1. 22 Aug, 2018 1 commit
  2. 08 Aug, 2018 1 commit
  3. 02 Aug, 2018 12 commits
  4. 25 Jun, 2018 1 commit
  5. 24 Jun, 2018 2 commits
  6. 23 Jun, 2018 3 commits
    • Victoria Zhislina's avatar
      hbd lpf sse2 perf and code quality improvement · 1c2430a2
      Victoria Zhislina authored
      aom_highbd_lpf_horizontal_14_sse2 -1.15x perf
      due to full sse2 register width usage for some ops
      highbd_hev_filter_mask_x_sse2 added for code quality
      if blocks in _4,_6,_8 and _14 eliminated
      Change-Id: Ie28a70798833c95fb21cac238ffdebfcead5f0a7
    • Xing Jin's avatar
      Speed up av1_optimize_b · bd91e946
      Xing Jin authored
      1. Remove calling of get_txb_ctx from av1_optimize_b
       to it's caller, outside of rdo loops.
      2. Remove calling of av1_get_tx_type, tx_type can be
       passed in by it's caller.
      3. For encoder, about 1.3% faster shows by encoding
      20 frame of BasketballDrill_832x480_50.y4m, with no
      coding loss.  ( 601278 ms -> 592634 ms)
      a) gcc (Ubuntu 5.4.0-6ubuntu1~16.04.9) 5.4.0 20160609
      b) CPU: Intel(R) Core(TM) i5-4590 CPU @ 3.30GHz
      c) Config cmd
      d) Test cmd:
      ./aomenc --cpu-used=1 --end-usage=vbr \
      --target-bitrate=800 --limit=20
      Change-Id: I755b337e29316f4ceed37c9b669aebb4ad2d5fac
    • Zoe Liu's avatar
      Merge two flags to one for hierarchical structure · e76041ec
      Zoe Liu authored
      Remove the extra flag to simply the decision making for the hierarchical
      structure construction.
      Change-Id: Iec6650270ee283d82bf51885efe432c79edcd710
  7. 22 Jun, 2018 14 commits
  8. 21 Jun, 2018 6 commits
    • Yue Chen's avatar
      Add wavelet energy based q modulation · 2dbdbc9d
      Yue Chen authored
      We introduce wavelet energy analysis for delta q modulation
      because wavelet energy might better reflects the complexity of
      image patches. This feature can be turned on by setting macro
      DELTAQ_MODULATION as 1 in encodeframe.h and specify "--deltaq-mode
      =1" from command line. It is not replacing variance based deltaq
      because tuning is needed to keep the bitrate closer to what
      is expected for the specified q level.
      BDRate compared to variance based deltaq under different metrics
      PSNR:-5.188% SSIM:-8.443% percept:-3.268%
      Change-Id: I307eb3ca5633704e31812c691dbc9149036951a4
    • Angie Chiang's avatar
      Turn off inter_mode_data_push on multi-tile mode · af2f1980
      Angie Chiang authored
      Change-Id: I6a035886c255f646d88a7606e3ac11cb23e07a89
    • David Barker's avatar
      [NORMATIVE] Fix random-access decoding · f6f8fa18
      David Barker authored
      To conform the the specification, we need to be able to decode
      sequences beginning with a forward keyframe, followed by a show-existing
      frame displaying that keyframe.
      However, we currently can't do that if frame IDs are in use. The code
      currently assumes that a previous frame ID is available for every frame
      that is not a shown keyframe. So for the initial forward keyframe,
      we think that there is a previous frame with ID 0. This causes things
      to break if the forward keyframe's ID is too far away from 0.
      The fix is to explicitly track whether we've decoded any frames yet,
      and to allow the first frame decoded to use any frame ID value.
      Change-Id: I6346dd070d7b3615c24bef63b332d8901bf13aef
    • Johann Koenig's avatar
      Revert "Fix bug in round shift array unit test" · 70bffc7d
      Johann Koenig authored
      This reverts commit e61e970e.
      Apparently the encoder has uses for this outside the decoder
      Change-Id: I6acef55dcf95d2dd35844ca7540d7ce203b5761f
    • Angie Chiang's avatar
      Implement handle_intra_mode · 798bc48f
      Angie Chiang authored
      Change-Id: I72e97fc8cdeb63cbbddb261bfca15cb2482eae18
    • Angie Chiang's avatar
      Move skip_repeated_mv into order indep skip func · 68356345
      Angie Chiang authored
      Change-Id: I6ba5948af9f56db0073ff0e081cbae9909732539