Enable splitting for EqualsSimpleNode

[Akirathan]

Fixes #9166

Pull Request Description

Fixes the regression introduced by #9070 in org.enso.benchmarks.generated.Collections.list_meta_fold benchmark.

Important Notes

As can be seen on the graph in IGV:

For some reason, EqualsSimpleNode is POLYMORPHIC. That seems to be the most visible performance problem.

First, I tried to introduce ConditionProfile with:

diff --git a/engine/runtime/src/main/java/org/enso/interpreter/node/expression/builtin/meta/EqualsNode.java b/engine/runtime/src/main/java/org/enso/interpreter/node/expression/builtin/meta/EqualsNode.java
index b368fb7fe..57274b37e 100644
--- a/engine/runtime/src/main/java/org/enso/interpreter/node/expression/builtin/meta/EqualsNode.java
+++ b/engine/runtime/src/main/java/org/enso/interpreter/node/expression/builtin/meta/EqualsNode.java
@@ -9,6 +9,7 @@ import com.oracle.truffle.api.dsl.Specialization;
 import com.oracle.truffle.api.frame.VirtualFrame;
 import com.oracle.truffle.api.interop.ArityException;
 import com.oracle.truffle.api.nodes.Node;
+import com.oracle.truffle.api.profiles.ConditionProfile;
 import org.enso.interpreter.dsl.AcceptsError;
 import org.enso.interpreter.dsl.BuiltinMethod;
 import org.enso.interpreter.node.EnsoRootNode;
@@ -46,6 +47,7 @@ public final class EqualsNode extends Node {
   @Child private EqualsSimpleNode node;
   @Child private TypeOfNode types;
   @Child private WithConversionNode convert;
+  private final ConditionProfile equalsProfile = ConditionProfile.create();
 
   private static final EqualsNode UNCACHED =
       new EqualsNode(EqualsSimpleNodeGen.getUncached(), TypeOfNode.getUncached(), true);
@@ -85,7 +87,7 @@ public final class EqualsNode extends Node {
   public boolean execute(
       VirtualFrame frame, @AcceptsError Object self, @AcceptsError Object other) {
     var areEqual = node.execute(frame, self, other);
-    if (!areEqual) {
+    if (!equalsProfile.profile(areEqual)) {
       var selfType = types.execute(self);
       var otherType = types.execute(other);
       if (selfType != otherType) {

But that did not resolve the issue.

My second attempt was to enable splitting for EqualsSimpleNode with @com.oracle.truffle.api.dsl.ReportPolymorphism annotation, which seems to resolve the issue. The benchmark is back to its original score, and EqualsSimpleNode is no longer POLYMORPHIC.

Checklist

Please ensure that the following checklist has been satisfied before submitting the PR:

• Run benchmarks and ensure that the regression is resolved, and no other introduced.

[Akirathan]

Stdlib benchmarks scheduled in https://github.com/enso-org/enso/actions/runs/8144579573

GitHub

Benchmark Standard Libraries · enso-org/enso@594a5f8

Hybrid visual and textual functional programming. Contribute to enso-org/enso development by creating an account on GitHub.

[Akirathan]

Collection.list_meta_fold is back to its original performance:

A similar case can be seen for many other benchmarks with "equality".

There are no regressions. Let's merge this PR.

[JaroslavTulach]

Great to hear the benchmarks are back to original speed. I still need to read about ReportPolymorphism.