Absolute imports

pyproject.toml

@@ -56,5 +56,8 @@ cunumpy = [ "py.typed", "*.pyi", "LLM_GUIDE.md", "cuda/include/cunumpy/*.cuh" ]
 # agent worktrees and scratch scripts are local copies, not part of the project
 extend-exclude = [ ".claude" ]
 # formatting: `ruff format`; import sorting: the isort rules of `ruff check`
-lint.extend-select = [ "I" ]
+lint.extend-select = [ "I", "TID252" ]
 lint.isort.known-first-party = [ "cunumpy" ]
+
+[tool.ruff.lint.flake8-tidy-imports]
+ban-relative-imports = "all"

src/cunumpy/__init__.py

@@ -3,9 +3,9 @@
 import warnings as _warnings
 from importlib.metadata import PackageNotFoundError, version
 
-from . import algorithms, cuda, kernels, memory, mpi, petsc, profiling, rng, xp
-from ._scipy_backend import scipy
-from .xp import (
+from cunumpy import algorithms, cuda, kernels, memory, mpi, petsc, profiling, rng, xp
+from cunumpy._scipy_backend import scipy
+from cunumpy.xp import (
     as_device_array,
     assert_same_backend,
     backend_info,

src/cunumpy/__init__.pyi

@@ -8,16 +8,16 @@ from typing import Any
 import numpy as np
 from numpy import *
 
-from . import algorithms as algorithms
-from . import cuda as cuda
-from . import kernels as kernels
-from . import memory as memory
-from . import mpi as mpi
-from . import petsc as petsc
-from . import profiling as profiling
-from . import rng as rng
-from . import xp as xp
-from ._scipy_backend import scipy as scipy
+from cunumpy import algorithms as algorithms
+from cunumpy import cuda as cuda
+from cunumpy import kernels as kernels
+from cunumpy import memory as memory
+from cunumpy import mpi as mpi
+from cunumpy import petsc as petsc
+from cunumpy import profiling as profiling
+from cunumpy import rng as rng
+from cunumpy import xp as xp
+from cunumpy._scipy_backend import scipy as scipy
 
 def to_numpy(array: Any) -> np.ndarray: ...
 def to_cupy(array: Any) -> Any: ...

src/cunumpy/_algorithms.py

@@ -8,7 +8,7 @@
 
 import array_api_compat.numpy as np
 
-from .xp import assert_same_backend, get_array_backend, get_array_module
+from cunumpy.xp import assert_same_backend, get_array_backend, get_array_module
 
 
 def _integer_keys(keys: Any) -> tuple[Any, Any]:
@@ -77,7 +77,7 @@ def cell_offsets(sorted_cells: Any, n_cells: int) -> Any:
 def _sum_kernel(dtype: Any) -> Any:
     name = np.dtype(dtype).name
     if name not in _SUM_KERNELS:
-        from ._cuda_kernel import CudaKernel
+        from cunumpy._cuda_kernel import CudaKernel
 
         ctype = "float" if name == "float32" else "double"
         _SUM_KERNELS[name] = CudaKernel(

src/cunumpy/_cuda_kernel.py

@@ -1362,7 +1362,7 @@ def verify_layout(
         RuntimeError
             If CuPy is not available.
         """
-        from .xp import cupy_available
+        from cunumpy.xp import cupy_available
 
         if not cupy_available():
             raise RuntimeError("verify_layout() compiles a CUDA kernel and needs CuPy")
@@ -1768,7 +1768,7 @@ def __host_args__(self) -> Any:
                         "or set host_copies = True for a read-only host evaluation "
                         "from host copies",
                     )
-                from .xp import to_numpy
+                from cunumpy.xp import to_numpy
 
                 value = to_numpy(value)
             host_values.append(value)
@@ -2093,7 +2093,7 @@ def debug_active(self) -> bool:
         """
         if self._debug is not None:
             return self._debug
-        from ._device import get_cuda_debug
+        from cunumpy._device import get_cuda_debug
 
         return get_cuda_debug()
 
@@ -2234,7 +2234,7 @@ def compile(self) -> Any:
             If CuPy or a GPU is not available.
         """
         if self._raw_kernel is None:
-            from .xp import cupy_available
+            from cunumpy.xp import cupy_available
 
             if not cupy_available():
                 raise RuntimeError(
@@ -2403,7 +2403,7 @@ def _opt_in_shared_memory(self, kernel: Any, shared_mem: int) -> None:
         attribute ``max_dynamic_shared_size_bytes``; it is set once (and again
         for a larger request) up to the device's opt-in limit.
         """
-        from ._device import (
+        from cunumpy._device import (
             DEFAULT_SHARED_MEMORY_PER_BLOCK,
             max_shared_memory_per_block,
         )

