Tweak performance of findRef #9870
Conversation
|
test performance please |
|
performance test scheduled: 6 job(s) in queue, 1 running. |
| sym.is(ExtensionMethod) && (sym.extensionParam.span == xmethod.extensionParam.span) | ||
| if !xmethod.exists || result.tpe.isError || canAccessUnqualified(result.symbol) then | ||
| result | ||
| if xmethod.exists then |
There was a problem hiding this comment.
I would have thought that the code here would be executed only rarely.
There was a problem hiding this comment.
For a simple program (see below), the speedup is about 24% (34ms/op VS 45ms/op):
class MyInt(val x: Int) {
def eq(that: MyInt): Boolean = this.x == that.x
}
class Test {
def foo(x: MyInt, y: MyInt): Boolean = x.eq(y)
val a = MyInt(2)
val b = MyInt(3)
foo(a, b)
}There was a problem hiding this comment.
Interesting. I just found out why: The code in question (including the expensive tryEither) is executed twice, once for each MyInt(...). I had not thought of that. So, yes, because of constructor syntax, failures in findRef are now not that uncommon. So this is a definite improvement. Unfortunately, the benchmarks are too noisy to see this immediately.
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/9870/ to see the changes. Benchmarks is based on merging with master (246fc89) |
|
retest performance please |
|
performance test scheduled: 3 job(s) in queue, 1 running. |
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/9870-1/ to see the changes. Benchmarks is based on merging with master (2ec7a6f) |
|
retest performance please |
|
performance test scheduled: 1 job(s) in queue, 0 running. |
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/9870-2/ to see the changes. Benchmarks is based on merging with master (f4528ce) |
Tweak performance of findRef