linux-armhf additions

[vb216]

Hopefully these fix the lot for arm-hf, seems to build OK with:
mvn install -Djavacpp.platform=linux-armhf -Dplatform=linux-armhf -Dplatform.compiler=arm-linux-gnueabihf-g++ -Djavacpp.platform.compiler=arm-linux-gnueabihf-g++

to build all the presets. Only bit to check, I modified cuda and took out the linux-arm - I'm not sure that would be used? Otherwise needs some way to not try and build it for linux-armhf?

[saudet]

You were not able to build for CUDA with their binaries for ARM? We might want to let them know...

[vb216]

I thought since 6.5 it was only 64 bit?

[saudet]

Oh yeah, looks like they dropped ARM. Ok, that's fine then.

[saudet]

Specifying "linux_x86" will also disable "linux-x86_64" and "linux-x86_64". It should be "linux-x86", with a hyphen, not an underscore.

The idea with FFTW is to build both fftw3 and fftw3f, so if we build both in the cppbuild.sh script (using the --enable-float flag to get fftw3f), we shouldn't need to modify the presets.

Otherwise looks good! When you get sometime to fix the above, I'll be merging this. Thanks

[vb216]

Ah sorry.. looks like a typo, should have spotted that one. Hopefully a CI system checking pull requests will pick that up ;)

Fixed the FFTW one too, looks better with the enable float flag.
Hope thats all OK now - I rebuilt and goes through fine, apart from OpenBLAS that we know about.

[saudet]

Thanks! AFAIK, we need to call the build for FFTW twice to get both libraries. Just do like linux-x86, it should work :) And then we don't need to modify the presets

[vb216]

Done! Rechecked it out + built fftw to check, looks OK. I guess it would have shown up if I did a clean rebuild on the last attempt, hopefully all good now.

[saudet]

Yes, looks good now. Thanks!