update LLVM easyblock to fix build on RISC-V systems and ignore a bunch of tests by julianmorillo · Pull Request #3676 · easybuilders/easybuild-easyblocks

julianmorillo · 2025-03-26T14:58:25Z

No description provided.

ocaisa · 2025-04-02T13:23:25Z

+        # glob_pattern = os.path.join(base_dir, 'lib', f'%s-{arch}')
+        glob_pattern = os.path.join(base_dir, 'lib', f'{arch}-*')


This doesn't look right, should this be arch-specific?

It worked for me in RISC-V.

I'm not sure what the original line does, but with the original code, glob_pattern was equal to /tmp/eb/easy build/build/LLVM/20.1.1/GCCcore-13.2.0/llvm.obj.1/lib/%s-riscv64. And with that, matches was null after matches = glob.glob(glob_pattern), so the ERROR "Could not find runtime library directory" was raised.

On the other hand, with my modification, matches[0] is /tmp/eb/easybuild/build/LLVM/20.1.1/GCCcore-13.2.0/llvm.obj.1/lib/riscv64-unknown-linux-gnu and the function finally returns lib/riscv64-unknown-linux-gnu

How does it work in other architectures?

@Crivella would know better than me...

There was an error introduced in #3674 when replacing the old python2 formatting with f-strings.
This is also currently being fixed in #3675, but it might take a bit before merging as there is also a lot of device testing going on there.

Regarding the location, during compilation LLVM puts the runtimes (specific for the host target) in $ROOT/lib/<target_triple>.
I am not aware of it ever compiling more than one set of runtimes (maybe if you are cross-compiling LLVM itself?), but that was the easiest way i had to find the runtime library directory

On my side i am ok with the glob pattern being fixed ASAP as it is a breaking change that will cause (probably) all multi-stage builds to fail using the current develop easyblock

@ocaisa

On second thought, should the get_arch_prefix introduced in #3674 be a framework function? (maybe renamed to get_compiler_arch_prefix)

@Crivella I think that could indeed go into systemtools

Crivella · 2025-04-30T11:27:47Z

@boegelbot please test @ jsc-zen3
EB_ARGS="LLVM-18.1.8-GCCcore-13.3.0-minimal.eb --installpath /tmp/$USER/ebpr-3676"

boegelbot · 2025-04-30T11:33:09Z

@Crivella: I noticed your comment, but I only dance when @akesandgren or @bartoldeman or @bedroge or @boegel or @branfosj or @casparvl or @jfgrimm or @lexming or @Micket or @migueldiascosta or @ocaisa or @SebastianAchilles or @smoors or @verdurin or @robert-mijakovic or @deniskristak or @ItIsI-Orient or @PetrKralCZ or @sassy-crick or @laraPPr or @pavelToman or @Louwrensth or @Thyre tells me (for now), I'm sorry...

Details

- notification for comment with ID 2841677283 processed

Message to humans: this is just bookkeeping information for me,
it is of no use to you (unless you think I have a bug, which I don't).

ocaisa · 2025-04-30T11:34:19Z

@boegelbot please test @ jsc-zen3
EB_ARGS="LLVM-18.1.8-GCCcore-13.3.0-minimal.eb --installpath /tmp/$USER/ebpr-3676"

boegelbot · 2025-04-30T11:53:08Z

@ocaisa: Request for testing this PR well received on jsczen3l1.int.jsc-zen3.fz-juelich.de

PR test command 'if [[ develop != 'develop' ]]; then EB_BRANCH=develop ./easybuild_develop.sh 2> /dev/null 1>&2; EB_PREFIX=/home/boegelbot/easybuild/develop source init_env_easybuild_develop.sh; fi; EB_PR=3676 EB_ARGS="LLVM-18.1.8-GCCcore-13.3.0-minimal.eb --installpath /tmp/$USER/ebpr-3676" EB_REPO=easybuild-easyblocks EB_BRANCH=develop /opt/software/slurm/bin/sbatch --job-name test_PR_3676 --ntasks=8 ~/boegelbot/eb_from_pr_upload_jsc-zen3.sh' executed!

exit code: 0
output:

Submitted batch job 6264

Test results coming soon (I hope)...

Details

- notification for comment with ID 2841691562 processed

Message to humans: this is just bookkeeping information for me,
it is of no use to you (unless you think I have a bug, which I don't).

