On 08.01.26 12:37, Sam James wrote:
> Frank Scheiner via Gcc <[email protected]> writes:
> 
>> Hi all,
>>
>> as Tomas' ran into an OOM situation for the testsuite, I took over and 
>> bootstrapped gcc-16-20251214 on my rx2800 i2 in ca. mid December.
>> Testsuite results are on [1]. I'll send a separate email for a possible
>> discussion of these to the gcc list.
>>
>> [1]: https://gcc.gnu.org/pipermail/gcc-testresults/2025-December/865397.html
>>
>> As shown in the testsuite results, the build time increased to some
>> degree compared to older runs. For bootstrap this took nearly an hour
>> longer (375 mins) compared to GCC 15.2.0 (319 mins; unfortunately
>> I don't have the timing data for GCC 15.1.0, which is directly from
>> trunk IINM).
>>
>> Together with the testsuite run, we are now close to 13 hours for the
>> whole process on a rx2800 i2 w/1 x Itanium 2 9320 (4 cores, 8 threads)
>> with everything done in `/dev/shm`.
>>
>> Only languages C and C++ were made. I used Linux v6.19-rc1 for
>> bootstrapping gcc-16-20251214 and v6.17-rc1 for GCC 15.2.0. I'm unsure
>> if the build time increase could be due to the used kernel, but I'll
>> also look if cross compilations on GHA and locally show a comparable
>> build time difference.
>>
>> Still, I wonder, do other arches show a similar increase in bootstrap
>> timings for gcc-16-[...] compared to GCC 15.x.x?
> 
> Bootstraps on trunk will take longer by default because we enable more
> assertions ("checking") for development, i.e. the default for releases
> is --enable-checking=release, while for development, it is
> --enable-checking=yes,extra.
> 
> But I note that your test-results link has --enable-checking=release
> which is curious, so that doesn't explain this.

There also were some reports about performance issues with Linux v6.19,
tough for pre-rc1 according to the date, see for example [2].

[2]: https://www.phoronix.com/review/linux-619-sched-regress

Not sure if this could have a play in this. I still need to check on
the cross-bootstrap runtimes on GitHub ([3]).

[3]: https://github.com/johnny-mnemonic/gcc-autobuilds/actions

> You could maybe use bootstrap-time from
> https://gcc.gnu.org/install/build.html but I admit I've never used it
> and I don't know if it'll be useful (or not) for debugging this.

That could be interesting! Maybe something for future builds. Thanks!

Cheers,
Frank

Reply via email to