8366017: Extend the set of inputs handled by fast paths in FloatingDecimal #26990

rgiulietti · 2025-08-28T15:24:21Z

Yet another step in modernizing FloatingDecimals floating-point parsing.

Progress

Change must be properly reviewed (1 review required, with at least 1 Reviewer)
Change must not contain extraneous whitespace
Commit message must refer to an issue

Issue

JDK-8366017: Extend the set of inputs handled by fast paths in FloatingDecimal (Enhancement - P4)

Reviewers

Joe Darcy (@jddarcy - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/26990/head:pull/26990
$ git checkout pull/26990

Update a local copy of the PR:
$ git checkout pull/26990
$ git pull https://git.openjdk.org/jdk.git pull/26990/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 26990

View PR using the GUI difftool:
$ git pr show -t 26990

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/26990.diff

Using Webrev

Link to Webrev Comment

…cimal

bridgekeeper · 2025-08-28T15:25:03Z

👋 Welcome back rgiulietti! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

openjdk · 2025-08-28T15:25:39Z

@rgiulietti This change now passes all automated pre-integration checks.

ℹ️ This project also has non-automated pre-integration requirements. Please see the file CONTRIBUTING.md for details.

After integration, the commit message for the final commit will be:

8366017: Extend the set of inputs handled by fast paths in FloatingDecimal

Reviewed-by: darcy

You can use pull request commands such as /summary, /contributor and /issue to adjust it as needed.

At the time when this comment was updated there had been 37 new commits pushed to the master branch:

d62553d: 8316274: javax/swing/ButtonGroup/TestButtonGroupFocusTraversal.java fails in Ubuntu 23.10 with Motif LAF
6080ccd: 8370797: Test runtime/ErrorHandling/AccessZeroNKlassHitsProtectionZone.java failed on macos 26
436dc68: 8367059: DTLS: loss of NewSessionTicket message results in handshake failure
... and 34 more: https://git.openjdk.org/jdk/compare/9625993611bb6acf84d428bea4a65d33b9d66e5f...master

As there are no conflicts, your changes will automatically be rebased on top of these commits when integrating. If you prefer to avoid this automatic rebasing, please check the documentation for the /integrate command for further details.

➡️ To integrate this PR with the above commit message to the master branch, type /integrate in a new comment.

openjdk · 2025-08-28T15:26:17Z

@rgiulietti The following label will be automatically applied to this pull request:

core-libs

When this pull request is ready to be reviewed, an "RFR" email will be sent to the corresponding mailing list. If you would like to change these labels, use the /label pull request command.

rgiulietti · 2025-08-28T15:48:16Z

Parsing proper has been the topic of a previous PR. However, the mathematical conversion from the decimal representation itself still relied on existing code.

The plan is to rework the conversion algorithms and end up with a 3 tiers schema.

The 1st tier implements some simple, efficient fast paths. They apply to a large number of simple decimals, like 1.2345 or 2.3456e-9 (see the code for the details).
The 2nd tier will implement some more involved fast paths with higher precision arithmetic. The hope is to be able to convert most decimals produced by Double.toString(double) in just the 1st or 2nd tier.
The 3rd is the fall back tier, using full precision arithmetic for long decimals. This tier will undergo a rehaul, reducing its complexity, and relying on [Mutable]BigInteger rather than FDBigInteger.

wenshao · 2025-09-14T10:08:21Z

src/java.base/share/classes/jdk/internal/math/FloatingDecimal.java

+    private static final int    EXP_SHIFT = DoubleConsts.SIGNIFICAND_WIDTH - 1;
+    private static final long   FRACT_HOB = 1L << EXP_SHIFT; // assumed High-Order bit
+    private static final long   EXP_ONE   = (long) DoubleConsts.EXP_BIAS << EXP_SHIFT; // exponent of 1.0
+    private static final int    MAX_SMALL_BIN_EXP = 62;
+    private static final int    MIN_SMALL_BIN_EXP = -63 / 3;
+    private static final int    MAX_DECIMAL_DIGITS = 15;  // max{n : 10^n <= 2^P}
+    private static final int    FLOG_10_MAX_LONG = 18;  // max{i : 10^i ≤ Long.MAX_VALUE}
+
+    private static final int    SINGLE_EXP_SHIFT  = FloatConsts.SIGNIFICAND_WIDTH - 1;
+    private static final int    SINGLE_FRACT_HOB  = 1 << SINGLE_EXP_SHIFT;
+    private static final int    SINGLE_MAX_DECIMAL_DIGITS = 7;


