Add cufile bindings

.github/actions/fetch_ctk/action.yml

@@ -17,18 +17,37 @@ inputs:
     description: "A list of the CTK components to install as a comma-separated list. e.g. 'cuda_nvcc,cuda_nvrtc,cuda_cudart'"
     required: false
     type: string
-    default: "cuda_nvcc,cuda_cudart,cuda_nvrtc,cuda_profiler_api,cuda_cccl,libnvjitlink"
+    default: "cuda_nvcc,cuda_cudart,cuda_nvrtc,cuda_profiler_api,cuda_cccl,libnvjitlink,libcufile"
 
 runs:
   using: composite
   steps:
     - name: Set up CTK cache variable
       shell: bash --noprofile --norc -xeuo pipefail {0}
       run: |
-        HASH=$(echo -n "${{ inputs.cuda-components }}" | sha256sum | awk '{print $1}')
+        # Pre-process the component list to ensure hash uniqueness
+        CTK_CACHE_COMPONENTS=${{ inputs.cuda-components }}
+        # Conditionally strip out libnvjitlink for CUDA versions < 12
+        CUDA_MAJOR_VER="$(cut -d '.' -f 1 <<< ${{ inputs.cuda-version }})"
+        if [[ "$CUDA_MAJOR_VER" -lt 12 ]]; then
+          CTK_CACHE_COMPONENTS="${CTK_CACHE_COMPONENTS//libnvjitlink/}"
+        fi
+        # Conditionally strip out libcufile since it does not support Windows
+        if [[ "${{ inputs.host-platform }}" == win-* ]]; then
+          CTK_CACHE_COMPONENTS="${CTK_CACHE_COMPONENTS//libcufile/}"
+        fi
+        # Conditionally strip out libcufile for CUDA versions < 12.2.0 + aarch64 (redist not available)
+        CUDA_MINOR_VER="$(cut -d '.' -f 2 <<< ${{ inputs.cuda-version }})"
+        if [[ ("$CUDA_MAJOR_VER" -lt 12 || "$CUDA_MINOR_VER" -lt 2) && "${{ inputs.host-platform }}" == "linux-aarch64" ]]; then
+          CTK_CACHE_COMPONENTS="${CTK_CACHE_COMPONENTS//libcufile/}"
+        fi
+        # Cleanup stray commas after removing components
+        CTK_CACHE_COMPONENTS="${CTK_CACHE_COMPONENTS//,,/,}"
+
+        HASH=$(echo -n "${CTK_CACHE_COMPONENTS}" | sha256sum | awk '{print $1}')
         echo "CTK_CACHE_KEY=mini-ctk-${{ inputs.cuda-version }}-${{ inputs.host-platform }}-$HASH" >> $GITHUB_ENV
         echo "CTK_CACHE_FILENAME=mini-ctk-${{ inputs.cuda-version }}-${{ inputs.host-platform }}-$HASH.tar.gz" >> $GITHUB_ENV
-        echo "CTK_CACHE_COMPONENTS=${{ inputs.cuda-components }}" >> $GITHUB_ENV
+        echo "CTK_CACHE_COMPONENTS=${CTK_CACHE_COMPONENTS}" >> $GITHUB_ENV
 
     - name: Install dependencies
       uses: ./.github/actions/install_unix_deps
@@ -94,12 +113,6 @@ runs:
           rm $CTK_COMPONENT_COMPONENT_FILENAME
         }
 
-        # Conditionally strip out libnvjitlink for CUDA versions < 12
-        if [[ "$(cut -d '.' -f 1 <<< ${{ inputs.cuda-version }})" -lt 12 ]]; then
-          CTK_CACHE_COMPONENTS="${CTK_CACHE_COMPONENTS//libnvjitlink/}"
-        fi
-        # Cleanup stray commas after removing components
-        CTK_CACHE_COMPONENTS="${CTK_CACHE_COMPONENTS//,,/,}"
         # Get headers and shared libraries in place
         for item in $(echo $CTK_CACHE_COMPONENTS | tr ',' ' '); do
             populate_cuda_path "$item"

cuda_bindings/cuda/bindings/_internal/cufile.pxd

@@ -0,0 +1,43 @@
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+#
+# SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
+#
+# This code was automatically generated with version 12.9.0. Do not modify it directly.
+
+from ..cycufile cimport *
+
+
+###############################################################################
+# Wrapper functions
+###############################################################################
+
+cdef CUfileError_t _cuFileHandleRegister(CUfileHandle_t* fh, CUfileDescr_t* descr) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef void _cuFileHandleDeregister(CUfileHandle_t fh) except* nogil
+cdef CUfileError_t _cuFileBufRegister(const void* bufPtr_base, size_t length, int flags) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileBufDeregister(const void* bufPtr_base) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+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_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
+cdef CUfileError_t _cuFileDriverSetPollMode(cpp_bool poll, size_t poll_threshold_size) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileDriverSetMaxDirectIOSize(size_t max_direct_io_size) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileDriverSetMaxCacheSize(size_t max_cache_size) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileDriverSetMaxPinnedMemSize(size_t max_pinned_size) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileBatchIOSetUp(CUfileBatchHandle_t* batch_idp, unsigned nr) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileBatchIOSubmit(CUfileBatchHandle_t batch_idp, unsigned nr, CUfileIOParams_t* iocbp, unsigned int flags) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileBatchIOGetStatus(CUfileBatchHandle_t batch_idp, unsigned min_nr, unsigned* nr, CUfileIOEvents_t* iocbp, timespec* timeout) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileBatchIOCancel(CUfileBatchHandle_t batch_idp) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef void _cuFileBatchIODestroy(CUfileBatchHandle_t batch_idp) except* nogil
+cdef CUfileError_t _cuFileReadAsync(CUfileHandle_t fh, void* bufPtr_base, size_t* size_p, off_t* file_offset_p, off_t* bufPtr_offset_p, ssize_t* bytes_read_p, CUstream stream) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileWriteAsync(CUfileHandle_t fh, void* bufPtr_base, size_t* size_p, off_t* file_offset_p, off_t* bufPtr_offset_p, ssize_t* bytes_written_p, CUstream stream) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileStreamRegister(CUstream stream, unsigned flags) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileStreamDeregister(CUstream stream) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileGetVersion(int* version) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileGetParameterSizeT(CUFileSizeTConfigParameter_t param, size_t* value) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileGetParameterBool(CUFileBoolConfigParameter_t param, cpp_bool* value) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+cdef CUfileError_t _cuFileGetParameterString(CUFileStringConfigParameter_t param, char* desc_str, int len) except?<CUfileError_t>CUFILE_LOADING_ERROR nogil
+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