Backport PRs #3201, #3523, #3547, #3580 to the 2.8.x branch. (#3536)

* [FEA]: Introduce Python module with CCCL headers (#3201)

* Add cccl/python/cuda_cccl directory and use from cuda_parallel, cuda_cooperative

* Run `copy_cccl_headers_to_aude_include()` before `setup()`

* Create python/cuda_cccl/cuda/_include/__init__.py, then simply import cuda._include to find the include path.

* Add cuda.cccl._version exactly as for cuda.cooperative and cuda.parallel

* Bug fix: cuda/_include only exists after shutil.copytree() ran.

* Use `f"cuda-cccl @ file://{cccl_path}/python/cuda_cccl"` in setup.py

* Remove CustomBuildCommand, CustomWheelBuild in cuda_parallel/setup.py (they are equivalent to the default functions)

* Replace := operator (needs Python 3.8+)

* Fix oversights: remove `pip3 install ./cuda_cccl` lines from README.md

* Restore original README.md: `pip3 install -e` now works on first pass.

* cuda_cccl/README.md: FOR INTERNAL USE ONLY

* Remove `$pymajor.$pyminor.` prefix in cuda_cccl _version.py (as suggested under #3201 (comment))

Command used: ci/update_version.sh 2 8 0

* Modernize pyproject.toml, setup.py

Trigger for this change:

* #3201 (comment)

* #3201 (comment)

* Install CCCL headers under cuda.cccl.include

Trigger for this change:

* #3201 (comment)

Unexpected accidental discovery: cuda.cooperative unit tests pass without CCCL headers entirely.

* Factor out cuda_cccl/cuda/cccl/include_paths.py

* Reuse cuda_cccl/cuda/cccl/include_paths.py from cuda_cooperative

* Add missing Copyright notice.

* Add missing __init__.py (cuda.cccl)

* Add `"cuda.cccl"` to `autodoc.mock_imports`

* Move cuda.cccl.include_paths into function where it is used. (Attempt to resolve Build and Verify Docs failure.)

* Add # TODO: move this to a module-level import

* Modernize cuda_cooperative/pyproject.toml, setup.py

* Convert cuda_cooperative to use hatchling as build backend.

* Revert "Convert cuda_cooperative to use hatchling as build backend."

This reverts commit 61637d6.

* Move numpy from [build-system] requires -> [project] dependencies

* Move pyproject.toml [project] dependencies -> setup.py install_requires, to be able to use CCCL_PATH

* Remove copy_license() and use license_files=["../../LICENSE"] instead.

* Further modernize cuda_cccl/setup.py to use pathlib

* Trivial simplifications in cuda_cccl/pyproject.toml

* Further simplify cuda_cccl/pyproject.toml, setup.py: remove inconsequential code

* Make cuda_cooperative/pyproject.toml more similar to cuda_cccl/pyproject.toml

* Add taplo-pre-commit to .pre-commit-config.yaml

* taplo-pre-commit auto-fixes

* Use pathlib in cuda_cooperative/setup.py

* CCCL_PYTHON_PATH in cuda_cooperative/setup.py

* Modernize cuda_parallel/pyproject.toml, setup.py

* Use pathlib in cuda_parallel/setup.py

* Add `# TOML lint & format` comment.

* Replace MANIFEST.in with `[tool.setuptools.package-data]` section in pyproject.toml

* Use pathlib in cuda/cccl/include_paths.py

* pre-commit autoupdate (EXCEPT clang-format, which was manually restored)

* Fixes after git merge main

* Resolve warning: AttributeError: '_Reduce' object has no attribute 'build_result'

```
=========================================================================== warnings summary ===========================================================================
tests/test_reduce.py::test_reduce_non_contiguous
  /home/coder/cccl/python/devenv/lib/python3.12/site-packages/_pytest/unraisableexception.py:85: PytestUnraisableExceptionWarning: Exception ignored in: <function _Reduce.__del__ at 0x7bf123139080>

  Traceback (most recent call last):
    File "/home/coder/cccl/python/cuda_parallel/cuda/parallel/experimental/algorithms/reduce.py", line 132, in __del__
      bindings.cccl_device_reduce_cleanup(ctypes.byref(self.build_result))
                                                       ^^^^^^^^^^^^^^^^^
  AttributeError: '_Reduce' object has no attribute 'build_result'

    warnings.warn(pytest.PytestUnraisableExceptionWarning(msg))

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
============================================================= 1 passed, 93 deselected, 1 warning in 0.44s ==============================================================
```

