From: Olzhas Zhumabek (anonymous.from.applecity_at_[hidden])
Date: 2021-06-14 17:51:53
Point two also requires alignment. Although not hard to do, might
complicate things a little bit.
About point three, Iâm unsure how widely used that will be. If somebody can
bring 3rd party library, one would just choose a library that does
convolution optimally out of the box. Anyway, fast single threaded
algorithm is a building block of the fast multithreaded implementation, the
latter just has task allocation added to it. Lets get the former to be fast
It seems like you had good progress. Please document the commit ref and
benchmark results for each complete experiment along with the hardware the
benchmark was run on. Note that things like YouTube video running in the
background or Steam game download that does parallel decompress might
affect the numbers.
Boost list run by Boost-Gil-Owners