1991501 - Investigate WeakSet overhead on babel-wtb and babel-minify-wtb tests in JetStream 3

Status: NEW

(Core :: JavaScript Engine, task, P2)

Description

[Jan de Mooij [:jandem]]

This test is adding a lot of values to a WeakSet.

I'm seeing some time in jemalloc allocating memory for the underlying hash table (using the buffer allocator for Weak{Map,Set} buffers might help there). Also some post barrier time and FULL_VALUE_BUFFER nursery GCs that I think are also related.

Comment 1

[Jan de Mooij [:jandem]]

The babel-wtb test is similar.

Comment 2

[Jan de Mooij [:jandem]]

Maybe nursery-allocating the WeakSet objects would help if we can avoid the store buffer entries in that case, but I haven't checked what the lifetime of these objects is.

Comment 3

[Jon Coppeard (:jonco)]

(In reply to Jan de Mooij [:jandem] from comment #0)

using the buffer allocator for Weak{Map,Set} buffers might help there

I had a go at doing this but it's complicated because sometimes weak maps are owned by GC things and have GC lifetime and sometimes they are embedded in data structures allocated on the C++ heap which don't. Giving the table GC lifetime will would allow the WeakMap destructor to observe dangling pointers in the latter case.

Comment 4

[Jan de Mooij [:jandem]]

Bug 1995021 improved these tests by 5-17%. Nice!