* Move `copy_cccl_headers_to_cuda_cccl_include()` functionality to `class CustomBuildPy`

* Introduce cuda_cooperative/constraints.txt

* Also add cuda_parallel/constraints.txt

* Add `--constraint constraints.txt` in ci/test_python.sh

* Update Copyright dates

* Switch to https://github.com/ComPWA/taplo-pre-commit (the other repo has been archived by the owner on Jul 1, 2024)

For completeness: The other repo took a long time to install into the pre-commit cache; so long it lead to timeouts in the CCCL CI.

* Remove unused cuda_parallel jinja2 dependency (noticed by chance).

* Remove constraints.txt files, advertise running `pip install cuda-cccl` first instead.

* Make cuda_cooperative, cuda_parallel testing completely independent.

* Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc]

* Try using another runner (because V100 runners seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc]

* Fix sign-compare warning (#3408) [skip-rapids][skip-matx][skip-docs][skip-vdc]

* Revert "Try using another runner (because V100 runners seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc]"

This reverts commit ea33a21.

Error message: #3201 (comment)

* Try using A100 runner (because V100 runners still seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc]

* Also show cuda-cooperative site-packages, cuda-parallel site-packages (after pip install) [skip-rapids][skip-matx][skip-docs][skip-vdc]

* Try using l4 runner (because V100 runners still seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc]

* Restore original ci/matrix.yaml [skip-rapids]

* Use for loop in test_python.sh to avoid code duplication.

* Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc][skip pre-commit.ci]

* Comment out taplo-lint in pre-commit config [skip-rapids][skip-matx][skip-docs][skip-vdc]

* Revert "Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc][skip pre-commit.ci]"

This reverts commit ec206fd.

* Implement suggestion by @shwina (#3201 (review))

* Address feedback by @leofang

---------

Co-authored-by: Bernhard Manfred Gruber <[email protected]>

* cuda.parallel: invoke pytest directly rather than via `python -m pytest` (#3523)

Co-authored-by: Ashwin Srinath <[email protected]>

* Copy file from PR #3547 (bugfix/drop_pipe_in_lit by @wmaxey)

* Revert "cuda.parallel: invoke pytest directly rather than via `python -m pytest` (#3523)"

This reverts commit a2e21cb.

* Replace pipes.quote with shlex.quote in lit config (#3547)

* Replace pipes.quote with shlex.quote

* Drop TBB run on windows to unblock CI

* Update ci/matrix.yaml

Co-authored-by: Michael Schellenberger Costa <[email protected]>
Co-authored-by: Bernhard Manfred Gruber <[email protected]>

* Remove nvks runners from testing pool. (#3580)

---------

Co-authored-by: Bernhard Manfred Gruber <[email protected]>
Co-authored-by: Ashwin Srinath <[email protected]>
Co-authored-by: Ashwin Srinath <[email protected]>
Co-authored-by: Wesley Maxey <[email protected]>
Co-authored-by: Michael Schellenberger Costa <[email protected]>
Co-authored-by: Allison Piper <[email protected]>

ci/matrix.yaml

@@ -219,13 +219,13 @@ projects:
 
 # testing -> Runner with GPU is in a nv-gh-runners testing pool
 gpus:
-  v100:     { sm: 70 }                # 32 GB,  40 runners
-  t4:       { sm: 75, testing: true } # 16 GB,   8 runners
-  rtx2080:  { sm: 75, testing: true } #  8 GB,   8 runners
-  rtxa6000: { sm: 86, testing: true } # 48 GB,  12 runners
-  l4:       { sm: 89, testing: true } # 24 GB,  48 runners
-  rtx4090:  { sm: 89, testing: true } # 24 GB,  10 runners
-  h100:     { sm: 90 }                # 80 GB,  16 runners
+  v100:     { sm: 70 } # 32 GB,  40 runners
+  t4:       { sm: 75 } # 16 GB,  10 runners
+  rtx2080:  { sm: 75 } #  8 GB,  12 runners
+  rtxa6000: { sm: 86 } # 48 GB,  12 runners
+  l4:       { sm: 89 } # 24 GB,  48 runners
+  rtx4090:  { sm: 89 } # 24 GB,  10 runners
+  h100:     { sm: 90 } # 80 GB,  16 runners
 
