RFR: 8339290: Optimize ClassFile Utf8EntryImpl#writeTo [v16]

[Shaojin Wen]

On Tue, 3 Sep 2024 15:24:02 GMT, Shaojin Wen <swen at openjdk.org> wrote:

>> Use fast path for ascii characters 1 to 127 to improve the performance of writing Utf8Entry to BufferWriter.
>
> Shaojin Wen has updated the pull request incrementally with two additional commits since the last revision:
> 
>  - suggestion from @cl4es
>  - Revert "countGreaterThanZero -> CountNonNegatives"
>    
>    This reverts commit 431f55b46daf86fe9f2e24a2587eba8f65d847ff.

Tests show that in intermediate mode and C1, the performance of `utf8_3_bytes` and `emoji` charType has regressed

## 1. -Xint

### 1.1 Script

git remote add wenshao git at github.com:wenshao/jdk.git
git fetch wenshao

# baseline
git checkout 5d2e88fd9d808ec636b3dc7eb3e642ee864b1655
make test TEST="micro:java.lang.classfile.Utf8EntryWriteTo" MICRO="VM_OPTIONS=-Xint"

# current
git checkout c0e425cd5d711d225d98ed3cd966f4a96caea7ab
make test TEST="micro:java.lang.classfile.Utf8EntryWriteTo" MICRO="VM_OPTIONS=-Xint"


### 1.2 MacBook M1 Pro Performance Numbers


-# baseline
-Benchmark                   (charType)  Mode  Cnt        Score       Error  Units
-Utf8EntryWriteTo.writeTo         ascii  avgt    9  1512116.497 ?  5557.804  ns/op
-Utf8EntryWriteTo.writeTo  utf8_2_bytes  avgt    9  1583141.289 ?  4445.117  ns/op
-Utf8EntryWriteTo.writeTo  utf8_3_bytes  avgt    9  2199600.958 ? 12867.974  ns/op
-Utf8EntryWriteTo.writeTo         emoji  avgt    9  2273588.004 ?  6239.290  ns/op

+# current
+Benchmark                   (charType)  Mode  Cnt        Score       Error  Units
+Utf8EntryWriteTo.writeTo         ascii  avgt    9   179355.324 ?   570.346  ns/op
+Utf8EntryWriteTo.writeTo  utf8_2_bytes  avgt    9   222915.457 ?   521.956  ns/op
+Utf8EntryWriteTo.writeTo  utf8_3_bytes  avgt    9  3090322.657 ?  7896.473  ns/op
+Utf8EntryWriteTo.writeTo         emoji  avgt    9  3142324.984 ? 10690.461  ns/op


|   | pattern | baseline  | current | delta |
| --- | --- | --- | --- | --- |
| Utf8EntryWriteTo.writeTo | ascii | 1512116.497 | 179355.324 | 743.08% |
| Utf8EntryWriteTo.writeTo | utf8_2_bytes | 1583141.289 | 222915.457 | 610.20% |
| Utf8EntryWriteTo.writeTo | utf8_3_bytes | 2199600.958 | 3090322.657 | -28.82% |
| Utf8EntryWriteTo.writeTo | emoji | 2273588.004 | 3142324.984 | -27.65% |


## 2. TieredStopAtLevel=1

### 2.1 Script

git remote add wenshao git at github.com:wenshao/jdk.git
git fetch wenshao

# baseline
git checkout 5d2e88fd9d808ec636b3dc7eb3e642ee864b1655
make test TEST="micro:java.lang.classfile.Utf8EntryWriteTo" MICRO="VM_OPTIONS=-Xint -XX:TieredStopAtLevel=1"

# current
git checkout c0e425cd5d711d225d98ed3cd966f4a96caea7ab
make test TEST="micro:java.lang.classfile.Utf8EntryWriteTo" MICRO="VM_OPTIONS=-Xint -XX:TieredStopAtLevel=1"



