[PATCH 0/2] x86/h264_intrapred: AVX2 for 10-bit intra 16x16 prediction

Two AVX2 kernels for 10-bit H.264 intra 16x16 prediction.

Patch 1 (dc, top/left/128 dc, vertical, horizontal) was sent a couple of
months ago and got no review; resending, rebased on current master.
Original: https://patchwork.ffmpeg.org/project/ffmpeg/patch/[email protected]/

Patch 2 adds plane prediction, which had no SIMD at all for 10-bit. It
depends on patch 1, they share the AVX2 dispatch block.

checkasm passes. Cycle counts from checkasm --bench:

  pred16x16_dc_10           35.6 -> 20.2   1.76x
  pred16x16_horizontal_10   32.4 -> 19.3   1.68x
  pred16x16_plane_10       453.5 -> 49.4   9.18x

ZaneHam (2):
  avcodec/x86/h264_intrapred: add AVX2 for 10-bit pred16x16 functions
  avcodec/x86/h264_intrapred: add AVX2 for 10-bit pred16x16 plane

 libavcodec/x86/h264_intrapred_10bit.asm | 283 ++++++++++++++++++++++++
 libavcodec/x86/h264_intrapred_init.c    |  16 ++
 2 files changed, 299 insertions(+)

-- 
2.51.0.windows.2

_______________________________________________
ffmpeg-devel mailing list -- [email protected]
To unsubscribe send an email to [email protected]