 # Tags are used to define a `matrix job` in the workflow section.
 #

ci/test_pycuda.sh

@@ -8,25 +8,28 @@ print_environment_details
 
 fail_if_no_gpu
 
-readonly prefix="${BUILD_DIR}/python/"
-export PYTHONPATH="${prefix}:${PYTHONPATH:-}"
+begin_group "⚙️ Existing site-packages"
+pip freeze
+end_group "⚙️ Existing site-packages"
 
-pushd ../python/cuda_cooperative >/dev/null
+for module in cuda_parallel cuda_cooperative; do
 
-run_command "⚙️  Pip install cuda_cooperative" pip install --force-reinstall --upgrade --target "${prefix}" .[test]
-run_command "🚀  Pytest cuda_cooperative" python -m pytest -v ./tests
+  pushd "../python/${module}" >/dev/null
 
-popd >/dev/null
+  TEMP_VENV_DIR="/tmp/${module}_venv"
+  rm -rf "${TEMP_VENV_DIR}"
+  python -m venv "${TEMP_VENV_DIR}"
+  . "${TEMP_VENV_DIR}/bin/activate"
+  echo 'cuda-cccl @ file:///home/coder/cccl/python/cuda_cccl' > /tmp/cuda-cccl_constraints.txt
+  run_command "⚙️  Pip install ${module}" pip install -c /tmp/cuda-cccl_constraints.txt .[test]
+  begin_group "⚙️ ${module} site-packages"
+  pip freeze
+  end_group "⚙️ ${module} site-packages"
+  run_command "🚀  Pytest ${module}" python -m pytest -v ./tests
+  deactivate
 
-pushd ../python/cuda_parallel >/dev/null
+  popd >/dev/null
 
-# Temporarily install the package twice to populate include directory as part of the first installation
-# and to let manifest discover these includes during the second installation. Do not forget to remove the
-# second installation after https://github.com/NVIDIA/cccl/issues/2281 is addressed.
-run_command "⚙️  Pip install cuda_parallel once" pip install --force-reinstall --upgrade --target "${prefix}" .[test]
-run_command "⚙️  Pip install cuda_parallel twice" pip install --force-reinstall --upgrade --target "${prefix}" .[test]
-run_command "🚀  Pytest cuda_parallel" python -m pytest -v ./tests
-
-popd >/dev/null
+done
 
 print_time_summary

ci/update_version.sh

@@ -103,6 +103,7 @@ update_file "$CUDAX_CMAKE_VERSION_FILE" "set(cudax_VERSION_MAJOR \([0-9]\+\))" "
 update_file "$CUDAX_CMAKE_VERSION_FILE" "set(cudax_VERSION_MINOR \([0-9]\+\))" "set(cudax_VERSION_MINOR $minor)"
 update_file "$CUDAX_CMAKE_VERSION_FILE" "set(cudax_VERSION_PATCH \([0-9]\+\))" "set(cudax_VERSION_PATCH $patch)"
 
+update_file "$CUDA_CCCL_VERSION_FILE" "^__version__ = \"\([0-9.]\+\)\"" "__version__ = \"$major.$minor.$patch\""
 update_file "$CUDA_COOPERATIVE_VERSION_FILE" "^__version__ = \"\([0-9.]\+\)\"" "__version__ = \"$pymajor.$pyminor.$major.$minor.$patch\""
 update_file "$CUDA_PARALLEL_VERSION_FILE" "^__version__ = \"\([0-9.]\+\)\"" "__version__ = \"$pymajor.$pyminor.$major.$minor.$patch\""
 

libcudacxx/test/utils/libcudacxx/test/config.py

@@ -1385,19 +1385,19 @@ def configure_modules(self):
 
