8266962: Add arch supporting check for "Op_VectorLoadConst" before creating the node

[XiaohongGong]

When creating the vector shuffle, the "VectorLoadConstNode" will be created to get an initial index vector. Before creating it, the compiler should check whether the current platform supports this opcode in case the jvm crashes with "bad ad file". The compiler should finish the intrinsification and go back to the default java implementation if the backend doesn't support it.

Tested tier1 and jdk::tier3.

---

Progress

• Change must not contain extraneous whitespace
• Commit message must refer to an issue
• Change must be properly reviewed

Issue

• JDK-8266962: Add arch supporting check for "Op_VectorLoadConst" before creating the node

Reviewers

• Vladimir Ivanov (@iwanowww - Reviewer)
• Nils Eliasson (@neliasso - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.java.net/jdk pull/4023/head:pull/4023
$ git checkout pull/4023

Update a local copy of the PR:
$ git checkout pull/4023
$ git pull https://git.openjdk.java.net/jdk pull/4023/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 4023

View PR using the GUI difftool:
$ git pr show -t 4023

Using diff file

Download this PR as a diff file:
https://git.openjdk.java.net/jdk/pull/4023.diff

[bridgekeeper]

👋 Welcome back xgong! 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]

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

• hotspot-compiler

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.

[mlbridge]

Webrevs

• 00: Full (8e8a6e8c)

[iwanowww]

The fix makes perfect sense, but I'm curious why do we have VectorLoadConst in the first place.

It exposes JVM support for the iota vector constant materialization, but it's not clear to me what benefits it brings compared to feeding the intrinsic with the vector materialized on JDK side.

[openjdk]

@XiaohongGong 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:

8266962: Add arch supporting check for "Op_VectorLoadConst" before creating the node

Reviewed-by: vlivanov, neliasso

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 72 new commits pushed to the master branch:

• 894547d: 8266897: com/sun/net/httpserver/FilterTest.java fails intermittently with AssertionError
• da7c846: 8264408: test_oopStorage no longer needs to disable some tests on WIN32
• f6c2891: 8267229: Split runtime/Metaspace/elastic test configurations for better scalability
• b60975d: 8267237: ARM32: bad AD file in matcher.cpp after 8266810
• 905b41a: 8265711: C1: Intrinsify Class.getModifier method
• 554caf3: 8251392: Consolidate Metaspace Statistics
• 3e97b07: 8267116: Lanai: Incorrect AlphaComposite for VolatileImage graphics
• cd1c17c: 8266404: Fatal error report generated with -XX:+CrashOnOutOfMemoryError should not contain suggestion to submit a bug report
• 2effdd1: 8267112: JVMCI compiler modules should be kept upgradable
• da4dfde: 8264777: Overload optimized FileInputStream::readAllBytes
• ... and 62 more: https://git.openjdk.java.net/jdk/compare/06d760283344a1d0fd510aed306e0efb76b51617...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 (@iwanowww, @neliasso) 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).

[XiaohongGong]

The fix makes perfect sense, but I'm curious why do we have VectorLoadConst in the first place.

It exposes JVM support for the iota vector constant materialization, but it's not clear to me what benefits it brings compared to feeding the intrinsic with the vector materialized on JDK side.

Thanks for looking at this PR @iwanowww ! As far as I know the VectorLoadConst is used here to get the initial shuffle iota of the vector. I'm not so clear about what the iota vector constant materialization you mean. Could you please elaborate more about it? Thanks so much!

[XiaohongGong]

/integrate

[openjdk]

@XiaohongGong
Your change (at version 8e8a6e8) is now ready to be sponsored by a Committer.

[neliasso]

Looks good.

[iwanowww]

As far as I know the VectorLoadConst is used here to get the initial shuffle iota of the vector. I'm not so clear about what the iota vector constant materialization you mean.

VectorLoadConst is backed by a constant in StubRoutines. Instead, the constant can be materialized as a on-heap ByteVector instance, cached in a static final field, and passed into the intrinsic.

An alternative approach would be to replace VectorLoadConst with a LoadVector which performs raw vector access at StubRoutines::_vector_iota_indices address.

