On Fri, 24 Jan 2025 08:10:21 GMT, Matthias Baesken <mbaes...@openjdk.org> wrote:
>> I agree with Magnus the files should only be excluded where there is a >> problem and for PPC the problem can be fixed by using the same `-m` flag (or >> else fixing the build logic to check for PPC as Magnus has). >> >> As far as I can see PCH has been used with these files on Linux and gcc with >> no problem for other architectures. I'm still unclear exactly what changed >> for PPC. > >> Note that OPT_SPEED_SRC contains a rather large set of files. They have >> previously been compiling using PCH just fine with gcc. I don't think it is >> motivated to exclude them everywhere, just because gcc on ppc64 has a >> problem. > > It seems usually it is empty, because in 99% of cases people will not set > the opt-size jvm feature > > > 193 ifeq ($(call check-jvm-feature, opt-size), true) > 194 JVM_OPTIMIZATION := SIZE > 195 OPT_SPEED_SRC := \ > 196 allocation.cpp \ > > > I haven't checked what happens with those recent PCH changes, when opt-size > is configured. If it works I am fine with your approach too that does not > add those to JVM_PRECOMPILED_HEADER_EXCLUDE . I must admit I did not realize this list was dependent on the `opt-size` feature, so I cannot say if my patch really works with that enabled. Maybe the OPT_SPEED_SRC list is needed then, after all. Nevertheless, I think we can limit the change for ppc on gcc. When I added the -W flag for more detailed warnings, I was taught that it was the value of some -mpower-..somethingsomething flag that was problematic for PCHs. So it seems like it could be a very much PPC specific problem. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/23268#discussion_r1928318138