed
867
      if distance >= self.config.enc.max_key_frame_interval {
Josh Holmer's avatar
Josh Holmer committed
868 869
        return FrameType::KEY;
      }
Josh Holmer's avatar
Josh Holmer committed
870
      if self.keyframe_detector.detect_scene_change(frame, frame_number as usize) {
Josh Holmer's avatar
Josh Holmer committed
871 872 873 874 875
        return FrameType::KEY;
      }
    }
    FrameType::INTER
  }
876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948

  // 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) {
      if let Some(fd) = self.frame_data.get(&idx) {
        if fd.frame_type == FrameType::KEY {
          collect_counts(nframes, &mut acc);
          prev_keyframe = idx;
          continue;
        }
      } else if idx == 0
        || idx - prev_keyframe >= self.config.enc.max_key_frame_interval
      {
        collect_counts(nframes, &mut acc);
        prev_keyframe = idx;
        continue;
      }
      // TODO: Implement golden P-frames.
      let mut fti = FRAME_SUBTYPE_P;
      if !self.config.enc.low_latency {
        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
949
}
950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968

#[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,
}

impl From<&FrameInvariants> for FirstPassFrame {
  fn from(fi: &FrameInvariants) -> FirstPassFrame {
    FirstPassFrame {
      number: fi.number,
      frame_type: fi.frame_type,
    }
  }
969
}