OS: Ubuntu 24.04.4 LTS
Kernel: Linux 6.17.0-1029-nvidia
When installing the Python version of clij2-fft via pip, it downloads some precompiled libraries -- libclFFT.so.2 and libclij2fft.so. When attempting to do this on a linux-aarch64 system, it will pull the x86_64-specific versions of those libraries, which do not run. Similarly, the clijx-deconvolution update site for Fiji provides a plugin JAR without ARM-compatible libraries and fails when it attempts to run after options have been entered.
I was able to work around the pip issue via installing an ARM version of libclfft2 via apt and then made some modifications to native/clij2fft/cppbuild.sh to compile a native libclij2fft.so (I could only get the build to work via specifying an absolute path to the system library -- this will likely need revision). From there, I could replace the pip-installed versions in my environment and run the examples with no issue.
Modifications to pom.xml from there allowed for a working build of the ImageJ/Fiji plugin that works with your provided examples; I also added a case to javacppbuild.sh to allow for combined native compilation and plugin building in this situation.
I've documented my changes at https://github.com/tanadx/clij2-fft, and included a copy of precompiled Linux ARM native libraries. Unfortunately, as I am unfamiliar with GitHub Actions, I do not know if these changes will be compatible with your automatic build system. I also do not know what modifications will be required for pip to download the correct versions of libclFFT.so.2 and libclij2fft.so based on architecture. As a result, I would like to request that you consider addition of linux-aarch64 compatibility to both the Python package and the ImageJ plugin.
OS: Ubuntu 24.04.4 LTS
Kernel: Linux 6.17.0-1029-nvidia
When installing the Python version of
clij2-fftviapip, it downloads some precompiled libraries --libclFFT.so.2andlibclij2fft.so. When attempting to do this on alinux-aarch64system, it will pull the x86_64-specific versions of those libraries, which do not run. Similarly, theclijx-deconvolutionupdate site for Fiji provides a plugin JAR without ARM-compatible libraries and fails when it attempts to run after options have been entered.I was able to work around the
pipissue via installing an ARM version oflibclfft2viaaptand then made some modifications tonative/clij2fft/cppbuild.shto compile a nativelibclij2fft.so(I could only get the build to work via specifying an absolute path to the system library -- this will likely need revision). From there, I could replace thepip-installed versions in my environment and run the examples with no issue.Modifications to
pom.xmlfrom there allowed for a working build of the ImageJ/Fiji plugin that works with your provided examples; I also added a case tojavacppbuild.shto allow for combined native compilation and plugin building in this situation.I've documented my changes at https://github.com/tanadx/clij2-fft, and included a copy of precompiled Linux ARM native libraries. Unfortunately, as I am unfamiliar with GitHub Actions, I do not know if these changes will be compatible with your automatic build system. I also do not know what modifications will be required for
pipto download the correct versions oflibclFFT.so.2andlibclij2fft.sobased on architecture. As a result, I would like to request that you consider addition oflinux-aarch64compatibility to both the Python package and the ImageJ plugin.