RFR: 8334328: Reduce object allocation for FloatToDecimal and DoubleToDecimal

[Shaojin Wen]

On Sat, 15 Jun 2024 01:59:42 GMT, Shaojin Wen <duke at openjdk.org> wrote:

> The current versions of FloatToDecimal and DoubleToDecimal allocate additional objects. Reducing these allocations can improve the performance of Float/Double.toString and AbstractStringBuilder's append(float/double).
> 
> This patch is just a code refactoring to reduce object allocation, but does not change the Float/Double to decimal algorithm.
> 
> The following code comments the allocated objects to be removed.
> 
> 
> class FloatToDecimal {
>     public static String toString(float v) {
>         // allocate object FloatToDecimal
>         return new FloatToDecimal().toDecimalString(v);
>     }
> 
>     public static Appendable appendTo(float v, Appendable app)
>             throws IOException {
>         // allocate object FloatToDecimal
>         return new FloatToDecimal().appendDecimalTo(v, app);
>     }
> 
>     private Appendable appendDecimalTo(float v, Appendable app)
>             throws IOException {
>         switch (toDecimal(v)) {
>             case NON_SPECIAL:
>                 // allocate object char[]
>                 char[] chars = new char[index + 1];
>                 for (int i = 0; i < chars.length; ++i) {
>                     chars[i] = (char) bytes[i];
>                 }
>                 if (app instanceof StringBuilder builder) {
>                     return builder.append(chars);
>                 }
>                 if (app instanceof StringBuffer buffer) {
>                     return buffer.append(chars);
>                 }
>                 for (char c : chars) {
>                     app.append(c);
>                 }
>                 return app;
>             // ...
>         }
>     }
> }
> 
> class DoubleToDecimal {
>     public static String toString(double v) {
>         // allocate object DoubleToDecimal
>         return new DoubleToDecimal(false).toDecimalString(v);
>     }
> 
>     public static Appendable appendTo(double v, Appendable app)
>             throws IOException {
>         // allocate object DoubleToDecimal
>         return new DoubleToDecimal(false).appendDecimalTo(v, app);
>     }
> 
>     private Appendable appendDecimalTo(double v, Appendable app)
>             throws IOException {
>         switch (toDecimal(v, null)) {
>             case NON_SPECIAL:
>                 // allocate object char[]
>                 char[] chars = new char[index + 1];
>                 for (int i = 0; i < chars.length; ++i) {
>                     chars[i] = (char) bytes[i];
>                 }
>                 if (app instanceof StringBuilder builder) {
>                     return builder.append(chars);
>                 }
>               ...

The performance numbers under `MacBookPro M1 Pro` & `MacBookPro 2018 i9` are as follows:

## Apple MacBookPro 2018
* CPU Intel i9 (x64)

-Benchmark                               Mode  Cnt    Score   Error  Units
-StringBuilders.appendWithFloat8Latin1   avgt   15  381.206 ? 8.681  ns/op
-StringBuilders.appendWithFloat8Utf16    avgt   15  366.549 ? 1.775  ns/op
-StringBuilders.appendWithDouble8Latin1  avgt   15  570.085 ? 8.040  ns/op
-StringBuilders.appendWithDouble8Utf16   avgt   15  501.409 ? 5.750  ns/op

+Benchmark                               Mode  Cnt    Score    Error  Units (Webrevs 00 4c810154)
+StringBuilders.appendWithFloat8Latin1   avgt   15  226.236 ?  3.350  ns/op +68.49%
+StringBuilders.appendWithFloat8Utf16    avgt   15  299.072 ? 19.895  ns/op +22.56%
+StringBuilders.appendWithDouble8Latin1  avgt   15  328.887 ?  5.865  ns/op +73.33%
+StringBuilders.appendWithDouble8Utf16   avgt   15  390.618 ? 30.142  ns/op +28.36%


## MacBook M1 Pro
* CPU Apple M1 Pro (aarch64)

-Benchmark                               Mode  Cnt    Score    Error  Units base
-StringBuilders.appendWithFloat8Latin1   avgt   15  317.144 ? 11.325  ns/op
-StringBuilders.appendWithFloat8Utf16    avgt   15  316.980 ? 17.955  ns/op
-StringBuilders.appendWithDouble8Latin1  avgt   15  440.853 ? 13.067  ns/op
-StringBuilders.appendWithDouble8Utf16   avgt   15  418.896 ?  4.610  ns/op


+Benchmark                               Mode  Cnt    Score    Error  Units (Webrevs 00 4c810154)
+StringBuilders.appendWithFloat8Latin1   avgt   15  168.231 ?  4.749  ns/op +88.51%
+StringBuilders.appendWithFloat8Utf16    avgt   15  213.981 ?  3.274  ns/op +48.13%
+StringBuilders.appendWithDouble8Latin1  avgt   15  241.536 ?  0.993  ns/op +82.52%
+StringBuilders.appendWithDouble8Utf16   avgt   15  284.863 ? 10.381  ns/op +47.05%

-------------

PR Comment: https://git.openjdk.org/jdk/pull/19730#issuecomment-2169061500