8365205: C2: Optimize popcount value computation using knownbits #27075

jatin-bhateja · 2025-09-03T16:10:43Z

This patch optimizes PopCount value transforms using KnownBits information.
Following are the results of the micro-benchmark included with the patch

System: 13th Gen Intel(R) Core(TM) i3-1315U

Baseline:
Benchmark                                      Mode  Cnt       Score   Error  Units
PopCountValueTransform.LogicFoldingKerenLong  thrpt    2  215460.670          ops/s
PopCountValueTransform.LogicFoldingKerenlInt  thrpt    2  294014.826          ops/s

Withopt:
Benchmark                                      Mode  Cnt       Score   Error  Units
PopCountValueTransform.LogicFoldingKerenLong  thrpt    2  389978.082          ops/s
PopCountValueTransform.LogicFoldingKerenlInt  thrpt    2  417261.583          ops/s

Kindly review and share your feedback.

Best Regards,
Jatin

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-8365205: C2: Optimize popcount value computation using knownbits (Enhancement - P4)

Reviewers

Hannes Greule (@SirYwell - Committer) 🔄 Re-review required (review applies to e206ccc3)
Quan Anh Mai (@merykitty - Committer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/27075/head:pull/27075
$ git checkout pull/27075

Update a local copy of the PR:
$ git checkout pull/27075
$ git pull https://git.openjdk.org/jdk.git pull/27075/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 27075

View PR using the GUI difftool:
$ git pr show -t 27075

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/27075.diff

Using Webrev

Link to Webrev Comment

jatin-bhateja · 2025-09-03T16:11:32Z

/label add hotspot-compiler-dev

bridgekeeper · 2025-09-03T16:11:40Z

👋 Welcome back jbhateja! 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 · 2025-09-03T16:12:38Z

❗ This change is not yet ready to be integrated.
See the Progress checklist in the description for automated requirements.

openjdk · 2025-09-03T16:13:08Z

@jatin-bhateja
The hotspot-compiler label was successfully added.

mlbridge · 2025-09-04T05:47:24Z

Webrevs

SirYwell · 2025-09-04T06:26:36Z

The change looks good, but I wonder:

if it makes sense to have some kind of IR tests (i.e., it's folded away when unneeded, when the input is a constant, ...)?
whether the explanation could be simplified: Assuming a correct implementation of the KnownBits canonicalization, we can argue
- _zeroes has the bits set that are known to be always 0. So BitsPer<Type> - popCount(x) gives you an upper limit of how many bits might be 1. And BitsPer<Type> - popCount(_zeroes) is equivalent to popCount(~_zeroes).
- _ones has the bits set that are known to be always 1. Trivially, popCount(_ones) is a valid lower bound.
- The rest repeats how adjust_bits_from_unsigned_bounds works, but that's not specific to the popcount nodes.

jatin-bhateja · 2025-09-09T02:09:53Z

Hi @TobiHartmann , @SirYwell , @eme64 , can you kindly verify the changes in the latest patch?

eme64

Very nice improvement @jatin-bhateja , thanks for working on it :)

test/hotspot/jtreg/compiler/intrinsics/TestPopCountValueTransforms.java

src/hotspot/share/opto/countbitsnode.cpp

test/hotspot/jtreg/compiler/intrinsics/TestPopCountValueTransforms.java

eme64 · 2025-09-09T11:44:59Z

test/micro/org/openjdk/bench/java/lang/PopCountValueTransform.java

+    @Benchmark
+    public int StockKernelInt() {
+        int res = 0;
+        for (int i = lower_bound; i < upper_bound; i++) {
+            int constrained_i = i & 0xFFFF;
+            res += constrained_i;
+        }
+        return res;
+    }
+
+    @Benchmark
+    public int LogicFoldingKerenlInt() {
+        int res = 0;
+        for (int i = lower_bound; i < upper_bound; i++) {
+            int constrained_i = i & 0xFFFF;
+            if (Integer.bitCount(constrained_i) > 16) {
+                throw new AssertionError("Uncommon trap");
+            }
+            res += constrained_i;
+        }
+        return res;
+    }
+
+    @Benchmark
+    public long StockKernelLong() {
+        long res = 0;
+        for (int i = lower_bound; i < upper_bound; i++) {
+            long constrained_i = i & 0xFFFFFFL;
+            res += constrained_i;
+        }
+        return res;
+    }
+
+    @Benchmark
+    public long LogicFoldingKerenLong() {
+        long res = 0;
+        for (int i = lower_bound; i < upper_bound; i++) {
+            long constrained_i = i & 0xFFFFFFL;
+            if (Long.bitCount(constrained_i) > 24) {
+                throw new AssertionError("Uncommon trap");
+            }
+            res += constrained_i;
+        }
+        return res;
+    }


I assume the stock kernels are there to show performance if there is no op, the folding kernels you hope have the same performance. It would be nice to have one where the bitCount does not fold away, just to keep that comparison :)

I see your point, on a second thought, since any benchmarks compare the performance of kernels with and without optimization it's better to do away with the stock variants and only retain folding kernels.

test/hotspot/jtreg/compiler/intrinsics/TestPopCountValueTransforms.java

Co-authored-by: Emanuel Peter <[email protected]>

test/hotspot/jtreg/compiler/intrinsics/TestPopCountValueTransforms.java

eme64 · 2025-09-18T12:55:16Z

@jatin-bhateja I'm going to be out of the office for about 3 weeks, so feel free to ask others for reviews!

openjdk · 2025-09-18T19:41:43Z

@jatin-bhateja 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.

jatin-bhateja · 2025-09-19T08:19:40Z

Hi @eme64 , @chhagedorn , @SirYwell , let me know if its good to land now.

eme64

As I'm about to board my plane for a 3-week vacation, I'm leaving a last little note for the reviewers.

