You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Improve decoding performance by always inlining Codec::index_of()
In the gaspardpetit/base64 test, perf showed a significant amount
of time spent in that function, which seemed off because that
function should not exist at all in a release build.
Making sure it's inline reduces decoding time for Gaspard's 256
workload by almost 30% (from approx. 2.38 to 1.69 on my machine).
It seems logical to assume that other codecs get a similar
performance boost this way, so I'm applying the change everywhere.
This puts performance characteristics around the same level of
adamvr/arduino-base64 or the Manuel Martinez one (but without
the latter's apparent setup overhead). Still not the best,
but now definitely in the "fair" spectrum for decoding.
0 commit comments