Merge remote-tracking branch 'upstream/branch-23.10' into cccl-update…

…-2.1.0

CHANGELOG.md

@@ -1,3 +1,38 @@
+# RMM 23.08.00 (9 Aug 2023)
+
+## 🚨 Breaking Changes
+
+- Stop invoking setup.py ([#1300](https://github.com/rapidsai/rmm/pull/1300)) [@vyasr](https://github.com/vyasr)
+- Remove now-deprecated top-level allocator functions ([#1281](https://github.com/rapidsai/rmm/pull/1281)) [@wence-](https://github.com/wence-)
+- Remove padding from device_memory_resource ([#1278](https://github.com/rapidsai/rmm/pull/1278)) [@vyasr](https://github.com/vyasr)
+
+## 🐛 Bug Fixes
+
+- Fix typo in wheels-test.yaml. ([#1310](https://github.com/rapidsai/rmm/pull/1310)) [@bdice](https://github.com/bdice)
+- Add a missing '#include <array>' in logger.hpp ([#1295](https://github.com/rapidsai/rmm/pull/1295)) [@valgur](https://github.com/valgur)
+- Use gbench `thread_index()` accessor to fix replay bench compilation ([#1293](https://github.com/rapidsai/rmm/pull/1293)) [@harrism](https://github.com/harrism)
+- Ensure logger tests don't generate temp directories in build dir ([#1289](https://github.com/rapidsai/rmm/pull/1289)) [@robertmaynard](https://github.com/robertmaynard)
+
+## 🚀 New Features
+
+- Remove now-deprecated top-level allocator functions ([#1281](https://github.com/rapidsai/rmm/pull/1281)) [@wence-](https://github.com/wence-)
+
+## 🛠️ Improvements
+
+- Switch to new CI wheel building pipeline ([#1305](https://github.com/rapidsai/rmm/pull/1305)) [@vyasr](https://github.com/vyasr)
+- Revert CUDA 12.0 CI workflows to branch-23.08. ([#1303](https://github.com/rapidsai/rmm/pull/1303)) [@bdice](https://github.com/bdice)
+- Update linters: remove flake8, add ruff, update cython-lint ([#1302](https://github.com/rapidsai/rmm/pull/1302)) [@vyasr](https://github.com/vyasr)
+- Adding identify minimum version requirement ([#1301](https://github.com/rapidsai/rmm/pull/1301)) [@hyperbolic2346](https://github.com/hyperbolic2346)
+- Stop invoking setup.py ([#1300](https://github.com/rapidsai/rmm/pull/1300)) [@vyasr](https://github.com/vyasr)
+- Use cuda-version to constrain cudatoolkit. ([#1296](https://github.com/rapidsai/rmm/pull/1296)) [@bdice](https://github.com/bdice)
+- Update to CMake 3.26.4 ([#1291](https://github.com/rapidsai/rmm/pull/1291)) [@vyasr](https://github.com/vyasr)
+- use rapids-upload-docs script ([#1288](https://github.com/rapidsai/rmm/pull/1288)) [@AyodeAwe](https://github.com/AyodeAwe)
+- Reorder parameters in RMM_EXPECTS ([#1286](https://github.com/rapidsai/rmm/pull/1286)) [@vyasr](https://github.com/vyasr)
+- Remove documentation build scripts for Jenkins ([#1285](https://github.com/rapidsai/rmm/pull/1285)) [@ajschmidt8](https://github.com/ajschmidt8)
+- Remove padding from device_memory_resource ([#1278](https://github.com/rapidsai/rmm/pull/1278)) [@vyasr](https://github.com/vyasr)
+- Unpin scikit-build upper bound ([#1275](https://github.com/rapidsai/rmm/pull/1275)) [@vyasr](https://github.com/vyasr)
+- RMM: Build CUDA 12 packages ([#1223](https://github.com/rapidsai/rmm/pull/1223)) [@bdice](https://github.com/bdice)
+
 # RMM 23.06.00 (7 Jun 2023)
 
