Various small optimizations #9605
Conversation
Thsi gets Context size down to 72 bytes from 80, assuming a 12 byte header and 8 byte rounding. In the type/* benchmark this gives a saving of ~700K contexts * 8 bytes = 5.4M vs a store allocation increase of 11K * ~90bytes -= 1M (approx).
It's fairly hot code, and eliminating the closure also avoids Int boxing.
The previously optimized apply function was not tail recursive since it was not final.
|
test performance please |
|
performance test scheduled: 13 job(s) in queue, 1 running. |
| def fold(x: X, trees: List[Tree]): X = trees match | ||
| case tree :: rest => fold(apply(x, tree), rest) | ||
| case Nil => x | ||
| fold(x, trees) |
There was a problem hiding this comment.
In ee4b125, I'm trying:
var acc = x
var list = trees
while (!list.isEmpty) do
acc = apply(acc, list.head)
list = list.tail
accAfter tail-call optimization, the two versions are almost the same. Let's see if there is difference in benchmarks.
There was a problem hiding this comment.
I don't think there will be. If anything, the tail recursive version should be faster since it does a single type test per iteration instead of one each in isEmpty, head, and tail.
There was a problem hiding this comment.
In #9565, we get a slight speedup for Dotty.
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/9605/ to see the changes. Benchmarks is based on merging with master (f2018f0) |
|
test performance please |
|
performance test scheduled: 15 job(s) in queue, 1 running. |
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/9605/ to see the changes. Benchmarks is based on merging with master (1431be2) |
Each of these will not change the needle much but they don't make the code more complicated either. So it is just attention to detail.