Update to support TIGRE v3

[gfardell]

Changes

Support for TIGRE 3.1.3.
Conda packages built and on anaconda.

ToDo:
Update in support table
Mark package as main
Update environment files

Testing you performed

Please add any demo scripts to https://github.com/TomographicImaging/CIL-Demos/tree/main/misc

Related issues/links

Checklist

• I have performed a self-review of my code
• I have added docstrings in line with the guidance in the developer guide
• I have updated the relevant documentation
• I have implemented unit tests that cover any new or modified functionality
• CHANGELOG.md has been updated with any functionality change
• Request review from all relevant developers

[casperdcl]

I suspect a problem with https://github.com/TomographicImaging/TIGRE-conda. Perhaps:

• we need to increase the nvidia channel priority (iirc the conda-forge cuda libs don't work with tigre)
• or we should simply have tigre & cuda statically linked?

@JeppeKlitgaard any thoughts?

[JeppeKlitgaard]

I suspect a problem with https://github.com/TomographicImaging/TIGRE-conda. Perhaps:

• we need to increase the nvidia channel priority (iirc the conda-forge cuda libs don't work with tigre)
• or we should simply have tigre & cuda statically linked?

@JeppeKlitgaard any thoughts?

I am a bit puzzled that the (3.10, 1.23) AND (3.12, 1.26) both work just fine, but somehow (3.11, 1.25) does not? Never mind, only (3.12, 1.26) actually tests CUDA!

I think some CUDA-based libraries do statically link it, but that seems like quite a cumbersome solution?

Is this just related to the runner not having CUDA installed? It could be related to the change to CUDA 0.12, which I guess is new in TIGREv3?

I suspect the runner might have an older, incompatible CUDA installed (0.10, 0.11?), which I guess was fine for TIGREv2

[casperdcl]

The runner has Driver Version: 580.126.09 (and CUDA Version: 13.0 though this shouldn't matter since conda should install the correct cuda toolkit for tigre).

[JeppeKlitgaard]

I am a bit pressed for time at the moment, but in a few weeks that should clear up. I can try and see if I can get TIGRE-conda to build locally on a machine with CUDA 13 and a recent driver, though that was already done as part of the PR that changed over the build system.

It might be difficult to debug without access to the runner, if it does turn out that it is related to the configuration there.

[lauramurgatroyd]

The 'conda-test' action is currently getting tigre: 2.6-np123py310_0 ccpi

This is because the dev channel hasn't been added in the action - I will do this

[lauramurgatroyd]

I have now updated so that all of the actions get the correct version of tigre: 3.1.3.

We should be able to import tigre whether we have cuda or not, and at the moment we are unable to import tigre on runners with or without cuda. So something is wrong with the build in that regard.

---

Docs action: this runs on a runner without GPU, however it should still be able to import the tigre codes and render them (as is done on master - docs rendered from master can be seen here:

https://tomographicimaging.github.io/CIL/nightly/plugins/#tigre

On this branch Tigre has been installed in the docs environment:
https://github.com/TomographicImaging/CIL/actions/runs/25121210631/job/73622135939?pr=2317#step:5:1451

However, the documentation for tigre does not render:

This is because it gets this issue on import:
https://github.com/TomographicImaging/CIL/actions/runs/25499785025/job/74829166123#step:10:334

Example of a run on master: https://github.com/TomographicImaging/CIL/actions/runs/25448205768

---

Conda-test: This is runs on a system without a GPU

We have errors with tests using tigre:
e.g.
test_PluginsTigre_FBP
ImportError: libcudart.so.12: cannot open shared object file: No such file or directory
https://github.com/TomographicImaging/CIL/actions/runs/25502156025/job/74838763248?pr=2317#step:6:3139

On master, this test is skipped in conda-test:
test_FBP (test_PluginsTigre_FBP.Test_Cone2D_FBP_tigre) ... skipped 'Requires TIGRE GPU'
https://github.com/TomographicImaging/CIL/actions/runs/25489114378/job/74794042165

I think on this branch, it fails on the tigre import before it even checks if it should be skipped:

CIL/Wrappers/Python/test/test_PluginsTigre_FBP.py

Lines 19 to 34 in 26882d8

	import unittest
	from utils_projectors import TestCommon_FBP_SIM
	from utils import has_tigre, has_nvidia, initialise_tests
	initialise_tests()
	if has_tigre:
	from cil.plugins.tigre import FBP
	def setup_parameters(self):
	self.backend = 'tigre'
	self.FBP = FBP
	self.FBP_args={}
	@unittest.skipUnless(has_tigre and has_nvidia, "Requires TIGRE GPU")

[lauramurgatroyd]

Note: the test actions pass because they do not install tigre
https://github.com/TomographicImaging/CIL/actions/runs/25502156025/job/74838763248?pr=2317#step:3:2

[lauramurgatroyd]

The docs issue was resolved by #2323

[JeppeKlitgaard]

I am happy to take a stab at trying to figure out whether this is related to the new TIGRE v3 builds made in https://github.com/TomographicImaging/TIGRE-conda, but it might take me couple of weeks before I have time to commit to this.

Thoughts:

The fact that this seemed also to be happening while the dependency was being resolved to TIGRE 2.6 might suggest that it is not related to the new TIGRE-conda builds.

The lack of a CUDA binary suggests something is off in the build system (either a lack of a dependency, or perhaps this should be declared via TIGRE-conda and fetched automatically?)

[lauramurgatroyd]

Hi @JeppeKlitgaard thank you very much for the offer to take a look, but we're trying to release very soon so we'll take a look at it before then.

The actions that were still on tigre 2.6 were passing - so the issue is only happening with this latest tigre version.

[JeppeKlitgaard]

Hi @JeppeKlitgaard thank you very much for the offer to take a look, but we're trying to release very soon so we'll take a look at it before then.

The actions that were still on tigre 2.6 were passing - so the issue is only happening with this latest tigre version.

Sorry, I misunderstood the problem in that case.

I can try and see if I can get the tigre build to install on a Linux machine with CUDA tomorrow to help pin down whether it is related to the packaging or environment.

Wednesday next week I would be able to take a more in-depth look if it hasn't been resolved by then?

Let me know if this would be helpful at all - I understand this may be too late for the release schedule

[casperdcl]

• tigre 2.* was built with https://github.com/TomographicImaging/TIGRE-conda/blob/5a14afb9a9a299ca01bea514731a91ee479efc00/recipe/meta.yaml and worked fine
• tigre 3.* is built with https://github.com/TomographicImaging/TIGRE-conda/blob/main/pixi.toml and has the installation problems