     def configure_substitutions(self):
         sub = self.config.substitutions
-        cxx_path = pipes.quote(self.cxx.path)
+        cxx_path = shlex.quote(self.cxx.path)
         # Configure compiler substitutions
         sub.append(('%cxx', cxx_path))
         sub.append(('%libcxx_src_root', self.libcudacxx_src_root))
         # Configure flags substitutions
-        flags_str = ' '.join([pipes.quote(f) for f in self.cxx.flags])
-        compile_flags_str = ' '.join([pipes.quote(f) for f in self.cxx.compile_flags])
-        link_flags_str = ' '.join([pipes.quote(f) for f in self.cxx.link_flags])
-        all_flags = '%s %s %s' % (flags_str, compile_flags_str, link_flags_str)
-        sub.append(('%flags', flags_str))
-        sub.append(('%compile_flags', compile_flags_str))
-        sub.append(('%link_flags', link_flags_str))
-        sub.append(('%all_flags', all_flags))
+        flags_str = " ".join([shlex.quote(f) for f in self.cxx.flags])
+        compile_flags_str = " ".join([shlex.quote(f) for f in self.cxx.compile_flags])
+        link_flags_str = " ".join([shlex.quote(f) for f in self.cxx.link_flags])
+        all_flags = "%s %s %s" % (flags_str, compile_flags_str, link_flags_str)
+        sub.append(("%flags", flags_str))
+        sub.append(("%compile_flags", compile_flags_str))
+        sub.append(("%link_flags", link_flags_str))
+        sub.append(("%all_flags", all_flags))
         if self.cxx.isVerifySupported():
             verify_str = ' ' + ' '.join(self.cxx.verify_flags) + ' '
             sub.append(('%verify', verify_str))
@@ -1422,11 +1422,11 @@ def configure_substitutions(self):
         # Configure run env substitution.
         sub.append(('%run', '%t.exe'))
         # Configure not program substitutions
-        not_py = os.path.join(self.libcudacxx_src_root, 'test', 'utils', 'not.py')
-        not_str = '%s %s ' % (pipes.quote(sys.executable), pipes.quote(not_py))
-        sub.append(('not ', not_str))
-        if self.get_lit_conf('libcudacxx_gdb'):
-            sub.append(('%libcxx_gdb', self.get_lit_conf('libcudacxx_gdb')))
+        not_py = os.path.join(self.libcudacxx_src_root, "test", "utils", "not.py")
+        not_str = "%s %s " % (shlex.quote(sys.executable), shlex.quote(not_py))
+        sub.append(("not ", not_str))
+        if self.get_lit_conf("libcudacxx_gdb"):
+            sub.append(("%libcxx_gdb", self.get_lit_conf("libcudacxx_gdb")))
 
     def can_use_deployment(self):
         # Check if the host is on an Apple platform using clang.

python/cuda_cooperative/.gitignore

@@ -1,3 +1,2 @@
-cuda/_include
 env
 *egg-info

python/cuda_cooperative/README.md

@@ -7,6 +7,7 @@ Please visit the documentation here: https://nvidia.github.io/cccl/python.html.
 ## Local development
 
 ```bash
+pip3 install -e ../cuda_cccl
 pip3 install -e .[test]
 pytest -v ./tests/
 ```

python/cuda_cooperative/pyproject.toml

@@ -1,7 +1,7 @@
-# Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
 #
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 [build-system]
-requires = ["packaging", "setuptools>=61.0.0", "wheel"]
+requires = ["setuptools>=61.0.0"]
 build-backend = "setuptools.build_meta"

python/cuda_cooperative/setup.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
 #
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 

python/cuda_parallel/.gitignore

@@ -1,4 +1,3 @@
-cuda/_include
 env
 *egg-info
 *so

python/cuda_parallel/README.md

@@ -7,6 +7,7 @@ Please visit the documentation here: https://nvidia.github.io/cccl/python.html.
 ## Local development
 
 ```bash
+pip3 install -e ../cuda_cccl
 pip3 install -e .[test]
 pytest -v ./tests/
 ```

python/cuda_parallel/pyproject.toml

@@ -1,7 +1,7 @@
-# Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
 #
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 [build-system]
-requires = ["packaging", "setuptools>=61.0.0", "wheel"]
+requires = ["setuptools>=61.0.0"]
 build-backend = "setuptools.build_meta"

python/cuda_parallel/setup.py

@@ -1,10 +1,9 @@
-# Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
 #
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
-import os
-import shutil
 import subprocess
+from pathlib import Path
 
 from setuptools import Command, Extension, setup, find_packages, find_namespace_packages
 from setuptools.command.build_py import build_py
@@ -84,8 +83,8 @@ def build_extension(self, ext):
             '-DCMAKE_BUILD_TYPE=Release',
         ]
 
-        if not os.path.exists(self.build_temp):
-            os.makedirs(self.build_temp)
+        build_temp_path = Path(self.build_temp)
+        build_temp_path.mkdir(parents=True, exist_ok=True)
 
         subprocess.check_call(['cmake', cccl_path] +
                               cmake_args, cwd=self.build_temp)