RFR: 8332528: Generate code in SwitchBootstraps.generateTypeSwitch that require fewer adaptations

Claes Redestad redestad at openjdk.org
Mon May 20 20:55:30 UTC 2024


On Mon, 20 May 2024 10:52:27 GMT, Claes Redestad <redestad at openjdk.org> wrote:

> We can fold the call to `Objects.checkIndex` into the code generated in generateTypeSwitchSkeleton instead of doing so by filtering the MH argument. This loads 9 less classes (of which 8 generated LFs and Species classes) on a minimal test, while being neutral on a throughput sanity test:
> 
> 
> Name                   Cnt  Base   Error   Test   Error  Unit  Change
> SwitchSanity.switchSum  15 8,162 ± 0,117  8,152 ± 0,131 ns/op   1,00x (p = 0,800 )
>   * = significant
>   ```
> 
> A few additional optimizations includes generating the switch method using the precise type (to avoid the need for an explicitCast adaptation), and moving some seldom used `findStatic` calls to a holder. All in all this means a reduction by 33-34M cycles to bootstrap a trivial switch expression on my M1.

To help evaluate and diagnose the startup overheads I added a `main` method to the provided `SwitchSanity` micro. This only runs a single invocation and is easy to run using your startup benchmarking script/tool of choice:


Name             Cnt          Base         Error           Test         Error         Unit  Change
Perfstartup-JMH   20        75,000 ±       4,455         62,000 ±       6,042        ms/op   1,21x (p = 0,000*)
  :.cycles           277393501,850 ± 4422757,249  253819501,350 ± 5482340,325       cycles   0,92x (p = 0,000*)
  :.instructions     701618995,500 ± 3774721,992  622929549,250 ± 3634060,803 instructions   0,89x (p = 0,000*)
  :.taskclock               82,000 ±       3,564         74,000 ±       4,365           ms   0,90x (p = 0,000*)
  * = significant

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

PR Comment: https://git.openjdk.org/jdk/pull/19307#issuecomment-2121179185


More information about the core-libs-dev mailing list