 ## 🚨 Breaking Changes

ci/build_docs.sh

@@ -37,11 +37,11 @@ popd
 
 rapids-logger "Build Python docs"
 pushd python/docs
-sphinx-build -b dirhtml . _html
-sphinx-build -b text . _text
+make dirhtml
+make text
 mkdir -p "${RAPIDS_DOCS_DIR}/rmm/"{html,txt}
-mv _html/* "${RAPIDS_DOCS_DIR}/rmm/html"
-mv _text/* "${RAPIDS_DOCS_DIR}/rmm/txt"
+mv _build/dirhtml/* "${RAPIDS_DOCS_DIR}/rmm/html"
+mv _build/text/* "${RAPIDS_DOCS_DIR}/rmm/txt"
 popd
 
 rapids-upload-docs

conda/environments/all_cuda-118_arch-x86_64.yaml

@@ -9,18 +9,28 @@ dependencies:
 - cuda-version=11.8
 - cudatoolkit
 - cython>=3.0.0
+- doxygen=1.8.20
 - fmt>=9.1.0,<10
 - gcovr>=5.0
+- graphviz
 - identify>=2.5.20
+- ipython
+- make
+- nbsphinx
 - ninja
 - numba>=0.57
 - numpy>=1.21
+- numpydoc
 - nvcc_linux-64=11.8
 - pre-commit
 - pytest
 - pytest-cov
 - python>=3.9,<3.11
 - scikit-build>=0.13.1
 - spdlog>=1.11.0,<1.12
+- sphinx
+- sphinx-copybutton
+- sphinx-markdown-tables
+- sphinx_rtd_theme
 - tomli
 name: all_cuda-118_arch-x86_64

conda/environments/all_cuda-120_arch-x86_64.yaml

@@ -9,17 +9,27 @@ dependencies:
 - cuda-python>=12.0,<13.0a0
 - cuda-version=12.0
 - cython>=3.0.0
+- doxygen=1.8.20
 - fmt>=9.1.0,<10
 - gcovr>=5.0
+- graphviz
 - identify>=2.5.20
+- ipython
+- make
+- nbsphinx
 - ninja
 - numba>=0.57
 - numpy>=1.21
+- numpydoc
 - pre-commit
 - pytest
 - pytest-cov
 - python>=3.9,<3.11
 - scikit-build>=0.13.1
 - spdlog>=1.11.0,<1.12
+- sphinx
+- sphinx-copybutton
+- sphinx-markdown-tables
+- sphinx_rtd_theme
 - tomli
 name: all_cuda-120_arch-x86_64

dependencies.yaml

@@ -10,6 +10,7 @@ files:
       - checks
       - cudatoolkit
       - develop
+      - docs
       - run
       - test_python
   test_python:
@@ -155,6 +156,7 @@ dependencies:
           - doxygen=1.8.20
           - graphviz
           - ipython
+          - make
           - nbsphinx
           - numpydoc
           - sphinx

python/docs/Makefile

@@ -3,7 +3,7 @@
 
 # You can set these variables from the command line, and also
 # from the environment for the first two.
-SPHINXOPTS    ?=
+SPHINXOPTS    = -n -v -W --keep-going
 SPHINXBUILD   ?= sphinx-build
 SOURCEDIR     = .
 BUILDDIR      = _build

python/docs/api.rst

@@ -15,6 +15,7 @@ Memory Resources
 
 .. automodule:: rmm.mr
    :members:
+   :inherited-members:
    :undoc-members:
    :show-inheritance:
 
@@ -28,6 +29,7 @@ Memory Allocators
 
 .. automodule:: rmm.allocators.numba
    :members:
+   :inherited-members:
    :undoc-members:
    :show-inheritance:
 

python/docs/basics.md

@@ -91,7 +91,7 @@ example, enabling the `ManagedMemoryResource` tells RMM to use
 > :warning: The default resource must be set for any device **before**
 > allocating any device memory on that device.  Setting or changing the
 > resource after device allocations have been made can lead to unexpected
-> behaviour or crashes. See [Multiple Devices](#multiple-devices)
+> behaviour or crashes.
 
