luctrudeau

#!/bin/bash


JOBS=(1 2 4 8 16 32 64)
ENCODE_THREADS=(1 2 4)
NUM_CPU_THREADS=64


for NUM_ENCODE_THREADS in ${ENCODE_THREADS[*]}; do
  for NUM_JOBS in ${JOBS[*]}; do

luctrudeau

# perf record ./vpxenc --good --cpu-used=2 --codec=vp9 --psnr -v -o test.webm ~/Videos/ducks_take_off_1080p50_60f.y4m 
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 1M of event 'cycles:ppp'
# Event count (approx.): 1615856152685
#
# Overhead  Command  Shared Object       Symbol                                                 

luctrudeau

# perf record ./vpxenc --good --cpu-used=2 --codec=vp9 --psnr -v -o /dev/null ~/Videos/ducks_take_off_1080p50_60f.y4m
# perf report --stdio > libVPXonPower.prof
#
# Overhead  Command  Shared Object        Symbol                                                 
# ........  .......  ...................  .......................................................
#
    23.35%  vpxenc   vpxenc               [.] vpx_convolve8_vsx
    11.09%  vpxenc   vpxenc               [.] vpx_sad16x16x4d_vsx
     6.30%  vpxenc   vpxenc               [.] vpx_convolve8_avg_vsx
     5.18%  vpxenc   vpxenc               [.] vpx_convolve8_vert_vsx

luctrudeau

./vpxenc --good --cpu-used=5 --codec=vp9 --psnr -v --threads=4 --tile-columns=4 -o /dev/null ~/Videos/big_buck_bunny_720p_h264.mov --width=1280 --height=720

# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 3M of event 'cycles:ppp'
# Event count (approx.): 3525565972681
#

luctrudeau

# Overhead  Command  Shared Object     Symbol                                                   
# ........  .......  ................  .........................................................
#
    26.47%  aomenc   aomenc            [.] av1_highbd_jnt_convolve_2d_c
    10.76%  aomenc   aomenc            [.] aom_highbd_convolve8_horiz_c
    10.44%  aomenc   aomenc            [.] aom_highbd_convolve8_vert_c
    10.27%  aomenc   aomenc            [.] av1_highbd_jnt_convolve_y_c
     6.68%  aomenc   aomenc            [.] av1_highbd_convolve_2d_sr_c
     4.87%  aomenc   aomenc            [.] av1_highbd_jnt_convolve_x_c
     3.89%  aomenc   aomenc            [.] av1_idct64_new

luctrudeau

# Overhead  Command  Shared Object     Symbol                                               
# ........  .......  ................  .....................................................
#
     9.17%  aomenc   aomenc            [.] cdef_filter_fb
     7.46%  aomenc   aomenc            [.] av1_idct32_new
     7.15%  aomenc   aomenc            [.] av1_idct64_new
     5.02%  aomenc   aomenc            [.] av1_highbd_quantize_fp_facade
     4.93%  aomenc   aomenc            [.] av1_round_shift_array_c
     4.10%  aomenc   aomenc            [.] av1_idct16_new
     3.89%  aomenc   aomenc            [.] search_txk_type.isra.61

luctrudeau

luctrudeau

luctrudeau

uint16_t hicolor_avg(uint16_t a, uint16_t b) {
  const uint16_t s = a ^ b;
  return ((s & 0xF7DEU) >> 1) + (a & b);
}

luctrudeau

uint16_t hicolor_avg(uint16_t a, uint16_t b) {
  const uint16_t s = a ^ b;
  return ((s & 0xF7DEU) >> 1) + (a & b) + (s & 0x0821U);
}