Bugzilla

This function is very hot during nursery collections because it's used to trace object slots/elements. It has branches to check for object/string/bigint values, but in practice about 50% of the values we see there aren't GC things and it would be nice to reduce the number of branches in this case.

We could check `isGCThing()` first to 'discard' about 48% of all values because they aren't gc things. After that we can check if the cell is tenured, this gets rid of another ~25% of values. That leaves about 27% of nursery-allocated values.

A Speedometer 3 comparison of just this change shows a number of medium/high 1-2% improvements on a number of sub tests.

https://treeherder.mozilla.org/perfherder/comparesubtest?originalProject=try&newProject=try&newRevision=0eae5ffd806a5fa22d393fb49684fee30673fb2c&originalSignature=4586009&newSignature=4586009&framework=13&application=firefox&originalRevision=827fe39e928d1a7b6cf118f335d8c46d85a81a22&page=1

This function is very hot during nursery collections because it's used to trace object slots/elements. It has branches to check for object/string/bigint values, but in practice about 50% of the values we see there aren't GC things and it would be nice to reduce the number of branches in this case.

We could check `isGCThing()` first to 'discard' about 48% of all values because they aren't gc things. After that we can check if the cell is tenured, this gets rid of another ~25% of values. That leaves about 27% of nursery-allocated values.

A Speedometer 3 comparison of just this change shows a number of medium/high confidence 1-2% improvements on a number of sub tests.

https://treeherder.mozilla.org/perfherder/comparesubtest?originalProject=try&newProject=try&newRevision=0eae5ffd806a5fa22d393fb49684fee30673fb2c&originalSignature=4586009&newSignature=4586009&framework=13&application=firefox&originalRevision=827fe39e928d1a7b6cf118f335d8c46d85a81a22&page=1