enforce a floor on libnvjitlink, build wheels with CUDA 13.0.x, test wheels against mix of CTK versions (#1862)

The changes from #1405 introduced linking against nvJitLink. nvJitLink has versioned symbols that are added in each new CTK release, and some of those are exposed in `libcuvs.so`.

`libcuvs` wheels are built against the latest CTK supported in RAPIDS (CUDA 13.1.1 as of this writing), so when those wheels are used in environments with older nvJitLink, runtime errors like this can happen:

> libcugraph.so: undefined symbol: __nvJitLinkGetErrorLog_13_1, version libnvJitLink.so.13

For more details, see rapidsai/cugraph#5443

This tries to fix that.

Contributes to rapidsai/build-planning#257

* builds CUDA 13 wheels with the 13.0 CTK
* ensures CUDA 13 wheels ship with a runtime dependency of `nvidia-nvjitlink>={whatever-minor-version-they-were-built-against}`

Contributes to rapidsai/build-planning#256

* updates wheel tests to cover a range of CTK versions (we previously, accidentally, were only testing the latest 12.x and 13.x)

Other changes

* ensures conda packages also take on floors of `libnvjitlink>={whatever-minor-version-they-were-built-against}`

## Notes for Reviewers

### How I tested this

This uses wheels from similar PRs from RAPIDS dependencies, at build and test time:

* NVIDIA/raft#2971
* rapidsai/rmm#2270
* rapidsai/ucxx#604

### Other Options

1. avoiding those versioned symbols with a build-time shim (#1855 does this, but hasn't been successful yet)
2. statically linking libnvJitLink (hasn't been successful yet)

Authors:
  - James Lamb (https://github.com/jameslamb)

Approvers:
  - Gil Forsyth (https://github.com/gforsyth)

URL: #1862

Author: jameslamb

ci/test_wheel_cuvs.sh

@@ -13,8 +13,18 @@ RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen "${RAPIDS_CUDA_VERSION}")"
 LIBCUVS_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcuvs_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
 CUVS_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" cuvs --stable --cuda "$RAPIDS_CUDA_VERSION")")
 
-# echo to expand wildcard before adding `[extra]` requires for pip
+# generate constraints (possibly pinning to oldest support versions of dependencies)
+rapids-generate-pip-constraints test_python "${PIP_CONSTRAINT}"
+
+# notes:
+#
+#   * echo to expand wildcard before adding `[test]` requires for pip
+#   * just providing --constraint="${PIP_CONSTRAINT}" to be explicit, and because
+#     that environment variable is ignored if any other --constraint are passed via the CLI
+#
 rapids-pip-retry install \
+    --prefer-binary \
+    --constraint "${PIP_CONSTRAINT}" \
     "${LIBCUVS_WHEELHOUSE}"/libcuvs*.whl \
     "$(echo "${CUVS_WHEELHOUSE}"/cuvs*.whl)[test]"
 

conda/recipes/libcuvs/recipe.yaml

@@ -134,9 +134,6 @@ outputs:
         - libcurand
         - libcusolver
         - libcusparse
-        - if: cuda_major == "13"
-          then:
-            - libnvjitlink
       ignore_run_exports:
         by_name:
           - cuda-cudart
@@ -150,9 +147,6 @@ outputs:
           - librmm
           - mkl
           - nccl
-          - if: cuda_major == "13"
-            then:
-              - libnvjitlink
     about:
       homepage: ${{ load_from_file("python/libcuvs/pyproject.toml").project.urls.Homepage }}
       license: ${{ load_from_file("python/libcuvs/pyproject.toml").project.license }}
@@ -203,9 +197,6 @@ outputs:
         - libcurand
         - libcusolver
         - libcusparse
-        - if: cuda_major == "13"
-          then:
-            - libnvjitlink
       ignore_run_exports:
         by_name:
           - cuda-cudart
@@ -219,9 +210,6 @@ outputs:
           - librmm
           - mkl
           - nccl
-          - if: cuda_major == "13"
-            then:
-              - libnvjitlink
     about:
       homepage: ${{ load_from_file("python/libcuvs/pyproject.toml").project.urls.Homepage }}
       license: ${{ load_from_file("python/libcuvs/pyproject.toml").project.license }}
@@ -270,9 +258,6 @@ outputs:
         - libcurand
         - libcusolver
         - libcusparse
-        - if: cuda_major == "13"
-          then:
-            - libnvjitlink
       ignore_run_exports:
         by_name:
           - cuda-cudart
@@ -283,9 +268,6 @@ outputs:
           - libcurand
           - libcusolver
           - libcusparse
-          - if: cuda_major == "13"
-            then:
-              - libnvjitlink
           - librmm
           - mkl
           - nccl
@@ -436,9 +418,6 @@ outputs:
         - libcurand
         - libcusolver
         - libcusparse
-        - if: cuda_major == "13"
-          then:
-            - libnvjitlink
       ignore_run_exports:
         by_name:
           - cuda-cudart
@@ -449,9 +428,6 @@ outputs:
           - libcurand
           - libcusolver
           - libcusparse
-          - if: cuda_major == "13"
-            then:
-              - libnvjitlink
           - librmm
           - mkl
           - nccl

