RFR: 8352316: More MergeStoreBench

[Emanuel Peter]

On Wed, 19 Mar 2025 06:02:17 GMT, Shaojin Wen <swen at openjdk.org> wrote:

>> Added performance tests related to String.getBytes/String.getChars/StringBuilder.append/System.arraycopy in constant scenarios to verify whether MergeStore works
>
> The performance numbers show that putNull_unsafePutInt and putNull_utf16_unsafePutLong perform more than 10 times better. It can be seen that MergeStore is very suitable for these scenarios.
> 
> # Scipt
> 
> git remote add wenshao git at github.com:wenshao/jdk.git
> git fetch wenshao
> git clone 23dba8c52454ae90eab4cb1b0a168c6e7249dd38
> make test TEST="micro:vm.compiler.MergeStoreBench.putNull"
> 
> 
> ## 2. aliyun_ecs_c8a_x64 (CPU AMD EPYC™ Genoa)
> 
> Benchmark                                     Mode  Cnt      Score      Error  Units
> MergeStoreBench.putNull_arraycopy             avgt    5   6715.041 ±   18.765  ns/op
> MergeStoreBench.putNull_getBytes              avgt    5   5880.725 ±   12.261  ns/op
> MergeStoreBench.putNull_getChars              avgt    5  11972.642 ±   24.990  ns/op
> MergeStoreBench.putNull_string_builder        avgt    5  15643.372 ± 4526.932  ns/op
> MergeStoreBench.putNull_unsafePutInt          avgt    5    280.570 ±    0.669  ns/op
> MergeStoreBench.putNull_utf16_arrayCopy       avgt    5  13053.191 ±   24.954  ns/op
> MergeStoreBench.putNull_utf16_string_builder  avgt    5  16349.747 ± 5029.799  ns/op
> MergeStoreBench.putNull_utf16_unsafePutLong   avgt    5    579.580 ±    0.710  ns/op
> 
> 
> 
> ## 3. aliyun_ecs_c8i_x64 (CPU Intel®Xeon®Emerald Rapids)
> 
> Benchmark                                     Mode  Cnt      Score      Error  Units
> MergeStoreBench.putNull_arraycopy             avgt    5   8029.622 ±   60.856  ns/op
> MergeStoreBench.putNull_getBytes              avgt    5   7444.635 ±   39.552  ns/op
> MergeStoreBench.putNull_getChars              avgt    5  16657.442 ±  147.301  ns/op
> MergeStoreBench.putNull_string_builder        avgt    5  23008.159 ± 6143.167  ns/op
> MergeStoreBench.putNull_unsafePutInt          avgt    5    235.302 ±    2.004  ns/op
> MergeStoreBench.putNull_utf16_arrayCopy       avgt    5  18330.317 ±  142.242  ns/op
> MergeStoreBench.putNull_utf16_string_builder  avgt    5  25843.593 ± 7089.392  ns/op
> MergeStoreBench.putNull_utf16_unsafePutLong   avgt    5   1860.076 ±   16.703  ns/op
> 
> 
> ## 4. aliyun_ecs_c8y_aarch64 (CPU Aliyun Yitian 710)
> 
> Benchmark                                     Mode  Cnt      Score      Error  Units
> MergeStoreBench.putNull_arraycopy             avgt    5   8114.176 ±   36.685  ns/op
> MergeStoreBench.putNull_getBytes              avgt    5   6171.538 ±    5.845  ns/op
> MergeStoreBench.putNull_getChars              avgt    5  10432.681 ±   26.401  ns/op
> MergeStoreBench.putNull_string_builder        avgt    5  21238.753 ± 1428.244  n...

@wenshao Do you have any insight from this benchmark? What was your motivation for it?

I also wonder if an IR test for some of the cases would be helpful. IR tests give us more info about what the compiler produced, and if there is a change in VM behaviour the IR test catches it in regular testing. Benchmarks are not run regularly, and regressions would therefore not be caught.

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

PR Comment: https://git.openjdk.org/jdk/pull/24108#issuecomment-2751928466