[STF] Rename places namespace from cuda::experimental::stf to cuda::experimental::places by caugonnet · Pull Request #8219 · NVIDIA/cccl

caugonnet · 2026-03-30T21:03:43Z

Move all __places/ headers from namespace cuda::experimental::stf to cuda::experimental::places. Each header imports needed stf utility types (pos4, dim4, cuda_safe_call, etc.) via using declarations and re-exports its public types back into cuda::experimental::stf for backward compat.

Hash specializations (stf::hash<data_place>, stf::hash<exec_place>, stf::hash<green_ctx_view>) remain in the stf namespace where the primary template is defined. The composite_slice cross-boundary function signature is updated to use the qualified places type.

Tests, examples, C API, and documentation are updated to use the new namespace.

Made-with: Cursor

Description

closes

Checklist

New or existing tests cover these changes.
The documentation is up to date with these changes.

…xperimental::places Move all __places/ headers from namespace cuda::experimental::stf to cuda::experimental::places. Each header imports needed stf utility types (pos4, dim4, cuda_safe_call, etc.) via using declarations and re-exports its public types back into cuda::experimental::stf for backward compat. Hash specializations (stf::hash<data_place>, stf::hash<exec_place>, stf::hash<green_ctx_view>) remain in the stf namespace where the primary template is defined. The composite_slice cross-boundary function signature is updated to use the qualified places type. Tests, examples, C API, and documentation are updated to use the new namespace. Made-with: Cursor

copy-pr-bot · 2026-03-30T21:03:46Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

cudax/include/cuda/experimental/__places/exec/cuda_stream.cuh

cudax/include/cuda/experimental/__places/exec/green_context.cuh

Move all stf namespace backward compatibility using-declarations from individual __places/ headers into the central places.cuh compat block. This makes places.cuh the single curated list of places types re-exported into the stf namespace. Made-with: Cursor

cudax/include/cuda/experimental/__places/places.cuh

caugonnet · 2026-03-30T21:31:53Z

cudax/include/cuda/experimental/__places/places.cuh

+};
+#endif // _CCCL_CTK_AT_LEAST(12, 4)
+
+// Backward compatibility: re-export places types into the stf namespace


We need to have a curated list ... and i hate this for now

caugonnet · 2026-03-30T21:32:50Z

docs/cudax/places.rst

+The places API is part of the ``cuda::experimental::places`` C++ namespace
 and can be used standalone via the ``cuda/experimental/places.cuh`` header,
-without pulling in the full CUDASTF task-graph framework.
+without pulling in the full CUDASTF task-graph framework. For backward


Be more specific about what exactly is made available

cudax/include/cuda/experimental/__places/places.cuh

Remove non-user-visible types from the places.cuh backward compat block and qualify them with places:: in the __stf/ files that use them. Move instance_id_t to its own __stf/internal/instance_id.cuh header since it conceptually belongs to the task framework, not the places library. The compat block now only re-exports user-facing types: data_place, exec_place, exec_place_scope, decorated_stream, make_grid, partition_cyclic, partition_tile, to_index, from_index, device_ordinal. Made-with: Cursor

Made-with: Cursor

cudax/include/cuda/experimental/__stf/localization/composite_slice.cuh

Remove redundant using places::decorated_stream from stream_ctx.cuh and remove place_partition backward compat block from place_partition.cuh since loop_dispatch.cuh now uses places:: qualified names directly. Made-with: Cursor

caugonnet · 2026-03-30T22:28:12Z

/ok to test 9ac579c

Forward declarations like class exec_place inside namespace stf conflict with the using places::exec_place from the compat block. Remove redundant forward declarations where places.cuh is transitively included, and for files without transitive inclusion, forward-declare in namespace places and add a local using. Made-with: Cursor

caugonnet · 2026-03-31T05:54:39Z

/ok to test a55db1e

Tests were using place_partition_scope, place_partition, place_partition_scope_to_string, and make_grid unqualified via `using namespace stf`, but these are no longer re-exported from the stf namespace. Also rename local variable `places` in gc_grid.cu to avoid shadowing the namespace. Made-with: Cursor

