On Fri, 8 Sep 2023 15:39:47 GMT, Chen Liang <li...@openjdk.org> wrote:
>> newStringLatin1NoRepl does not significantly help performance, but it >> simplifies the code. >> >> Of course, there is another advantage: because the call will be simpler, >> making it easier for the caller method to implement codeSize smaller than >> the default value of FreqInlineSize 325, and there are more opportunities to >> be inlined. > > Notice there is patch https://github.com/openjdk/jdk/pull/14655 (bug 8310901) > for converting these usages to a new `newStringLatin1NoRepl`. Going from `jla.newStringNoRepl(bytes, ISO_...)` to `jla.newStringLatin1NoRepl(bytes)` isn't much of a code simplification, so this distracts a bit from the bulk of the changes in this PR. I agree that it *might* help inlining since the former calls into a large method, but I think we ought to focus on one thing at a time so if this doesn't significantly help on the benchmarks we have available it would be better to leave it out for now. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/15555#discussion_r1320030130
