8353290: C2: Refactor PhaseIdealLoop::is_counted_loop() #24458

tabjy · 2025-04-04T20:33:47Z

This PR refactors PhaseIdealLoop::is_counted_loop() into (mostly) CountedLoopConverter::is_counted_loop() and CountedLoopConverter::convert() to decouple the detection and conversion code. This enables us to try different loop configurations easily and finally convert once a counted loop is found.

A nested PhaseIdealLoop::CountedLoopConverter class is created to handle the context, but I'm not if this is the best name or place for it. Please let me know what you think.

Blocks JDK-8336759.

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-8353290: C2: Refactor PhaseIdealLoop::is_counted_loop() (Enhancement - P4)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/24458/head:pull/24458
$ git checkout pull/24458

Update a local copy of the PR:
$ git checkout pull/24458
$ git pull https://git.openjdk.org/jdk.git pull/24458/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 24458

View PR using the GUI difftool:
$ git pr show -t 24458

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/24458.diff

Using Webrev

Link to Webrev Comment

bridgekeeper · 2025-04-04T20:34:18Z

👋 Welcome back kxu! 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 · 2025-04-04T20:34:59Z

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

openjdk · 2025-04-04T20:35:29Z

@tabjy 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.

tabjy · 2025-04-07T15:25:41Z

/reviewers 2

C2 changes

openjdk · 2025-04-07T15:25:57Z

@tabjy
The total number of required reviews for this PR (including the jcheck configuration and the last /reviewers command) is now set to 2 (with at least 1 Reviewer, 1 Author).

This reverts commit fd60717.

mlbridge · 2025-04-09T15:52:55Z

Webrevs

01: Full - Incremental (b72e7714)
00: Full (3db745d2)

chhagedorn

That's a good idea to refactor this, thanks for tackling it! Some initial general comments, I will have a closer look at your patch later again.

chhagedorn · 2025-04-10T07:03:20Z

src/hotspot/share/opto/loopnode.cpp

@@ -310,7 +310,7 @@ IdealLoopTree* PhaseIdealLoop::insert_outer_loop(IdealLoopTree* loop, LoopNode*
 // clone the outer loop as well as the inner, unrolling needs to only
 // clone the inner loop etc. No optimizations need to change the outer
 // strip mined loop as it is only a skeleton.
-IdealLoopTree* PhaseIdealLoop::create_outer_strip_mined_loop(BoolNode *test, Node *cmp, Node *init_control,
+IdealLoopTree* PhaseIdealLoop::create_outer_strip_mined_loop(Node *init_control,


Generally, for the touched code, you can also fix the * placement to be at the type.

Suggested change

IdealLoopTree* PhaseIdealLoop::create_outer_strip_mined_loop(Node *init_control,

IdealLoopTree* PhaseIdealLoop::create_outer_strip_mined_loop(Node* init_control,

Good point. I tried to fix them as I go but missed this one.

With the x -> head change, there are some more opportunities :-)

chhagedorn · 2025-04-10T07:04:47Z

src/hotspot/share/opto/loopnode.cpp

@@ -364,7 +364,7 @@ void PhaseIdealLoop::insert_loop_limit_check_predicate(ParsePredicateSuccessProj
 #endif
 }

-Node* PhaseIdealLoop::loop_exit_control(Node* x, IdealLoopTree* loop) {
+Node* PhaseIdealLoop::loop_exit_control(const Node* x, const IdealLoopTree* loop) const {


There are quite some places where we use x to denote a loop_head. Maybe you can also fix that with this patch.

I would prefer a simple head for function named .*loop.*. Updated all relevant functions I touched.

chhagedorn · 2025-04-10T07:10:38Z

src/hotspot/share/opto/loopnode.cpp

-  Node* phi_incr = nullptr;
-  incr = loop_iv_incr(incr, x, loop, phi_incr);
-  assert(incr != nullptr && incr->Opcode() == Op_Add(bt), "no incr");
+  LoopExitTest exit_test = loop_exit_test(back_control, loop);


Just an idea: Would it make sense to have a single LoopStructure (or another name) class that contains all the information stored now with separate structs LoopExitTest, LoopIvIncr and LoopIvStride? Then we could offer query methods for validation like is_valid(), is_stride_valid() etc., or check for existence like has_stride(), or access specific nodes like stride(), incr() etc.

