build and test against CUDA 13.1.0

[jameslamb]

Contributes to rapidsai/build-planning#236

Tests that CI here will work with the changes from rapidsai/shared-workflows#483,
switches CUDA 13 builds to CUDA 13.1.0 and adds some CUDA 13.1.0 test jobs.

[jameslamb]

Builds are failing like this:

  -- Could NOT find Doxygen (missing: DOXYGEN_EXECUTABLE) (Required is at least version "1.8.11")
  -- Configuring done (33.6s)
  CMake Error in build/py3-none-linux_x86_64/_deps/cuvs-src/cpp/CMakeLists.txt:
    export called with target "cuvs_static" which requires target "CUTLASS"
    that is not in any export set.

(build link)

Something similar's happening in cuML: rapidsai/cuml#7650 (comment)

The root cause is probably rapidsai/raft#2916

[jakirkham]

Just noting this for visibility (James already knows this), we likely first need PR: rapidsai/cuvs#1686

[jameslamb]

pip devcontainer builds (but only pip devcontainers, not conda or wheels or conda devcontainers) are failing like this:

/usr/bin/sccache /usr/local/cuda/bin/nvcc -forward-unknown-to-host-compiler -ccbin=/usr/bin/g++ ...flags... -c /home/coder/cugraph/cpp/tests/utilities/property_generator_utilities_sg.cu -o tests/CMakeFiles/cugraphtestutil.dir/utilities/property_generator_utilities_sg.cu.o

In file included from /home/coder/cugraph/cpp/src/prims/update_edge_src_dst_property.cuh:7,
from /home/coder/cugraph/cpp/tests/utilities/property_generator_utilities_impl.cuh:8,
from /home/coder/cugraph/cpp/tests/utilities/property_generator_utilities_sg.cu:5:
/home/coder/cugraph/cpp/src/detail/graph_partition_utils.cuh:18:10: fatal error: cuco/hash_functions.cuh: No such file or directory
18 | #include <cuco/hash_functions.cuh>
| ^~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.

(build link)

Probably a bit more fallout from these changes:

• Remove cutlass and cuco dependencies raft#2916
• prefer CUDA 13.1 devcontainers, react to some cutlass removals in RAFT cuvs#1686

I'll try to fix that here.

[jameslamb]

Similar to #5383 (comment)

FAILED: CMakeFiles/cugraph_etl.dir/src/renumbering.cu.o 

/usr/bin/sccache /usr/local/cuda/bin/nvcc ...flags... -c /home/coder/cugraph/cpp/libcugraph_etl/src/renumbering.cu -o CMakeFiles/cugraph_etl.dir/src/renumbering.cu.o

In file included from /home/coder/cugraph/cpp/libcugraph_etl/build/pip/cuda-13.1/release/_deps/cudf-src/cpp/include/cudf/hashing/detail/default_hash.cuh:8,
                 from /home/coder/cugraph/cpp/libcugraph_etl/include/hash/concurrent_unordered_map.cuh:13,
                 from /home/coder/cugraph/cpp/libcugraph_etl/src/renumbering.cu:24:
/home/coder/cugraph/cpp/libcugraph_etl/build/pip/cuda-13.1/release/_deps/cudf-src/cpp/include/cudf/hashing/detail/murmurhash3_x86_32.cuh:16:10: fatal error: cuco/hash_functions.cuh: No such file or directory
   16 | #include <cuco/hash_functions.cuh>
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.

(build link)

libcugraph_etl needs to explicitly declare its usage of cuco::cuco, now that it's not getting it transitively from raft -> cugraph.

[jameslamb]

Just noticed while looking at this file, seems non-controversial 😅

[bdice]

Style checks were blocking (missed a copyright year update). I pushed a fix and triggered CI.

[jameslamb]

Since this will need a rebuild anyway, I cancelled CI to free up some CI resources for the all-of-RAPIDS rebuild happening at https://github.com/rapidsai/workflows/actions/runs/20937345532/job/60163076795

[jameslamb]

/ok to test

[bdice]

[ RUN      ] file_test/Tests_ODShortestDistances_File.CheckInt32Int32Float/0
CMake Error at run_gpu_test.cmake:26 (execute_process):
  execute_process failed command indexes:

    1: "Abnormal exit with child return code: Segmentation fault"

I'm seeing this failure in the CI. I don't know if this is related but I saw a similar error during CCCL 3.2 bringup and fixed an out-of-bounds memory access in this test in #5375. Just mentioning in case it helps with finding a solution.

[jameslamb]

Thanks! I'll also try re-running the wheel test jobs that timed out after 6 hours.

I'll put this up for review and go ask for some help from cuGraph folks.

[jameslamb]

It looks like those same C++ test failures happened on the latest nightly run: https://github.com/rapidsai/cugraph/actions/runs/20947945296/job/60194614337

So probably not related to the changes in this PR.

[seunghwak]

LGTM

[bdice]

/merge

[jameslamb]

Admin-merging, based on offline conversation with @bdice @rlratzel @seunghwak , to unblock at least C++ / devcontainers CI.