VaVDA: split kReduced and kSuperReduced modes

H264 GetNumReferenceFrames() was increased by 1 [1] for all platforms to
circumvent a limitation during introduction of the kReduced mode; this
- Renames VaVDA::BufferAllocationMode::kReduced to kSuperReduced.
- Takes that +1 out of h264_decoder
- Introduces kReduced to represent this H264 case where we +1 the
  amount of media pipeline frames allocated.

Note that this CL doesn't change ToT: there's an extra allocation
in the # codec reference frames _and_ in the amount of pipeline

Test results ToT:
With this change: and (2nd round) - similar.

This way, we can have all buffer amount calculations in one place,
specially with a view to introducing kReduced for platforms < SkyLake.

[1] (bc of

Test: tast run DEVICE_IP video.PlaybackPerfH2642160P60FPS on a SkyLake device (cave)
Bug: 912295, 920510
Change-Id: Ice286b1bd4db21f353d86cb9c65467e1322963be
Reviewed-by: Miguel Casas <>
Commit-Queue: Miguel Casas <>
Cr-Commit-Position: refs/heads/master@{#629774}
4 files changed