Crivella · 2025-04-30T11:53:28Z

As a note on the test, i do not think we have a test runner to properly test the arch string, but I do not see any particular exceptions looking at

https://github.com/llvm/llvm-project/blob/120e940356561035bf37bc4fcb6fab771e7a9b7d/llvm/cmake/config.guess#L1006-L1029

Maybe we can also add the string for 32bit systems, but this might be as discussed better moved to a framework implementation.

@julianmorillo Just in case you have the easybuild github integration installed would you mind running a minimal build uploading the report? EG eb LLVM-18.1.8-GCCcore-13.3.0-minimal.eb --include-easyblocks-from-pr 3676 --upload-test-report

boegelbot · 2025-04-30T12:24:48Z

Test report by @boegelbot

Overview of tested easyconfigs (in order)

SUCCESS LLVM-18.1.8-GCCcore-13.3.0-minimal.eb

Build succeeded for 1 out of 1 (1 easyconfigs in total)
jsczen3c1.int.jsc-zen3.fz-juelich.de - Linux Rocky Linux 9.5, x86_64, AMD EPYC-Milan Processor (zen3), Python 3.9.21
See https://gist.github.com/boegelbot/11a978ff93e236792718e4ef3f5ba170 for a full test report.

Thyre · 2025-05-01T09:34:01Z

We need to test a non-minimal build for this, as the arch prefix is only relevant if runtimes are built (or tests in a multi-stage build) IIRC.

The new LLVM-EasyConfig PRs would be ideal.

Crivella · 2025-05-01T10:36:07Z

Agree the test on RISCV should be on a full build uploaded the message and also retyping here:

For the RISCV test eb --from-pr 22517 --include-easyblocks-from-pr 3676 --upload-test-report

I tested the minimal as self._set_gcc_prefix() is still called during the configure step, but it would be ideal to test runtime builds on RISCV

Crivella · 2025-05-01T10:38:42Z

Now also considering the changes in #3706 I am starting to wonder if it would be better to have get_arch_prefix at the framework level in general, or for every toolchain, or possibly at the easyblock level ... 🤔

jfgrimm · 2025-05-01T10:51:45Z

@Crivella perhaps throw it in framework's systemtools?
https://github.com/easybuilders/easybuild-framework/blob/develop/easybuild/tools/systemtools.py

boegel · 2025-06-10T16:30:36Z

This looks really trivial now...

@julianmorillo Is there really all that's necessary to fix the installation of LLVM on RISC-V?

julianmorillo · 2025-06-13T07:51:15Z

@boegel ,

eb --optarch=GENERIC -r LLVM-20.1.5-GCCcore-13.2.0.eb --from-commit a0e254bc8a04d20dcd97ded5e339214628074ff3 --include-easyblocks-from-pr=3676

Gives these results:

Testing Time: 11664.78s

Total Discovered Tests: 139851
  Skipped          :     32 (0.02%)
  Unsupported      :   5751 (4.11%)
  Passed           : 133367 (95.36%)
  Expectedly Failed:    269 (0.19%)
  Timed Out        :      1 (0.00%)
  Failed           :    431 (0.31%)

The build process works fine, but since only 10 failing tests are allowed in the easyconfig, it does not get finally installed. This is the complete log:
easybuild-LLVM-20.1.5-20250612.091751.cYUNa.log.zip

Crivella · 2025-06-16T12:57:46Z

Many of the failing tests seems to be due to errors of the kind

# .---command stderr------------
# | error: OpenMP target is invalid: 'riscv64-unknown-linux-gnu'
# `-----------------------------

So i would guess just changing that is not enough of a fix.
It is possible is just a problem with the tests not recognizing differences in targets between stuff like -pc-linux- and -unknown-linux

We might have to add a different patch similar to https://github.com/easybuilders/easybuild-easyconfigs/blob/develop/easybuild/easyconfigs/l/LLVM/LLVM-19.1.7_libomptarget_tests.patch or modify it to catch this case

… it is not supported

…chpoints

…clang: error: -gsplit-dwarf is unsupported with RISC-V linker relaxation (-mrelax)"

…, which is not supported)

…n LLVM JIT

…ch: 243" in RISC-V

…into llvm

…uption of mismatch, likely due to incomplete or buggy support in the LLVM runtime for RISC-V under '-std=c++26'

