8385648: PPC64: Improve receiver type profiling reliability by TheRealMDoerr · Pull Request #31325 · openjdk/jdk

TheRealMDoerr · 2026-05-29T19:38:15Z

This is the PPC64 implementation. x64 and aarch64 are already integrated.
Unfortunately, I had one register too few in C1, so I have reused the recv register only for C1. I've kept the better code for the interpreter. Maybe we can clean up the C1 version if we free up another register in the future.

I confirm that I make this contribution in accordance with the OpenJDK Interim AI Policy.

Progress

Change must not contain extraneous whitespace
Commit message must refer to an issue
Change must be properly reviewed (2 reviews required, with at least 1 Reviewer, 1 Author)

Issue

JDK-8385648: PPC64: Improve receiver type profiling reliability (Enhancement - P4)

Reviewers

Aleksey Shipilev (@shipilev - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/31325/head:pull/31325
$ git checkout pull/31325

Update a local copy of the PR:
$ git checkout pull/31325
$ git pull https://git.openjdk.org/jdk.git pull/31325/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 31325

View PR using the GUI difftool:
$ git pr show -t 31325

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/31325.diff

Using Webrev

Link to Webrev Comment

bridgekeeper · 2026-05-29T19:39:50Z

👋 Welcome back mdoerr! 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 · 2026-05-29T19:40:25Z

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

openjdk · 2026-05-29T19:41:19Z

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

hotspot

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.

openjdk · 2026-05-29T19:41:21Z

The total number of required reviews for this PR has been set to 2 based on the presence of this label: hotspot. This can be overridden with the /reviewers command.

mlbridge · 2026-05-29T19:44:59Z

Webrevs

01: Full - Incremental (4aa1597d)
00: Full (52e1fc40)

shipilev

It seems okay, but I suspect that killing recv is not really fine for this path:

 if (op->should_profile_receiver_type()) {
   assert(op->recv()->is_single_cpu(), "recv must be allocated");
   Register recv = op->recv()->as_register(); // <--- not a temp
   ...
    type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1); // kills recv
 }

Can you instead ask for another tmp?

TheRealMDoerr · 2026-05-30T10:21:08Z

It seems okay, but I suspect that killing recv is not really fine for this path:

 if (op->should_profile_receiver_type()) {
   assert(op->recv()->is_single_cpu(), "recv must be allocated");
   Register recv = op->recv()->as_register(); // <--- not a temp
   ...
    type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1); // kills recv
 }

Can you instead ask for another tmp?

Thank you for reviewing!

I had the same concern initially. However, recv has a temp and no output effect:

jdk/src/hotspot/share/c1/c1_LIR.cpp

Line 880 in d3073b5

    
           if (opProfileCall->_recv->is_valid())              do_temp(opProfileCall->_recv);

The register is allocated and the receiver value copied here:

jdk/src/hotspot/share/c1/c1_LIRGenerator.cpp

Line 3056 in d3073b5

recv = new_register(T_OBJECT);

Here's an example: The new sequence is ...0868 to ...08d4. recv = r4 is loaded before it and immediately reused for something else after it.

  0x0000668958ff0860:   lis     r4,2820
  0x0000668958ff0864:   addi    r4,r4,15360
  0x0000668958ff0868:   li      r0,2
  0x0000668958ff086c:   mtctr   r0
  0x0000668958ff0870:   li      r6,248
  0x0000668958ff0874:   ldx     r0,r6,r5
  0x0000668958ff0878:   cmpd    r0,r4
  0x0000668958ff087c:   beq-    0x0000668958ff08c8,bo=0b01100[no_hint]
  0x0000668958ff0880:   addi    r6,r6,16
  0x0000668958ff0884:   bdnz+   0x0000668958ff0874,bo=0b10000[no_hint]
  0x0000668958ff0888:   li      r0,2
  0x0000668958ff088c:   mtctr   r0
  0x0000668958ff0890:   li      r6,248
  0x0000668958ff0894:   ldx     r0,r6,r5
  0x0000668958ff0898:   cmpdi   r0,0
  0x0000668958ff089c:   beq-    0x0000668958ff08b0,bo=0b01100[no_hint]
  0x0000668958ff08a0:   addi    r6,r6,16
  0x0000668958ff08a4:   bdnz+   0x0000668958ff0894,bo=0b10000[no_hint]
  0x0000668958ff08a8:   li      r6,240
  0x0000668958ff08ac:   b       0x0000668958ff08cc
  0x0000668958ff08b0:   add     r6,r5,r6
  0x0000668958ff08b4:   ldarx   r0,0,r6
  0x0000668958ff08b8:   cmpdi   r0,0
  0x0000668958ff08bc:   bne-    0x0000668958ff08c4,bo=0b00110[not_taken]
  0x0000668958ff08c0:   stdcx.  r4,0,r6
  0x0000668958ff08c4:   b       0x0000668958ff0868
  0x0000668958ff08c8:   addi    r6,r6,8
  0x0000668958ff08cc:   ldx     r4,r6,r5
  0x0000668958ff08d0:   addi    r4,r4,1
  0x0000668958ff08d4:   stdx    r4,r6,r5
  0x0000668958ff08d8:   lis     r4,2820
  0x0000668958ff08dc:   addi    r4,r4,-9128

