On Mon, 25 Aug 2025 17:56:18 GMT, Erik Joelsson <er...@openjdk.org> wrote:
>>> Also, as a side note, I think we need to look more into -ftime-trace. We >>> most likely have some code that is very heavy to compile that can be >>> simplified for the compiler without sacrificing runtime performance. If >>> would be interesting to compile all JDK native code with that flag... >> >> I created a ticket with some details to follow up: >> https://bugs.openjdk.org/browse/JDK-8366111 >> >>> Apart from that, I still share Kim's worry that the platform with the most >>> potential impact is Windows, and we have no numbers at all there. Can you >>> arrange to test so we at least to not have any regressions there? And also >>> check how the performance changes if you also manually include the same set >>> of inline files as is currently in the PCH. >> >> Might take a while as I don't really work on Windows, I'll try to set up >> some EC2 with the tooling in the coming days. > >> Apart from that, I still share Kim's worry that the platform with the most >> potential impact is Windows, and we have no numbers at all there. > > I did run a quick comparison on Windows with an early version of this patch > and it was basically a wash compared to before (slightly slower but could > just be variance). I just reran with the latest patch and saw a slight > improvement. Could still just be variance, but here are the numbers for > running `make hotspot` immediately after `make clean-hotspot` to really > isolate just the hotspot build. I warmed up caches by first building hotspot > from scratch before cleaning. > > Baseline (@57434c73eac9bd6557b09d4a057e3a2a18f382b4): 4m38 > This branch (@2f97d57b55fc838b30cf0d730c6119bf97749c6b): 4m30, 4m33 > > I only did one baseline run and then 2 on the branch. It's enough to show no > regression. > I added them back in > [4a2ff56](https://github.com/openjdk/jdk/commit/4a2ff5649ba9dc6a024aeb7577d630a3fefcbe4f), > there's only two of them since the others are now always included. @erikj79 > any chance you could try once again and see if > [4a2ff56](https://github.com/openjdk/jdk/commit/4a2ff5649ba9dc6a024aeb7577d630a3fefcbe4f) > has any effect on build time? I ran two more times: This branch (@4a2ff5649ba9dc6a024aeb7577d630a3fefcbe4f): 4m37, 4m29 So basically unchanged. ------------- PR Comment: https://git.openjdk.org/jdk/pull/26681#issuecomment-3221794432
