Merge branch 'main' of https://github.com/rapidsai/cugraph into enh_fix_mg_test_failures

Author: seunghwak

.github/workflows/test.yaml

@@ -50,6 +50,8 @@ jobs:
       date: ${{ inputs.date }}
       script: ci/test_python.sh
       sha: ${{ inputs.sha }}
+      # Skip failing tests on RTX PRO 6000 (Blackwell). xref: https://github.com/rapidsai/cugraph/issues/5421
+      matrix_filter: map(select(.GPU != "rtxpro6000"))
   wheel-tests-pylibcugraph:
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@main
@@ -68,3 +70,5 @@ jobs:
       date: ${{ inputs.date }}
       sha: ${{ inputs.sha }}
       script: ci/test_wheel_cugraph.sh
+      # Skip failing tests on RTX PRO 6000 (Blackwell). xref: https://github.com/rapidsai/cugraph/issues/5421
+      matrix_filter: map(select(.GPU != "rtxpro6000"))

…marks/shared/build_cugraph_ucx/README.MD …marks/shared/build_cugraph_ucx/README.mdbenchmarks/shared/build_cugraph_ucx/README.MD renamed to benchmarks/shared/build_cugraph_ucx/README.md benchmarks/shared/build_cugraph_ucx/README.MD renamed to benchmarks/shared/build_cugraph_ucx/README.md

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2022-2025, NVIDIA CORPORATION.
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION.
 # SPDX-License-Identifier: Apache-2.0
 
 # rdma build instructions from here:
@@ -47,7 +47,7 @@ RUN conda config --set ssl_verify false
 RUN conda install -c gpuci gpuci-tools
 RUN gpuci_conda_retry install -c conda-forge mamba
 
-RUN gpuci_mamba_retry install -y -c pytorch -c rapidsai-nightly -c rapidsai -c conda-forge -c nvidia \
+RUN gpuci_mamba_retry install -y -c rapidsai-nightly -c rapidsai -c conda-forge \
     cugraph=$RAPIDS_VER \
     pytorch=$PYTORCH_VER \
     python=$PYTHON_VER \
@@ -59,8 +59,6 @@ RUN chmod 744 /root/build-ucx.sh & bash /root/build-ucx.sh
 
 ADD test_client_bandwidth.py  /root/test_client_bandwidth.py
 RUN chmod 777 /root/test_client_bandwidth.py
-ADD test_cugraph_sampling.py  /root/test_cugraph_sampling.py
-RUN chmod 777 /root/test_cugraph_sampling.py
 
 ENV PATH /opt/conda/env/bin:$PATH
 WORKDIR /root/

benchmarks/shared/build_cugraph_ucx/cugraph_ucx.dockerfile

@@ -32,6 +32,7 @@ dependencies:
 - libcusolver-dev
 - libcusparse-dev
 - libcuvs==26.4.*,>=0.0.0a0
+- libnvjitlink-dev
 - libraft==26.4.*,>=0.0.0a0
 - librmm==26.4.*,>=0.0.0a0
 - nbsphinx

conda/environments/all_cuda-131_arch-aarch64.yaml

@@ -32,6 +32,7 @@ dependencies:
 - libcusolver-dev
 - libcusparse-dev
 - libcuvs==26.4.*,>=0.0.0a0
+- libnvjitlink-dev
 - libraft==26.4.*,>=0.0.0a0
 - librmm==26.4.*,>=0.0.0a0
 - nbsphinx

conda/environments/all_cuda-131_arch-x86_64.yaml

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2018-2025, NVIDIA CORPORATION.
+# SPDX-FileCopyrightText: Copyright (c) 2018-2026, NVIDIA CORPORATION.
 # SPDX-License-Identifier: Apache-2.0
 schema_version: 1
 
@@ -85,6 +85,9 @@ cache:
       - libcurand-dev
       - libcusolver-dev
       - libcusparse-dev
+      - if: cuda_major == "13"
+        then:
+          - libnvjitlink-dev
 
 outputs:
   - package:
@@ -114,6 +117,9 @@ outputs:
         - libcurand-dev
         - libcusolver-dev
         - libcusparse-dev
+        - if: cuda_major == "13"
+          then:
+            - libnvjitlink-dev
       run:
         - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="x") }}
         - libraft =${{ minor_version }}
@@ -124,6 +130,9 @@ outputs:
         - libcurand
         - libcusolver
         - libcusparse
+        - if: cuda_major == "13"
+          then:
+            - libnvjitlink
       ignore_run_exports:
         by_name:
           - cuda-nvtx
@@ -137,6 +146,9 @@ outputs:
           - libucxx
           - nccl
           - openmpi
+          - if: cuda_major == "13"
+            then:
+              - libnvjitlink
     about:
       homepage: ${{ load_from_file("python/libcugraph/pyproject.toml").project.urls.Homepage }}
       license: ${{ load_from_file("python/libcugraph/pyproject.toml").project.license }}
@@ -183,6 +195,9 @@ outputs:
           - libucxx
           - nccl
           - openmpi
+          - if: cuda_major == "13"
+            then:
+              - libnvjitlink
     about:
       homepage: ${{ load_from_file("python/libcugraph/pyproject.toml").project.urls.Homepage }}
       license: ${{ load_from_file("python/libcugraph/pyproject.toml").project.license }}
@@ -228,6 +243,9 @@ outputs:
           - libucxx
           - nccl
           - openmpi
+          - if: cuda_major == "13"
+            then:
+              - libnvjitlink
     about:
       homepage: ${{ load_from_file("python/libcugraph/pyproject.toml").project.urls.Homepage }}
       license: ${{ load_from_file("python/libcugraph/pyproject.toml").project.license }}

conda/recipes/libcugraph/recipe.yaml

@@ -1,5 +1,5 @@
 /*
- * SPDX-FileCopyrightText: Copyright (c) 2024-2025, NVIDIA CORPORATION.
+ * SPDX-FileCopyrightText: Copyright (c) 2024-2026, NVIDIA CORPORATION.
  * SPDX-License-Identifier: Apache-2.0
  */
 
@@ -88,10 +88,10 @@ normalize_biases(raft::handle_t const& handle,
     // be part of the upper-most range.  We'll compute the last non-zero value in the
     // gpu_biases array here and below we will fill it with a value larger than 1.0
     size_t trailing_zeros = cuda::std::distance(
-      thrust::make_reverse_iterator(gpu_biases->end()),
+      cuda::std::make_reverse_iterator(gpu_biases->end()),
       thrust::find_if(handle.get_thrust_policy(),
-                      thrust::make_reverse_iterator(gpu_biases->end()),
-                      thrust::make_reverse_iterator(gpu_biases->begin()),
+                      cuda::std::make_reverse_iterator(gpu_biases->end()),
+                      cuda::std::make_reverse_iterator(gpu_biases->begin()),
                       cuda::proclaim_return_type<bool>(
                         [] __device__(weight_t bias) { return bias > weight_t{0}; })));
 

cpp/src/sampling/negative_sampling_impl.cuh

@@ -252,6 +252,16 @@ dependencies:
           - libcurand-dev
           - libcusolver-dev
           - libcusparse-dev
+    specific:
+      - output_types: [conda]
+        matrices:
+          - matrix:
+              cuda: "13.*"
+            packages:
+              - libnvjitlink-dev
+          - matrix:
+              cuda: "12.*"
+            packages:
   common_build:
     common:
       - output_types: [conda, requirements, pyproject]