8345125: Aarch64: Add aarch64 backend for Float16 scalar operations #23748

Bhavana-Kilambi · 2025-02-24T12:09:57Z

This patch adds aarch64 backend for scalar FP16 operations namely - add, subtract, multiply, divide, fma, sqrt, min and max.

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-8345125: Aarch64: Add aarch64 backend for Float16 scalar operations (Enhancement - P4)

Reviewers

Andrew Haley (@theRealAph - Reviewer)
Hao Sun (@shqking - Committer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/23748/head:pull/23748
$ git checkout pull/23748

Update a local copy of the PR:
$ git checkout pull/23748
$ git pull https://git.openjdk.org/jdk.git pull/23748/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 23748

View PR using the GUI difftool:
$ git pr show -t 23748

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/23748.diff

Using Webrev

Link to Webrev Comment

This patch adds aarch64 backend for scalar FP16 operations namely - add, subtract, multiply, divide, fma, sqrt, min and max.

bridgekeeper · 2025-02-24T12:10:45Z

👋 Welcome back bkilambi! 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 · 2025-02-24T12:11:11Z

@Bhavana-Kilambi 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:

8345125: Aarch64: Add aarch64 backend for Float16 scalar operations

Reviewed-by: aph, haosun

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 18 new commits pushed to the master branch:

5c06723: 8355074: RISC-V: C2: Support Vector-Scalar version of Zvbb Vector And-Not instruction
a2f9c24: 8354327: Rewrite runtime/LoadClass/LoadClassNegative.java
e6cea40: 8355475: UNCTest should use an existing UNC path
... and 15 more: https://git.openjdk.org/jdk/compare/0edd018a48c202a6da4afe80e245799b47000885...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.

As you do not have Committer status in this project an existing Committer must agree to sponsor your change. Possible candidates are the reviewers of this PR (@theRealAph, @shqking) but any other Committer may sponsor as well.

➡️ To flag this PR as ready for integration with the above commit message, type /integrate in a new comment. (Afterwards, your sponsor types /sponsor in a new comment to perform the integration).

openjdk · 2025-02-24T12:11:41Z

@Bhavana-Kilambi The following labels will be automatically applied to this pull request:

graal
hotspot

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

mlbridge · 2025-02-24T12:14:03Z

Webrevs

src/hotspot/cpu/aarch64/aarch64.ad

theRealAph · 2025-02-25T13:37:51Z

test/hotspot/gtest/aarch64/aarch64-asmtest.py

-        0xff03, 0xfffe]
+        0x7e0, 0xfc0, 0x1f80, 0x3ff0, 0x7e00, 0x8000,
+        0x81ff, 0xc1ff, 0xc003, 0xc7ff, 0xdfff, 0xe03f,
+        0xe1ff, 0xf801, 0xfc00, 0xfc07, 0xff03, 0xfffe]


So here you've deleted the duplicated 0x7e00 (good) but also the not-duplicated 0xe10f. Is 0xe10f not valid?

Hi, yes 0xe10f does not seem to be valid. While I tried generating the asmtest.out.h I ran into errors with this value -

aarch64ops.s:1105: Error: immediate out of range at operand 3 -- eor z6.h,z6.h,#0xe10f aarch64ops.s:1123: Error: immediate out of range at operand 3 -- eor z3.h,z3.h,#0xe10f

So I looked it up here - https://gist.github.com/dinfuehr/51a01ac58c0b23e4de9aac313ed6a06a to see if this number is a legal immediate and looks like it isn't. Maybe it's just chance that this number wasn't generated before as an immediate operand and these errors didn't show up till now.

theRealAph · 2025-02-25T13:44:44Z

Overall, this looks like a great pice of work. I only have a few changes in comments and a question, then we're good to go.

shqking · 2025-02-26T08:26:57Z

src/hotspot/cpu/aarch64/assembler_aarch64.hpp

+  INSN(fnmuld, 0b000, 0b01, 0b100010, 0b1);
+
+  // Half-precision floating-point instructions
+  INSN(fabdh,  0b011, 0b11, 0b000101, 0b0);


I suppose fadbh and fnmulh are added to keep aligned with the float and double ones, i.e. fabd(s|d) and fnmul(s|d).

I noticed that there are matching rules for fabd(s|d), i.e. absd(F|D)_reg. I wonder if we need add the corresponding rule for fp16 here?

