Compiling errors using ROCM7.0.x and clang 20.0.x

[denisbertini]

Trying to compile picongpu (dev branch ) example with ROCM7.0 i got the following compiling errors ( in attachment in this post )
Any idea what can go wrong here?

errors.txt

[psychocoderHPC]

Thanks for reporting the issue. The issue is in mallocMC

https://github.com/alpaka-group/mallocMC/blob/be3ac39bae74c61f0ff4283d7967f328c235048d/include/mallocMC/creationPolicies/Scatter.hpp#L1297

We use the template keyword without calling the method with a explicit template argument.
The fix gales some days to propergate into piconhpu because we need to fix first mallocMC and than update the subtree in PIConGPU.

[chillenzer]

Relatedly, FYI: llvm/llvm-project#94194

[chillenzer]

This should fix it: alpaka-group/mallocMC#295

It will take a day or two to make it into PIConGPU mainline. If you're eager to get going, you can copy the content from that PR into your thirdParty/mallocMC or use the corresponding CMake flags to point your build to a separate clone of mallocMC. (Would also be nice to get the feedback before merging.)

[denisbertini]

still compilation problems

In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/plugins/EnergyFields.x.cpp:22:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/defines.hpp:24:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../pmacc/types.hpp:33:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/thirdParty/mallocMC/include/mallocMC/mallocMC.hpp:43:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/thirdParty/mallocMC/include/mallocMC/mallocMC_hostclass.hpp:31:
/lustre/rz/dbertini/hydra/pog/picongpu/thirdParty/mallocMC/include/mallocMC/allocator.hpp:143:70: warning: volatile-qualified parameter type 'volatile size_t' (aka 'volatile unsigned long') is deprecated [-Wdeprecated-volatile]
  143 |         alloc(AlpakaDevice& dev, AlpakaQueue& queue, volatile size_t size)
      |                                                                      ^
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/plugins/BinEnergyParticles.x.cpp:22:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/algorithms/Gamma.hpp:22:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/algorithms/Gamma.def:22:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/defines.hpp:24:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../pmacc/types.hpp:33:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/thirdParty/mallocMC/include/mallocMC/mallocMC.hpp:43:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/thirdParty/mallocMC/include/mallocMC/mallocMC_hostclass.hpp:31:
/lustre/rz/dbertini/hydra/pog/picongpu/thirdParty/mallocMC/include/mallocMC/allocator.hpp:143:70: warning: volatile-qualified parameter type 'volatile size_t' (aka 'volatile unsigned long') is deprecated [-Wdeprecated-volatile]
  143 |         alloc(AlpakaDevice& dev, AlpakaQueue& queue, volatile size_t size)
      |                                                                      ^
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/plugins/CountParticles.x.cpp:22:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/defines.hpp:24:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../pmacc/types.hpp:33:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/thirdParty/mallocMC/include/mallocMC/mallocMC.hpp:43:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/thirdParty/mallocMC/include/mallocMC/mallocMC_hostclass.hpp:31:
/lustre/rz/dbertini/hydra/pog/picongpu/thirdParty/mallocMC/include/mallocMC/allocator.hpp:143:70: warning: volatile-qualified parameter type 'volatile size_t' (aka 'volatile unsigned long') is deprecated [-Wdeprecated-volatile]
  143 |         alloc(AlpakaDevice& dev, AlpakaQueue& queue, volatile size_t size)
      |                                                                      ^
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/plugins/EnergyFields.x.cpp:22:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/defines.hpp:37:
/lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/param/mallocMC.param:42:39: error: no member named 'FlatterScatterAlloc' in namespace 'mallocMC::CreationPolicies'
   42 |         = mallocMC::CreationPolicies::FlatterScatterAlloc::DefaultHeapConfig<blockSize, pageSize, wastefactor>;
      |           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/plugins/CountParticles.x.cpp:22:
In file included from /lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/defines.hpp:37:
/lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/param/mallocMC.param:42:39: error: no member named 'FlatterScatterAlloc' in namespace 'mallocMC::CreationPolicies'
   42 |         = mallocMC::CreationPolicies::FlatterScatterAlloc::DefaultHeapConfig<blockSize, pageSize, wastefactor>;
      |           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
/lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/param/mallocMC.param:46:37: error: no member named 'FlatterScatter' in namespace 'mallocMC::CreationPolicies'
   46 |         mallocMC::CreationPolicies::FlatterScatter<DeviceHeapConfig>,
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
/lustre/rz/dbertini/hydra/pog/picongpu/include/picongpu/../picongpu/param/mallocMC.param:46:37: error: no member named 'FlatterScatter' in namespace 'mallocMC::CreationPolicies'
   46 |         mallocMC::CreationPolicies::FlatterScatter<DeviceHeapConfig>,
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^

[chillenzer]

Thanks for the quick feedback. Will take a look!

[chillenzer]

Prematurely closed by "closing" keyword. This is not yet done.

[chillenzer]

Sorry for the delay, having a closer look at the errors you've posted, that can't possibly be right. I was already wondering about the volatile warning which stems from this line but was fixed last year already. The actual errors make no sense because the namespace and the struct both exist. Could it be that you have checkout out the dev branch of my fork (which is unused and points to an ancient version of the code) instead of the PR branch's version? By now, the fix is mainline (in mallocMC), so you can simply checkout mainline/dev of mallocMC for testing.

[denisbertini]

yes ok i will try again with the picongpu updated dev branch

[denisbertini]

is it enough to checkout the main picongpu directory or should i checkout precisely the mallocMC dir?

[chillenzer]

Thanks! Sorry again for the confusion. Just to be extra sure: The fix is not yet mainline in PIConGPU. You still have to patch in the new mallocMC version.

git clone https://github.com/ComputationalRadiationPhysics/picongpu
cd picongpu/thirdParty
rm mallocMC -rf
# quick and dirty, you'd never want to put a repo into another repo!
git clone https://github.com/alpaka-group/mallocMC

Should be roughly what you want (UNTESTED!).

[denisbertini]

I have done this but same errors at the compilation ( see attachment )

errors2.txt

[denisbertini]

i rolled back to ROCM 6.2 and all of the sudden everything compiled and work on AMD GPUs. So i confirm that the latest PiconGPU base code ( cf dev branch ) still has compilation issue with the latest ROCM 7.0.1 library