Let me see what I can do! :)

chhagedorn · 2025-04-10T07:13:35Z

src/hotspot/share/opto/loopnode.hpp

                                               IdealLoopTree* loop, float cl_prob, float le_fcnt,
                                               Node*& entry_control, Node*& iffalse);

+  class CountedLoopConverter {


Probably quite a subjective matter, but what about just naming it CountedLoop? Then you have counted_loop.convert().

I disagree: I don't want to confuse it with CountedLoopNode or making it looks like extending from IdealLoopTree. Besides it wouldn't not necessarily guarantee a counted loop until confirmed by is_counted_loop()

What I had in mind was something like that (I'm also fine with CountedLoopConverter:

CountedLoop counted_loop(...); counted_loop.convert(); return counted_loop.is_valid();

But maybe you can explain in more detail what the follow-up work will be and how you use this class again later?

About the placement of this class. You could probably also move it out of the already huge PhaseIdealLoop class to make it a non-inner class, if there is not some strong coupling that you absolutely need.

chhagedorn · 2025-04-10T07:14:20Z

src/hotspot/share/opto/loopnode.hpp

                                               IdealLoopTree* loop, float cl_prob, float le_fcnt,
                                               Node*& entry_control, Node*& iffalse);

+  class CountedLoopConverter {
+    PhaseIdealLoop* _phase;


I suggest to add const whenever you can. For example, here you will probably not change the _phase pointer anymore:

Suggested change

PhaseIdealLoop* _phase;

PhaseIdealLoop* const _phase;

That was my initial thought, too; however, PhaseIdealLoop::insert_loop_limit_check_predicate(), ::lazy_replace(), ::set_subtree_ctrl() and many other mutations make this impossible.

You might be confusing it with having:

const PhaseIdealLoop* _phase

where we cannot call any non-const methods. That's indeed not possible. But what I mean was to make the pointer const:

PhaseIdealLoop* const _phase;

such that you cannot do _phase = xyz later. You would probably not do that anyway but it's an easy addition and safety for all fields not being reassigned again. It also helps to see which fields are going to be updated as part of the mutable state and which fields are not.

chhagedorn · 2025-04-10T07:15:43Z

src/hotspot/share/opto/loopnode.hpp

+    Node* _phi_incr;
+    Node* _stride;
+    bool _includes_limit;
+    BoolTest::mask _mask;
+    Node* _increment;


Here you have (_phi_)incr and increment. I suggest to make the naming consistent. I personally prefer full names whenever we can if it's not a universal abbreviation like cmp or iv. But what we count as well-known abbreviation is debatable and also depends on personal taste.

Updated s/_phi_incr/_phi_increment

chhagedorn · 2025-04-10T07:18:45Z

src/hotspot/share/opto/loopnode.hpp

+    float _cl_prob;
+    Node* _sfpt;
+    jlong _final_correction;
+    Node* _trunc1;


trunc1 is a little hard to understand. Can we find a better name?

Droped Node* _trunc1 in favour of TypeInteger* _increment_truncation_type. Added a note in should_stress_long_counted_loop().

chhagedorn · 2025-04-11T08:18:23Z

src/hotspot/share/opto/loopnode.cpp

+  return false;
+}
+
+bool PhaseIdealLoop::CountedLoopConverter::is_counted_loop() {


is_counted_loop() is still huge. When trying to refactor it anyway, it might be a good idea to further split the code up into separate methods for all the different checks. That would increase the readability since we have a lot of different checks.

chhagedorn · 2025-04-11T08:20:22Z

src/hotspot/share/opto/loopnode.hpp

                                               IdealLoopTree* loop, float cl_prob, float le_fcnt,
                                               Node*& entry_control, Node*& iffalse);

