Optimize StringSequence

[dreis2211]

Hi,

as mentioned earlier today in #15465 this is the PR for some optimizations of StringSequence. More specifically for StringSequence.subSequence() in cases where the starting offset is 0 (much like in String.substring()). Additionally, for StringSequence.startsWith() by avoiding potential allocations of subSequence() - even though there is a good chance escape analysis avoids them.

See the following results with the patch:

Benchmark                                                         Mode  Cnt          Score          Error   Units
MyBenchmark.testStartsWith                                    thrpt   10   67431840,006 ±   791645,524   ops/s
MyBenchmark.testStartsWith:·gc.alloc.rate                     thrpt   10          0,001 ±        0,001  MB/sec
MyBenchmark.testStartsWith:·gc.alloc.rate.norm                thrpt   10         ? 10??                   B/op
MyBenchmark.testStartsWith:·gc.count                          thrpt   10            ? 0                 counts
MyBenchmark.testSubsequence                                   thrpt   10  316711796,197 ± 10992166,059   ops/s
MyBenchmark.testSubsequence:·gc.alloc.rate                    thrpt   10          0,001 ±        0,001  MB/sec
MyBenchmark.testSubsequence:·gc.alloc.rate.norm               thrpt   10         ? 10??                   B/op
MyBenchmark.testSubsequence:·gc.count                         thrpt   10            ? 0                 counts

vs. the old code:

Benchmark                                                         Mode  Cnt          Score          Error   Units
MyBenchmark.testStartsWith                                    thrpt   10   63029809,023 ±  1172502,340   ops/s
MyBenchmark.testStartsWith:·gc.alloc.rate                     thrpt   10          0,001 ±        0,001  MB/sec
MyBenchmark.testStartsWith:·gc.alloc.rate.norm                thrpt   10         ? 10??                   B/op
MyBenchmark.testStartsWith:·gc.count                          thrpt   10            ? 0                 counts
MyBenchmark.testSubsequence                                   thrpt   10  123833544,662 ±  9933577,230   ops/s
MyBenchmark.testSubsequence:·gc.alloc.rate                    thrpt   10       3777,956 ±      302,934  MB/sec
MyBenchmark.testSubsequence:·gc.alloc.rate.norm               thrpt   10         32,000 ±        0,001    B/op
MyBenchmark.testSubsequence:·gc.churn.PS_Eden_Space           thrpt   10       3769,693 ±      330,836  MB/sec
MyBenchmark.testSubsequence:·gc.churn.PS_Eden_Space.norm      thrpt   10         31,948 ±        2,016    B/op
MyBenchmark.testSubsequence:·gc.churn.PS_Survivor_Space       thrpt   10          0,187 ±        0,074  MB/sec
MyBenchmark.testSubsequence:·gc.churn.PS_Survivor_Space.norm  thrpt   10          0,002 ±        0,001    B/op
MyBenchmark.testSubsequence:·gc.count                         thrpt   10        141,000                 counts
MyBenchmark.testSubsequence:·gc.time                          thrpt   10         93,000                     ms

For completeness reasons here is the benchmark for the old startsWith without escape analysis kicking in (by disabling it via -XX:-DoEscapeAnalysis):

MyBenchmark.testStartsWith                                    thrpt   10   53268054,305 ± 3264411,681   ops/s
MyBenchmark.testStartsWith:·gc.alloc.rate                     thrpt   10       1625,151 ±      99,597  MB/sec
MyBenchmark.testStartsWith:·gc.alloc.rate.norm                thrpt   10         32,000 ±       0,001    B/op
MyBenchmark.testStartsWith:·gc.churn.PS_Eden_Space            thrpt   10       1678,723 ±     156,577  MB/sec
MyBenchmark.testStartsWith:·gc.churn.PS_Eden_Space.norm       thrpt   10         33,066 ±       2,719    B/op
MyBenchmark.testStartsWith:·gc.churn.PS_Survivor_Space        thrpt   10          0,131 ±       0,077  MB/sec
MyBenchmark.testStartsWith:·gc.churn.PS_Survivor_Space.norm   thrpt   10          0,003 ±       0,001    B/op
MyBenchmark.testStartsWith:·gc.count                          thrpt   10         81,000                counts
MyBenchmark.testStartsWith:·gc.time                           thrpt   10         58,000                    ms

Let me know what you think.
Cheers,
Christoph

[philwebb]

Excellent changes! Thanks once again @dreis2211