On Fri, 14 Jun 2024 01:17:29 GMT, Shaojin Wen <d...@openjdk.org> wrote:
>> After PR https://github.com/openjdk/jdk/pull/16245, C2 optimizes stores into >> primitive arrays by combining values into larger stores. >> >> This PR rewrites the code of appendNull and append(boolean) methods so that >> these two methods can be optimized by C2. > > Shaojin Wen has updated the pull request incrementally with one additional > commit since the last revision: > > optimization for x64 The performance regression issue under x64 has been resolved. The following code will affect the running performance under x64: * x64 slow version int count = this.count; ensureCapacityInternal(count + 4); * x64 faster version, WebRev 08: [Full](https://webrevs.openjdk.org/?repo=jdk&pr=19626&range=08) - [Incremental](https://webrevs.openjdk.org/?repo=jdk&pr=19626&range=07-08) ([b5ad8e70](https://git.openjdk.org/jdk/pull/19626/files/b5ad8e70928c547d134d0e4a532441cad9a7e4a2)) ensureCapacityInternal(count + 4); int count = this.count; I don't know why, but this solved the problem. The performance numbers for various platforms are as follows ## aliyun ecs.c8a * CPU AMD EPYCTM Genoa * Platform x64 -Benchmark Mode Cnt Score Error Units #master (a6fc2f8) -StringBuilders.appendWithBool8Latin1 avgt 15 7.270 ± 0.013 ns/op -StringBuilders.appendWithBool8Utf16 avgt 15 10.860 ± 0.044 ns/op -StringBuilders.appendWithNull8Latin1 avgt 15 5.490 ± 0.007 ns/op -StringBuilders.appendWithNull8Utf16 avgt 15 26.203 ± 0.294 ns/op +Benchmark Mode Cnt Score Error Units # Web 08 (b5ad8e70) +StringBuilders.appendWithBool8Latin1 avgt 15 6.035 ± 0.062 ns/op +20% +StringBuilders.appendWithBool8Utf16 avgt 15 10.072 ± 0.032 ns/op +7.82% +StringBuilders.appendWithNull8Latin1 avgt 15 5.491 ± 0.021 ns/op -0.01$ +StringBuilders.appendWithNull8Utf16 avgt 15 7.701 ± 0.036 ns/op +240.25% ## aliyun ecs.c8i * CPU Intel® Xeon® Emerald * Platform x64 -Benchmark Mode Cnt Score Error Units #master (a6fc2f8) -StringBuilders.appendWithBool8Latin1 avgt 15 7.466 ± 0.012 ns/op -StringBuilders.appendWithBool8Utf16 avgt 15 15.220 ± 0.005 ns/op -StringBuilders.appendWithNull8Latin1 avgt 15 22.084 ± 0.331 ns/op -StringBuilders.appendWithNull8Utf16 avgt 15 25.717 ± 0.106 ns/op +Benchmark Mode Cnt Score Error Units # Web 08 (b5ad8e70) +StringBuilders.appendWithBool8Latin1 avgt 15 6.706 ± 0.011 ns/op +11.33% +StringBuilders.appendWithBool8Utf16 avgt 15 11.610 ± 0.140 ns/op +31.09% +StringBuilders.appendWithNull8Latin1 avgt 15 19.345 ± 0.050 ns/op +14.15% +StringBuilders.appendWithNull8Utf16 avgt 15 22.765 ± 0.045 ns/op +12.96% ## MacBook M1 Max * CPU Apple M1 Max * Platform aarch64 -Benchmark Mode Cnt Score Error Units #master (a6fc2f8) -StringBuilders.appendWithBool8Latin1 avgt 15 7.647 ? 0.014 ns/op -StringBuilders.appendWithBool8Utf16 avgt 15 9.620 ? 0.022 ns/op -StringBuilders.appendWithNull8Latin1 avgt 15 7.208 ? 0.127 ns/op -StringBuilders.appendWithNull8Utf16 avgt 15 9.455 ? 0.167 ns/op +Benchmark Mode Cnt Score Error Units # Web 08 (b5ad8e70) +StringBuilders.appendWithBool8Latin1 avgt 15 5.932 ? 0.026 ns/op +28.91% +StringBuilders.appendWithBool8Utf16 avgt 15 7.432 ? 0.015 ns/op +29.44$ +StringBuilders.appendWithNull8Latin1 avgt 15 5.461 ? 0.015 ns/op +31.99% +StringBuilders.appendWithNull8Utf16 avgt 15 6.966 ? 0.015 ns/op +35.73% ## aliyun ecs.c8y * CPU Yitian710 aarch64 * Platform aarch64 -Benchmark Mode Cnt Score Error Units -StringBuilders.appendWithBool8Latin1 avgt 15 12.127 ± 1.241 ns/op -StringBuilders.appendWithBool8Utf16 avgt 15 20.954 ± 1.668 ns/op -StringBuilders.appendWithNull8Latin1 avgt 15 10.838 ± 0.244 ns/op -StringBuilders.appendWithNull8Utf16 avgt 15 14.391 ± 0.063 ns/op +Benchmark Mode Cnt Score Error Units +StringBuilders.appendWithBool8Latin1 avgt 15 9.299 ± 0.079 ns/op +30.41% +StringBuilders.appendWithBool8Utf16 avgt 15 11.688 ± 0.004 ns/op +79.27% +StringBuilders.appendWithNull8Latin1 avgt 15 8.516 ± 0.258 ns/op +27.26% +StringBuilders.appendWithNull8Utf16 avgt 15 11.020 ± 0.004 ns/op +30.58% ## aws.c5 * Platform aarch64 -Benchmark Mode Cnt Score Error Units -StringBuilders.appendWithBool8Latin1 avgt 15 10.278 ± 0.071 ns/op -StringBuilders.appendWithBool8Utf16 avgt 15 24.355 ± 0.418 ns/op -StringBuilders.appendWithNull8Latin1 avgt 15 33.104 ± 0.589 ns/op -StringBuilders.appendWithNull8Utf16 avgt 15 35.762 ± 7.580 ns/op +Benchmark Mode Cnt Score Error Units +StringBuilders.appendWithBool8Latin1 avgt 15 9.812 ± 0.075 ns/op +StringBuilders.appendWithBool8Utf16 avgt 15 17.812 ± 0.006 ns/op +StringBuilders.appendWithNull8Latin1 avgt 15 34.773 ± 2.897 ns/op +StringBuilders.appendWithNull8Utf16 avgt 15 33.187 ± 1.604 ns/op ------------- PR Comment: https://git.openjdk.org/jdk/pull/19626#issuecomment-2167566396
