Merge remote-tracking branch 'upstream/main' into auto-release-toc

Author: mdboom

.github/workflows/bandit.yml

@@ -24,6 +24,8 @@ jobs:
 
       - name: Install uv
         uses: astral-sh/setup-uv@d0cc045d04ccac9d8b7881df0226f9e82c39688e  # v6.8.0
+        with:
+          enable-cache: false
 
       - name: Get ignore codes
         id: ignore-codes

.github/workflows/build-wheel.yml

@@ -33,7 +33,6 @@ jobs:
           - "3.11"
           - "3.12"
           - "3.13"
-          - "3.13t"
           - "3.14"
           - "3.14t"
     name: py${{ matrix.python-version }}
@@ -287,9 +286,7 @@ jobs:
           cuda-path: "./cuda_toolkit_prev"
 
       - name: Download cuda.bindings build artifacts from the prior branch
-        if: ${{ matrix.python-version == '3.13t'
-                || matrix.python-version == '3.14'
-                || matrix.python-version == '3.14t' }}
+        if: startsWith(matrix.python-version, '3.14')
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         run: |

.github/workflows/ci.yml

@@ -202,4 +202,28 @@ jobs:
       - doc
     steps:
       - name: Exit
-        run: exit 0
+        run: |
+          # if any dependencies were cancelled, that's a failure
+          #
+          # see https://docs.github.com/en/actions/reference/workflows-and-actions/expressions#always
+          # and https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/collaborating-on-repositories-with-code-quality-features/troubleshooting-required-status-checks#handling-skipped-but-required-checks
+          # for why this cannot be encoded in the job-level `if:` field
+          #
+          # TL; DR: `$REASONS`
+          #
+          # The intersection of skipped-as-success and required status checks
+          # creates a scenario where if you DON'T `always()` run this job, the
+          # status check UI will block merging and if you DO `always()` run and
+          # a dependency is _cancelled_ (due to a critical failure, which is
+          # somehow not considered a failure ¯\_(ツ)_/¯) then the critically
+          # failing job(s) will timeout causing a cancellation here and the
+          # build to succeed which we don't want (originally this was just
+          # 'exit 0')
+          if ${{ needs.test-linux-64.result == 'cancelled' ||
+                 needs.test-linux-aarch64.result == 'cancelled' ||
+                 needs.test-windows.result == 'cancelled' ||
+                 needs.doc.result == 'cancelled' }}; then
+            exit 1
+          else
+            exit 0
+          fi

.pre-commit-config.yaml

@@ -73,11 +73,11 @@ repos:
         files: ^cuda_pathfinder/cuda/.*\.py$  # Exclude tests directory
         args: [--config-file=cuda_pathfinder/pyproject.toml]
 
-  # TODO: Reinstate after the workflows have been corrected.  See #1146
-  # - repo: https://github.com/rhysd/actionlint
-  #   rev: "03d0035246f3e81f36aed592ffb4bebf33a03106"  # frozen: v1.7.7
-  #   hooks:
-  #     - id: actionlint
+  - repo: https://github.com/rhysd/actionlint
+    rev: "03d0035246f3e81f36aed592ffb4bebf33a03106"  # frozen: v1.7.7
+    hooks:
+      - id: actionlint
+        args: ["-shellcheck="]
 
   - repo: https://github.com/MarcoGorelli/cython-lint
     rev: "d9ff7ce99ef4f2ae8fba93079ca9d76c4651d4ac"  # frozen: v0.18.0

ci/test-matrix.json

@@ -14,7 +14,6 @@
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "13.0.2", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
-      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.14", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.14t", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.9",  "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
@@ -27,7 +26,6 @@
       { "ARCH": "arm64", "PY_VER": "3.12", "CUDA_VER": "13.0.2", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
-      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.14", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.14t", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
     ],
@@ -95,8 +93,6 @@
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "t4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "13.0.2", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
-      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.2", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
-      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.14", "CUDA_VER": "13.0.2", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.14", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.14t", "CUDA_VER": "13.0.2", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
@@ -106,9 +102,7 @@
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "11.8.0", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "11.8.0", "LOCAL_CTK": "1", "GPU": "t4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
-      { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
-      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.2", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
-      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.2", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" }
+      { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" }
     ]
   }
 }

cuda_bindings/cuda/bindings/_internal/cufile.pxd

@@ -2,7 +2,7 @@
 #
 # SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 #
-# This code was automatically generated across versions from 12.9.0 to 13.0.2. Do not modify it directly.
+# This code was automatically generated across versions from 12.9.1 to 13.0.2. Do not modify it directly.
 
