RFR: 8327858: Improve spliterator and forEach for single-element immutable collections [v2]

[Chen Liang]

On Thu, 21 Mar 2024 20:09:23 GMT, Viktor Klang <vklang at openjdk.org> wrote:

>> Chen Liang has updated the pull request with a new target base due to a merge or a rebase. The pull request now contains 13 commits:
>> 
>>  - Use the improved form in forEach
>>  - Merge branch 'master' of https://github.com/openjdk/jdk into feature/imm-coll-stream
>>  - Null checks should probably be in the beginning...
>>  - mark implicit null checks
>>  - Merge branch 'master' of https://github.com/openjdk/jdk into feature/imm-coll-stream
>>  - Copyright year, revert changes for non-few element collections
>>  - Merge branch 'master' of https://github.com/openjdk/jdk into feature/imm-coll-stream
>>  - Merge branch 'feature/imm-coll-stream' of https://github.com/liachmodded/jdk into feature/imm-coll-stream
>>  - Spliterator for 12, iterate/forEach benchmark
>>  - fix comments
>>  - ... and 3 more: https://git.openjdk.org/jdk/compare/d5b95a0e...69bd0e9c
>
> src/java.base/share/classes/java/util/ImmutableCollections.java line 924:
> 
>> 922:                 action.accept(REVERSE ? (E)e1 : e0); // implicit null check
>> 923:                 action.accept(REVERSE ? e0 : (E)e1);
>> 924:             }
> 
> Out of curiosity, how does the following fare performance-wise?
> 
> Suggestion:
> 
>                 action.accept((!REVERSE || e1 == EMPTY) ? e0 : (E)e1); // implicit null check
>                 if (e1 != EMPTY)
>                     action.accept(!REVERSE ? (E)e1 : e0);

Benchmark                        Mode  Cnt    Score   Error   Units
ImmutableColls.forEachOverList  thrpt   15  361.423 ± 8.751  ops/us
ImmutableColls.forEachOverSet   thrpt   15   79.158 ± 5.064  ops/us
ImmutableColls.getOrDefault     thrpt   15  244.012 ± 0.943  ops/us
ImmutableColls.iterateOverList  thrpt   15  152.598 ± 3.687  ops/us
ImmutableColls.iterateOverSet   thrpt   15   61.969 ± 4.453  ops/us

The 3 results are also available at https://gist.github.com/f0b4336e5b1cf9c5299ebdbcd82232bf, where baseline is the master this patch currently is based on (which has WhiteBoxResizeTest failures), patch-0 being the current code, and patch-1 being your proposal (uncommited patch below).

diff --git a/src/java.base/share/classes/java/util/ImmutableCollections.java b/src/java.base/share/classes/java/util/ImmutableCollections.java
index fc232a521fb..f38b093cf60 100644
--- a/src/java.base/share/classes/java/util/ImmutableCollections.java
+++ b/src/java.base/share/classes/java/util/ImmutableCollections.java
@@ -916,12 +916,9 @@ public <T> T[] toArray(T[] a) {
         @Override
         @SuppressWarnings("unchecked")
         public void forEach(Consumer<? super E> action) {
-            if (e1 == EMPTY) {
-                action.accept(e0); // implicit null check
-            } else {
-                action.accept(REVERSE ? (E)e1 : e0); // implicit null check
-                action.accept(REVERSE ? e0 : (E)e1);
-            }
+            action.accept((!REVERSE || e1 == EMPTY) ? e0 : (E) e1); // implicit null check
+            if (e1 != EMPTY)
+                action.accept(!REVERSE ? (E) e1 : e0);
         }
 
         @Override



My testing shows that the existing version I have is most likely faster than your proposed version.

Also note that the test failures are from WhiteBoxResizeTest that's fixed in latest master; I decide not to pull as not to invalidate the existing benchmark baselines.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/15834#discussion_r1534886983