On Tue, 24 May 2022, J. Dekker wrote:
libavcodec/aarch64/Makefile | 1 +
libavcodec/aarch64/hevcdsp_init_aarch64.c | 43 +-
libavcodec/aarch64/hevcdsp_qpel_neon.S| 520 ++
3 files changed, 563 insertions(+), 1 deletion(-)
create mode 100644 libavcodec/aarch6
checkasm --benchmark on Ampere Altra (Neoverse N1):
put_hevc_qpel_bi_h4_8_c: 173.7
put_hevc_qpel_bi_h4_8_neon: 77.0
put_hevc_qpel_bi_h6_8_c: 385.7
put_hevc_qpel_bi_h6_8_neon: 125.7
put_hevc_qpel_bi_h8_8_c: 680.7
put_hevc_qpel_bi_h8_8_neon: 137.5
put_hevc_qpel_bi_h12_8_c: 1480.0
put_hevc_qpel_bi_h1