src/cunumpy/_device.py

@@ -9,8 +9,8 @@
 
 import array_api_compat.numpy as np
 
-from ._mpi_serial import local_rank
-from .xp import array_backend, cupy_available
+from cunumpy._mpi_serial import local_rank
+from cunumpy.xp import array_backend, cupy_available
 
 
 def set_device(device_id: int) -> None:

src/cunumpy/_dispatch.py

@@ -34,16 +34,16 @@
 
 import array_api_compat
 
-from ._cuda_kernel import CudaKernel, _compile_in_threads
-from ._kernel import (
+from cunumpy._cuda_kernel import CudaKernel, _compile_in_threads
+from cunumpy._kernel import (
     CompiledHostKernel,
     HostImplementations,
     PyccelKernel,
     resolve_host_args,
 )
-from ._transfers import _ACTIVE as _COUNTERS
-from ._transfers import _record
-from .xp import get_backend
+from cunumpy._transfers import _ACTIVE as _COUNTERS
+from cunumpy._transfers import _record
+from cunumpy.xp import get_backend
 
 __all__ = ["Kernel", "KernelCatalog"]
 

src/cunumpy/_emulation.py

@@ -55,7 +55,7 @@
 
 import numpy as np
 
-from ._cuda_kernel import (
+from cunumpy._cuda_kernel import (
     CudaKernel,
     CudaParameter,
     _scalar_checker,

src/cunumpy/_fusion.py

@@ -32,7 +32,7 @@ def pressure(rho, T, gamma):
 import array_api_compat
 import numpy
 
-from .xp import use_backend
+from cunumpy.xp import use_backend
 
 __all__ = ["fuse"]
 

src/cunumpy/_kernel.py

@@ -34,9 +34,9 @@
 import array_api_compat
 import numpy as np
 
-from ._transfers import _ACTIVE as _COUNTERS
-from ._transfers import _record
-from .xp import _cupy_backend, _to_cupy, _to_numpy, is_gpu, to_cupy, to_numpy
+from cunumpy._transfers import _ACTIVE as _COUNTERS
+from cunumpy._transfers import _record
+from cunumpy.xp import _cupy_backend, _to_cupy, _to_numpy, is_gpu, to_cupy, to_numpy
 
 __all__ = ["CompiledHostKernel", "KernelArguments", "PyccelKernel", "resolve_host_args"]
 

src/cunumpy/_mirror.py

@@ -18,7 +18,7 @@
 
 import numpy as np
 
-from .xp import _cupy_backend, to_cupy
+from cunumpy.xp import _cupy_backend, to_cupy
 
 __all__ = ["DeviceMirror"]
 

src/cunumpy/_morton.py

@@ -27,7 +27,7 @@
 
 import numpy as np
 
-from ._philox import _module
+from cunumpy._philox import _module
 
 __all__ = [
     "MAX_MORTON_LEVELS",

src/cunumpy/_mpi.py

@@ -11,12 +11,12 @@
 import array_api_compat
 import array_api_compat.numpy as np
 
-from ._mpi_serial import (
+from cunumpy._mpi_serial import (
     _LOCAL_RANK_VARIABLES,  # noqa: F401 - re-exported
 )
-from ._transfers import _ACTIVE as _COUNTERS
-from ._transfers import _describe, _record
-from .xp import array_backend, cupy_available, to_numpy
+from cunumpy._transfers import _ACTIVE as _COUNTERS
+from cunumpy._transfers import _describe, _record
+from cunumpy.xp import array_backend, cupy_available, to_numpy
 
 _logger = logging.getLogger(__name__)
 
@@ -43,7 +43,7 @@ def synchronize_for_mpi(*arrays: Any, stream: Any = None, event: Any = None) ->
     devices = {a.device.id for a in arrays if array_api_compat.is_cupy_array(a)}
     if not devices:
         return
-    from ._streams import HostEvent, HostStream
+    from cunumpy._streams import HostEvent, HostStream
 
     if isinstance(event, HostEvent) or isinstance(stream, HostStream):
         raise TypeError("device buffers require a CUDA producer stream or event")

src/cunumpy/_profiling.py

@@ -9,7 +9,7 @@
 from dataclasses import dataclass
 from types import ModuleType
 
-from .xp import array_backend, synchronize
+from cunumpy.xp import array_backend, synchronize
 
 
 def _nvtx_module() -> ModuleType | None:

src/cunumpy/_random_streams.py

@@ -34,7 +34,7 @@
 
 import numpy as np
 
-from .xp import get_backend
+from cunumpy.xp import get_backend
 
 __all__ = ["BIT_GENERATORS", "RandomStreams", "random_streams"]
 

src/cunumpy/_scipy_backend.py

@@ -28,7 +28,7 @@
 from types import ModuleType
 from typing import Any
 
-from .xp import get_backend
+from cunumpy.xp import get_backend
 
 __all__ = ["SUBMODULES", "ScipyNamespace", "scipy"]
 

src/cunumpy/_staging.py

@@ -34,8 +34,8 @@
 import array_api_compat
 import numpy as np
 
-from ._transfers import _ACTIVE as _COUNTERS
-from ._transfers import _record
+from cunumpy._transfers import _ACTIVE as _COUNTERS
+from cunumpy._transfers import _record
 