### 2.2 MacBook M1 Pro Performance Numbers


-# baseline
-Benchmark                   (charType)  Mode  Cnt      Score     Error  Units
-Utf8EntryWriteTo.writeTo         ascii  avgt    9  23465.291 ? 136.422  ns/op
-Utf8EntryWriteTo.writeTo  utf8_2_bytes  avgt    9  24892.661 ?  90.412  ns/op
-Utf8EntryWriteTo.writeTo  utf8_3_bytes  avgt    9  24555.321 ?  96.907  ns/op
-Utf8EntryWriteTo.writeTo         emoji  avgt    9  25322.357 ? 145.033  ns/op

+# current
+Benchmark                   (charType)  Mode  Cnt      Score     Error  Units
+Utf8EntryWriteTo.writeTo         ascii  avgt    9   9309.663 ? 122.261  ns/op
+Utf8EntryWriteTo.writeTo  utf8_2_bytes  avgt    9  10037.304 ?  61.700  ns/op
+Utf8EntryWriteTo.writeTo  utf8_3_bytes  avgt    9  31145.324 ? 168.506  ns/op
+Utf8EntryWriteTo.writeTo         emoji  avgt    9  31804.202 ?  66.719  ns/op


|   | charType | baseline  | current | delta |
| --- | --- | --- | --- | --- |
| Utf8EntryWriteTo.writeTo | ascii | 23465.291 | 9309.663 | 152.05% |
| Utf8EntryWriteTo.writeTo | utf8_2_bytes | 24892.661 | 10037.304 | 148.00% |
| Utf8EntryWriteTo.writeTo | utf8_3_bytes | 24555.321 | 31145.324 | -21.16% |
| Utf8EntryWriteTo.writeTo | emoji | 25322.357 | 31804.202 | -20.38% |



## 3. Non-JVM Options

### 3.1 Script

git remote add wenshao git at github.com:wenshao/jdk.git
git fetch wenshao

# baseline
git checkout 5d2e88fd9d808ec636b3dc7eb3e642ee864b1655
make test TEST="micro:java.lang.classfile.Utf8EntryWriteTo"

# current
git checkout c0e425cd5d711d225d98ed3cd966f4a96caea7ab
make test TEST="micro:java.lang.classfile.Utf8EntryWriteTo"


### 3.2 MacBook M1 Pro Performance Numbers


-# baseline
-Benchmark                   (charType)  Mode  Cnt      Score     Error  Units
-Utf8EntryWriteTo.writeTo         ascii  avgt    9  15179.059 ? 152.360  ns/op
-Utf8EntryWriteTo.writeTo  utf8_2_bytes  avgt    9  16189.050 ?  70.459  ns/op
-Utf8EntryWriteTo.writeTo  utf8_3_bytes  avgt    9  15974.934 ?  61.737  ns/op
-Utf8EntryWriteTo.writeTo         emoji  avgt    9  16549.691 ? 119.606  ns/op

+# current
+Benchmark                   (charType)  Mode  Cnt      Score     Error  Units
+Utf8EntryWriteTo.writeTo         ascii  avgt    9   4243.038 ?  86.923  ns/op
+Utf8EntryWriteTo.writeTo  utf8_2_bytes  avgt    9   5050.839 ?  24.070  ns/op
+Utf8EntryWriteTo.writeTo  utf8_3_bytes  avgt    9  13142.966 ? 164.620  ns/op
+Utf8EntryWriteTo.writeTo         emoji  avgt    9  13483.680 ? 244.955  ns/op


|   | charType | baseline  | current | delta |
| --- | --- | --- | --- | --- |
| Utf8EntryWriteTo.writeTo | ascii | 15179.059 | 4243.038 | 257.74% |
| Utf8EntryWriteTo.writeTo | utf8_2_bytes | 16189.050 | 5050.839 | 220.52% |
| Utf8EntryWriteTo.writeTo | utf8_3_bytes | 15974.934 | 13142.966 | 21.55% |
| Utf8EntryWriteTo.writeTo | emoji | 16549.691 | 13483.680 | 22.74% |

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

PR Comment: https://git.openjdk.org/jdk/pull/20772#issuecomment-2326859712