Hi @shqking , thanks for your review comments. Yes I added fabdh and fnmulh to keep aligned with float and double types.
For adding support for FP16 absd we need AbsHF to be supported (along with SubHF) but AbsHF node is not implemented currently. abs operation is directly executed from the java code here -

jdk/src/jdk.incubator.vector/share/classes/jdk/incubator/vector/Float16.java

Line 1464 in 037e471

public static Float16 abs(Float16 f16) {

and is not intrinsified or pattern matched like other FP16 operations. Same with negate operation for FP16 -

jdk/src/jdk.incubator.vector/share/classes/jdk/incubator/vector/Float16.java

Line 1449 in 037e471

return shortBitsToFloat16((short)(f16.value ^ (short)0x0000_8000));

On the Valhalla repo, while these operation were being developed, I tried adding support for AbsHF/NegHF which emitted fabs and fneg instructions but the performance with the direct java code(bit manipulation operations) was much faster (sorry don't remember the exact number) so we decided to go with the java implementation instead.
I still added fabd here because op21 is 0 only in fabd H variant and felt that it'd be better to handle it here as it belongs to this group of instructions. Please let me know your thoughts.

According to the RM, fabd is in Advanced SIMD scalar three same FP16, but the rest are in Floating-point data-processing (2 source). The decoding scheme looks rather different.fabd, then, doesn't really fit here, but in a section with the rest of the three same FP16 instructions.
The encoding scheme for Advanced SIMD scalar three same FP16 is pretty simple, so I suggest you create a new group for them, and put fabd in there.

@Bhavana-Kilambi Thanks for your explanation for the missing AbsHF. It's okay to me to have fadbh and fnmulh in this patch.

Overall it's good to me except aph's comment above.

Hi @theRealAph Thanks again for the review and apologies for the delay in responding.
I moved the three fabd instructions out of their current place and added them in two separate sections - one for the single and double precision (Advanced SIMD scalar three same) and another for FP16 (Advanced SIMD scalar three same FP16). Please review the changes. Thank you!

bridgekeeper · 2025-03-26T21:24:27Z

@Bhavana-Kilambi This pull request has been inactive for more than 4 weeks and will be automatically closed if another 4 weeks passes without any activity. To avoid this, simply add a new comment to the pull request. Feel free to ask for assistance if you need help with progressing this pull request towards integration!

Bhavana-Kilambi · 2025-03-27T08:10:31Z

Hello @shqking @theRealAph , sincere apologies for the delay in addressing the review comments. I am planning on uploading a patch soon addressing all review comments. Thank you !

Bhavana-Kilambi · 2025-03-31T09:51:49Z

Hello, I would not be able to respond to comments until the next couple months or so due to some urgent tasks at work. Until then, I'd move this PR to draft status so that it would not be closed due to lack of activity. Thank you for the review!

openjdk · 2025-03-31T09:53:28Z

@Bhavana-Kilambi this pull request can not be integrated into master due to one or more merge conflicts. To resolve these merge conflicts and update this pull request you can run the following commands in the local repository for your personal fork:

git checkout JDK-8345125
git fetch https://git.openjdk.org/jdk.git master
git merge FETCH_HEAD
# resolve conflicts and follow the instructions given by git merge
git commit -m "Merge master"
git push

mhaessig · 2025-04-16T07:47:02Z

Can you please uncomment the following tests using aarch64 float16 in this PR?

https://github.com/mhaessig/jdk/blob/cec48ed270d3bdf704c389a091b42a32c2ed6440/test/hotspot/jtreg/compiler/floatingpoint/TestSubNodeFloatDoubleNegation.java#L58-L60

Bhavana-Kilambi · 2025-04-24T15:47:17Z

Can you please uncomment the following tests using aarch64 float16 in this PR?

https://github.com/mhaessig/jdk/blob/cec48ed270d3bdf704c389a091b42a32c2ed6440/test/hotspot/jtreg/compiler/floatingpoint/TestSubNodeFloatDoubleNegation.java#L58-L60

Done. Thanks for notifying.

shqking · 2025-04-25T03:26:26Z

Hi @Bhavana-Kilambi, I noticed there exists inconsistency between test/hotspot/gtest/aarch64/asmtest.out.h and test/hotspot/gtest/aarch64/aarch64-asmtest.py in the latest commit. We should resolve that.

Bhavana-Kilambi · 2025-04-25T08:18:39Z

Hi @Bhavana-Kilambi, I noticed there exists inconsistency between test/hotspot/gtest/aarch64/asmtest.out.h and test/hotspot/gtest/aarch64/aarch64-asmtest.py in the latest commit. We should resolve that.

Thanks. I think I missed generating the asmtest.out.h file in the new commit. I'll update it.

Bhavana-Kilambi · 2025-04-25T08:28:25Z

Hi @shqking I have regenerated the test/hotspot/gtest/aarch64/asmtest.out.h to keep it consistent with the instructions in test/hotspot/gtest/aarch64/aarch64-asmtest.py. Can you please review now? Thanks!

theRealAph

Looks good to me. Please summarize the tests that you've run before committing.