In general, I'm not very happy with C1's register usage for some LIR instructions, either. Especially that the temp effect is missing for some LIR nodes here:

jdk/src/hotspot/share/c1/c1_LIR.cpp

Line 807 in d3073b5

if (op->code() == lir_store_check && opTypeCheck->_object->is_valid()) {

That required a terrible hack for some architectures because obj may be same register as one of the temps.
In addition, reg2mem and emit_profile_type could be implemented better if we had one more temp register.

So, I think that may be worth a separate RFE.
Initial test results for this PR look good so far, but I'll run more tests.
What do you think?

TheRealMDoerr · 2026-06-01T09:52:40Z

I've run millions of tests over the weekend plus some tests with -XX:-C1OptimizeVirtualCallProfiling and the results are good.

shipilev

Looks okay, with nits.

shipilev · 2026-06-01T13:39:02Z

+  }
+
+  Label L_loop_search_receiver, L_loop_search_empty;
+  Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;


L_polymorphic is dead. It is actually smarter than the original version I did in x86 -- saves a branch, right? -- so we might want to touch up those, separately.

Exactly. Yeah, improving it for x86 would be nice, too.

https://bugs.openjdk.org/browse/JDK-8385702

shipilev · 2026-06-01T13:39:45Z

+
+  // Atomically swing receiver slot: null -> recv.
+  //
+  // The update uses CAS, which clobbers tmp. Therefore, rscratch2


I don't think this is about rscratch2 anymore?

Good catch! I've removed the whole comment. It's already described in the other comment that we kill the offset.

TheRealMDoerr · 2026-06-01T14:02:15Z

Thanks for a lot for reviewing! I appreciate it.

8385648: PPC64: Improve receiver type profiling reliability

52e1fc4

openjdk Bot added the hotspot hotspot-dev@openjdk.org label May 29, 2026

openjdk Bot added the rfr Pull request is ready for review label May 29, 2026

shipilev reviewed May 29, 2026

View reviewed changes

shipilev approved these changes Jun 1, 2026

View reviewed changes

Review comments.

4aa1597

shipilev approved these changes Jun 1, 2026

View reviewed changes

Conversation

TheRealMDoerr commented May 29, 2026 • 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 May 29, 2026

Uh oh!

openjdk Bot commented May 29, 2026

Uh oh!

openjdk Bot commented May 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

openjdk Bot commented May 29, 2026

Uh oh!

mlbridge Bot commented May 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Webrevs

Uh oh!

shipilev left a comment • 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!

TheRealMDoerr commented May 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

TheRealMDoerr commented Jun 1, 2026

Uh oh!

shipilev left a comment

Choose a reason for hiding this comment

Uh oh!

shipilev Jun 1, 2026

Choose a reason for hiding this comment

Uh oh!

TheRealMDoerr Jun 1, 2026

Choose a reason for hiding this comment

Uh oh!

shipilev Jun 1, 2026

Choose a reason for hiding this comment

Uh oh!

shipilev Jun 1, 2026

Choose a reason for hiding this comment

Uh oh!

TheRealMDoerr Jun 1, 2026

Choose a reason for hiding this comment

Uh oh!

TheRealMDoerr commented Jun 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

2 participants

TheRealMDoerr commented May 29, 2026 •

edited by openjdk Bot

Loading

openjdk Bot commented May 29, 2026 •

edited

Loading

mlbridge Bot commented May 29, 2026 •

edited

Loading

shipilev left a comment •

edited

Loading

TheRealMDoerr commented May 30, 2026 •

edited

Loading