Fix: CMakeLists: removed manual CUDA path and upgrade Catch2 version

Context

CUDA is not found on Archlinux which install CUDA to /opt and Conda on its dedicated environment. But standard CMake files are populated, which allows find_package to function correctly.

When AIDGE_INSTALL variable is set, manual override of the CUDA path prevents finding the library.

Catch2 does not compile under GCC 14.2, upgrading to 3.7.1 fix the problem.

Modified files

• CMakeLists.txt
• unit_tests/CMakeLists.txt

Detailed major modifications

find_package work well under Archlinux and Conda environment. Specific scenario would require a FindCUDA.cmake file approach to search for the correct path instead of manual overrides.

Catch2 upgraded from V3.0.1 -> V3.7.1 for GCC 14.2 compatibility.

assigned to @olivierbichler

assigned to @gregkub

unassigned @olivierbichler

AIDGE_INSTALL variable is a variable used to detect if we are building in the CI environement for building wheels.

I do agree it is weird that it is written as such in the CMake @gregkub why haven't you just set CUDA path in the docker ?

Why not define a specific variable for CI building? I use AIDGE_INSTALL to define the location of my aidge folder, maybe it's a wrong practice

I do agree

This was a typo, it has been corrected here : a26cfae7

We just override the cuda path if we are in CIBuildWheel as we set it up @cmoineau.

assigned to @silvanosky and unassigned @gregkub

requested review from @gregkub

This MR is obsolete, the fixes have either already been applied or in process of being applied . I'll close this MR.

closed