RFR: 8347555: [REDO] C2: implement optimization for series of Add of unique value [v7]

[Kangcheng Xu]

On Tue, 11 Mar 2025 19:49:41 GMT, Kangcheng Xu <kxu at openjdk.org> wrote:

>> [JDK-8347555](https://bugs.openjdk.org/browse/JDK-8347555) is a redo of [JDK-8325495](https://bugs.openjdk.org/browse/JDK-8325495) was [first merged](https://git.openjdk.org/jdk/pull/20754) then backed out due to a regression. This patch redos the feature and fixes the bit shift overflow problem. For more information please refer to the previous PR.
>> 
>> When constanlizing multiplications (possibly in forms on `lshifts`), the multiplier is upgraded to long and then later narrowed to int if needed. However, when a `lshift` operand is exactly `32`, overflowing an int, using long has an unexpected result. (i.e., `(1 << 32) = 1` and `(int) (1L << 32) = 0`)
>> 
>> The following was implemented to address this issue.
>> 
>> if (UseNewCode2) {
>>     *multiplier = bt == T_INT
>>         ? (jlong) (1 << con->get_int()) // loss of precision is expected for int as it overflows
>>         : ((jlong) 1) << con->get_int();
>> } else {
>>     *multiplier = ((jlong) 1 << con->get_int());
>> }
>> 
>> 
>> Two new bitshift overflow tests were added.
>
> Kangcheng Xu has updated the pull request incrementally with one additional commit since the last revision:
> 
>   add micro benchmark

I did some micro benchmark. The result is expected: unoptimized baseline scales linearly as the # of terms, while the optimized remains constant. 

@eme64 Please see [`test/micro/org/openjdk/bench/vm/compiler/SerialAdditions.java`](https://github.com/openjdk/jdk/pull/23506/files#:~:text=test/micro/org/openjdk/bench/vm/compiler/SerialAdditions.java) for details. This is my first time using the JMH framework, please let me know if I made any mistake. Thanks!

---

![](https://github.com/user-attachments/assets/f24c4443-16cc-4efc-9e41-6888f75114a8)

**baseline**: 


$ CONF=linux-x86_64-server-release make test TEST=micro:org.openjdk.bench.vm.compiler.SerialAdditions



Benchmark                      Mode  Cnt  Score   Error  Units
SerialAdditions.addIntsMixed   avgt   12  8.231 ± 0.767  ns/op
SerialAdditions.addIntsTo02    avgt   12  0.516 ± 0.074  ns/op
SerialAdditions.addIntsTo04    avgt   12  0.496 ± 0.033  ns/op
SerialAdditions.addIntsTo05    avgt   12  0.492 ± 0.031  ns/op
SerialAdditions.addIntsTo06    avgt   12  0.571 ± 0.029  ns/op
SerialAdditions.addIntsTo08    avgt   12  0.711 ± 0.038  ns/op
SerialAdditions.addIntsTo16    avgt   12  1.245 ± 0.049  ns/op
SerialAdditions.addIntsTo23    avgt   12  1.793 ± 0.054  ns/op
SerialAdditions.addIntsTo32    avgt   12  2.635 ± 0.033  ns/op
SerialAdditions.addIntsTo42    avgt   12  3.726 ± 0.092  ns/op
SerialAdditions.addIntsTo64    avgt   12  6.468 ± 0.215  ns/op
SerialAdditions.addLongsMixed  avgt   12  6.462 ± 0.202  ns/op
SerialAdditions.addLongsTo02   avgt   12  0.467 ± 0.014  ns/op
SerialAdditions.addLongsTo04   avgt   12  0.466 ± 0.022  ns/op
SerialAdditions.addLongsTo05   avgt   12  0.495 ± 0.024  ns/op
SerialAdditions.addLongsTo06   avgt   12  0.553 ± 0.023  ns/op
SerialAdditions.addLongsTo08   avgt   12  0.688 ± 0.032  ns/op
SerialAdditions.addLongsTo16   avgt   12  1.201 ± 0.034  ns/op
SerialAdditions.addLongsTo23   avgt   12  1.736 ± 0.053  ns/op
SerialAdditions.addLongsTo32   avgt   12  2.679 ± 0.088  ns/op
SerialAdditions.addLongsTo42   avgt   12  3.742 ± 0.081  ns/op
SerialAdditions.addLongsTo64   avgt   12  6.359 ± 0.056  ns/op


**with this patch**: 


$ CONF=linux-x86_64-server-release make test TEST=micro:org.openjdk.bench.vm.compiler.SerialAdditions TEST_VM_OPTS="-XX:+UnlockDiagnosticVMOptions -XX:+UseNewCode"



Benchmark                      Mode  Cnt  Score   Error  Units
SerialAdditions.addIntsMixed   avgt   12  0.933 ± 0.045  ns/op
SerialAdditions.addIntsTo02    avgt   12  0.459 ± 0.015  ns/op
SerialAdditions.addIntsTo04    avgt   12  0.468 ± 0.018  ns/op
SerialAdditions.addIntsTo05    avgt   12  0.467 ± 0.022  ns/op
SerialAdditions.addIntsTo06    avgt   12  0.455 ± 0.020  ns/op
SerialAdditions.addIntsTo08    avgt   12  0.475 ± 0.027  ns/op
SerialAdditions.addIntsTo16    avgt   12  0.470 ± 0.018  ns/op
SerialAdditions.addIntsTo23    avgt   12  0.469 ± 0.028  ns/op
SerialAdditions.addIntsTo32    avgt   12  0.474 ± 0.017  ns/op
SerialAdditions.addIntsTo42    avgt   12  0.476 ± 0.012  ns/op
SerialAdditions.addIntsTo64    avgt   12  0.480 ± 0.017  ns/op
SerialAdditions.addLongsMixed  avgt   12  1.051 ± 0.040  ns/op
SerialAdditions.addLongsTo02   avgt   12  0.460 ± 0.027  ns/op
SerialAdditions.addLongsTo04   avgt   12  0.473 ± 0.028  ns/op
SerialAdditions.addLongsTo05   avgt   12  0.463 ± 0.024  ns/op
SerialAdditions.addLongsTo06   avgt   12  0.457 ± 0.022  ns/op
SerialAdditions.addLongsTo08   avgt   12  0.523 ± 0.074  ns/op
SerialAdditions.addLongsTo16   avgt   12  0.460 ± 0.013  ns/op
SerialAdditions.addLongsTo23   avgt   12  0.468 ± 0.018  ns/op
SerialAdditions.addLongsTo32   avgt   12  0.523 ± 0.078  ns/op
SerialAdditions.addLongsTo42   avgt   12  0.470 ± 0.019  ns/op
SerialAdditions.addLongsTo64   avgt   12  0.475 ± 0.032  ns/op


---

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

PR Comment: https://git.openjdk.org/jdk/pull/23506#issuecomment-2715539309