On Fri, 17 Feb 2023 17:31:09 GMT, Eirik Bjorsnos <d...@openjdk.org> wrote:
> This PR suggests we speed up Character.toUpperCase and Character.toLowerCase > for latin1 code points by applying the 'oldest ASCII trick in the book'. > > This takes advantage of the fact that latin1 uppercase code points are always > 0x20 lower than their lowercase (with the exception of two code points which > uppercase out of latin1). > > To verify the correctness of the new implementation, the test > `Latin1CaseConversion` is added with an exhaustive verification of > toUpperCase/toLowerCase for all latin1 code points. > > The implementation needs to balance the performance of the various ranges in > latin1. An effort has been made to favour operations on ASCII code points, > without causing excessive regression for higher code points. > > Performance is benchmarked for 7 chosen sample code points, each representing > a range or a special-case. Results in the first comment. Benchmark results: Baseline: Benchmark (codePoint) Mode Cnt Score Error Units Characters.Latin1CaseConversion.toLowerCase low avgt 15 1.267 ± 0.013 ns/op Characters.Latin1CaseConversion.toLowerCase A avgt 15 1.657 ± 0.011 ns/op Characters.Latin1CaseConversion.toLowerCase a avgt 15 1.258 ± 0.005 ns/op Characters.Latin1CaseConversion.toLowerCase A-grave avgt 15 1.656 ± 0.011 ns/op Characters.Latin1CaseConversion.toLowerCase a-grave avgt 15 1.270 ± 0.023 ns/op Characters.Latin1CaseConversion.toLowerCase mu avgt 15 1.261 ± 0.006 ns/op Characters.Latin1CaseConversion.toLowerCase yD avgt 15 1.260 ± 0.005 ns/op Characters.Latin1CaseConversion.toUpperCase low avgt 15 1.284 ± 0.043 ns/op Characters.Latin1CaseConversion.toUpperCase A avgt 15 1.264 ± 0.008 ns/op Characters.Latin1CaseConversion.toUpperCase a avgt 15 1.818 ± 0.016 ns/op Characters.Latin1CaseConversion.toUpperCase A-grave avgt 15 1.261 ± 0.015 ns/op Characters.Latin1CaseConversion.toUpperCase a-grave avgt 15 1.822 ± 0.013 ns/op Characters.Latin1CaseConversion.toUpperCase mu avgt 15 1.823 ± 0.006 ns/op Characters.Latin1CaseConversion.toUpperCase yD avgt 15 1.822 ± 0.008 ns/op PR: Benchmark (codePoint) Mode Cnt Score Error Units Characters.Latin1CaseConversion.toLowerCase low avgt 15 0.878 ± 0.005 ns/op Characters.Latin1CaseConversion.toLowerCase A avgt 15 1.038 ± 0.009 ns/op Characters.Latin1CaseConversion.toLowerCase a avgt 15 1.036 ± 0.007 ns/op Characters.Latin1CaseConversion.toLowerCase A-grave avgt 15 1.357 ± 0.015 ns/op Characters.Latin1CaseConversion.toLowerCase a-grave avgt 15 1.352 ± 0.003 ns/op Characters.Latin1CaseConversion.toLowerCase mu avgt 15 1.273 ± 0.002 ns/op Characters.Latin1CaseConversion.toLowerCase yD avgt 15 1.352 ± 0.004 ns/op Characters.Latin1CaseConversion.toUpperCase low avgt 15 0.880 ± 0.013 ns/op Characters.Latin1CaseConversion.toUpperCase A avgt 15 0.920 ± 0.071 ns/op Characters.Latin1CaseConversion.toUpperCase a avgt 15 1.055 ± 0.013 ns/op Characters.Latin1CaseConversion.toUpperCase A-grave avgt 15 1.394 ± 0.010 ns/op Characters.Latin1CaseConversion.toUpperCase a-grave avgt 15 1.391 ± 0.009 ns/op Characters.Latin1CaseConversion.toUpperCase mu avgt 15 1.597 ± 0.021 ns/op Characters.Latin1CaseConversion.toUpperCase yD avgt 15 1.354 ± 0.003 ns/op ------------- PR: https://git.openjdk.org/jdk/pull/12623
