perf(runtime): optimize PermissionState::check

[AaronO]

By converting log_perm_access() to a macro, so the message sub-expression is dropped at compile time in release mode. Otherwise it made allocations and added substantial overhead to opcalls.

Benches

Measured on perf_now (op_now) from deno_common:

Before:
perf_now:                n = 5000000, dt = 2.429s, r = 2058460/s, t = 485ns/op

After:
perf_now:                n = 5000000, dt = 1.773s, r = 2820079/s, t = 354ns/op

Flamegraph views

Before:

After

[ry]

Cool - seems simple enough.

Out of curiosity wouldn't #[inline] also do the trick?

[AaronO]

Cool - seems simple enough.

Out of curiosity wouldn't #[inline] also do the trick?

I've refactored this so it no longer uses a macro, it instead inlines it and calls Self::fmt_access inside debug!(), which will cause a double alloc of the access msg in the prompt path, but only when debug logging is enabled which is fine IMO (since debug logging + prompt isn't a hotpath).

[AaronO]

Both the flamegraph and the deno_common bench confirm this optimization:

Before (main):
perf_now:            	n = 500000, dt = 0.155s, r = 3225806/s, t = 310ns/op

After (this PR):
perf_now:            	n = 500000, dt = 0.094s, r = 5319149/s, t = 188ns/op

[bartlomieju]

LGTM