+  class CountedLoopConverter {


About the placement of this class. You could probably also move it out of the already huge PhaseIdealLoop class to make it a non-inner class, if there is not some strong coupling that you absolutely need.

tabjy added 8 commits March 12, 2025 13:58

add const where possible

443cef0

update a caller

9b7b7aa

wip

e8016ce

wip: create CountedLoopConverter

e730665

fix TestMultiversionRemoveUselessSlowLoop

f3305a4

remove debug lines

f349693

extract stress_long_counted_loop()

c221986

clean up code

8a5aba4

openjdk bot added the hotspot-compiler [email protected] label Apr 4, 2025

tabjy added 7 commits April 7, 2025 11:36

rearrange code for a more sensible diff

cef7b08

improve formatting, naming, comments

c4e2dff

Merge branch 'master' into counted-loop-refactor

80df6d4

improve formatting, naming, comments

fd60717

Revert "improve formatting, naming, comments"

fff8aee

This reverts commit fd60717.

remove TODOs

5c1eb79

line break

3db745d

tabjy marked this pull request as ready for review April 9, 2025 15:49

openjdk bot added the rfr Pull request is ready for review label Apr 9, 2025

chhagedorn reviewed Apr 10, 2025

View reviewed changes

reviewer suggested changes

b72e771

chhagedorn reviewed Apr 11, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

8353290: C2: Refactor PhaseIdealLoop::is_counted_loop() #24458

8353290: C2: Refactor PhaseIdealLoop::is_counted_loop() #24458

tabjy commented Apr 4, 2025 •

edited by openjdk bot

Loading

bridgekeeper bot commented Apr 4, 2025

openjdk bot commented Apr 4, 2025

openjdk bot commented Apr 4, 2025

tabjy commented Apr 7, 2025

openjdk bot commented Apr 7, 2025

mlbridge bot commented Apr 9, 2025 •

edited

Loading

chhagedorn left a comment

chhagedorn Apr 10, 2025

tabjy Apr 10, 2025

chhagedorn Apr 11, 2025

chhagedorn Apr 10, 2025 •

edited

Loading

tabjy Apr 10, 2025

chhagedorn Apr 10, 2025

tabjy Apr 10, 2025

chhagedorn Apr 10, 2025

tabjy Apr 10, 2025

chhagedorn Apr 11, 2025

chhagedorn Apr 11, 2025

chhagedorn Apr 10, 2025

tabjy Apr 10, 2025 •

edited

Loading

chhagedorn Apr 11, 2025

chhagedorn Apr 10, 2025 •

edited

Loading

tabjy Apr 10, 2025

chhagedorn Apr 10, 2025

tabjy Apr 10, 2025

chhagedorn Apr 11, 2025

chhagedorn Apr 11, 2025

	IdealLoopTree* PhaseIdealLoop::create_outer_strip_mined_loop(Node *init_control,
	IdealLoopTree* PhaseIdealLoop::create_outer_strip_mined_loop(Node* init_control,

8353290: C2: Refactor PhaseIdealLoop::is_counted_loop() #24458

Are you sure you want to change the base?

8353290: C2: Refactor PhaseIdealLoop::is_counted_loop() #24458

Conversation

tabjy commented Apr 4, 2025 • edited by openjdk bot Loading

Progress

Issue

Reviewing

bridgekeeper bot commented Apr 4, 2025

openjdk bot commented Apr 4, 2025

openjdk bot commented Apr 4, 2025

tabjy commented Apr 7, 2025

openjdk bot commented Apr 7, 2025

mlbridge bot commented Apr 9, 2025 • edited Loading

Webrevs

chhagedorn left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

chhagedorn Apr 10, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tabjy Apr 10, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

chhagedorn Apr 10, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tabjy commented Apr 4, 2025 •

edited by openjdk bot

Loading

mlbridge bot commented Apr 9, 2025 •

edited

Loading

chhagedorn Apr 10, 2025 •

edited

Loading

tabjy Apr 10, 2025 •

edited

Loading

chhagedorn Apr 10, 2025 •

edited

Loading