1934123 - Add a fuse for the existence of user access to Symbol.isConcatSpreadable

Status: NEW

(Core :: JavaScript Engine, task, P5)

Description

[Matthew Gaudet (he/him) [:mgaudet]]

If we add a fuse for this and add fuse telemetry (Bug 1877258) we can decide if we want a fuse based optimization for this.

(The check for concat spread has already been optimized, but with a fuse we could make a runtime-wide optimization)

Comment 1

[Matthew Gaudet (he/him) [:mgaudet]]

(I suppose the alternative definition: "If user code ever creates a new property with the key Symbol.isConcatSpreadable" would also work)

Comment 2

[Matthew Gaudet (he/him) [:mgaudet]]

(Evidence this could be valuable is that during an Sp3 run, there are 2.4 million checks involved in "IsConcatSpreadable")

Comment 3

[Matthew Gaudet (he/him) [:mgaudet]]

Oh and out of all these checks, in Sp3 precisely zero ever said "yes, we have an isconcatspreadable"

Comment 4

[Matthew Gaudet (he/him) [:mgaudet]]

Attachment: stats.patch

(Current stats patch for the curious)

Comment 5

[Matthew Gaudet (he/him) [:mgaudet]]

It's also maybe interesting to note that toStringTag and toPrimitive never seem to take this path?

Comment 6

[Matthew Gaudet (he/him) [:mgaudet]]

To experiment if getting rid of the checks for IsConcatSpreadable would be valuable, I did a perf compare where I just commented out the checks. This would be the ideal outcome of a fuse optimization.

Results are pretty underwhelming -- almost no impact.

It seems that our current HasInterestingSymbol opt is pretty darn good, 2.4 million checks and all.

Comment 7

[Matthew Gaudet (he/him) [:mgaudet]]

Would be interesting to see if this helped a microbenchmark.