Hip updates main

[StewMH]

As proposed by @krasznaa in acts-project/vecmem#261, extend the host and device macros for HIP.

Clang wanted TRACCC_HOST_DEVICE to be added to the function definition as well in a couple of places. I could not get compilation to work with gcc. Wasn't clear to me if this was supported at all.

I now get as far as a puzzling linker error:

[100%] Linking CXX executable ../../../bin/traccc_seeding_example_alpaka
/cvmfs/sft.cern.ch/lcg/releases/binutils/2.39-393d1/x86_64-centos7/bin/ld: CMakeFiles/traccc_seeding_example_alpaka.dir/seeding_example_alpaka.cpp.o: relocation R_X86_64_32S against symbol `_ZTVN6vecmem4copyE' can not be used when making a PIE object; recompile with -fPIE
/cvmfs/sft.cern.ch/lcg/releases/binutils/2.39-393d1/x86_64-centos7/bin/ld: failed to set dynamic section sizes: bad value

but I think this is separate and that these changes are safe.

[krasznaa]

I am very surprised about the TRACCC_HOST_DEVICE additions. 😕 These functions are clearly not needed by the CUDA code. (Otherwise that build would've failed.) So how are they needed by Alpaka when using the HIP backend? 😕

I knew for a while already that a couple more "common functions" will need to become TRACCC_HOST_AND_DEVICE from their current TRACCC_DEVICE status when finally we pick those up with Kokkos and Alpaka. But I did not expect any function, which was so far not marked as having to be a "device function" at all, to need such additions.

I guess I'll need to have a closer look at your Alpaka code... 🤔

[StewMH]

Hi Attila,

I can reproduce the host device issue standalone:
https://godbolt.org/z/qcrcM5E85

You can see that rocm needs the declaration to match the definition (i.e. __host__ and __device__ in both), while nvcc does not.

[krasznaa]

Hi Attila,

I can reproduce the host device issue standalone: https://godbolt.org/z/qcrcM5E85

You can see that rocm needs the declaration to match the definition (i.e. __host__ and __device__ in both), while nvcc does not.

Ahh... I didn't appreciate that what you're doing here is to fix the inconsistency between the declaration and implementation of these functions. 🤔 Yes, that's definitely a bug in our current code...

[krasznaa]

As long as you run the changes through clang-format, I'll be okay with this. 😉

I.e.

docker run -it --rm -v $PWD:$PWD -w $PWD --user "$(id -u):$(id -g)" ghcr.io/acts-project/format10:v11
./traccc/.github/check_format.sh ./traccc/

[StewMH]

@krasznaa should I merge from main again?

[StewMH]

Merged into main to try out my new CI bridge privileges in any case

[krasznaa]

My own preference is to use rebasing in such cases, but this works fine as well...

[beomki-yeo]

Looks good

[beomki-yeo]

I don't know how to remove the WIP block :(