Hi Alexandros,
Linaro benchmarking CI flagged this patch. After it clang seems to miscompile 602.gcc_s from SPEC CPU2017 for "-O3 -flto" on aarch64-linux-gnu. Also, it appears that 600.perlbench_s slows down by 9%.
Could you investigate, please? Let me know if you need any assistance in reproducing the problem.
Our Benchmarking CI is still in active development, and there are false positives, but this report seems to be legit [1].
Kind regards,
[1] https://ci.linaro.org/job/tcwg_bmk-code_speed-cpu2017speed--llvm-aarch64-ma…
--
Maxim Kuvyrkov
https://www.linaro.org
Hello,
[GNU-796] Stabilize GDB testsuite results in the CI
- Committed ABE change suggested by Maxim to mention failures that cause
additional testsuite runs.
- Tried to update Python MinGW build to 3.x, but it's a rabbit hole.
Proposing to disable Python support when cross-building GDB with MinGW
instead.
- First regression detected by the tcwg_gnu_native_check_gdb CI job
reported upstream¹. Started investigating it.
[GNU-767] Support changing SVE vector length in remote debugging
- Created a testcase to exercise the corner case mentioned by Pedro.
Found out that GDB is also susceptible to it. Fixed both GDB and
gdbserver. Posted patch series upstream².
--
Thiago
¹ https://inbox.sourceware.org/gdb-patches/871ql6dhf4.fsf@linaro.org/
² https://inbox.sourceware.org/gdb-patches/20230331034432.3037148-1-thiago.ba…
Progress (short week, holiday and recovering from covid...):
* UM-2 [QEMU upstream maintainership]
* 8.0 release related work rumbles on
* debugged and fixed a regression caused by my work on HSTR_EL2 traps
* QEMU-530 [QEMU ARM v9.4 Baseline CPU for TCG]
* implemented FEAT_PAN3 (and fixed a minor bug in
syndrome reporting that I noticed in the process)
-- PMM
Hello,
[GNU-796] Stabilize GDB testsuite results in the CI:
- Enabled gdb.gdb/unittest.exp in fast_check_gdb job. It should be
stable now that tcwg-jade-02's kernel has been upgraded.
- Enabled bisections for check_gdb jobs.
- Implemented Maxim's idea to handle GDB testsuite's tests that can FAIL
but are silent when they pass (i.e., the "gdb_test -nopass"
statements).
[GNU-767] Support changing SVE vector length in remote debugging
- Back in February I upstreamed a couple of preparation patches and
Pedro noticed an unintended change in behaviour that can affect a
corner case, so I am working on a fix for that now.
--
Thiago
Hello,
[GNU-796] Stabilize GDB testsuite results in the CI:
- Finished implementing ABE's support for rerunning failed tests using
Maxim's idea of leveraging validate_failures.py to determine which
tests need to be rerun. Submitted v3 and v4 versions, and committed
v4.
- Investigated why tcwg-abet-tested jobs have been failing in the CI.
Found out that Jenkins sets a bogus core.hooksPath in the git repo
config. Submitted and merged a gerrit request fixing the problem.
- Investigated why cross-build gdbserver needs GMP and MPFR (and is thus
failing to build) in tcwg-gnu-build jobs. Submitted Gerrit review
request to fix it.
- Started looking into why re-enabling GDB testsuite parallelism makes
it run fewer tests.
--
Thiago
Progress:
* UM-2 [QEMU upstream maintainership]
- went through and made estimates for a pile of JIRA tasks
relating to implementing features we're missing to get to ARMv9.4
- diagnosed why --enable-werror wasn't affecting warnings from
the kerneldoc docs generator, and sent a patch fixing it
- some bug triage, looking for issues that ought to be fixed for 8.0
- fixed a division-by-zero bug in the cadence UART model
- fixed a documentation markup mistake that made the docs look odd
- investigating/fixing some more CI failures
- minor bits of travel/conference related admin
-- PMM
Hello,
[GNU-796] Stabilize GDB testsuite results in the CI:
- Submitted and merged a couple of review requests removing obsolete CI
jobs related to release automation and release regression detection
- Finished version of scripts that have most of the logic for rerunning
testsuites in ABE. Submitted as a review request. Addressed review
comments and submitted v2.
- Maxim found out a way to simplify the logic of determining whether a
new testsuite run is warranted by using validate_failures.py, which
I'm now implementing.
--
Thiago
Progress:
* UM-2 [QEMU upstream maintainership]
- Softfreeze was this Tuesday; lots of wrangling of pull requests
- Code review; last arm pullreq before softfreeze
- various admin type bits and pieces
-- PMM