dependencies.yaml

@@ -286,6 +286,8 @@ dependencies:
           - clang==20.1.4
           - clang-tools==20.1.4
           - libclang==20.1.4
+  # 'cuda_version' intentionally does not contain fallback entries... we want
+  # a loud error if an unsupported 'cuda' value is passed
   cuda_version:
     specific:
       - output_types: conda
@@ -314,6 +316,51 @@ dependencies:
               cuda: "13.1"
             packages:
               - cuda-version=13.1
+      - output_types: requirements
+        matrices:
+          # if use_cuda_wheels=false is provided, do not add dependencies on any CUDA wheels
+          # (e.g. for DLFW and pip devcontainers)
+          - matrix:
+              use_cuda_wheels: "false"
+            packages:
+          - matrix:
+              arch: aarch64
+              cuda: "12.2"
+              use_cuda_wheels: "true"
+            packages:
+              # some components (like nvidia-cublas-cu12 and nvidia-cuda-nvcc-cu12) didn't have
+              # aarch64 wheels until CTK 12.3, so allow a slightly looser bound here
+              - cuda-toolkit>=12.2,<12.4
+          - matrix:
+              cuda: "12.2"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.2.*
+          - matrix:
+              cuda: "12.5"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.5.*
+          - matrix:
+              cuda: "12.8"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.8.*
+          - matrix:
+              cuda: "12.9"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.9.*
+          - matrix:
+              cuda: "13.0"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==13.0.*
+          - matrix:
+              cuda: "13.1"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==13.1.*
   cuda:
     common:
       - output_types: [conda]
@@ -337,26 +384,40 @@ dependencies:
             packages:
   cuda_wheels:
     specific:
+      # cuVS needs 'nvJitLink>={whatever-cuvs-was-built-against}' at runtime, and mixing
+      # old-CTK with new-nvJitLink is supported, so maintain 1 entry here per CUDA major.minor
+      # cuVS is built against.
+      #
+      # nvJitLink has historically roughly followed the CTK's minor versioning scheme,
+      # but check at https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
       - output_types: [requirements, pyproject]
         matrices:
+          # skip this whole group if explicitly opting out of CUDA wheels
           - matrix:
-              cuda: "12.*"
-              use_cuda_wheels: "true"
+              use_cuda_wheels: "false"
             packages:
-              - cuda-toolkit[cublas,curand,cusolver,cusparse,nvjitlink]==12.*
+          # We always want nvJitLink >= whatever minor version was available in the build environment
+          # nvJitLink tends to follow the CTK's minor versions, but check at https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
+          #
+          # ref: https://github.com/rapidsai/cudf/pull/20873
           - matrix:
-              cuda: "13.*"
+              cuda: "12.9"
               use_cuda_wheels: "true"
             packages:
-              - cuda-toolkit[cublas,curand,cusolver,cusparse,nvjitlink]==13.*
+              - cuda-toolkit[cublas,curand,cusolver,cusparse]==12.*
+              - nvidia-nvjitlink-cu12>=12.9,<13
           - matrix:
-              use_cuda_wheels: "false"
+              cuda: "13.*"
+              use_cuda_wheels: "true"
             packages:
-          # if no matching matrix selectors passed, list a range
+              - &ctk_cu13 cuda-toolkit[cublas,curand,cusolver,cusparse]==13.*
+              - &nvjitlink_cu13 nvidia-nvjitlink>=13.0,<14
+          # if no matching matrix selectors passed, list the CUDA 13 requirement
           # (just as a source of documentation, as this populates pyproject.toml in source control)
           - matrix:
             packages:
-              - cuda-toolkit[cublas,curand,cusolver,cusparse,nvjitlink]>=12,<14
+              - *ctk_cu13
+              - *nvjitlink_cu13
   depends_on_cupy:
     common:
       - output_types: conda

python/libcuvs/pyproject.toml

@@ -19,9 +19,10 @@ authors = [
 license = "Apache-2.0"
 requires-python = ">=3.11"
 dependencies = [
-    "cuda-toolkit[cublas,curand,cusolver,cusparse,nvjitlink]>=12,<14",
+    "cuda-toolkit[cublas,curand,cusolver,cusparse]==13.*",
     "libraft==26.4.*,>=0.0.0a0",
     "librmm==26.4.*,>=0.0.0a0",
+    "nvidia-nvjitlink>=13.0,<14",
 ] # This list was generated by `rapids-dependency-file-generator`. To make changes, edit ../../dependencies.yaml and run `rapids-dependency-file-generator`.
 classifiers = [
     "Intended Audience :: Developers",