On Tue, 3 Sep 2024 17:46:17 GMT, Kevin Rushforth <k...@openjdk.org> wrote:
> Btw, I ran another CI build with the `NUM_COMPILE_THREADS=1` workaround for
> Mac removed, and the build times are good again. In fact, the x64 build runs
> even faster than before -- possibly because WebKit build defaults to a lower
> number than num_processors, and it happens to be 12 on that x64 build system.
>
> Here are the updated numbers:
>
> OS / Arch Current This PR PR + don't limit num threads on macOS
> macOS / aarch64 13m:41s 56m:36s 13m:43s
> macOS / x64 28m:30s 2h:24m:52s 23m:01s
> I should also add that the Linux and Windows build times are largely
> unaffected by this PR (which is good). They do run slightly faster with your
> patch than current, but the percentage gain is small.
Interesting.
The default NUM_COMPILE_THREADS comes from what Gradle gives us, using
`${Runtime.runtime.availableProcessors()}`
The default threads that webkit uses `numberOfCPUs()` which is on mac
implemented as `sysctl -n hw.ncpu`
On my macOS/aarch64, both implementations give "10", but this can indeed give
different values.
-------------
PR Comment: https://git.openjdk.org/jfx/pull/1554#issuecomment-2327173344