Suggested change

private static final int EXP_SHIFT = DoubleConsts.SIGNIFICAND_WIDTH - 1;

private static final long FRACT_HOB = 1L << EXP_SHIFT; // assumed High-Order bit

private static final long EXP_ONE = (long) DoubleConsts.EXP_BIAS << EXP_SHIFT; // exponent of 1.0

private static final int MAX_SMALL_BIN_EXP = 62;

private static final int MIN_SMALL_BIN_EXP = -63 / 3;

private static final int MAX_DECIMAL_DIGITS = 15; // max{n : 10^n <= 2^P}

private static final int FLOG_10_MAX_LONG = 18; // max{i : 10^i ≤ Long.MAX_VALUE}

private static final int SINGLE_EXP_SHIFT = FloatConsts.SIGNIFICAND_WIDTH - 1;

private static final int SINGLE_FRACT_HOB = 1 << SINGLE_EXP_SHIFT;

private static final int SINGLE_MAX_DECIMAL_DIGITS = 7;

private static final int

EXP_SHIFT = DoubleConsts.SIGNIFICAND_WIDTH - 1,

MAX_SMALL_BIN_EXP = 62,

MIN_SMALL_BIN_EXP = -63 / 3,

MAX_DECIMAL_DIGITS = 15, // max{n : 10^n <= 2^P}

FLOG_10_MAX_LONG = 18, // max{i : 10^i ≤ Long.MAX_VALUE}

SINGLE_EXP_SHIFT = FloatConsts.SIGNIFICAND_WIDTH - 1,

SINGLE_FRACT_HOB = 1 << SINGLE_EXP_SHIFT,

SINGLE_MAX_DECIMAL_DIGITS = 7;

private static final long

FRACT_HOB = 1L << EXP_SHIFT, // assumed High-Order bit

EXP_ONE = (long) DoubleConsts.EXP_BIAS << EXP_SHIFT; // exponent of 1.0

How about this style?

wenshao · 2025-09-14T10:10:58Z

src/java.base/share/classes/jdk/internal/math/FloatingDecimal.java

+    private static final BinaryToASCIIConverter B2AC_POSITIVE_INFINITY = new ExceptionalBinaryToASCIIBuffer(INFINITY_REP);
+    private static final BinaryToASCIIConverter B2AC_NEGATIVE_INFINITY = new ExceptionalBinaryToASCIIBuffer("-" + INFINITY_REP);
+    private static final BinaryToASCIIConverter B2AC_NOT_A_NUMBER = new ExceptionalBinaryToASCIIBuffer(NAN_REP);
+    private static final BinaryToASCIIConverter B2AC_POSITIVE_ZERO = new BinaryToASCIIBuffer(false, new byte[] {'0'});
+    private static final BinaryToASCIIConverter B2AC_NEGATIVE_ZERO = new BinaryToASCIIBuffer(true,  new byte[] {'0'});


Suggested change

private static final BinaryToASCIIConverter B2AC_POSITIVE_INFINITY = new ExceptionalBinaryToASCIIBuffer(INFINITY_REP);

private static final BinaryToASCIIConverter B2AC_NEGATIVE_INFINITY = new ExceptionalBinaryToASCIIBuffer("-" + INFINITY_REP);

private static final BinaryToASCIIConverter B2AC_NOT_A_NUMBER = new ExceptionalBinaryToASCIIBuffer(NAN_REP);

private static final BinaryToASCIIConverter B2AC_POSITIVE_ZERO = new BinaryToASCIIBuffer(false, new byte[] {'0'});

private static final BinaryToASCIIConverter B2AC_NEGATIVE_ZERO = new BinaryToASCIIBuffer(true, new byte[] {'0'});

