RFR: 8254807: Optimize startsWith() for String.substring()

[Xin Liu]

On Sat, 31 Oct 2020 05:22:39 GMT, Xin Liu <xliu at openjdk.org> wrote:

> the optimization transforms code from s=substring(base, beg, end); s.startsWith(prefix)
> to substring(base, beg, end) | base.startsWith(prefix, beg). 
> 
> it reduces an use of substring. hopefully c2 optimizer can remove the used substring()

$make test TEST="micro:SubstringAndStartsWith"  CONF=linux-x86_64-server-release

Benchmark                                         (substrLength)   Mode  Cnt       Score      Error   Units
SubstringAndStartsWith.substr2StartsWith                       1  thrpt  100  140630.609 ?    9.404  ops/ms
SubstringAndStartsWith.substr2StartsWith                       8  thrpt  100  140634.988 ?    7.704  ops/ms
SubstringAndStartsWith.substr2StartsWith                      32  thrpt  100  135812.111 ? 4687.410  ops/ms
SubstringAndStartsWith.substr2StartsWith                     128  thrpt  100  140644.499 ?    7.856  ops/ms
SubstringAndStartsWith.substr2StartsWith                     256  thrpt  100  140640.309 ?    8.535  ops/ms
SubstringAndStartsWith.substr2StartsWith                     512  thrpt  100  140643.976 ?    8.168  ops/ms
SubstringAndStartsWith.substr2StartsWith_noalloc               1  thrpt  100  158672.117 ?    8.510  ops/ms
SubstringAndStartsWith.substr2StartsWith_noalloc               8  thrpt  100  158673.989 ?   10.198  ops/ms
SubstringAndStartsWith.substr2StartsWith_noalloc              32  thrpt  100  158677.660 ?    8.987  ops/ms
SubstringAndStartsWith.substr2StartsWith_noalloc             128  thrpt  100  153677.095 ? 5418.718  ops/ms
SubstringAndStartsWith.substr2StartsWith_noalloc             256  thrpt  100  158668.766 ?    9.138  ops/ms
SubstringAndStartsWith.substr2StartsWith_noalloc             512  thrpt  100  152544.827 ? 5922.157  ops/ms
Finished running test 'micro:SubstringAndStartsWith'
Test report is stored in build/linux-x86_64-server-release/test-results/micro_SubstringAndStartsWith

The micro demonstrates the optimization removes substring() and the throughput is constant value. 
it's still slight lower than handcrafted version `substr2StartsWith_noalloc` because the later doesn't need to check complete bound check.

-------------

PR: https://git.openjdk.java.net/jdk/pull/974