Closed airhorns closed 4 months ago
Comparing faster-get-5
(58da4e6) with main
(f183291)
⚡ 4
improvements
❌ 1
regressions
✅ 3
untouched benchmarks
:warning: Please fix the performance issues or acknowledge them on CodSpeed.
Benchmark | main |
faster-get-5 |
Change | |
---|---|---|---|---|
⚡ | instantiating a large root |
4 ms | 3.5 ms | +14.55% |
⚡ | accessing memoized getter properties of a class model |
15.2 µs | 10.5 µs | +44.43% |
⚡ | accessing memoized null property of a class model |
14.4 µs | 11.2 µs | +28.64% |
⚡ | accessing unmemoized getter properties of a class model |
66.6 µs | 44.9 µs | +48.33% |
❌ | accessing unmemoized null property of a class model |
17.5 µs | 24.1 µs | -27.67% |
This implements a better performing getter function for readonly instances. Before this, we used the same definition of the getter function for all views, which ended up megamorphic, because it accessed a very wide variety of properties for every instance! Instead, this evals a monomorphic getter for each one. I also changed the object that stores the memos to have a fixed shape from birth by evaling it out as well.
I also changed us to use one object to store all the memoized values, instead of two. We were previously using two in order to implement memoization of undefined and null correctly, but with the new strategy to initialize an object with slots for every memo from the start, we can populate it with a
$notYetMemoized
symbol that indicates if we have memoized or not.