I don't know what a v8::EntryScope is, but clarifying the specifics of your plan in the form of uploading a patch to Gerrit is probably a good idea.
I'm guessing an absl::InlinedVector is like a v8::base::SmallVector? From a 20-second glance, it looks like maybe DetachableVector could derive from that? (I don't know this code well and have no strong opinions.) On Tue, Jul 11, 2023 at 9:43 AM 'Konnor Krupp' via v8-dev < v8-dev@googlegroups.com> wrote: > Checking in here, I may jump to the "Submit your code" section of the > contributor docs and rely on code review to vet the approach and/or get a > yes/no? > > Thanks, > > Konnor > > On Friday, June 30, 2023 at 1:02:44 AM UTC+2 Konnor Krupp wrote: > >> Hey all, >> >> (~beginner to V8 internals!) >> >> We have a use case which does high frequency lambda-like JS function >> execution across any/multiple threads in a constrained environment >> (Android) which has heavy cost for malloc. >> >> In local on-device benchmarking, we noticed grabbing the V8::EntryScope >> on every lambda-like function call does a few malloc/free pairs, >> specifically: >> * Context::Enter uses two DetachableVectors for the first time (once >> each) which mallocs space >> * HandleScope::CreateHandle uses a DetachableVector for the first time >> (once) which mallocs space >> * Destruction of the V8::EntryScope frees this space >> >> For our case which rarely nests v8::EntryScopes and creates/frees them in >> pairs at high frequency, it would be more efficient (overall) to use an >> absl::InlinedVector-like object which assumes at least one (but up to N) >> Handles will be created / Contexts will be entered/saved. This shaved a few >> percent off our minimal overhead numbers. Naturally this would use up more >> initial space, though I expect the situations where a V8::EntryScope are >> created and *never* used are rare? >> >> I am happy to make this contribution, but want to validate I'm not >> missing some subtlety, and wanted to generally ask for permission before >> going through the process! I expect this would be generally useful for >> other lambda-like use cases such as edge functions, Cloudflare Workers, etc. >> >> My intended implementation would: >> * have DetachableVector be backed by an absl::InlinedVector of size >> kMinimumCapacity (8) >> * Optional?: have minimum capacity controlled by compile-time flag (could >> be default 0 and not inline anything!) >> >> Thanks! >> >> Konnor >> >> >> -- > > -- -- v8-dev mailing list v8-dev@googlegroups.com http://groups.google.com/group/v8-dev --- You received this message because you are subscribed to the Google Groups "v8-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to v8-dev+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/v8-dev/CAKSzg3Q3oFLhsOrQRVmdq-Jz3atp8xNezKTW-iN0HzB-P1dW6g%40mail.gmail.com.