 __all__ = ["HostStaging", "StagedCopy"]
 

src/cunumpy/_streams.py

@@ -4,7 +4,7 @@
 
 from typing import Any
 
-from .xp import get_backend
+from cunumpy.xp import get_backend
 
 
 class HostEvent:

src/cunumpy/algorithms.py

@@ -10,14 +10,14 @@
     charge = xp.algorithms.segment_sum(q, cell, n_cells)
 """
 
-from ._algorithms import (
+from cunumpy._algorithms import (
     SegmentPlan,
     cell_offsets,
     segment_boundaries,
     segment_sum,
     sort_by_key,
 )
-from ._morton import (
+from cunumpy._morton import (
     MAX_MORTON_LEVELS,
     morton_decode,
     morton_encode,

src/cunumpy/cuda/__init__.py

@@ -24,7 +24,7 @@
 use ``import cupy; cupy.cuda`` for CuPy's module.
 """
 
-from .._cuda_kernel import (
+from cunumpy._cuda_kernel import (
     DEBUG_OPTIONS,
     CudaArguments,
     CudaKernel,
@@ -41,7 +41,7 @@
     resolve_includes,
     write_cuda_header,
 )
-from .._device import (
+from cunumpy._device import (
     DEFAULT_SHARED_MEMORY_PER_BLOCK,
     bind_local_device,
     cuda_debug,
@@ -56,7 +56,7 @@
     set_device_for_rank,
     stream,
 )
-from .._streams import (
+from cunumpy._streams import (
     HostEvent,
     HostStream,
     create_event,

src/cunumpy/cuda_kernel.py

@@ -3,6 +3,6 @@
 The public names are in :mod:`cunumpy.cuda`.
 """
 
-from ._deprecated import alias_module
+from cunumpy._deprecated import alias_module
 
 alias_module(__name__, "cunumpy._cuda_kernel", "cunumpy.cuda")

src/cunumpy/dispatch.py

@@ -3,6 +3,6 @@
 The public names are in :mod:`cunumpy.kernels`.
 """
 
-from ._deprecated import alias_module
+from cunumpy._deprecated import alias_module
 
 alias_module(__name__, "cunumpy._dispatch", "cunumpy.kernels")

src/cunumpy/kernel.py

@@ -3,6 +3,6 @@
 The public names are in :mod:`cunumpy.kernels`.
 """
 
-from ._deprecated import alias_module
+from cunumpy._deprecated import alias_module
 
 alias_module(__name__, "cunumpy._kernel", "cunumpy.kernels")

src/cunumpy/kernel_testing.py

@@ -56,8 +56,8 @@ def test_parity(name, kernel):
 import array_api_compat
 import numpy as np
 
-from . import _fake_cupy
-from ._cuda_kernel import (
+from cunumpy import _fake_cupy
+from cunumpy._cuda_kernel import (
     CudaKernel,
     CudaParameter,
     CudaStructArguments,
@@ -66,9 +66,9 @@ def test_parity(name, kernel):
     _split_top_level,
     _strip_comments,
 )
-from ._dispatch import Kernel
-from ._emulation import emulate_cuda_kernel, emulation_compiler
-from .xp import cupy_available, get_backend, to_numpy, use_backend
+from cunumpy._dispatch import Kernel
+from cunumpy._emulation import emulate_cuda_kernel, emulation_compiler
+from cunumpy.xp import cupy_available, get_backend, to_numpy, use_backend
 
 # the pytest objects are created on first access, see __getattr__
 __all__ = [

src/cunumpy/kernels.py

@@ -19,10 +19,10 @@
 :mod:`cunumpy.kernel_testing`.
 """
 
-from ._cuda_kernel import PyccelStructArguments
-from ._dispatch import Kernel, KernelCatalog
-from ._fusion import fuse
-from ._kernel import (
+from cunumpy._cuda_kernel import PyccelStructArguments
+from cunumpy._dispatch import Kernel, KernelCatalog
+from cunumpy._fusion import fuse
+from cunumpy._kernel import (
     HOST_IMPLEMENTATIONS,
     CompiledHostKernel,
     HostImplementations,

src/cunumpy/memory.py

@@ -16,7 +16,7 @@
 :mod:`cunumpy.cuda`.
 """
 
-from ._mirror import DeviceMirror
-from ._staging import HostStaging, StagedCopy
+from cunumpy._mirror import DeviceMirror
+from cunumpy._staging import HostStaging, StagedCopy
 
 __all__ = ["DeviceMirror", "HostStaging", "StagedCopy"]