extend the image if its size is not modulo 16
give up on weird shit
exit cleanly; print average fps when debug is enabled
copy_wh: specify for w == 4
add arm64 builtins
treat a8r8g8b8 as x8r8g8b8
fix flush on note and remove fps printout
start time from epoch
allow non-multiple-of-16 sizes (fixes cropping too) since we're allocating frame stuff properly anyway
default GOP to fps×2
change -O to -o: disable optimization by default (twitch stream craps out)
ignore closed pipe writes to exit properly
procrfork & threadexitsall
not WIP anymore, seems to work fine
add -O option to disable optimizations
encoder thread: compare each frame to the previous one and discard if the contents are the same
builtins/amd64: sync clz, use MOVL in bswap32
default to lowest quality and gop to every two seconds
ivf: write header and flush first, then proceed with reading images
use procdata instead of threaddata
use IVF by default
assign timestamp to each frame
take yuv logic out; add -G option to specify gop; -d → -D (denoise), -d is debug
forgot to add builtins.h (thanks kemal)
only one file is required to be specified as argument
remove 10s timer, use npe_nanosec, shut down better on Del
flush the reminding frames properly
do not use memdraw
add clz and bswap32 builtins
inline small functions
build amd64-specific stuff if possible
use faster (on new cpus) memcpy
set max long term ref frames number and set frames channel size to 1
capture 10 seconds; reuse input buffer; produce human readable error reports
use full swing conversion
read image directly, bypassing memimage and header generated by rio
add another FIXME
limit fps (in a dumb way)
encode in a separate thread
make rgb→yuv more readable
THAT'S A FIRST