Run Core tests and CoreValidation with IAR toolchain

[JonatanAntoni]

No description provided.

[Copilot]

Pull Request Overview

This PR adds support for running Core tests with the IAR toolchain by updating several test check patterns and configuration files. Key changes include:

• Updated CHECK patterns in test source files to distinguish between IAR and non-IAR output.
• Added a new Toolchain_IAR class and modified configuration settings in lit.cfg.py.
• Updated build scripts and workflow configurations to support IAR.

Reviewed Changes

Copilot reviewed 9 out of 10 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
CMSIS/Core/Test/src/simd.c	Updated CHECK patterns to separate IAR and non-IAR cases.
CMSIS/Core/Test/src/rrx.c	Modified CHECK patterns to include both rrx and rrxs instructions.
CMSIS/Core/Test/src/ror.c	Revised CHECK patterns to update regex for THUMB and ARM variants.
CMSIS/Core/Test/src/noreturn.c	Adjusted regex checks to support both b and bl branch instructions.
CMSIS/Core/Test/src/cpsr.c	Minor regex update to allow optional 's' in the AND operation check.
CMSIS/Core/Test/lit.cfg.py	Added Toolchain_IAR class with new configuration settings; updated mfpu.
CMSIS/Core/Test/build.py	Removed obsolete IAR matrix filter support.
CMSIS/Core/Include/m-profile/cmsis_iccarm_m.h	Simplified preprocessor conditions regarding architecture features.
.github/workflows/core.yml	Added IAR token environment variable for CI integration.

Files not reviewed (1)

• CMSIS/Core/Test/vcpkg-configuration.json: Language not supported

Comments suppressed due to low confidence (2)

CMSIS/Core/Test/lit.cfg.py:891

• The variable 'device' is used instead of 'self.device' when accessing the DEVICES configuration. Update to 'DEVICES[self.device]["header"]' for consistency.

f'--cpu={self.cpu()}', f'--fpu={self.fpu()}', '-I', os.path.abspath('../Include'), '-c', '-D', f'CORE_HEADER="{DEVICES[device]["header"]}"'

CMSIS/Core/Test/lit.cfg.py:892

• The variable 'device' is undefined in this context; use 'self.device' to ensure the correct configuration is referenced.

if device.endswith('S') and not device.endswith('NS'):

[JonatanAntoni]

@RobinKastberg, I enhanced CI tests with IAR toolchain. Could you please cross check at least the modification to IAR compiler header?

Looks like many test cases do not pass with IAR compiler. Could you have a look as well? We might need to enhance the test expectations as IAR compiler generates another assembly variant.

[github-actions]

Test Results

   500 files  +  128     500 suites  +128   11m 52s ⏱️ + 3m 24s
    49 tests ±    0      27 ✅  -    17      5 💤 ±    0   17 ❌ + 17 
24 500 runs  +6 272  15 787 ✅ +3 823  8 400 💤 +2 136  313 ❌ +313 

For more details on these failures, see this check.

Results for commit e224a92. ± Comparison against base commit 6f0a58d.

♻️ This comment has been updated with latest results.

[RobinKastberg]

@RobinKastberg, I enhanced CI tests with IAR toolchain. Could you please cross check at least the modification to IAR compiler header?

Great work!

Looks like many test cases do not pass with IAR compiler. Could you have a look as well? We might need to enhance the test expectations as IAR compiler generates another assembly variant.

I will take a look, but I am quite busy with other things currently.
I did go through this locally some time ago finding/fixing some critical bugs in the process. I think the remaining bugs where not major (i.e. mainly related to how the tests are setup). Of course my memory is quite fuzzy now and things might have changed.

[JonatanAntoni]

@RobinKastberg, I am now about to run CoreValidation tests. But I am seeing a weird issue on GitHub. Despite having set IAR_LMS_BEARER_TOKEN I am getting errors running IAR Toolchain commands:

> iccarm --version

Fatal error[LMSC4050]: product field missing in checkout request
   IAR ANSI C/C++ Compiler V9.60.4.438/LNX for ARM
   Copyright 1999-2025 IAR Systems AB.
Fatal error detected, aborting.

This is surprising, because in within the Core test job the toolchain has no issues.
Can you help us figuring out what is going wrong?

[RobinKastberg]

@JonatanAntoni where did you get toolchain from? ARM artifactory?

[JonatanAntoni]

where did you get toolchain from? ARM artifactory?

Yes, its installed to the build machine via vcpkg-shell activate.
I.e., its extracted to /home/runner/work/CMSIS_6/CMSIS_6/.vcpkg/downloads/artifacts/2139c4c6/compilers.iar.cxarm/9.60.4/, see https://github.com/ARM-software/CMSIS_6/pull/229/checks#step:8:147.

To be exact, its downloaded from https://github.com/iarsystems/arm/releases/download/9.60.4/cxarm-9.60.4-linux-x86_64-base.tar.bz2 by the vcpkg manifest distributed via Arm Artifactory.

It is working for the Core test job, without issues. In there, the compiler is launched directly via lit. But when used via cbuild/cmake in CoreValidation it is failing. I can reproduce the problem in an interactive Codespace session. On my local machine I do not see this issue. But there I have a hack anyway, because I am on a Mac and due to the lack of a Mac distribution I don't receive the asset via vcpkg. Instead, I built a Docker wrapper around the toolchain bundle.

[RobinKastberg]

@JonatanAntoni Issue is currently being looked into. We don't know details yet.

Can you try rerunning the job a few times. Does it happen every time?

[JonatanAntoni]

Can you try rerunning the job a few times. Does it happen every time?

Yes, the behaviour is reproducible each time.

[JonatanAntoni]

@RobinKastberg, now the toolchain seems to run correctly. We need to focus on fixing the tests, now. Can you help with that?

[RobinKastberg]

@RobinKastberg, now the toolchain seems to run correctly. We need to focus on fixing the tests, now. Can you help with that?

Nice! I will get started on it next week!

[RobinKastberg]

@JonatanAntoni
Hi!

I took a look at build errors for CoreValidation to start with, couldn't get FVP running this time due to licensing issues (I mailed you regarding this)

• NS builds fail because bootloader.c writes to MSP_NS. Root cause seems to be that this should be compiled with --cmse.
• It looks like a bunch of Cortex-A fails due to having d16 FPUs AND NEON, so the FPU code expects >16 registers since there is NEON but there is only 16 due to the FPU type mentioned. This seems to be a build system inconsistency, maybe we should set .no_neon in devtools cmake? Looks like the other compilers does this for some reason?
• Cortex-A9 startup doesn't support IAR:

  CMSIS_6/CMSIS/CoreValidation/Layer/Target/CA9/RTE/Device/ARMCA9/startup_ARMCA9.c

  Line 123 in b2c51f0

  #error Unknown compiler.

  
  I see both of these issues as mostly artifacts of the testing, and not real issues.

I will take a look at the lit tests next.

[JonatanAntoni]

Thanks for spotting such issues. Yes, I'd expect most of the issues to be some toolchain integration issues, such as wrong/misaligned compiler flags. We need to sort them out gradually.

[KeilChris]

Do not forget to update the README.md file with instructions on how to build with IAR. Currently, it is missing.