inline several methods to make tracing::event! smaller

[ldm0]

Motivation

Make tracing::event! codegen smaller

Solution

Add inline to several functions called by tracing::event!.

Simple example: https://github.com/ldm0/tracing_test

After inlining, executable size drops from 746kb to 697kb(cargo build --release + strip), saves 50 bytes per event!.

Test environment:

toolchain: nightly-aarch64-apple-darwin
rustc-version: rustc 1.70.0-nightly (88fb1b922 2023-04-10)

[hawkw]

It's interesting that inlining these functions actually improves binary size — they must all compile to fewer instructions than a function call, in release mode.

Out of curiosity, would you mind running the tracing crate's benchmarks before and after this change? It would be interesting to see if there's a meaningful performance delta in those micro benchmarks.

[ldm0]

Out of curiosity, would you mind running the tracing crate's benchmarks before and after this change? It would be interesting to see if there's a meaningful performance delta in those micro benchmarks.

Yes, there are some perf win:

event/scoped [-40.689% -40.475% -40.228%]
event/scoped_recording [-14.972% -14.685% -14.410%]
event/global [-48.412% -48.217% -48.010%]
span_fields/scoped [-25.317% -24.876% -24.494%]
span_fields/global [-39.695% -39.488% -39.242%]
span_repeated/global [-27.514% -26.633% -25.298%]
static/baseline_single_threaded [-32.275% -32.032% -31.808%]
static/single_threaded [-29.628% -29.376% -29.156%]
static/enabled_one [-29.777% -29.544% -29.305%]
static/enabled_many [-30.901% -30.504% -30.140%]
dynamic/baseline_single_threaded [-20.157% -19.880% -19.603%]

I retried benchmark several times and the improvements seem to be fairly stable.

raw log: https://gist.github.com/ldm0/6573935f4979d2645fbcf5bde7361386

[hawkw]

Thanks, that looks great! I'll merge this once CI passes. :)