 from ..cycufile cimport *
 
@@ -18,6 +18,7 @@ cdef CUfileError_t _cuFileBufDeregister(const void* bufPtr_base) except?<CUfileE
 cdef ssize_t _cuFileRead(CUfileHandle_t fh, void* bufPtr_base, size_t size, off_t file_offset, off_t bufPtr_offset) except* nogil
 cdef ssize_t _cuFileWrite(CUfileHandle_t fh, const void* bufPtr_base, size_t size, off_t file_offset, off_t bufPtr_offset) except* nogil
 cdef CUfileError_t _cuFileDriverOpen() except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileDriverClose() except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
 cdef CUfileError_t _cuFileDriverClose_v2() except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
 cdef long _cuFileUseCount() except* nogil
 cdef CUfileError_t _cuFileDriverGetProperties(CUfileDrvProps_t* props) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
@@ -41,7 +42,6 @@ cdef CUfileError_t _cuFileGetParameterString(CUFileStringConfigParameter_t param
 cdef CUfileError_t _cuFileSetParameterSizeT(CUFileSizeTConfigParameter_t param, size_t value) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
 cdef CUfileError_t _cuFileSetParameterBool(CUFileBoolConfigParameter_t param, cpp_bool value) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
 cdef CUfileError_t _cuFileSetParameterString(CUFileStringConfigParameter_t param, const char* desc_str) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
-cdef CUfileError_t _cuFileDriverClose() except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
 cdef CUfileError_t _cuFileGetParameterMinMaxValue(CUFileSizeTConfigParameter_t param, size_t* min_value, size_t* max_value) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
 cdef CUfileError_t _cuFileSetStatsLevel(int level) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
 cdef CUfileError_t _cuFileGetStatsLevel(int* level) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil

cuda_bindings/cuda/bindings/_internal/cufile_linux.pyx

@@ -2,7 +2,7 @@
 #
 # SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 #
-# This code was automatically generated across versions from 12.9.0 to 13.0.2. Do not modify it directly.
+# This code was automatically generated across versions from 12.9.1 to 13.0.2. Do not modify it directly.
 
 from libc.stdint cimport intptr_t, uintptr_t
 import threading
@@ -68,6 +68,7 @@ cdef void* __cuFileBufDeregister = NULL
 cdef void* __cuFileRead = NULL
 cdef void* __cuFileWrite = NULL
 cdef void* __cuFileDriverOpen = NULL
+cdef void* __cuFileDriverClose = NULL
 cdef void* __cuFileDriverClose_v2 = NULL
 cdef void* __cuFileUseCount = NULL
 cdef void* __cuFileDriverGetProperties = NULL
@@ -91,7 +92,6 @@ cdef void* __cuFileGetParameterString = NULL
 cdef void* __cuFileSetParameterSizeT = NULL
 cdef void* __cuFileSetParameterBool = NULL
 cdef void* __cuFileSetParameterString = NULL
-cdef void* __cuFileDriverClose = NULL
 cdef void* __cuFileGetParameterMinMaxValue = NULL
 cdef void* __cuFileSetStatsLevel = NULL
 cdef void* __cuFileGetStatsLevel = NULL
@@ -167,6 +167,13 @@ cdef int __check_or_init_cufile() except -1 nogil:
                 handle = load_library()
             __cuFileDriverOpen = dlsym(handle, 'cuFileDriverOpen')
 
+        global __cuFileDriverClose
+        __cuFileDriverClose = dlsym(RTLD_DEFAULT, 'cuFileDriverClose')
+        if __cuFileDriverClose == NULL:
+            if handle == NULL:
+                handle = load_library()
+            __cuFileDriverClose = dlsym(handle, 'cuFileDriverClose')
+
         global __cuFileDriverClose_v2
         __cuFileDriverClose_v2 = dlsym(RTLD_DEFAULT, 'cuFileDriverClose_v2')
         if __cuFileDriverClose_v2 == NULL:
@@ -328,13 +335,6 @@ cdef int __check_or_init_cufile() except -1 nogil:
                 handle = load_library()
             __cuFileSetParameterString = dlsym(handle, 'cuFileSetParameterString')
 
-        global __cuFileDriverClose
-        __cuFileDriverClose = dlsym(RTLD_DEFAULT, 'cuFileDriverClose')
-        if __cuFileDriverClose == NULL:
-            if handle == NULL:
-                handle = load_library()
-            __cuFileDriverClose = dlsym(handle, 'cuFileDriverClose')
-
         global __cuFileGetParameterMinMaxValue
         __cuFileGetParameterMinMaxValue = dlsym(RTLD_DEFAULT, 'cuFileGetParameterMinMaxValue')
         if __cuFileGetParameterMinMaxValue == NULL:
@@ -462,6 +462,9 @@ cpdef dict _inspect_function_pointers():
     global __cuFileDriverOpen
     data["__cuFileDriverOpen"] = <intptr_t>__cuFileDriverOpen
 