I think this is a really nice addition, so thanks for doing it @jatin-bhateja 😊 . Though it will only reach its full potential once we implement more "basic" KnownBits optimizations such as JDK-8367341.

Please make sure you test it, and make sure the random values generated with the Generators in test/hotspot/jtreg/compiler/intrinsics/TestPopCountValueTransforms.java make sense. Currently, there is for example a 32 bit range for a 64 bit long value, which is not correct, I think.

By default, my recommendation is to not constrain the Generators ranges, unless there is a really good reason. Generators are already built to produce values close to zero at an over-proportional rate. But by not restricting we may at some point also hit cases that we did not anticipate, and catch bugs that way.

test/hotspot/jtreg/compiler/intrinsics/TestPopCountValueTransforms.java

jatin-bhateja · 2025-09-19T09:49:36Z

As I'm about to board my plane for a 3-week vacation, I'm leaving a last little note for the reviewers.

I think this is a really nice addition, so thanks for doing it @jatin-bhateja 😊 . Though it will only reach its full potential once we implement more "basic" KnownBits optimizations such as JDK-8367341.

Correct, currently KnownBits information is constrained as they are generated for limited value ranges, as discussed in
#27075 (comment)

SirYwell · 2025-09-19T11:16:26Z

src/hotspot/share/opto/countbitsnode.cpp

+// We use the KnownBits information from the integer types to derive how many one bits
+// we have at least and at most.
+// From the definition of KnownBits, we know:
+//   zeros: Indicates which bits must be 0: ones[i] =1 -> t[i]=0


I'm a bit confused by this, is ones[i] mixed up with zeros[i]? I.e., t[i]=0 if zeros[i]=1

@SirYwell , comment updated.

Links to formal z3 proofs for this:-

#25928 (comment)

https://bugs.openjdk.org/browse/JDK-8365205?focusedId=14807707&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-14807707:~:text=C%3A%5CGithub%5Csoftwares%5Cz3%5Cz3%2D4.15.2%2Dx64%2Dwin%5Cbin%5Cpython%3Epython3%20known_bits_popcount.py%0AMain%20constraints%20satisfiable.%0AConstraints%20are%20valid%20(negation%20unsatisfiable).

jatin-bhateja · 2025-09-25T15:58:27Z

Hi @SirYwell , @chhagedorn , @eme64 , I have addressed your comments. Let me know if this is good to land in.

SirYwell

Looks good to me now, although I'm not exactly sure about the semantics of widen and whether Type::WidenMax is the right choice here. Someone else has to look at that.

Thanks for the work!

liach

A suggestion, we might do this as a later RFE.

src/hotspot/share/opto/countbitsnode.cpp

jatin-bhateja · 2025-09-29T17:16:13Z

Looks good to me now, although I'm not exactly sure about the semantics of widen and whether Type::WidenMax is the right choice here. Someone else has to look at that.

Thanks for the work!

Hi @TobiHartmann @chhagedorn , can you kindly run this through your testing and share if its good to land in.

TobiHartmann · 2025-10-03T08:38:40Z

I submitted testing for this and will report back once it finished.

merykitty · 2025-10-03T12:14:46Z

src/hotspot/share/opto/countbitsnode.cpp

+    return Type::TOP;
+  }
+  KnownBits<juint> bits = t->isa_int()->_bits;
+  return TypeInt::make(population_count(bits._ones), population_count(~bits._zeros), Type::WidenMax);


The widen of the output should be the same as the widen of the input, not WidenMax here.

Thanks @merykitty, widen is mainly used for optimistic data flow analysis pass like CCP where type analysis begins with TOP and progressively grows the value range till convergence / fixed point.
it's good to preserve the widen of input to delay eager convergence.

TobiHartmann · 2025-10-06T07:50:41Z

Testing all passed. I'll pass the review to someone else.

eirbjo · 2025-10-06T08:07:14Z

Is the core-libs label appropriate for this PR? Looks hotspot specific?

SirYwell · 2025-10-06T08:12:32Z

Is the core-libs label appropriate for this PR? Looks hotspot specific?

That label was added automatically, closely after https://mail.openjdk.org/pipermail/jdk-dev/2025-September/010486.html. Not sure why, but the change is definitely hotspot specific.

eirbjo · 2025-10-06T08:17:29Z

That label was added automatically, closely after https://mail.openjdk.org/pipermail/jdk-dev/2025-September/010486.html. Not sure why, but the change is definitely hotspot specific.

