RFR: 8320921: GHA: Parallelize hotspot_compiler test jobs

[Aleksey Shipilev]

On Tue, 28 Nov 2023 18:15:56 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

> In current GHA, `hotspot_compiler` testing takes a long time, and often takes the longest. On MacOS and Windows it routinely takes 60..80 minutes, while other test groups run in 30..40 minutes. This often drags the total wall clock time to run the GHA up. Fortunately, there are already several parts in `hotspot_compiler` test group, which we might just run separately:
> https://github.com/openjdk/jdk/blob/dc256fbc6490f8163adb286dbb7380c10e5e1e06/test/hotspot/jtreg/TEST.groups#L142-L146
> 
> There is a bit of awkwardness about `not_xcomp` subgroup, which does not have lots of tests. Instead of doing another short job, I think we can just meld it into the quickest "part 2".
> 
> 
> # Before
> linux-x64 / test (hs/tier1 compiler) 38m 27s 	
> windows-x64 / test (hs/tier1 compiler): 1h 17m 30s 	
> macos-x64 / test (hs/tier1 compiler): 1h 2m 53s 
> 
> # After
> linux-x64 / test (hs/tier1 compiler part 1): 11m 38s 	
> linux-x64 / test (hs/tier1 compiler part 2): 9m 34s 	
> linux-x64 / test (hs/tier1 compiler part 3): 21m 40s 	
> 
> windows-x64 / test (hs/tier1 compiler part 1): 24m 19s 	
> windows-x64 / test (hs/tier1 compiler part 2): 18m 54s 	
> windows-x64 / test (hs/tier1 compiler part 3): 43m 32s 	
> 
> macos-x64 / test (hs/tier1 compiler part 1): 26m 44s 	
> macos-x64 / test (hs/tier1 compiler part 2): 20m 40s	
> macos-x64 / test (hs/tier1 compiler part 3): 42m 10s
> 
> 
> Additional testing:
>  - [x] GHA (current run is in progress, but I had clean runs before)

Thanks!

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

PR Comment: https://git.openjdk.org/jdk/pull/16860#issuecomment-1833470270