private static final BinaryToASCIIConverter

B2AC_POSITIVE_INFINITY = new ExceptionalBinaryToASCIIBuffer(INFINITY_REP),

B2AC_NEGATIVE_INFINITY = new ExceptionalBinaryToASCIIBuffer("-" + INFINITY_REP),

B2AC_NOT_A_NUMBER = new ExceptionalBinaryToASCIIBuffer(NAN_REP),

B2AC_POSITIVE_ZERO = new BinaryToASCIIBuffer(false, new byte[] {'0'}),

B2AC_NEGATIVE_ZERO = new BinaryToASCIIBuffer(true, new byte[] {'0'});

wenshao · 2025-09-14T10:14:17Z

src/java.base/share/classes/jdk/internal/math/FloatingDecimal.java

+    private static final PreparedASCIIToBinaryBuffer A2BC_POSITIVE_INFINITY =
+            new PreparedASCIIToBinaryBuffer(Double.POSITIVE_INFINITY, Float.POSITIVE_INFINITY);
+    private static final PreparedASCIIToBinaryBuffer A2BC_NEGATIVE_INFINITY =
+            new PreparedASCIIToBinaryBuffer(Double.NEGATIVE_INFINITY, Float.NEGATIVE_INFINITY);
+    private static final PreparedASCIIToBinaryBuffer A2BC_NOT_A_NUMBER  =
+            new PreparedASCIIToBinaryBuffer(Double.NaN, Float.NaN);
+    private static final PreparedASCIIToBinaryBuffer A2BC_POSITIVE_ZERO =
+            new PreparedASCIIToBinaryBuffer(0.0d, 0.0f);
+    private static final PreparedASCIIToBinaryBuffer A2BC_NEGATIVE_ZERO =
+            new PreparedASCIIToBinaryBuffer(-0.0d, -0.0f);


Suggested change

private static final PreparedASCIIToBinaryBuffer A2BC_POSITIVE_INFINITY =

new PreparedASCIIToBinaryBuffer(Double.POSITIVE_INFINITY, Float.POSITIVE_INFINITY);

private static final PreparedASCIIToBinaryBuffer A2BC_NEGATIVE_INFINITY =

new PreparedASCIIToBinaryBuffer(Double.NEGATIVE_INFINITY, Float.NEGATIVE_INFINITY);

private static final PreparedASCIIToBinaryBuffer A2BC_NOT_A_NUMBER =

new PreparedASCIIToBinaryBuffer(Double.NaN, Float.NaN);

private static final PreparedASCIIToBinaryBuffer A2BC_POSITIVE_ZERO =

new PreparedASCIIToBinaryBuffer(0.0d, 0.0f);

private static final PreparedASCIIToBinaryBuffer A2BC_NEGATIVE_ZERO =

new PreparedASCIIToBinaryBuffer(-0.0d, -0.0f);

private static final PreparedASCIIToBinaryBuffer

A2BC_POSITIVE_INFINITY = new PreparedASCIIToBinaryBuffer(Double.POSITIVE_INFINITY, Float.POSITIVE_INFINITY),

A2BC_NEGATIVE_INFINITY = new PreparedASCIIToBinaryBuffer(Double.NEGATIVE_INFINITY, Float.NEGATIVE_INFINITY),

A2BC_NOT_A_NUMBER = new PreparedASCIIToBinaryBuffer(Double.NaN, Float.NaN),

A2BC_POSITIVE_ZERO = new PreparedASCIIToBinaryBuffer(0.0d, 0.0f),

A2BC_NEGATIVE_ZERO = new PreparedASCIIToBinaryBuffer(-0.0d, -0.0f);

mlbridge · 2025-10-14T17:57:05Z

Webrevs

rgiulietti · 2025-10-14T17:58:40Z

This work's focus is on:

Code reduction.
Extensive documentation of mathematical aspects.
Enhanced performance.

In particular, parsing outcomes of the Double.toString(double) and Float.toString(float) methods should be faster.

