This PR fixes an oversight with the inlining pass I wrote a bit ago. I thought the data I was calculating the inlining information from included the builtins, but apparently it didn't. So, this adds a predefined map with all of them, and adds it into the generated stuff.
Again, the gains aren't extreme, but a lot of key builtins aren't actually inlinable without stuff that Chris is working on.
fib1
1.373914ms -> 1.231279ms
fib2
5.164774ms -> 4.507025ms
fib3
5.133732ms -> 5.056868ms
Decode Nat
889ns -> 858ns
Generate 100 random numbers
600.667µs -> 569.827µs
List.foldLeft
4.278845ms -> 4.023873ms
Count to 1 million
490.727883ms -> 466.292019ms
Json parsing (per document)
582.807µs -> 451.109µs
Count to N (per element)
641ns -> 595ns
Count to 1000
629.742µs -> 602.579µs
Mutate a Ref 1000 times
1.001907ms -> 947.564µs
CAS an IO.ref 1000 times
1.480368ms -> 1.204739ms
List.range (per element)
832ns -> 716ns
List.range 0 1000
888.355µs -> 757.605µs
Set.fromList (range 0 1000)
3.71154ms -> 3.647534ms
Map.fromList (range 0 1000)
3.189949ms -> 2.841982ms
NatMap.fromList (range 0 1000)
14.157325ms -> 14.01479ms
Map.lookup (1k element map)
7.379µs -> 7.099µs
Map.insert (1k element map)
18.488µs -> 18.224µs
List.at (1k element list)
777ns -> 771ns
Text.split /
63.731µs -> 59.457µs
aryairani
commented
1 week ago
This PR fixes an oversight with the inlining pass I wrote a bit ago. I thought the data I was calculating the inlining information from included the builtins, but apparently it didn't. So, this adds a predefined map with all of them, and adds it into the generated stuff.
Again, the gains aren't extreme, but a lot of key builtins aren't actually inlinable without stuff that Chris is working on.