Add back, with some minor editing, the comments about branch prediction,
when to use it, etc.
Requested by: jhb
Sponsored by: Netflix
Details
Details
Diff Detail
Diff Detail
- Repository
- rG FreeBSD src repository
- Lint
Lint Skipped - Unit
Tests Skipped - Build Status
Buildable 58457 Build 55345: arc lint + arc unit
Event Timeline
Comment Actions
It did occur to me after I sent my earlier e-mail that while it's probably good to document this here, it probably belongs in style(9) in some form as well.
| sys/sys/cdefs.h | ||
|---|---|---|
| 326 | maybe? | |
| sys/sys/cdefs.h | ||
|---|---|---|
| 326 | Yup. | |
| share/man/man9/style.9 | ||
|---|---|---|
| 909 | I would keep some of the old language about only using in paths that run frequently (not sure that "performance critical" alone implies that?). | |
| 910 | I feel like some guidance on how to (not) nest these is also useful, that is I think the best practice (but this wasn't in the old comment) is to use a single hint for an entire if () expression, not multiple hints joined with operators, etc. Maybe amend this to: The .Fn __predict_true and .Fn __predict_false branch prediction hints should... and then start this sentence with: When using branch prediction hints, atypical error conditions should use .Fn __predict_false (document any exceptions). ... Currently I think it reads like it wants to encourage __predict_false. | |
| sys/sys/cdefs.h | ||
|---|---|---|
| 329 | Consider adding a slight emphasis on measurement rather than just vibes | |