Prepare for cuda_bindings 3.15 release (#3368)

* git merge --squash forge/ctk-next-public-main-prs/gha-41760261196-1/merge-pr; preserve excluded paths (NO MANUAL CHANGES)

* Update versions

* Update generated files for CTK 3.15

* Update versions of the CTK

* Update release notes

* Revert cuda-core cuda-bindings floor update

* Make the transition of updating the bindings floor work

* Fix type of constants to prevent overflow

* Fix cyclical imports

* Also fix same bug in CUDA_EGL_INFINITE_TIMEOUT

* Update NVML docs

---------

Co-authored-by: cuda-python-private-automation[bot] <cuda-python-private-automation[bot]@users.noreply.github.com>

Author: mdboom

.github/workflows/test-wheel-linux.yml

@@ -203,7 +203,7 @@ jobs:
 
       - name: Download cuda.bindings build artifacts
         if: ${{ (env.TEST_BINDINGS == 'true' || env.TEST_CORE == 'true' || env.TEST_PYTHON == 'true') &&
-                env.BINDINGS_SOURCE == 'main' }}
+                (env.BINDINGS_SOURCE == 'main' || (env.BINDINGS_SOURCE == 'floor' && env.TEST_CORE == 'true')) }}
         uses: $/.github/actions/download-artifact-retry
         with:
           name: ${{ env.CUDA_BINDINGS_ARTIFACT_NAME }}

.github/workflows/test-wheel-windows.yml

@@ -194,7 +194,7 @@ jobs:
 
       - name: Download cuda.bindings build artifacts
         if: ${{ (env.TEST_BINDINGS == 'true' || env.TEST_CORE == 'true' || env.TEST_PYTHON == 'true') &&
-                env.BINDINGS_SOURCE == 'main' }}
+                (env.BINDINGS_SOURCE == 'main' || (env.BINDINGS_SOURCE == 'floor' && env.TEST_CORE == 'true')) }}
         uses: $/.github/actions/download-artifact-retry
         with:
           name: ${{ env.CUDA_BINDINGS_ARTIFACT_NAME }}

.gitignore

@@ -8,8 +8,8 @@ __pycache__/
 
 # CUDA Python specific
 .cache/
-.cython-stdlib
-.cython-bindings
+.cython-stdlib*
+.cython-bindings*
 .lycheecache
 .lychee-rendered/
 .pytest_cache/

.pre-commit-config.yaml

@@ -162,7 +162,7 @@ repos:
     rev: "3e8a8703264a2f4a69428a0aa4dcb512790b2c8c"  # frozen: v6.0.0
     hooks:
     - id: check-added-large-files
-      exclude: '^(?:cuda_bindings_13/cuda/bindings/(?:nvml\.pyx|_v2/driver\.pyx)|cuda_bindings_12/cuda/bindings/(?:driver\.pyx|runtime\.pyx\.in|nvml\.pyx))$'
+      exclude: '^(?:cuda_bindings_13/cuda/bindings/(?:nvml\.pyx|_v2/driver\.pyx)|cuda_bindings_12/cuda/bindings/(?:driver\.pyx|runtime\.pyx\.in|nvml\.pyx)|qa/.*\.svg)$'
     - id: check-case-conflict
     - id: check-docstring-first
     - id: check-merge-conflict

.spdx-ignore

@@ -22,7 +22,5 @@ cuda_core/third_party/**
 # Generated by stubgen-pyx; regenerated on every commit so a header would be lost
 cuda_core/cuda/**/*.pyi
 
-qa/ctk-next.drawio.svg
-
 # Third-party eMudhra disclosure data, kept byte-identical to the published files
 ci/tools/runner_trust_probe_chains/**

ci/test-matrix.yml

@@ -132,6 +132,8 @@ elif [[ "${1}" == "test" ]]; then
     echo "SKIP_CYTHON_TEST=${SKIP_CYTHON_TEST}"
     echo "TEST_CUDA_MAJOR=${TEST_CUDA_MAJOR}"
     echo "TEST_CUDA_MINOR=${TEST_CUDA_MINOR}"
+    echo "BUILD_CUDA_MAJOR=${BUILD_CUDA_MAJOR}"
+    echo "BUILD_CUDA_MINOR=${BUILD_CUDA_MINOR}"
   } >> $GITHUB_ENV
 fi
 

