Hi,
following up on JDK-8213478[1], this patch consolidates the use of
"Stringifiers" in StringConcatFactory to use a single Stringifier for
all String and Object arguments. This has the effect that repeated
filters become much more common with the default implementation, and can
reduce the number of species and LF classes generated by almost a third
in extreme cases[2], but still around a 10-15% reduction on average in
any test that mix Strings and Object arguments.
Webrev: http://cr.openjdk.java.net/~redestad/8213741/jdk.00/
Bug: https://bugs.openjdk.java.net/browse/JDK-8213741
The consolidation theoretically means an extra branch in the case of
String arguments, but any cost from that appears lost in the noise on my
measurements. OTOH, the simplified shape for Object arguments (a simple
MH rather than a chained one) means a significant speedup (up to around
1.3x) in microbenchmarks with Object arguments.
Thanks!
/Claes
[1] https://bugs.openjdk.java.net/browse/JDK-8213478
[2] http://cr.openjdk.java.net/~redestad/8213741/BadStringConcat.java
$ time $BASELINE/bin/java -Xlog:class+load BadStringConcat | grep
"MH\|Species" | wc -l
660
real 0m1.285s
user 0m6.024s
sys 0m0.216s
$ time $TEST/bin/java -Xlog:class+load BadStringConcat | grep
"MH\|Species" | wc -l
446
real 0m0.759s
user 0m4.092s
sys 0m0.136s
