Looks good.
It’s tempting to do something like this to de-dup the code with less potential
for mistakes:
String s = null;
// Select the singular non-null value, if any
If (s0 != null && s1 == null) s = s0;
else if (s0 == null && s1 != null) s = s1;
If (s != null) {
if (s.isEmpty()) {
…
} else if (…) { … }
}
Paul.
> On Jun 16, 2020, at 6:00 AM, Claes Redestad <claes.redes...@oracle.com> wrote:
>
> Hi,
>
> this patch specializes bootstrapping of unary concatenation expressions.
>
> Such expressions can be reduced to the canonical String.valueOf
> "stringifier" for any primitive argument, but needs a special
> stringifier for reference arguments since we need to produce a new
> String to be compliant with JLS.
>
> Bug: https://bugs.openjdk.java.net/browse/JDK-8247681
> Webrev: http://cr.openjdk.java.net/~redestad/8247681/open.00/
>
> This ensures we get a similar speed-up as JDK-8247605 for
> "" + fooString (and fooString + "").
>
> Also speeds up bootstrapping for all such simple unary concatenation
> expressions.
>
> Testing: tier1+2
>
> Before:
>
> Benchmark (intValue) Mode Cnt Score Error
> Units
> StringConcat.concatEmptyConstInt 4711 avgt 5 15.539 ± 0.831
> ns/op
> StringConcat.concatEmptyConstString 4711 avgt 5 17.046 ± 1.047
> ns/op
> StringConcat.concatEmptyLeft 4711 avgt 5 7.506 ± 0.588
> ns/op
> StringConcat.concatEmptyRight 4711 avgt 5 7.890 ± 0.314
> ns/op
>
> After:
> Benchmark (intValue) Mode Cnt Score Error
> Units
> StringConcat.concatEmptyConstInt 4711 avgt 5 15.410 ± 0.944
> ns/op
> StringConcat.concatEmptyConstString 4711 avgt 5 7.397 ± 0.384
> ns/op
> StringConcat.concatEmptyLeft 4711 avgt 5 7.465 ± 0.328
> ns/op
> StringConcat.concatEmptyRight 4711 avgt 5 7.857 ± 0.355
> ns/op
>
> Thanks!
>
> /Claes