ci/tools/env-vars

@@ -98,7 +98,24 @@ elif [[ "${test_module}" == "core" || "${test_module}" == nightly-* ]]; then
     CORE_WHL_FOR_FLOOR=("${CUDA_CORE_ARTIFACTS_DIR}"/*.whl)
     BINDINGS_FLOOR="$(python ci/tools/cuda_core_bindings_floor.py --wheel "${CORE_WHL_FOR_FLOOR[0]}" --major "${TEST_CUDA_MAJOR}")"
     echo "cuda-bindings floor of ${CORE_WHL_FOR_FLOOR[0]##*/} for CUDA ${TEST_CUDA_MAJOR}: ${BINDINGS_FLOOR}"
-    BINDINGS_ARGS+=("cuda-bindings==${BINDINGS_FLOOR}")
+    BINDINGS_FLOOR_MINOR="$(cut -d '.' -f 2 <<< "${BINDINGS_FLOOR}")"
+    if [[ "${test_module}" == "core" && "${TEST_CUDA_MAJOR}" == "${BUILD_CUDA_MAJOR}" && "${BINDINGS_FLOOR_MINOR}" -lt "${BUILD_CUDA_MINOR}" ]]; then
+      # Every CTK minor bump (see "CUDA Toolkit minor bumps" in
+      # cuda_core/AGENTS.md) opens a window in which the floor's minor is below
+      # the minor the wheel was compiled against. No released cuda-bindings
+      # at the floor can pass cuda.core's header check then, because it was
+      # generated from older headers. Use the cuda-bindings built from this
+      # commit instead. The branch is taken exactly while that window is open
+      # and stops being taken once the floor is bumped to the new minor, so it
+      # needs no per-release edit or revert.
+      echo "Floor ${BINDINGS_FLOOR} predates CUDA ${BUILD_CUDA_MAJOR}.${BUILD_CUDA_MINOR} headers; using the just-built cuda-bindings wheel"
+      BINDINGS_ARGS=("${CUDA_BINDINGS_ARTIFACTS_DIR}"/*.whl)
+      if [[ "${LOCAL_CTK}" != 1 ]]; then
+        BINDINGS_ARGS=("${BINDINGS_ARGS[0]}[all]")
+      fi
+    else
+      BINDINGS_ARGS+=("cuda-bindings==${BINDINGS_FLOOR}")
+    fi
   else
     BINDINGS_ARGS=("${CUDA_BINDINGS_ARTIFACTS_DIR}"/*.whl)
     if [[ "${LOCAL_CTK}" != 1 ]]; then
@@ -128,7 +145,14 @@ elif [[ "${test_module}" == "core" || "${test_module}" == nightly-* ]]; then
     echo "Installing core wheel"
     # Constrain cuda-toolkit to the requested CTK version to avoid
     # pip pulling in a newer nvidia-cuda-runtime that conflicts with it.
-    pip install "${CORE_WHL[@]}" --group "test-cu${TEST_CUDA_MAJOR}" "${TEST_FT_GROUP[@]}" "cuda-toolkit==${TEST_CUDA_MAJOR_MINOR}.*"
+    # Skip the constraint for a local CTK of the build's own (newest) minor:
+    # its toolkit wheels may not be on PyPI yet, and the local CTK is what the
+    # job tests.
+    TOOLKIT_PIN=("cuda-toolkit==${TEST_CUDA_MAJOR_MINOR}.*")
+    if [[ "${LOCAL_CTK}" == 1 && "${TEST_CUDA_MAJOR}" == "${BUILD_CUDA_MAJOR}" && "${TEST_CUDA_MINOR}" == "${BUILD_CUDA_MINOR}" ]]; then
+      TOOLKIT_PIN=()
+    fi
+    pip install "${CORE_WHL[@]}" --group "test-cu${TEST_CUDA_MAJOR}" "${TEST_FT_GROUP[@]}" "${TOOLKIT_PIN[@]}"
     # Optional deps needed by some cuda.core samples exercised in this suite.
     # CuPy already comes from the
     # test-cuXX group; install the rest best-effort so those samples run

ci/tools/run-tests

@@ -8,7 +8,7 @@ backport_branch: "12.9.x"  # keep in sync with target-branch in .github/dependab
 # solely determined by git tags via setuptools_scm.
 cuda:
   build:
-    version: "13.4.2"
+    version: "13.5.0"
   prev_build:
     version: "12.9.1"
 

ci/versions.yml

@@ -1,8 +1,8 @@
 # SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
-# This code was automatically generated across versions from 1.5.0 to 13.4.1. Do not modify it directly.
-# CYTHON-BINDINGS-GENERATED-DO-NOT-MODIFY-THIS-FILE: format=1; content-sha256=6866708da9b12dc99b597cd56196b66a1f2645198821b572326bb0858b5906d3
+# This code was automatically generated across versions from 1.5.0 to 13.5.0. Do not modify it directly.
+# CYTHON-BINDINGS-GENERATED-DO-NOT-MODIFY-THIS-FILE: format=1; content-sha256=3e8adcab381ef60e6cce3139d67c8430f058bd1f9d5fedd28ab42f1897113bef
 
 
 # <<<< PREAMBLE CONTENT >>>>