Right, makes sense. I'll go ahead and remove that label.

/label remove core-libs

openjdk · 2025-10-06T08:18:27Z

@eirbjo
The core-libs label was successfully removed.

jatin-bhateja · 2025-10-06T08:22:52Z

Hi @merykitty , Can you kindly be the second reviewer?

liach · 2025-10-06T15:20:52Z

FYI this patch was marked core-libs because of the benchmark addition in java/lang. I wonder if it belongs to vm/compiler instead.

merykitty

LGTM otherwise

merykitty · 2025-10-06T15:53:35Z

src/hotspot/share/opto/countbitsnode.cpp

+  if (t == Type::TOP) {
+    return Type::TOP;
+  }
+  const TypeInt* tint = t->isa_int();


This should be is_int. isa_int is fine, but in cases of unexpected types, it will SIGSEGV instead of throwing an assertion, which is more difficult to debug.

8365205: C2: Optimize popcount value computation using knownbits

c83be33

openjdk bot added the hotspot-compiler [email protected] label Sep 3, 2025

jatin-bhateja marked this pull request as ready for review September 4, 2025 05:42

openjdk bot added the rfr Pull request is ready for review label Sep 4, 2025

jbhateja and others added 2 commits September 5, 2025 22:37

New IR test addition and review resolutions

a68fbc0

Update countbitsnode.cpp

52ae6bc

eme64 suggested changes Sep 9, 2025

View reviewed changes

jatin-bhateja and others added 4 commits September 10, 2025 17:52

Update src/hotspot/share/opto/countbitsnode.cpp

36ecb5d

Co-authored-by: Emanuel Peter <[email protected]>

review resoultions

f1095b5

Update TestPopCountValueTransforms.java

9e3957d

Adding random bound test point

a7f9b79

eme64 reviewed Sep 12, 2025

View reviewed changes

test/hotspot/jtreg/compiler/intrinsics/TestPopCountValueTransforms.java Outdated Show resolved Hide resolved

Extending the random ranges

278f1dc

eme64 suggested changes Sep 16, 2025

View reviewed changes

test/hotspot/jtreg/compiler/intrinsics/TestPopCountValueTransforms.java Outdated Show resolved Hide resolved

test/hotspot/jtreg/compiler/intrinsics/TestPopCountValueTransforms.java Outdated Show resolved Hide resolved

openjdk bot added the core-libs [email protected] label Sep 18, 2025

Review comments resolutions

367622b

eme64 suggested changes Sep 19, 2025

View reviewed changes

test/hotspot/jtreg/compiler/intrinsics/TestPopCountValueTransforms.java Outdated Show resolved Hide resolved

Update TestPopCountValueTransforms.java

92cf2fa

SirYwell reviewed Sep 19, 2025

View reviewed changes

Update countbitsnode.cpp

e206ccc

SirYwell approved these changes Sep 25, 2025

View reviewed changes

liach reviewed Sep 25, 2025

View reviewed changes

src/hotspot/share/opto/countbitsnode.cpp Show resolved Hide resolved

merykitty reviewed Oct 3, 2025

View reviewed changes

Review comments resolution

85b10e8

openjdk bot removed the core-libs [email protected] label Oct 6, 2025

merykitty approved these changes Oct 6, 2025

View reviewed changes

8365205: C2: Optimize popcount value computation using knownbits #27075

Are you sure you want to change the base?

8365205: C2: Optimize popcount value computation using knownbits #27075

Conversation

jatin-bhateja commented Sep 3, 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!

jatin-bhateja commented Sep 3, 2025

Uh oh!

bridgekeeper bot commented Sep 3, 2025

Uh oh!

openjdk bot commented Sep 3, 2025

Uh oh!

openjdk bot commented Sep 3, 2025

Uh oh!

mlbridge bot commented Sep 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Webrevs

Uh oh!

SirYwell commented Sep 4, 2025

Uh oh!

jatin-bhateja commented Sep 9, 2025

Uh oh!

eme64 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

eme64 Sep 9, 2025

Choose a reason for hiding this comment

Uh oh!

jatin-bhateja Sep 10, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

eme64 commented Sep 18, 2025

Uh oh!

openjdk bot commented Sep 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jatin-bhateja commented Sep 19, 2025

Uh oh!

eme64 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jatin-bhateja commented Sep 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

SirYwell Sep 19, 2025

Choose a reason for hiding this comment

Uh oh!

jatin-bhateja Sep 19, 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!

jatin-bhateja commented Sep 25, 2025

Uh oh!

SirYwell left a comment

Choose a reason for hiding this comment

Uh oh!

liach left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jatin-bhateja commented Sep 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jatin-bhateja commented Sep 3, 2025 •

edited by openjdk bot

Loading

mlbridge bot commented Sep 4, 2025 •

edited

Loading

openjdk bot commented Sep 18, 2025 •

edited

Loading

jatin-bhateja commented Sep 19, 2025 •

edited

Loading

jatin-bhateja Sep 19, 2025 •

edited

Loading

jatin-bhateja commented Sep 29, 2025 •

edited

Loading

jatin-bhateja commented Oct 6, 2025 •

edited

Loading