On Fri, 25 Feb 2022 09:41:09 GMT, Andrew Haley <[email protected]> wrote:
>> Claes Redestad has updated the pull request incrementally with one
>> additional commit since the last revision:
>>
>> Brent review
>
> test/micro/org/openjdk/bench/java/lang/StringDecode.java line 72:
>
>> 70: public void setup() {
>> 71: charset = Charset.forName(charsetName);
>> 72: asciiString = LOREM.substring(0, 32).getBytes(charset);
>
> This is problematic IMO in that it's missing short strings such as "Claes".
> Average Java strings are about 32 bytes long AFAICR, and people writing
> (vectorized) ijntrinsics have a nasty habit of optimizing for long strings,
> to the detriment of typical-length ones.
> Whether we like it or not, people will optimize for benchmarks, so it's
> important that benchmark data is realistic. The shortest here is 15 bytes, as
> far as I can see. I'd certainly include a short string of just a few bytes so
> that intrinsics don't cause regressions in important cases.
All good points. I've added a number of such short variants to all(?) relevant
microbenchmarks. The tests should now better cover a mix of input lengths and
encodings.
-------------
PR: https://git.openjdk.java.net/jdk/pull/7516
