oh wow complete with custom CharSequence.parseInt(...
# advent-of-codek
keturn
12/08/2018, 9:03 PMoh wow complete with custom CharSequence.parseInt(start)
j
joelpedraza
12/08/2018, 9:05 PMhttps://i.kym-cdn.com/entries/icons/original/000/006/360/gottago.jpgâ–ľ
o
orangy
12/08/2018, 9:08 PM
(nitpick) no check for int overflow 🙂
j
joelpedraza
12/08/2018, 9:10 PMThat would make it less fast =P
k
karelpeeters
12/08/2018, 9:10 PMparseNodeinlinekarelpeeters
12/08/2018, 9:10 PMYou'd avoid the first call to
onMetadataj
joelpedraza
12/08/2018, 9:11 PM@karelpeeters Large method body though…wouln’t it bloat code
k
karelpeeters
12/08/2018, 9:11 PMYeah you're probably right.
karelpeeters
12/08/2018, 9:11 PMYou can remove the
if (c in '0'..'9')j
joelpedraza
12/08/2018, 9:11 PMTrue! Lets live dangerously!
k
keturn
12/08/2018, 9:12 PMand maybe another nanosecond from assuming that the only kind of whitespace is
j
joelpedraza
12/08/2018, 9:12 PMShaved 0.09 ms. WORTH IT
joelpedraza
12/08/2018, 9:13 PMCould probably save a few more by assuming theres no repeating whitespace also
k
karelpeeters
12/08/2018, 9:13 PMAnd what do default parameters look like in bytecode again? Maybe they slow it down a bit too.
o
orangy
12/08/2018, 9:13 PM
yes, if they some are ommitted, there is a bit check
j
joelpedraza
12/08/2018, 9:13 PMTrue, I dont think I ever use the end arg of
CharSequence.indexOfFirstk
karelpeeters
12/08/2018, 9:14 PMYou could also make
parseIntendj
joelpedraza
12/08/2018, 9:15 PMHmm ya, and return an inlined class instance to pull the value and next index out
joelpedraza
12/08/2018, 9:16 PMI think could also be a bit faster by using
StringCharSequenceo
orangy
12/08/2018, 9:16 PM
Yes, Strings are handled specially by JVM
j
joelpedraza
12/08/2018, 9:17 PMAlso should be
invokestaticinvokeinterfaceo
orangy
12/08/2018, 9:17 PM
JIT will handle polymorphic with only one real class pretty good (after warmup)
k
karelpeeters
12/08/2018, 9:18 PMBiggest performance difference would probably be to drop the
onMetaDataj
joelpedraza
12/08/2018, 9:20 PMAnd accumulate metadata sum in an argument?
k
karelpeeters
12/08/2018, 9:20 PMAren't there some more high bits left in the return of
parseNodej
joelpedraza
12/08/2018, 9:21 PMHaha
joelpedraza
12/08/2018, 9:23 PMOn second glance you’re right, in fact it HAS to be in the return because we dont know where the metadata are until after we parse children
joelpedraza
12/08/2018, 10:37 PM@karelpeeters Set up JMH, tried with your suggested bit twiddles
joelpedraza
12/08/2018, 10:38 PMBit twiddling version was 1.04x function args vers
k
karelpeeters
12/08/2018, 10:55 PMHuh that's interesting.
j
joelpedraza
12/08/2018, 10:56 PMVery!
k
karelpeeters
12/08/2018, 11:00 PMWould it be possible to quickly put my code though JMH as well? It runs in 0.5ms for me, but maybe that's just the pc?
j
joelpedraza
12/08/2018, 11:57 PM@karelpeeters
Copy code
Day8Bench.day8part1joel thrpt 25 6715.561 ± 84.500 ops/s
Day8Bench.day8part1karel thrpt 25 1853.961 ± 5.016 ops/sjoelpedraza
12/08/2018, 11:58 PMI removed the value calculation and prints to make it an even comparison
k
karelpeeters
12/09/2018, 12:00 AMCool, thanks for running it!
j
joelpedraza
12/09/2018, 12:00 AMSure thing
k
karelpeeters
12/09/2018, 12:01 AMSo you got a 10x improvement between this and the code you posted (that was 1.79ms)? Or is that mostly JMH measuring better?
j
joelpedraza
12/09/2018, 12:01 AMbetter mesaurements
2 Views