String indexOf Performance

[Aleksey Shipilev]

On 01/27/2017 09:21 PM, Scott Palmer wrote:
> http://pastebin.com/zRXKwD64

Okay:

Benchmark                              Mode  Cnt       Score      Error  Units

# JDK 8u121
IndexOfBenchmark.StringIndexOfChar    thrpt    5  141857.332 ± 5530.472  ops/s
IndexOfBenchmark.StringIndexOfString  thrpt    5  113091.517 ± 2241.533  ops/s

# JDK 9b152
IndexOfBenchmark.StringIndexOfChar    thrpt    5  154525.343 ± 3796.818  ops/s
IndexOfBenchmark.StringIndexOfString  thrpt    5  185917.059 ± 3391.230  ops/s

Perfasm for 9b152:
  http://cr.openjdk.java.net/~shade/scratch/indexOf.perfasm

It does indeed look like indexOf(String) is backed by the intrinsic (notice
"vpcmpestri $0xc,(%rbx),%xmm0"), while indexOf(char) does the Java code (notice
how real locations in bytecode are mapped onto generated code). This is further
corroborated by StringLatin1.indexOf not having @HotspotIntrinsicCandidate:

http://hg.openjdk.java.net/jdk9/jdk9/jdk/file/e4b19b8d4bbf/src/java.base/share/classes/java/lang/StringLatin1.java#l181

Submitted:
 https://bugs.openjdk.java.net/browse/JDK-8173585

Think about it this way: you have the improvement for both cases in JDK 9 due to
Compact Strings, but it somewhat twisted the performance model to be less
intuitive. Continue using indexOf(char) in the code, and wait for JDK to catch up.

Thanks,
-Aleksey