On Thu, 24 Feb 2022 19:01:32 GMT, Brent Christian <[email protected]> wrote:
>> Splitting out these micro changes from #7231
>>
>> - Clean up and simplify setup and code
>> - Add variants with different inputs with varying lengths and encoding
>> weights, but also relevant mixes of each so that we both cover interesting
>> corner cases but also verify that performance behave when there's a
>> multitude of input shapes. Both simple and mixed variants are interesting
>> diagnostic tools.
>> - Drop all charsets from the default run configuration except UTF-8.
>> Motivation: Defaults should give good coverage but try to keep runtimes at
>> bay. Additionally if the charset tested can't encode the higher codepoints
>> used in these micros the results might be misleading. If you for example
>> test using ISO-8859-1 the UTF-16 bytes in StringDecode.decodeUTF16 will have
>> all been replaced by `?`, so the test is effectively the same as testing
>> ASCII-only.
>
> test/micro/org/openjdk/bench/java/lang/StringDecode.java line 93:
>
>> 91: public void decodeAsciiLong(Blackhole bh) throws Exception {
>> 92: bh.consume(new String(longAsciiString, charset));
>> 93: bh.consume(new String(longAsciiString, 0, 1024 + 31, charset));
>
> I imagine the 1024+31 addition gets compiled down, and is not executed during
> the test, right?
Yes, adding two integer literals will be constant folded already by javac:
public static void main(String...args) {
int foo = 1024 + 31;
}
javap -v output:
``` stack=1, locals=2, args_size=1
0: sipush 1055
3: istore_1
4: return
-------------
PR: https://git.openjdk.java.net/jdk/pull/7516
