Add Linux support for loading libcupti.so.12 and libcupti.so.13

This commit adds support for finding and loading CUPTI libraries on Linux
through cuda.pathfinder. It implements support for all enumerated installation
methods:

- Site-packages: nvidia/cuda_cupti/lib (CUDA 12) and nvidia/cu13/lib (CUDA 13)
- Conda: $CONDA_PREFIX/lib (colocated with other CUDA libraries)
- CTK via CUDA_HOME: $CUDA_HOME/extras/CUPTI/lib64
- CTK via canary probe: system CTK root discovery (similar to nvvm)

Changes:
- Add 'cupti' to supported library names and SONAMEs
- Add site-packages paths for CUDA 12 and 13
- Add cupti to CTK root canary discoverable libraries
- Update find_nvidia_dynamic_lib to handle extras/CUPTI/lib64 path
- Add logic to distinguish CTK (extras/CUPTI/lib64) vs conda (lib) paths
- Update _find_so_using_lib_dir to support versioned libraries via glob
- Add comprehensive mock tests covering all installation methods

Fixes #1572 (Linux support)

Made-with: Cursor

Author: rwgk

cuda_pathfinder/cuda/pathfinder/_dynamic_libs/find_nvidia_dynamic_lib.py

@@ -80,7 +80,7 @@ def _find_dll_using_nvidia_bin_dirs(
 
 def _find_lib_dir_using_anchor_point(libname: str, anchor_point: str, linux_lib_dir: str) -> str | None:
     # Resolve paths for the four cases:
-    #    Windows/Linux x nvvm yes/no
+    #    Windows/Linux x nvvm/cupti yes/no
     if IS_WINDOWS:
         if libname == "nvvm":  # noqa: SIM108
             rel_paths = [
@@ -93,8 +93,15 @@ def _find_lib_dir_using_anchor_point(libname: str, anchor_point: str, linux_lib_
                 "bin",  # CTK 12
             ]
     else:
-        if libname == "nvvm":  # noqa: SIM108
+        if libname == "nvvm":
             rel_paths = ["nvvm/lib64"]
+        elif libname == "cupti":
+            # cupti is in extras/CUPTI/lib64 for CTK, but in lib for conda
+            # Check if this looks like a CTK root (has extras/CUPTI) or conda (uses linux_lib_dir)
+            if os.path.isdir(os.path.join(anchor_point, "extras", "CUPTI")):
+                rel_paths = ["extras/CUPTI/lib64"]
+            else:
+                rel_paths = [linux_lib_dir]
         else:
             rel_paths = [linux_lib_dir]
 
@@ -128,7 +135,12 @@ def _find_so_using_lib_dir(
     so_name = os.path.join(lib_dir, so_basename)
     if os.path.isfile(so_name):
         return so_name
-    error_messages.append(f"No such file: {so_name}")
+    # Look for a versioned library using glob (e.g., libcupti.so.12, libcupti.so.13)
+    file_wild = so_basename + "*"
+    for so_name in sorted(glob.glob(os.path.join(lib_dir, file_wild))):
+        if os.path.isfile(so_name):
+            return so_name
+    error_messages.append(f"No such file: {file_wild}")
     attachments.append(f'  listdir("{lib_dir}"):')
     if not os.path.isdir(lib_dir):
         attachments.append("    DIRECTORY DOES NOT EXIST")

cuda_pathfinder/cuda/pathfinder/_dynamic_libs/supported_nvidia_libs.py

@@ -51,6 +51,7 @@
 SUPPORTED_LIBNAMES_LINUX_ONLY = (
     "cufile",
     # "cufile_rdma",  # Requires libmlx5.so
+    "cupti",
 )
 SUPPORTED_LIBNAMES_LINUX = SUPPORTED_LIBNAMES_COMMON + SUPPORTED_LIBNAMES_LINUX_ONLY
 
@@ -201,6 +202,10 @@
         "libnvrtc.so.13",
     ),
     "nvvm": ("libnvvm.so.4",),
+    "cupti": (
+        "libcupti.so.12",
+        "libcupti.so.13",
+    ),
 }
 SUPPORTED_LINUX_SONAMES_OTHER = {
     "cublasmp": ("libcublasmp.so.0",),
@@ -363,7 +368,7 @@
 #   CTK root in an isolated child process.
 # - discoverable libs: libs that are allowed to use the CTK-root canary fallback.
 _CTK_ROOT_CANARY_ANCHOR_LIBNAMES = ("cudart",)
-_CTK_ROOT_CANARY_DISCOVERABLE_LIBNAMES = ("nvvm",)
+_CTK_ROOT_CANARY_DISCOVERABLE_LIBNAMES = ("nvvm", "cupti")
 
 # Based on output of toolshed/make_site_packages_libdirs_linux.py
 SITE_PACKAGES_LIBDIRS_LINUX_CTK = {
@@ -395,6 +400,7 @@
     "nvjpeg": ("nvidia/cu13/lib", "nvidia/nvjpeg/lib"),
     "nvrtc": ("nvidia/cu13/lib", "nvidia/cuda_nvrtc/lib"),
     "nvvm": ("nvidia/cu13/lib", "nvidia/cuda_nvcc/nvvm/lib64"),
+    "cupti": ("nvidia/cu13/lib", "nvidia/cuda_cupti/lib"),
 }
 SITE_PACKAGES_LIBDIRS_LINUX_OTHER = {
     "cublasmp": ("nvidia/cublasmp/cu13/lib", "nvidia/cublasmp/cu12/lib"),