Before

Benchmark                               Mode  Cnt     Score    Error  Units
FloatingPointParse.parseDoubleS1        avgt    9   268.213 ±  0.901  ns/op
FloatingPointParse.parseDoubleS10       avgt    9  1091.924 ± 52.040  ns/op
FloatingPointParse.parseDoubleS2        avgt    9   340.953 ± 10.139  ns/op
FloatingPointParse.parseDoubleS4        avgt    9   499.272 ± 13.940  ns/op
FloatingPointParse.parseDoubleToString  avgt    9   272.870 ±  1.677  ns/op
FloatingPointParse.parseFloatS1         avgt    9    72.291 ±  1.574  ns/op
FloatingPointParse.parseFloatS10        avgt    9   321.380 ± 10.046  ns/op
FloatingPointParse.parseFloatS2         avgt    9   104.538 ±  2.701  ns/op
FloatingPointParse.parseFloatS4         avgt    9   159.879 ±  6.711  ns/op
FloatingPointParse.parseFloatToString   avgt    9   140.413 ±  1.844  ns/op

After

Benchmark                               Mode  Cnt    Score   Error  Units
FloatingPointParse.parseDoubleS1        avgt    9   77.753 ± 1.437  ns/op
FloatingPointParse.parseDoubleS10       avgt    9  368.850 ± 0.974  ns/op
FloatingPointParse.parseDoubleS2        avgt    9  112.624 ± 2.094  ns/op
FloatingPointParse.parseDoubleS4        avgt    9  177.373 ± 1.137  ns/op
FloatingPointParse.parseDoubleToString  avgt    9   76.904 ± 1.152  ns/op
FloatingPointParse.parseFloatS1         avgt    9   53.126 ± 0.793  ns/op
FloatingPointParse.parseFloatS10        avgt    9  300.140 ± 1.330  ns/op
FloatingPointParse.parseFloatS2         avgt    9   83.329 ± 0.251  ns/op
FloatingPointParse.parseFloatS4         avgt    9  137.046 ± 2.160  ns/op
FloatingPointParse.parseFloatToString   avgt    9   56.267 ± 0.698  ns/op```

jddarcy · 2025-10-29T04:05:03Z

src/java.base/share/classes/jdk/internal/math/FDBigInteger.java

        CDS.initializeFromArchive(FDBigInteger.class);
        Object[] caches = archivedCaches;
        if (caches == null) {
            long[] long5pow = {


Would it be worthwhile to make these arrays @Stable?

In the future, might be a candidate for StableConstant/LazyValue functionality.

Would it be worthwhile to make these arrays @Stable?

In the future, might be a candidate for StableConstant/LazyValue functionality.

PS If each array is read only once, then perhaps the other data structures would be better initialized in a loop.

jddarcy · 2025-10-29T04:16:37Z

test/jdk/java/lang/Double/ParseDouble.java


+    private static void testFastPaths() {
+        /* Exercises the fast paths in jdk.internal.math.FloatingDecimal. */
+        check("1", 1.0);


Something to consider: specify the double values in hex literals -- this would exercise entirely different code paths than the decimal ones.

Good point, I fully agree.
Updating the test to use hexadecimals...

rgiulietti · 2025-11-03T09:47:03Z

/integrate

openjdk · 2025-11-03T09:49:00Z

Going to push as commit deb7edb.
Since your change was applied there have been 94 commits pushed to the master branch:

400f51f: 8369913: Parallel: Refactor PSAdjustTask
89b492f: 8371018: Remove unused CollectedHeap::fill_with_object
10ea585: 8370345: Parallel: Rework TLAB accounting in MutableNUMASpace
... and 91 more: https://git.openjdk.org/jdk/compare/9625993611bb6acf84d428bea4a65d33b9d66e5f...master

Your commit was automatically rebased without conflicts.

openjdk · 2025-11-03T09:49:10Z

@rgiulietti Pushed as commit deb7edb.

💡 You may see a message that your pull request was closed with unmerged commits. This can be safely ignored.

rgiulietti added 10 commits August 22, 2025 20:08

8366017: Extend the set of inputs handled by fast paths in FloatingDe…

d0cee8a

…cimal

Small simplifications.

ed5ef90

First set of fast paths.

d82d687

Merge branch 'master' into 8366017

0114c26

Some simplifications.

617cb2f

Minor corrections.

970c896

Merge branch 'master' into 8366017

59e716b

Renaming qe to qp.

fd74b8e

Merge branch 'master' into 8366017

85d5108

Mainly enhanced comments.

