Reduce some allocations #5748
Reduce some allocations #5748
Conversation
Add a miniphase, enabled by -Yinstrument-closures that counts closure allocation by invoking Stats.doRecord. Todo: Find a generally shippable way to use this. Right now it depends on the compiler-internal utility class Stats.
Instead use NoAddr as a sentinel for None.
Avpid creation of moeratly hot closure
Without the optimization a comparison of value classes always boxing, since the rhs operand is cast to Any. Most likely these are eliminated by escape analysis, but by specializing here we make the job of the JOT compiler easier.
Again, these might be eliminated by escape analysis, but it does not hurt to not allocate in the first place.
|
test performance please |
|
performance test scheduled: 1 job(s) in queue, 0 running. |
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/5748/ to see the changes. Benchmarks is based on merging with master (e724058) |
|
Some smallish improvements. About 1-2% on compiler, other tests go up to 10%. So, this looks like a worthwhile thing to do. |
Co-Authored-By: odersky <[email protected]>
|
test performance please |
|
performance test scheduled: 2 job(s) in queue, 1 running. |
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/5748/ to see the changes. Benchmarks is based on merging with master (9b74ac1) |
| envelope(source, span.startPos) // fill in children spans | ||
| () // Note: the `()` is there to prevent some inefficient code from being generated. | ||
| // Without it we get an allocation of a span here since the result type of the `if` |
There was a problem hiding this comment.
I don't think this changes whether an allocation happens or not but it does make the code more efficient, I've opened an issue to track this that you could refer to in this comment: #5750
Co-Authored-By: odersky <[email protected]>
No description provided.