 As another example, `PoolMemoryResource` allows you to allocate a
 large "pool" of device memory up-front. Subsequent allocations will

python/docs/conf.py

@@ -35,6 +35,7 @@
 # ones.
 
 extensions = [
+    "sphinxcontrib.jquery",
     "sphinx.ext.intersphinx",
     "sphinx.ext.autodoc",
     "sphinx.ext.autosummary",
@@ -113,7 +114,7 @@
 # Add any paths that contain custom static files (such as style sheets) here,
 # relative to this directory. They are copied after the builtin static files,
 # so a file named "default.css" will overwrite the builtin "default.css".
-html_static_path = ["_static"]
+html_static_path = []
 
 
 # -- Options for HTMLHelp output ------------------------------------------
@@ -179,14 +180,22 @@
 
 
 # Example configuration for intersphinx: refer to the Python standard library.
-intersphinx_mapping = {"https://docs.python.org/": None}
+intersphinx_mapping = {
+    "python": ("https://docs.python.org/3", None),
+    "numba": ("https://numba.readthedocs.io/en/stable", None),
+}
 
 # Config numpydoc
 numpydoc_show_inherited_class_members = True
 numpydoc_class_members_toctree = False
 
 autoclass_content = "init"
 
+nitpick_ignore = [
+    ("py:class", "size_t"),
+    ("py:class", "void"),
+]
+
 
 def setup(app):
     app.add_js_file("copybutton_pydocs.js")

python/rmm/_lib/memory_resource.pyx

@@ -217,18 +217,41 @@ cdef extern from "rmm/mr/device/failure_callback_resource_adaptor.hpp" \
 cdef class DeviceMemoryResource:
 
     cdef device_memory_resource* get_mr(self):
+        """Get the underlying C++ memory resource object."""
         return self.c_obj.get()
 
     def allocate(self, size_t nbytes):
+        """Allocate ``nbytes`` bytes of memory.
+
+        Parameters
+        ----------
+        nbytes : size_t
+            The size of the allocation in bytes
+        """
         return <uintptr_t>self.c_obj.get().allocate(nbytes)
 
     def deallocate(self, uintptr_t ptr, size_t nbytes):
+        """Deallocate memory pointed to by ``ptr`` of size ``nbytes``.
+
+        Parameters
+        ----------
+        ptr : uintptr_t
+            Pointer to be deallocated
+        nbytes : size_t
+            Size of the allocation in bytes
+        """
         self.c_obj.get().deallocate(<void*>(ptr), nbytes)
 
 
 # See the note about `no_gc_clear` in `device_buffer.pyx`.
 @cython.no_gc_clear
 cdef class UpstreamResourceAdaptor(DeviceMemoryResource):
+    """Parent class for all memory resources that track an upstream.
+
+    Upstream resource tracking requires maintaining a reference to the upstream
+    mr so that it is kept alive and may be accessed by any downstream resource
+    adaptors.
+    """
 
     def __cinit__(self, DeviceMemoryResource upstream_mr, *args, **kwargs):
 

python/rmm/allocators/numba.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import ctypes
+import inspect
 
 from cuda.cuda import CUdeviceptr, cuIpcGetMemHandle
 from numba import config, cuda
@@ -112,13 +113,29 @@ def get_ipc_handle(self, memory):
         )
 
     def get_memory_info(self):
+        """Returns ``(free, total)`` memory in bytes in the context.
+
+        This implementation raises `NotImplementedError` because the allocation
+        will be performed using rmm's currently set default mr, which may be a
+        pool allocator.
+        """
         raise NotImplementedError()
 
     @property
     def interface_version(self):
         return 1
 
 
+# The parent class docstrings contain references without fully qualified names,
+# so we need to replace them here for our Sphinx docs to render properly.
+for _, method in inspect.getmembers(RMMNumbaManager, inspect.isfunction):
+    if method.__doc__ is not None:
+        method.__doc__ = method.__doc__.replace(
+            ":class:`BaseCUDAMemoryManager`",
+            ":class:`numba.cuda.BaseCUDAMemoryManager`",
+        )
+
+
 # Enables the use of RMM for Numba via an environment variable setting,
 # NUMBA_CUDA_MEMORY_MANAGER=rmm. See:
 # https://numba.readthedocs.io/en/stable/cuda/external-memory.html#environment-variable

python/rmm/mr.py

@@ -24,6 +24,7 @@
     PoolMemoryResource,
     StatisticsResourceAdaptor,
     TrackingResourceAdaptor,
+    UpstreamResourceAdaptor,
     _flush_logs,
     _initialize,
     disable_logging,
@@ -51,6 +52,7 @@
     "StatisticsResourceAdaptor",
     "TrackingResourceAdaptor",
     "FailureCallbackResourceAdaptor",
+    "UpstreamResourceAdaptor",
     "_flush_logs",
     "_initialize",
     "set_per_device_resource",