Co-authored-by: Davide Grassano <[email protected]>

julianmorillo · 2025-07-25T08:59:37Z

eb --optarch=GENERIC -r LLVM-20.1.5-GCCcore-13.2.0.eb --from-pr 22656 --include-easyblocks-from-pr=3676
Successfully builds on RISC-V. This is the log:
easybuild-LLVM-20.1.5-20250725.103240.log.zip

Crivella · 2025-07-25T12:41:00Z

@boegelbot please test @ jsc-zen3
EB_ARGS="--installpath /tmp/$USER/ebpr-3676 LLVM-18.1.8-GCCcore-13.3.0.eb LLVM-19.1.7-GCCcore-13.3.0.eb LLVM-20.1.5-GCCcore-13.3.0.eb"
CORE_CNT=16

boegelbot · 2025-07-25T12:53:08Z

@Crivella: Request for testing this PR well received on jsczen3l1.int.jsc-zen3.fz-juelich.de

PR test command 'if [[ develop != 'develop' ]]; then EB_BRANCH=develop ./easybuild_develop.sh 2> /dev/null 1>&2; EB_PREFIX=/home/boegelbot/easybuild/develop source init_env_easybuild_develop.sh; fi; EB_PR=3676 EB_ARGS="--installpath /tmp/$USER/ebpr-3676 LLVM-18.1.8-GCCcore-13.3.0.eb LLVM-19.1.7-GCCcore-13.3.0.eb LLVM-20.1.5-GCCcore-13.3.0.eb" EB_REPO=easybuild-easyblocks EB_BRANCH=develop /opt/software/slurm/bin/sbatch --job-name test_PR_3676 --ntasks="16" ~/boegelbot/eb_from_pr_upload_jsc-zen3.sh' executed!

exit code: 0
output:

Submitted batch job 7359

Test results coming soon (I hope)...

Details

- notification for comment with ID 3117641641 processed

Message to humans: this is just bookkeeping information for me,
it is of no use to you (unless you think I have a bug, which I don't).

boegelbot · 2025-07-25T20:13:51Z

Test report by @boegelbot

Overview of tested easyconfigs (in order)

SUCCESS LLVM-18.1.8-GCCcore-13.3.0.eb
SUCCESS LLVM-19.1.7-GCCcore-13.3.0.eb
SUCCESS LLVM-20.1.5-GCCcore-13.3.0.eb

Build succeeded for 3 out of 3 (3 easyconfigs in total)
jsczen3c2.int.jsc-zen3.fz-juelich.de - Linux Rocky Linux 9.5, x86_64, AMD EPYC-Milan Processor (zen3), Python 3.9.21
See https://gist.github.com/boegelbot/31dde3dc21b118b07776e32c7d369f28 for a full test report.

Crivella

LGTM

in the future we might want some way to also versioncontrol which tests are ignored, but that is a problem for another PR (or even a new feature in framework)

Crivella · 2025-07-31T13:58:15Z

Going in, thanks @julianmorillo!

julianmorillo added 6 commits March 26, 2025 15:57

Fix LLVM build on RISCV

9aa9461

Give more informative warning

2e80fd1

Give more informative warning in get_runtime_lib_path

f8edb20

Added print_warning messages in get_runtime_lib_path

c0a07ed

Fix bug in get_runtime_lib_path function

9e959d8

Fix style typo

0be9f92

ocaisa reviewed Apr 2, 2025

View reviewed changes

Merge branch 'develop' into llvm

f020531

Merge branch 'easybuilders:develop' into llvm

5ff298a

boegel changed the title ~~Fix LLVM build on RISCV~~ update LLVM easyblock to fix build on RISC-V systems Jun 10, 2025

boegel added enhancement riscv labels Jun 10, 2025

boegel added this to the next release (5.1.1?) milestone Jun 10, 2025

Crivella mentioned this pull request Jun 16, 2025

Update easyconfig LLVM-20.1.5-GCCcore-13.2.0.eb for RISC-V easybuilders/easybuild-easyconfigs#22656