Bhavana-Kilambi · 2025-04-25T09:16:39Z

Looks good to me. Please summarize the tests that you've run before committing.

Thank you for the approval.
All hotspot (hotspot_all), jdk (tiers 1-3) and langtools (tier1) pass on N1, V1 and V2 architectures.

shqking

Thanks for your update. Looks good to me.

theRealAph · 2025-04-25T10:07:16Z

On 4/25/25 10:17, bkilambi wrote: Bhavana-Kilambi left a comment (openjdk/jdk#23748) > Looks good to me. Please summarize the tests that you've run before committing. Thank you for the approval. All hotspot (hotspot_all), jdk (tiers 1-3) and langtools (tier1) pass on N1, V1 and V2 architectures.

But do any of these tests use Float16 ?

…

-- Andrew Haley (he/him) Java Platform Lead Engineer Red Hat UK Ltd. <https://www.redhat.com> https://keybase.io/andrewhaley EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671

Bhavana-Kilambi · 2025-04-25T10:31:16Z

On 4/25/25 10:17, bkilambi wrote: Bhavana-Kilambi left a comment (openjdk/jdk#23748) > Looks good to me. Please summarize the tests that you've run before committing. Thank you for the approval. All hotspot (hotspot_all), jdk (tiers 1-3) and langtools (tier1) pass on N1, V1 and V2 architectures.
But do any of these tests use Float16 ?
…
-- Andrew Haley (he/him) Java Platform Lead Engineer Red Hat UK Ltd. https://www.redhat.com https://keybase.io/andrewhaley EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671

Yes they do.
The following FP16 tests are included in them and have also been tested separately (these are the ones which have been added/modified in the commit which added initial scalar FP16 support in JDK mainline (4b463ee) -


test/hotspot/jtreg/compiler/c2/irTests/ConvF2HFIdealizationTests.java
test/hotspot/jtreg/compiler/c2/irTests/MulHFNodeIdealizationTests.java
test/hotspot/jtreg/compiler/c2/irTests/TestFloat16ScalarOperations.java
test/hotspot/jtreg/compiler/vectorization/TestFloat16VectorConvChain.java
test/jdk/jdk/incubator/vector/ScalarFloat16OperationsTest.java

Please let me know if you'd like me to test anything else. Thanks!

Bhavana-Kilambi · 2025-04-25T13:52:35Z

Hi @theRealAph , is it ok if I integrate this or would you like me to do any other testing on the patch?

theRealAph · 2025-04-25T15:12:25Z

Hi @theRealAph , is it ok if I integrate this or would you like me to do any other testing on the patch?

As long as you've got test coverage for everything here, go ahead.

Bhavana-Kilambi · 2025-04-25T15:37:40Z

Thanks a lot! Can I ask you to please sponsor this patch?
/integrate

openjdk · 2025-04-25T15:38:26Z

@Bhavana-Kilambi
Your change (at version 6a49a11) is now ready to be sponsored by a Committer.

shqking · 2025-04-28T08:07:58Z

/sponsor

openjdk · 2025-04-28T08:08:45Z

Going to push as commit 3140de4.
Since your change was applied there have been 39 commits pushed to the master branch:

a05ff55: 8355657: RISC-V: Improve PrintOptoAssembly output of vector-scalar instructions
7df1bfe: 8354811: clock_tics_per_sec code duplication between os_linux and os_posix
40e7986: 8355611: Get rid of SurfaceManagerFactory
... and 36 more: https://git.openjdk.org/jdk/compare/0edd018a48c202a6da4afe80e245799b47000885...master

Your commit was automatically rebased without conflicts.

openjdk · 2025-04-28T08:08:53Z

@shqking @Bhavana-Kilambi Pushed as commit 3140de4.

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

TobiHartmann · 2025-04-28T13:12:27Z

This caused a regression: JDK-8355708

8345125: Aarch64: Add aarch64 backend for Float16 operations

a608a03

This patch adds aarch64 backend for scalar FP16 operations namely - add, subtract, multiply, divide, fma, sqrt, min and max.

openjdk bot added the rfr Pull request is ready for review label Feb 24, 2025

openjdk bot added graal graal-dev@openjdk.org hotspot hotspot-dev@openjdk.org labels Feb 24, 2025

