deno
1e349864 - perf(napi): misc napi optimizations (#30291)

Commit
1 year ago
perf(napi): misc napi optimizations (#30291) before ``` ❯ deno run -A perf-turso.js cpu: Apple M3 Max runtime: deno 2.4.3+6ae4eda (aarch64-apple-darwin) benchmark time (avg) (min … max) p75 p99 p999 --------------------------------------------------------------------------------------- ----------------------------- Statement.get() with bind parameters [expanded] 4'544 ns/iter (4'424 ns … 4'801 ns) 4'599 ns 4'801 ns 4'801 ns Statement.get() with bind parameters [raw] 4'484 ns/iter (4'409 ns … 4'667 ns) 4'527 ns 4'667 ns 4'667 ns Statement.run() with bind parameters 5'424 ns/iter (4'833 ns … 4'518 µs) 5'292 ns 8'042 ns 15'292 ns ``` after ``` ❯ devdeno run -A perf-turso.js cpu: Apple M3 Max runtime: deno 2.4.3 (aarch64-apple-darwin) benchmark time (avg) (min … max) p75 p99 p999 --------------------------------------------------------------------------------------- ----------------------------- Statement.get() with bind parameters [expanded] 1'141 ns/iter (1'057 ns … 3'223 ns) 1'126 ns 2'910 ns 3'223 ns Statement.get() with bind parameters [raw] 1'015 ns/iter (975 ns … 1'112 ns) 1'017 ns 1'093 ns 1'112 ns Statement.run() with bind parameters 1'560 ns/iter (1'295 ns … 8'840 ns) 1'433 ns 8'752 ns 8'840 ns ``` Optimizations are: - don't log napi entry/exit in release builds (this ends up being very expensive, as it checks whether logging is enabled twice per call into NAPI) - use internalized strings to speed up property accesses (matches node's behavior as well) - change order we check value types (matches node now) - don't unnecessarily validate utf8 - don't get handle scopes when an isolate will do - reuse handle scope within a function - use the faster APIs that don't require handle scopes for getting uint32 / int32 values Co-authored-by: Bartek Iwańczuk <biwanczuk@gmail.com>
Author
Parents
Loading