lavc/aarch64: add pred16x16 10-bit functions
authorMikhail Nitenko <mnitenko@gmail.com>
Thu, 15 Apr 2021 09:13:52 +0000 (12:13 +0300)
committerJosh Dekker <josh@itanimul.li>
Mon, 19 Apr 2021 07:01:14 +0000 (09:01 +0200)
commit84ac1440b2a726b8a989dafdf634081651de1c10
tree43fc1ed474f170aa4dd1452583f7acf59efc568c
parent6b2e7dc82862c6f87517c928c3408c3f8d62d671
lavc/aarch64: add pred16x16 10-bit functions

Benchmarks:                      A53     A72
pred16x16_dc_10_c:              136.0   124.0
pred16x16_dc_10_neon:           121.2   106.0
pred16x16_horizontal_10_c:      155.0    73.2
pred16x16_horizontal_10_neon:    82.2    67.7
pred16x16_top_dc_10_c:          106.0    93.7
pred16x16_top_dc_10_neon:        87.7    77.2
pred16x16_vertical_10_c:         83.0    67.7
pred16x16_vertical_10_neon:      54.2    61.7

Some functions work slower than C and are left commented out.
libavcodec/aarch64/h264pred_init.c
libavcodec/aarch64/h264pred_neon.S