Age | Commit message (Collapse) | Author |
|
This reverts commit 5de42d593e2441b0e4cecba83319c593ab0e7491.
|
|
This reverts commit bd361bdd07f33854ca608d3d36b270a8048108e5.
|
|
Change-Id: If2cc8ec4baba5506a6f71169ffa3d3de5a0b5ee2
|
|
Change-Id: I4cece8bec814f2154d04540b8872488f2aeebc05
|
|
Change-Id: If13a975bb9d814e14ae29d6ed792c03617d5e3ce
|
|
Change-Id: I71acf4fad8055f8f5fea9ab58a4585fd8bea7428
|
|
Before the cmake min version bump, the libxml2 detection in clang/CMakeLists.txt
probably failed, and so it didn't try to use libxml2 for c-index-test at all.
We should just build with -DCLANG_ENABLE_LIBXML2=NO to keep the status quo.
Ref: https://github.com/llvm/llvm-project/issues/64263
Change-Id: Ia538181c6a03698c1c6b4586850f46cbb776ac1f
|
|
This patch adds some tweaks to vanilla LLVM Windows release script for
making WoA release.
Following changes have been made:
1) Enable running tests with ninja check-all
2) Only build clang and lld for first stage build
3) Include flang and mlir to projects to be built
4) Disable running LLDB watchpoint tests
5) Run LLDB with LLDB server enabled.
6) Reduce no of parallel running lit tests to 4. (For surface X)
Change-Id: I2b141bf3ea562edfbabf02c6d7be8b5c642bfd32
|
|
This patch adds upstream LLVM release script to make WoA LLVM release.
We are copying the latest version of the script as there are still
some changes we need to make before making LLVM WoA release. This patch
only adds the current tot version of the release scripts. Any changes
will be made separtely.
Change-Id: I74626cb578f886ea1fa1d6327323aeeb580a3230
|
|
Add -j2 to LLVM_LIT_ARGS to avoid tests running out of memory and
being killed.
Add -DLIBCXXABI_USE_LLVM_UNWINDER=ON to configure-flags to fix
several libc++/libc++abi tests that use exceptions. Without this,
compiler-rt would try to use libunwind while libc++abi would try
to use libgcc_s for the unwind related functions. This caused
crashes on ARMv7.
|
|
The use of xz to compress the release tarball causes a
"Cannot allocate memory" error on ARMv7, so use gzip instead.
|
|
Before the cmake min version bump, the libxml2 detection in clang/CMakeLists.txt
probably failed, and so it didn't try to use libxml2 for c-index-test at all.
We should just build with -DCLANG_ENABLE_LIBXML2=NO to keep the status quo.
Ref: https://github.com/llvm/llvm-project/issues/64263
Change-Id: I504581f19e0c44f5d57ac03a5f3c661c24ad8432
|
|
... for tcwg_gcc_check--master-arm to test history rewrite.
Rewriting got to a point where an old result was removed, and that
removed result had a flaky test detected. Consequently, some
subsequent results now fail to identify that result as flaky,
and fail. Increasing rr[major] will cause those results to be
removed from history.
Change-Id: I19362af03af266ae93f94aa313b27a24e1baa74c
|
|
Change-Id: I4adbfe5a5ca69d6b5beeb9347081ea02400511b3
|
|
... so that they are included in the files uploaded to jenkins.
This simplifies troubleshooting of failures in check_regression()
step -- we will have logs uploaded to jenkins.
Change-Id: I022d8f7df15585652f88fa0d95ae6078a6d61ce5
|
|
We cannot use both --with-cpu and --with-arch, so drop
--with-arch=armv8-m.main+dsp for consistency with the other
configurations.
Change-Id: Idc413138131504db1b6941391b7c9106605338d1
|
|
Change-Id: Iebe85747cdf764e91b72ff2580b1dfa18d2dfbe2
|
|
First baseline run (successful) needs to be used as a reference.
but the results, are not pushed to bkp-01.
In that case, in tested runs, we want to use reference bmk results
directly from base-artifacts directory
Change-Id: If9500bfd9a040ee36289ca31cfa98c37e9cef7d0
|
|
Complete the list of configurations we want to monitor in our CI.
When we use --with-mode=thumb/-mthumb, the config is named thumb*_eabi
instead of arm*_eabi.
For M-profile, we use the processor name as a shortcut, but this makes
it less obvious which architecture level we are testing.
We add the equivalent of the following target-boards:
-mthumb/-march=armv8-a/-mfpu=crypto-neon-fp-armv8/-mfloat-abi=hard
-mthumb/-mcpu=cortex-m0
-mthumb/-mcpu=cortex-m3
-mthumb/-mcpu=cortex-m7
-mthumb/-mcpu=cortex-m23
-mthumb/-mcpu=cortex-m33/-march=armv8-m.main+dsp
-mthumb/-march=armv8.1-m.main+mve/-mfloat-abi=hard
Change-Id: I632c4617e2356566b66ebadfeb175cb97f4663b8
|
|
Change-Id: Ib1793805e3dfd374be25f2f2ff2caa32d86bd518
|
|
Change-Id: Icb7fb9f8a55c5e0907c0f16d368ba8a78fc73f95
|
|
Change-Id: I34cff1f808e67b3e99a7e76d6a88707b77fa242e
|
|
Change-Id: Iae6ae617ee8522f722e593eccd58688b76752b81
|
|
Our handling of "queued" bundles (which we can delete and re-create)
confuses patchwork. I'm now seeing dozens of broken links to "queued"
bundles in web-ui. We don't really need queued bundles, so remove
them.
Change-Id: Ide31c24d9b2dea0d3bb68829abb8dac6c585957a
|
|
Using patchwork bundles to store state of patches quickly became
a problem. Bundles are designed to contain tens of patches, but not
hundreds. Given that we, add every single tested patch to a bundle,
these bundles grown with no bound.