All in all, I don't see VectorLoadConst well-justified.

src/hotspot/cpu/aarch64/aarch64_neon.ad:

  3369 instruct loadcon8B(vecD dst, immI0 src)
  3374   match(Set dst (VectorLoadConst src));
  3378     __ lea(rscratch1, ExternalAddress(StubRoutines::aarch64::vector_iota_indices()));


src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp:

  7055     StubRoutines::aarch64::_vector_iota_indices    = generate_iota_indices("iota_indices");

   618   // Generate indices for iota vector.
   619   address generate_iota_indices(const char *stub_name) {
   620     __ align(CodeEntryAlignment);
   621     StubCodeMark mark(this, "StubRoutines", stub_name);
   622     address start = __ pc();
   623     __ emit_data64(0x0706050403020100, relocInfo::none);
   624     __ emit_data64(0x0F0E0D0C0B0A0908, relocInfo::none);
   625     return start;
   626   }

[XiaohongGong]

As far as I know the VectorLoadConst is used here to get the initial shuffle iota of the vector. I'm not so clear about what the iota vector constant materialization you mean.

VectorLoadConst is backed by a constant in StubRoutines. Instead, the constant can be materialized as a on-heap ByteVector instance, cached in a static final field, and passed into the intrinsic.

An alternative approach would be to replace VectorLoadConst with a LoadVector which performs raw vector access at StubRoutines::_vector_iota_indices address.

All in all, I don't see VectorLoadConst well-justified.

src/hotspot/cpu/aarch64/aarch64_neon.ad:

  3369 instruct loadcon8B(vecD dst, immI0 src)
  3374   match(Set dst (VectorLoadConst src));
  3378     __ lea(rscratch1, ExternalAddress(StubRoutines::aarch64::vector_iota_indices()));


src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp:

  7055     StubRoutines::aarch64::_vector_iota_indices    = generate_iota_indices("iota_indices");

   618   // Generate indices for iota vector.
   619   address generate_iota_indices(const char *stub_name) {
   620     __ align(CodeEntryAlignment);
   621     StubCodeMark mark(this, "StubRoutines", stub_name);
   622     address start = __ pc();
   623     __ emit_data64(0x0706050403020100, relocInfo::none);
   624     __ emit_data64(0x0F0E0D0C0B0A0908, relocInfo::none);
   625     return start;
   626   }

Thanks for your explanation! Yes, currently VectorLoadConst is implemented by loading the constant in StubRoutines. And the alternative way that passing a static final instance (like ByteMaxShuffle.IOTA) to the intrinisc makes sense to me. Maybe we can revisit it as a kind of optimization in future? Thanks!

[iwanowww]

Maybe we can revisit it as a kind of optimization in future?

Sure. Please, file an RFE.

[XiaohongGong]

Maybe we can revisit it as a kind of optimization in future?

Sure. Please, file an RFE.

Sure, please see: https://bugs.openjdk.java.net/browse/JDK-8267366. Thanks!

[nsjian]

/sponsor

[openjdk]

@nsjian @XiaohongGong Since your change was applied there have been 89 commits pushed to the master branch:

• 4954383: 8267364: Remove mask.incr which is introduced by JDK-8256973
• c2b50f9: 8266480: Implicit null check optimization does not update control of hoisted memory operation
• 3f883e8: 8267351: runtime/cds/SharedBaseAddress.java fails on x86_32 due to Unrecognized VM option 'UseCompressedOops'
• 7aa6568: 8256973: Intrinsic creation for VectorMask query (lastTrue,firstTrue,trueCount) APIs
• 65a8bf5: 8265126: [REDO] unified handling for VectorMask object re-materialization during de-optimization
• ff84577: 8267098: AArch64: C1 StubFrames end confusingly
• 0daec49: 8267246: -XX:MaxRAMPercentage=0 is unreasonable for jtreg tests on many-core machines
• 324defe: 8267212: test/jdk/java/util/Collections/FindSubList.java intermittent crash with "no reachable node should have no use"
• bdbe23b: 8265462: Handle multiple slots in the NSS Internal Module from SunPKCS11's Secmod
• 10236e7: 8263242: serviceability/sa/ClhsdbFindPC.java cannot find MaxJNILocalCapacity with ASLR
• ... and 79 more: https://git.openjdk.java.net/jdk/compare/06d760283344a1d0fd510aed306e0efb76b51617...master

Your commit was automatically rebased without conflicts.

Pushed as commit 2563a6a.

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