Merged

boegel modified the milestones: 5.1.1, release after 5.1.1 Jul 2, 2025

Merge branch 'easybuilders:develop' into llvm

86516c9

julianmorillo and others added 18 commits July 14, 2025 17:23

In RISC-V, ignore all tests that try to create watchpoints in lldb as…

161a719

… it is not supported

Fix "TabError: inconsistent use of tabs and spaces in indentation"

ce3f464

Clarify that what it is not supported is the creation of HARDWARE wat…

ff10cb1

…chpoints

Fix line too long and erroneous insertion

6e47727

Remove trailing whitespace

06ce699

Ignore tests that use the flag -gsplit-dwarf. They cause the error: "…

ac883cb

…clang: error: -gsplit-dwarf is unsupported with RISC-V linker relaxation (-mrelax)"

Fix "line too long" and "trailing whitespace"

52c19fa

Ignore tests that fail in RISC-V (they use '-fsanitize=memory,fuzzer'…

39af2d4

…, which is not supported)

Ignore tests due to Missing support for RISC-V relocation type '55' i…

8b1bf32

…n LLVM JIT

Avoid tests that produce the error "unsupported 64-bit ELF machine ar…

4e9284b

…ch: 243" in RISC-V

Fix formatting errors

d4917a4

Merge branch 'easybuilders:develop' into llvm

7e52f7a

Move ignore_patterns from Easyconfig to Easyblock

a9fc2c3

Merge branch 'llvm' of github.com:julianmorillo/easybuild-easyblocks …

a77488b

…into llvm

Remove whitespace in blank line

be24c87

Adde ignore patterns for tests that fail due to memory allocator corr…

a011665

…uption of mismatch, likely due to incomplete or buggy support in the LLVM runtime for RISC-V under '-std=c++26'

Fix bug getting binutils version

dfe7432

Check binutils version fix

debb88f

Crivella reviewed Jul 24, 2025

View reviewed changes

Comment thread easybuild/easyblocks/l/llvm.py Outdated

Update easybuild/easyblocks/l/llvm.py

0901c94

Co-authored-by: Davide Grassano <[email protected]>

Crivella approved these changes Jul 31, 2025

View reviewed changes

Crivella merged commit faa1044 into easybuilders:develop Jul 31, 2025
17 checks passed

julianmorillo deleted the llvm branch July 31, 2025 13:59

boegel changed the title ~~update LLVM easyblock to fix build on RISC-V systems~~ update LLVM easyblock to fix build on RISC-V systems and ignore a bunch of tests Aug 12, 2025

		# glob_pattern = os.path.join(base_dir, 'lib', f'%s-{arch}')
		glob_pattern = os.path.join(base_dir, 'lib', f'{arch}-*')

Conversation

julianmorillo commented Mar 26, 2025

Uh oh!

ocaisa Apr 2, 2025

Choose a reason for hiding this comment

Uh oh!

julianmorillo Apr 2, 2025

Choose a reason for hiding this comment

Uh oh!

ocaisa Apr 2, 2025

Choose a reason for hiding this comment

Uh oh!

Crivella Apr 2, 2025

Choose a reason for hiding this comment

Uh oh!

Crivella Apr 2, 2025

Choose a reason for hiding this comment

Uh oh!

Crivella Apr 2, 2025

Choose a reason for hiding this comment

Uh oh!

ocaisa Apr 3, 2025

Choose a reason for hiding this comment

Uh oh!

Crivella commented Apr 30, 2025

Uh oh!

boegelbot commented Apr 30, 2025

Uh oh!

ocaisa commented Apr 30, 2025

Uh oh!

boegelbot commented Apr 30, 2025

Uh oh!

Crivella commented Apr 30, 2025

Uh oh!

boegelbot commented Apr 30, 2025

Overview of tested easyconfigs (in order)

Uh oh!

Thyre commented May 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Crivella commented May 1, 2025

Uh oh!

Crivella commented May 1, 2025

Uh oh!

jfgrimm commented May 1, 2025

Uh oh!

boegel commented Jun 10, 2025

Uh oh!

julianmorillo commented Jun 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Crivella commented Jun 16, 2025

Uh oh!

Uh oh!

julianmorillo commented Jul 25, 2025

Uh oh!

Crivella commented Jul 25, 2025

Uh oh!

boegelbot commented Jul 25, 2025

Uh oh!

boegelbot commented Jul 25, 2025

Overview of tested easyconfigs (in order)

Uh oh!

Crivella left a comment

Choose a reason for hiding this comment

Uh oh!

Crivella commented Jul 31, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

Thyre commented May 1, 2025 •

edited

Loading

julianmorillo commented Jun 13, 2025 •

edited

Loading