Mastodon Feed
slightlyoff@toot.cafe ("Alex Russell") wrote:
V8 will go faster with naive Float32Array accessors than classes w/ all the vtable dispatch, but once you manually unroll your loops, classes (with hot IC data) just scream.
Bun/JSC seem to have faster paths for raw TypedArray access; I can get them (with unrolling) to nearly 2x faster than classes w/ unrolling across larger collections and 1.5x faster than the fastest Node/v8 approach. Fascinating.