Since we really are only using "triggered" bundle for decision making,
this patch replaces it with looking at states of the checks for
a particular patch.
Follow up patches will remove use of other remaining bundles.
Change-Id: I4e7487f95f3773c45162728cac56da9e8c2fa94b
|
|
Change-Id: Ib99adbcf0ec93acf4b59d385fb0182bf866ca6d0
|
|
Change-Id: Iebb32b89a1d498c69f8f8c9d897abe41b52c4c1b
|
|
Change-Id: Ifac3b7bb62a52fb2892c4c6674b9520a4091fe0f
|
|
Move logic to create $jira_dir from post_interesting_commits() to
update_interesting_commits(). This ensures that $jira_dir is created
regardless of whether the jira card exists. This is for the case
of processing old results before support for creating jira cards.
Change-Id: Ibab1bd49fa5ced899c65801804251203713d7b13
|
|
I've finished my experiments.
Change-Id: I0d95f9541d8dba4c95819ed93fb79e27a19b71cc
|
|
Change-Id: Ibf530ba4a348cfb440df4f6f7a1db8517f13a394
|
|
different container
When starting the lldb arm bot on jade-04 we want to use jammy not focal.
Previously, any bot after that would also use focal which is not what we want.
Explicitly set the default distro each time, and space the code out so it's
clear what gets what.
Change-Id: Icf8fd0a83d6c19897427378658f767ba32786ef9
|
|
I had a request from Arm to speed up this bot some as it is a lot slower
than even the 32 bit 2 stage builds. Likely due to building flang.
We have spare capacity on jade-04 so lets use it. I will also be allocating
30 cores to it, vs. the previous 15.
Change-Id: Id1959fce935ef94275da8342a52a1ad8219437f5
|
|
I suspect there is a performance degredation on jade-04, as libcxx builds
went from around 1/2 an hour to an hour without a good explanation.
Could be my faulty memory, so put one of them on jade-01 to compare with.
Change-Id: Ie88a1e94323640fb9fcba10faa256608ae0aba18
|
|
--security
The --docket_opts option is currently used by glibc config to add
extra security options, while --default_security disables the default
one. To simplify both start-container-docker.sh and container config
that wantso to use a different security options, remove both option
in favor of a new --security one.
If not set, the default security option is used; otherwise it overrides
the default one. The idea is if the configuration sets it, it knows
exactly which security it requires.
Change-Id: Ie29ea23d8d029efc9a56eec0d9ec5df776037337
|
|
Change-Id: Ic9ebfea312bb05c6e78fa061ecfe617db66e4bed
|
|
- Also transfer links to bkp-01
- Adjust permission of the remote dir created by tcwg-buildslave,
It is sometimes created without possibility for tcwg-benchmark to read it
Change-Id: Iedd8cfae67f61a93e4fe06ae4f68790bf345aa99
|
|
The rsync was not transfering links, therefore internal perf record
directory was uncomplete. (i.e, .build-id sym links).
This rsync should now also copies these relative symlinks.
These links pointing on the elf file are important for having funtion
names in the generated perf report.
Change-Id: I292947fa037f8af8b32b2c695923fed4237d6b35
|
|
Change-Id: I17635e2a85033d69ef901b45f15e07b5bda5b2b1
|
|
Change-Id: I1026938f63228686e1e788486c63274ddd7b6abe
|
|
Change-Id: Ib825c79d67ed76b5e7681464ebfb069eb99201f9
|
|
bmk-results getting&storing
tcwg-benchmark-results: -input/-output results are content of the
results_id files
round-robin-baseline: fix .gitignore
Change-Id: Ia695f951fc725c6f11e30c874bb0b51b50300828
|
|
Change-Id: Ia72e5a31b311bed9e44b7b317f8cd4d41f527fe3
|
|
.. now store in /home/tcwg-benchmark/results-xxx only if successful (in
round-robin-baseline.sh)
Change-Id: Ic700c3c8dbfefff84cb386259f54cbd70147c3f7
|
|
... like "{notify-}" fetched in round-robin-baseline.sh. Make
sure to unset any local definitions and return manifest value.
Change-Id: I72b4eb15c6069cfee039a52e1aa946495ba06fd1
|
|
Change-Id: I0b9489c36bd3360b88343902871c55e59271a558
|
|
We run reboot either from host container (where sudo is unrestricted)
or as "root" user on the bare machine (where sudo is unrestricted).
Change-Id: I5b78704f59ce064f97a975c3edea4b6b46217565
|
|
The string arm_v7a_softfp is a shortcut to mean testing an arm-eabi toolchain with:
-marm -march=armv7-a -mfpu=vfpv3-d16 -mfloat-abi=softfp
We could build the toolchain with the default configuration flags, but
that would mean building a lot of useless multilibs, since we want to
force the above flags when running the tests.
So we also configure & build GCC with
--disable-multilib --with-arch=armv7-a --with-fpu=vfpv3-d16 --with-mode=arm --with-float=softfp
Change-Id: Ieb6c40ad56a3ae3795ad5a74faeb5d0add16a977
|
|
Change-Id: Ib876e4eb1ad4620e8c9746ea5648a4011a839402
|
|
When re-writing history we set the oldest commit to initialize
the baseline (update_baseline==init), and, if we had more than
200 commits in the history, trim_base_artifacts() removed
the new "init" commit. Then rewrite "promoted" next commit
to be the "init" one, and trim_base_artifacts() removed that
commit as well. This continued until we reached 200 commits
in history, and deleted the oldest "forced" commits.
Change-Id: I1b2b0f48220ab84564ac5361f5f642a37b51cc70
|