Files
muh-bot 2a7ca101d7 feat(cccl): integrate missing CCCL directories — python/, ci/, .agent/, docs/, test/
Sparse-checkout from NVIDIA/cccl main branch to complete cccl_upstream:

Added:
- python/cuda_cccl/ (226 files) — Python bindings for device-level algorithms
  Critical for muh toolchain: cuda.compute.reduce_into, scan, radix_sort, etc.
  Includes 204 .py files with full test coverage for all 27 algorithms
- ci/ (163 files) — Build/test infrastructure
  build_cub.sh, test_cub.sh, build_and_test_targets.sh, matrix.yaml
  Directly maps to our [INFRA-CI] and [INFRA-BUILD] items
- .agent/skills/ (7 files) — NVIDIA's own agent skills for CCCL
  cccl-style/SKILL.md, cccl-test/SKILL.md, sass-diff/SKILL.md
- docs/ (491 files) — Official CCCL documentation
  CI references, CMake guides, Python compute docs, libcudacxx PTX docs
- test/ (12 files) — Top-level integration tests (cuda_smoke, stdpar)
- Root configs: .clang-format, .clang-tidy, CONTRIBUTING.md, pyproject.toml
- CLAUDE.md symlink → AGENTS.md (NVIDIA's standard)

cccl_upstream now mirrors full NVIDIA/cccl structure:
  Before: 42M (cub + thrust + libcudacxx + cudax + c + examples + benchmarks)
  After:  53M (+python +ci +docs +.agent +test +configs)

This completes the CCCL base needed for:
- [muh-bench] items: ci/util/build_and_test_targets.sh for targeted builds
- [CCCL-verify] items: python/cuda_cccl/tests/ as reference implementations
- [CCCL-test] items: ci/test_cub.sh, ci/test_thrust.sh
- Agent workflow: .agent/skills/ for consistent style and test patterns
2026-08-07 02:34:33 +00:00

195 lines
6.0 KiB
Python

# Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
#
#
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
from __future__ import annotations
import functools
from types import new_class
from typing import Any, ClassVar, TypeGuard, Union, cast, get_type_hints
import numpy as np
from . import types
"""
This module provides `gpu_struct`, a factory for producing struct types.
"""
def gpu_struct(
field_dict: Union[dict, np.dtype, type],
name: str = "AnonymousStruct",
):
"""
A factory for creating struct types.
Args:
field_dict
A dictionary, numpy dtype, or annotated class providing the
mapping of field names to data types.
name
The name of the struct type that will be returned.
Returns:
A struct class helpful for writing operations on struct values.
"""
# Handle numpy dtype input
if isinstance(field_dict, np.dtype):
if field_dict.type != np.void or field_dict.fields is None:
field_dict = {}
else:
field_dict = {
name: field_info[0] for name, field_info in field_dict.fields.items()
}
# Handle annotated class (decorator usage)
if isinstance(field_dict, type) and hasattr(field_dict, "__annotations__"):
name = field_dict.__name__
field_dict = get_type_hints(field_dict)
# At this point, field_dict must be a dict
assert isinstance(field_dict, dict)
# Validate field names are valid Python identifiers
for key in field_dict:
if not isinstance(key, str) or not key.isidentifier():
raise ValueError(
f"gpu_struct field name {key!r} is not a valid Python identifier"
)
# Normalize fields for storage on the struct class
field_spec = {}
for key, val in field_dict.items():
if _is_struct_type(val):
field_spec[key] = val
elif isinstance(val, dict):
# Nested struct definition - recursively create inner struct
field_spec[key] = gpu_struct(val, name=key)
else:
field_spec[key] = val
# Create a simple Python class for user-facing struct values
struct_class = cast(type[_Struct], new_class(name, bases=(_Struct,)))
struct_class._field_spec = field_spec
struct_class._type_descriptor = _get_struct_type_descriptor(struct_class) # type: ignore[arg-type]
struct_class.dtype = _get_struct_record_dtype(struct_class) # type: ignore[arg-type]
return struct_class
class _Struct:
"""Internal base class for all gpu_structs."""
_field_spec: ClassVar[dict[str, Any]]
_type_descriptor: ClassVar[types.StructTypeDescriptor]
dtype: ClassVar[np.dtype]
_fields: dict[str, Any]
@classmethod
def _fields_from_args(cls, *args, **kwargs):
field_spec = cls._field_spec
if args and isinstance(args[0], dict):
fields = args[0]
elif args:
assert len(args) == len(field_spec), (
f"Expected {len(field_spec)} arguments, got {len(args)}"
)
fields = dict(zip(field_spec.keys(), args))
else:
fields = kwargs
assert fields.keys() == field_spec.keys()
return {
name: _coerce_value(field_spec[name], fields[name]) for name in field_spec
}
def __init__(self, *args, **kwargs):
"""Supporting construction from positional, keyword, and dict arguments."""
self._fields = self._fields_from_args(*args, **kwargs)
for name, value in self._fields.items():
setattr(self, name, value)
# NumPy array representation:
self._data = np.asarray(_as_numpy_record_value(self))
self.__array_interface__ = self._data.__array_interface__
def _as_numpy_record_value(val) -> np.void:
"""Convert a gpu_struct *value* to a numpy record."""
def _fields_to_tuples(fields_dict: dict[str, Any]) -> tuple[Any, ...]:
return tuple(
_fields_to_tuples(v._fields) if isinstance(v, _Struct) else v
for v in fields_dict.values()
)
return np.void(
_fields_to_tuples(val._fields),
dtype=_get_struct_record_dtype(type(val)), # type: ignore[arg-type]
)
@functools.cache
def _get_struct_record_dtype(struct_class: type) -> np.dtype:
return _get_struct_type_descriptor(struct_class).dtype
def _coerce_value(field_type, value: Any) -> Any:
if isinstance(value, _Struct):
return value
if isinstance(field_type, np.dtype):
return field_type.type(value)
if isinstance(field_type, type) and issubclass(field_type, np.generic):
return field_type(value) # type: ignore[call-arg]
if isinstance(value, tuple):
return field_type(*value)
if isinstance(value, dict):
return field_type(**value)
# field_type is a class (e.g., another gpu_struct)
raise TypeError(f"Cannot coerce {type(value).__name__} into {field_type.__name__}")
def _is_struct_type(typ: Any) -> TypeGuard[type[_Struct]]:
"""Check if a type is a GPU struct class."""
return isinstance(typ, type) and issubclass(typ, _Struct)
@functools.cache
def _get_struct_type_descriptor(
struct_class: type,
) -> types.StructTypeDescriptor:
type_descriptors = _field_spec_to_type_descriptors(
struct_class._field_spec # type: ignore[attr-defined]
)
return types.struct(type_descriptors, name=struct_class.__name__)
def _field_spec_to_type_descriptors(
field_spec: dict[str, Any],
) -> dict[str, types.TypeDescriptor]:
type_descriptors = {}
for key, val in field_spec.items():
if isinstance(val, types.TypeDescriptor):
type_descriptors[key] = val
elif _is_struct_type(val):
type_descriptors[key] = val._type_descriptor
elif isinstance(val, np.dtype):
type_descriptors[key] = types.from_numpy_dtype(val)
else:
type_descriptors[key] = types.from_numpy_dtype(np.dtype(val))
return type_descriptors