Merge branch 'main' into nvml-example

Author: mdboom

AGENTS.md

@@ -23,6 +23,12 @@ guide for package-specific conventions and workflows.
   (search), `read_file`, `list_dir`, `glob_file_search`, `apply_patch`,
   `todo_write/update_plan`. Use `cmd`/`run_terminal_cmd` only when no listed
   tool can perform the action.
+- If `pixi` is available for this repo, prefer `pixi run ...` or the matching
+  `pixi` task over invoking raw `python`, `pytest`, `pip`, or similar tools
+  directly so commands run in the repository-managed environment.
+- When extracting or transforming JSON in shell workflows, prefer `jq` over
+  one-off Python parsing. For `gh` commands that return JSON, prefer the
+  built-in `--jq` flag instead of piping the output into `python`.
 - When multiple tool calls can be parallelized (e.g., todo updates with other
   actions, file searches, reading files), make these tool calls in parallel
   instead of sequential. Avoid single calls that might not yield a useful

cuda_bindings/cuda/bindings/_internal/utils.pyx

@@ -120,7 +120,14 @@ cdef int get_nested_resource_ptr(nested_resource[ResT] &in_out_ptr, object obj,
         nested_ptr.reset(nested_vec, True)
         for i, obj_i in enumerate(obj):
             if ResT is char:
-                obj_i_bytes = (<str?>(obj_i)).encode()
+                obj_i_type = type(obj_i)
+                if obj_i_type is str:
+                    obj_i_bytes = obj_i.encode("utf-8")
+                elif obj_i_type is bytes:
+                    obj_i_bytes = obj_i
+                else:
+                    raise TypeError(
+                        f"Expected str or bytes, got {obj_i_type.__name__}")
                 str_len = <size_t>(len(obj_i_bytes)) + 1  # including null termination
                 deref(nested_res_vec)[i].resize(str_len)
                 obj_i_ptr = <char*>(obj_i_bytes)

cuda_bindings/cuda/bindings/nvml.pyx

@@ -27471,3 +27471,9 @@ cpdef str vgpu_type_get_name(unsigned int vgpu_type_id):
         __status__ = nvmlVgpuTypeGetName(<nvmlVgpuTypeId_t>vgpu_type_id, vgpu_type_name, <unsigned int*>size)
     check_status(__status__)
     return cpython.PyUnicode_FromStringAndSize(vgpu_type_name, size[0])
+
+
+# Cleanup some docstrings that don't parse as rst.
+device_get_virtualization_mode.__doc__ = device_get_virtualization_mode.__doc__.replace("NVML_GPU_VIRTUALIZATION_?", "``NVML_GPU_VIRTUALIZATION_?``")
+device_set_virtualization_mode.__doc__ = device_set_virtualization_mode.__doc__.replace("NVML_GPU_VIRTUALIZATION_?", "``NVML_GPU_VIRTUALIZATION_?``")
+GpmMetricId.GPM_METRIC_DRAM_BW_UTIL.__doc__ = "Percentage of DRAM bw used vs theoretical maximum. ``0.0 - 100.0 *\u200d/``."

cuda_bindings/docs/source/api.rst

@@ -16,4 +16,5 @@ CUDA Python API Reference
    module/nvvm
    module/nvfatbin
    module/cufile
+   module/nvml
    module/utils

cuda_bindings/docs/source/conf.py

@@ -37,10 +37,10 @@
     "sphinx.ext.napoleon",
     "sphinx.ext.intersphinx",
     "myst_nb",
-    "enum_tools.autoenum",
     "sphinx_copybutton",
     "release_toc",
     "release_date",
+    "enum_documenter",
 ]
 
 nb_execution_mode = "off"