+    global __cuFileDriverClose
+    data["__cuFileDriverClose"] = <intptr_t>__cuFileDriverClose
+
     global __cuFileDriverClose_v2
     data["__cuFileDriverClose_v2"] = <intptr_t>__cuFileDriverClose_v2
 
@@ -531,9 +534,6 @@ cpdef dict _inspect_function_pointers():
     global __cuFileSetParameterString
     data["__cuFileSetParameterString"] = <intptr_t>__cuFileSetParameterString
 
-    global __cuFileDriverClose
-    data["__cuFileDriverClose"] = <intptr_t>__cuFileDriverClose
-
     global __cuFileGetParameterMinMaxValue
     data["__cuFileGetParameterMinMaxValue"] = <intptr_t>__cuFileGetParameterMinMaxValue
 
@@ -656,6 +656,16 @@ cdef CUfileError_t _cuFileDriverOpen() except?<CUfileError_t>CUFILE_LOADING_ERRO
         )
 
 
+cdef CUfileError_t _cuFileDriverClose() except?<CUfileError_t>CUFILE_LOADING_ERROR nogil:
+    global __cuFileDriverClose
+    _check_or_init_cufile()
+    if __cuFileDriverClose == NULL:
+        with gil:
+            raise FunctionNotFoundError("function cuFileDriverClose is not found")
+    return (<CUfileError_t (*)() noexcept nogil>__cuFileDriverClose)(
+        )
+
+
 cdef CUfileError_t _cuFileDriverClose_v2() except?<CUfileError_t>CUFILE_LOADING_ERROR nogil:
     global __cuFileDriverClose_v2
     _check_or_init_cufile()
@@ -887,16 +897,6 @@ cdef CUfileError_t _cuFileSetParameterString(CUFileStringConfigParameter_t param
         param, desc_str)
 
 
-cdef CUfileError_t _cuFileDriverClose() except?<CUfileError_t>CUFILE_LOADING_ERROR nogil:
-    global __cuFileDriverClose
-    _check_or_init_cufile()
-    if __cuFileDriverClose == NULL:
-        with gil:
-            raise FunctionNotFoundError("function cuFileDriverClose is not found")
-    return (<CUfileError_t (*)() noexcept nogil>__cuFileDriverClose)(
-        )
-
-
 cdef CUfileError_t _cuFileGetParameterMinMaxValue(CUFileSizeTConfigParameter_t param, size_t* min_value, size_t* max_value) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil:
     global __cuFileGetParameterMinMaxValue
     _check_or_init_cufile()

cuda_bindings/cuda/bindings/_version.py

@@ -1,4 +1,4 @@
 # SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 
-__version__ = "13.0.2"
+__version__ = "13.0.3"

cuda_bindings/cuda/bindings/cufile.pxd

@@ -2,7 +2,7 @@
 #
 # SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 #
-# This code was automatically generated across versions from 12.9.0 to 13.0.2. Do not modify it directly.
+# This code was automatically generated across versions from 12.9.1 to 13.0.2. Do not modify it directly.
 
 from libc.stdint cimport intptr_t
 
@@ -18,12 +18,7 @@ ctypedef CUfileBatchHandle_t BatchHandle
 ctypedef CUfileError_t Error
 ctypedef cufileRDMAInfo_t RDMAInfo
 ctypedef CUfileFSOps_t FSOps
-ctypedef CUfileOpCounter_t OpCounter
-ctypedef CUfilePerGpuStats_t PerGpuStats
 ctypedef CUfileDrvProps_t DrvProps
-ctypedef CUfileStatsLevel1_t StatsLevel1
-ctypedef CUfileStatsLevel2_t StatsLevel2
-ctypedef CUfileStatsLevel3_t StatsLevel3
 
 
 ###############################################################################
@@ -77,3 +72,15 @@ cpdef str get_parameter_string(int param, int len)
 cpdef set_parameter_size_t(int param, size_t value)
 cpdef set_parameter_bool(int param, bint value)
 cpdef set_parameter_string(int param, intptr_t desc_str)
+cpdef tuple get_parameter_min_max_value(int param)
+cpdef set_stats_level(int level)
+cpdef int get_stats_level() except? 0
+cpdef stats_start()
+cpdef stats_stop()
+cpdef stats_reset()
+cpdef get_stats_l1(intptr_t stats)
+cpdef get_stats_l2(intptr_t stats)
+cpdef get_stats_l3(intptr_t stats)
+cpdef size_t get_bar_size_in_kb(int gpu_ind_ex) except? 0
+cpdef set_parameter_posix_pool_slab_array(intptr_t size_values, intptr_t count_values, int len)
+cpdef get_parameter_posix_pool_slab_array(intptr_t size_values, intptr_t count_values, int len)