Compilation error on WarpX 25.08 with CUDA toolkit 13.0

[denisbertini]

Hi there,
trying to compile the latest vesion of WarpX ( 25.08 ) with CUDA toolkit 13.0 i receive this king of compilation errors:

  2%] Building CXX object _deps/fetchedopenpmd-build/CMakeFiles/openPMD.dir/src/Iteration.cpp.o
/tmp/warp/WarpX/build/_deps/fetchedamrex-src/Src/Base/AMReX_GpuReduce.H(429): error: namespace "cub" has no member "Min"
      return BlockReduce(temp_storage).Reduce(source, cub::Min());
                                                           ^

/tmp/warp/WarpX/build/_deps/fetchedamrex-src/Src/Base/AMReX_GpuReduce.H(483): error: namespace "cub" has no member "Max"
      return BlockReduce(temp_storage).Reduce(source, cub::Max());
                                                           ^

/tmp/warp/WarpX/build/_deps/fetchedamrex-src/Src/Base/AMReX_GpuReduce.H(429): error: namespace "cub" has no member "Min"
      return BlockReduce(temp_storage).Reduce(source, cub::Min());
                                                           ^

/tmp/warp/WarpX/build/_deps/fetchedamrex-src/Src/Base/AMReX_GpuReduce.H(429): error: namespace "cub" has no member "Min"
      return BlockReduce(temp_storage).Reduce(source, cub::Min());
                                                           ^

/tmp/warp/WarpX/build/_deps/fetchedamrex-src/Src/Base/AMReX_GpuReduce.H(483): error: namespace "cub" has no member "Max"
      return BlockReduce(temp_storage).Reduce(source, cub::Max());
                                                           ^

/tmp/warp/WarpX/build/_deps/fetchedamrex-src/Src/Base/AMReX_GpuReduce.H(483): error: namespace "cub" has no member "Max"
      return BlockReduce(temp_storage).Reduce(source, cub::Max());
                                                           ^

2 errors detected in the compilation of "/tmp/warp/WarpX/build/_deps/fetchedamrex-src/Src/Base/AMReX_BlockMutex.cpp".
2 errors detected in the compilation of "/tmp/warp/WarpX/build/_deps/fetchedamrex-src/Src/Base/AMReX_BlockMutex.cpp".
2 errors detected in the compilation of "/tmp/warp/WarpX/build/_deps/fetchedamrex-src/Src/Base/AMReX_BlockMutex.cpp".

which seems to be linked to the underlying AMReX code base. Any idea how i can solve that ?

[denisbertini]

it compiles with CUDA toolkit 12 when using

AMReX_CUDA_ARCH=80

[EZoni]

Looking at some documentation online, e.g., https://nvidia.github.io/cccl/cub/api/structcub_1_1DeviceReduce.html, it seems that maybe it should be something like cub::DeviceReduce::Min instead of cub::Min, etc.

@atmyers
It looks like something that should be fixed on the AMReX end, correct?

[atmyers]

I think this fixes it: AMReX-Codes/amrex#4648

[EZoni]

@denisbertini

Could you try after the AMReX fix in AMReX-Codes/amrex#4648?

Until we update the central dependency next week, you would have to modify by hand the AMReX commit that we fetch.

You can do that by setting

    "commit_amrex": "7b690df3062286dd1cbeeec581d56a0e28af792d",

in the file dependencies.json:

warpx/dependencies.json

Line 6 in 738d7bd

"commit_amrex": "25.09",

[denisbertini]

this is now perfeclty working,, thanks !

[EZoni]

Great, next week we will update the central dependency file, so the fix will be available without manual tweaks.