4f49c5d

openjdk bot added the core-libs core-libs-dev@openjdk.org label Aug 28, 2025

Merge branch 'master' into 8366017

3197144

j3graham mentioned this pull request Sep 5, 2025

8367324: Avoid redundant parsing when formatting with DigitList #27118

Closed

3 tasks

First high precision fast path.

80d3f90

wenshao reviewed Sep 14, 2025

View reviewed changes

rgiulietti added 9 commits September 30, 2025 14:53

Restored some methods.

4551916

Merge branch 'master' into 8366017

2bf8943

Implemented 1st version of highPrecisionDoubleValue()

2b3fda4

Merge master

ac839bd

Minors.

f364ada

Reorganization of tables at the end.

0d03797

Merge branch 'master' into 8366017

88c88a3

Complete solution for doubles.

6ac8df7

Merge branch 'master' into 8366017

11bb6cd

rgiulietti added 3 commits October 14, 2025 17:30

Adapted tests to more restricted access.

966d619

JMH benchmarks.

7019239

More complete benchmarks.

5a4d7a6

rgiulietti marked this pull request as ready for review October 14, 2025 17:52

openjdk bot added the rfr Pull request is ready for review label Oct 14, 2025

rgiulietti added 6 commits October 22, 2025 12:19

Merge branch 'master' into 8366017

74536da

Fix a rare bug.

7c5cd92

Added tests to exercise fast paths.

21118e9

Merge branch 'master' into 8366017

314c984

Added test for all float->String->float roundtrips.

024ed76

Merge branch 'master' into 8366017

c348586

jddarcy reviewed Oct 29, 2025

View reviewed changes

jddarcy approved these changes Oct 29, 2025

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Oct 29, 2025

Replace decimal with hexadecimal floating-point literals in test.

07e08ac

openjdk bot removed the ready Pull request is ready to be integrated label Oct 29, 2025

Make arrays stable and remove synchronisation.

81b3c64

rgiulietti requested a review from jddarcy October 29, 2025 13:58

jddarcy approved these changes Oct 29, 2025

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Oct 29, 2025

openjdk bot added the integrated Pull request has been integrated label Nov 3, 2025

openjdk bot closed this Nov 3, 2025

openjdk bot removed ready Pull request is ready to be integrated rfr Pull request is ready for review labels Nov 3, 2025

8366017: Extend the set of inputs handled by fast paths in FloatingDecimal #26990

8366017: Extend the set of inputs handled by fast paths in FloatingDecimal #26990

Uh oh!

Conversation

rgiulietti commented Aug 28, 2025 • edited by openjdk bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Progress

Issue

Reviewers

Reviewing

Uh oh!

bridgekeeper bot commented Aug 28, 2025

Uh oh!

openjdk bot commented Aug 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

openjdk bot commented Aug 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rgiulietti commented Aug 28, 2025

Uh oh!

wenshao Sep 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

wenshao Sep 14, 2025

Choose a reason for hiding this comment

Uh oh!

wenshao Sep 14, 2025

Choose a reason for hiding this comment

Uh oh!

mlbridge bot commented Oct 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Webrevs

Uh oh!

rgiulietti commented Oct 14, 2025

Uh oh!

jddarcy Oct 29, 2025

Choose a reason for hiding this comment

Uh oh!

jddarcy Oct 29, 2025

Choose a reason for hiding this comment

Uh oh!

jddarcy Oct 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rgiulietti Oct 29, 2025

Choose a reason for hiding this comment

Uh oh!

rgiulietti commented Nov 3, 2025

Uh oh!

openjdk bot commented Nov 3, 2025

Uh oh!

openjdk bot commented Nov 3, 2025

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

3 participants

rgiulietti commented Aug 28, 2025 •

edited by openjdk bot

Loading

openjdk bot commented Aug 28, 2025 •

edited

Loading

openjdk bot commented Aug 28, 2025 •

edited

Loading

wenshao Sep 14, 2025 •

edited

Loading

mlbridge bot commented Oct 14, 2025 •

edited

Loading

jddarcy Oct 29, 2025 •

edited

Loading