Bhavana-Kilambi changed the title ~~8345125: Aarch64: Add aarch64 backend for Float16 operations~~ 8345125: Aarch64: Add aarch64 backend for Float16 scalar operations Feb 24, 2025

theRealAph reviewed Feb 24, 2025

View reviewed changes

src/hotspot/cpu/aarch64/aarch64.ad Outdated Show resolved Hide resolved

theRealAph reviewed Feb 24, 2025

View reviewed changes

src/hotspot/cpu/aarch64/aarch64.ad Outdated Show resolved Hide resolved

theRealAph reviewed Feb 25, 2025

View reviewed changes

graalvmbot mentioned this pull request Feb 25, 2025

[JDK-8350672] Adapt JDK-8345125: Aarch64: Add aarch64 backend for Float16 scalar operations oracle/graal#10762

Merged

Address review comments

4d69974

shqking reviewed Feb 26, 2025

View reviewed changes

Bhavana-Kilambi marked this pull request as draft March 31, 2025 09:52

openjdk bot added merge-conflict Pull request has merge conflict with target branch and removed rfr Pull request is ready for review labels Mar 31, 2025

mhaessig mentioned this pull request Apr 10, 2025

8353730: TestSubNodeFloatDoubleNegation.java fails with native Float16 support #24565

Closed

3 tasks

Merge master

db9114a

openjdk bot removed the merge-conflict Pull request has merge conflict with target branch label Apr 24, 2025

Addressed review comments

eb667fc

Bhavana-Kilambi marked this pull request as ready for review April 24, 2025 15:53

openjdk bot added the rfr Pull request is ready for review label Apr 24, 2025

Update asmtest.out.h to keep it consistent with aarch64-asmtest.py

6a49a11

theRealAph approved these changes Apr 25, 2025

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Apr 25, 2025

shqking approved these changes Apr 25, 2025

View reviewed changes

openjdk bot added the sponsor Pull request is ready to be sponsored label Apr 25, 2025

openjdk bot added the integrated Pull request has been integrated label Apr 28, 2025

openjdk bot closed this Apr 28, 2025

openjdk bot removed ready Pull request is ready to be integrated rfr Pull request is ready for review sponsor Pull request is ready to be sponsored labels Apr 28, 2025

graalvmbot mentioned this pull request May 5, 2025

[GR-64700] Update labsjdk to 25+21-jvmci-b01 oracle/graal#11129

Merged

8345125: Aarch64: Add aarch64 backend for Float16 scalar operations #23748

8345125: Aarch64: Add aarch64 backend for Float16 scalar operations #23748

Uh oh!

Conversation

Bhavana-Kilambi commented Feb 24, 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 Feb 24, 2025

Uh oh!

openjdk bot commented Feb 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

openjdk bot commented Feb 24, 2025

Uh oh!

mlbridge bot commented Feb 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Webrevs

Uh oh!

Uh oh!

Uh oh!

theRealAph Feb 25, 2025

Choose a reason for hiding this comment

Uh oh!

Bhavana-Kilambi Feb 25, 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!

theRealAph commented Feb 25, 2025

Uh oh!

shqking Feb 26, 2025

Choose a reason for hiding this comment

Uh oh!

Bhavana-Kilambi Feb 26, 2025

Choose a reason for hiding this comment

Uh oh!

theRealAph Feb 26, 2025

Choose a reason for hiding this comment

Uh oh!

shqking Feb 26, 2025

Choose a reason for hiding this comment

Uh oh!

Bhavana-Kilambi Apr 24, 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!

bridgekeeper bot commented Mar 26, 2025

Uh oh!

Bhavana-Kilambi commented Mar 27, 2025

Uh oh!

Bhavana-Kilambi commented Mar 31, 2025

Uh oh!

openjdk bot commented Mar 31, 2025

Uh oh!

mhaessig commented Apr 16, 2025

Uh oh!

Bhavana-Kilambi commented Apr 24, 2025

Uh oh!

shqking commented Apr 25, 2025

Uh oh!

Bhavana-Kilambi commented Apr 25, 2025

Uh oh!

Bhavana-Kilambi commented Apr 25, 2025

Uh oh!

theRealAph left a comment

Choose a reason for hiding this comment

Uh oh!

Bhavana-Kilambi commented Apr 25, 2025

Uh oh!

shqking left a comment

Choose a reason for hiding this comment

Uh oh!

theRealAph commented Apr 25, 2025 via email

Uh oh!

Bhavana-Kilambi commented Apr 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Bhavana-Kilambi commented Feb 24, 2025 •

edited by openjdk bot

Loading

openjdk bot commented Feb 24, 2025 •

edited

Loading

mlbridge bot commented Feb 24, 2025 •

edited

Loading

Bhavana-Kilambi Feb 25, 2025 •

edited

Loading

Bhavana-Kilambi Apr 24, 2025 •

edited

Loading

Bhavana-Kilambi commented Apr 25, 2025 •

edited

Loading