Split partition-related usings into stf_places_partition_into_stf.cuh to avoid include cycle with async_resources_handle. Extend stf_places_extended_exports with localized_array, stream ids, hash keys, and machine.

andralex · 2026-04-02T02:32:08Z

/ok to test b3b4d1d

…APIs Fix CI after partition usings moved out of stf.cuh umbrella.

andralex · 2026-04-02T03:03:21Z

/ok to test 79b3ded

- stf_places_into_stf_core: include places.cuh for isolated header tests - stf_places_extended_exports: include hash.cuh so SortIncludes cannot break specialization order; drop redundant hash includes from callers - composite_slice: extended + into_stf_core only (sorted order safe)

andralex · 2026-04-02T05:18:17Z

/ok to test 25d6260

andralex

More to come

andralex · 2026-04-02T15:08:35Z

/ok to test 25d6260

copy-pr-bot · 2026-04-02T15:08:39Z

/ok to test 25d6260

@andralex, there was an error processing your request: E2

See the following link for more information: https://docs.gha-runners.nvidia.com/cpr/e/2/

andralex · 2026-04-02T15:30:59Z

/ok to test 88b034b

github-actions · 2026-04-02T16:30:06Z

🥳 CI Workflow Results

🟩 Finished in 57m 18s: Pass: 100%/52 | Total: 22h 13m | Max: 43m 35s | Hits: 59%/26965

See results here.

…xperimental::places (NVIDIA#8219) * [STF] Rename places namespace from cuda::experimental::stf to cuda::experimental::places Move all __places/ headers from namespace cuda::experimental::stf to cuda::experimental::places. Each header imports needed stf utility types (pos4, dim4, cuda_safe_call, etc.) via using declarations and re-exports its public types back into cuda::experimental::stf for backward compat. Hash specializations (stf::hash<data_place>, stf::hash<exec_place>, stf::hash<green_ctx_view>) remain in the stf namespace where the primary template is defined. The composite_slice cross-boundary function signature is updated to use the qualified places type. Tests, examples, C API, and documentation are updated to use the new namespace. Made-with: Cursor * remove some redundant using * consolidate places backward compat into places.cuh Move all stf namespace backward compatibility using-declarations from individual __places/ headers into the central places.cuh compat block. This makes places.cuh the single curated list of places types re-exported into the stf namespace. Made-with: Cursor * qualify internal places types with places:: in __stf/ headers Remove non-user-visible types from the places.cuh backward compat block and qualify them with places:: in the __stf/ files that use them. Move instance_id_t to its own __stf/internal/instance_id.cuh header since it conceptually belongs to the task framework, not the places library. The compat block now only re-exports user-facing types: data_place, exec_place, exec_place_scope, decorated_stream, make_grid, partition_cyclic, partition_tile, to_index, from_index, device_ordinal. Made-with: Cursor * remove make_grid from places backward compat block Made-with: Cursor * clang-format * comments about a TODO * qualify place_partition types with places:: in loop_dispatch.cuh Remove redundant using places::decorated_stream from stream_ctx.cuh and remove place_partition backward compat block from place_partition.cuh since loop_dispatch.cuh now uses places:: qualified names directly. Made-with: Cursor * fix forward declarations of places types in stf headers Forward declarations like class exec_place inside namespace stf conflict with the using places::exec_place from the compat block. Remove redundant forward declarations where places.cuh is transitively included, and for files without transitive inclusion, forward-declare in namespace places and add a local using. Made-with: Cursor * qualify places:: types in test files after compat block cleanup Tests were using place_partition_scope, place_partition, place_partition_scope_to_string, and make_grid unqualified via `using namespace stf`, but these are no longer re-exported from the stf namespace. Also rename local variable `places` in gc_grid.cu to avoid shadowing the namespace. Made-with: Cursor * clang-format * use 'using namespace places' in tests instead of explicit qualification Replace `places::` qualifiers with `using namespace cuda::experimental::places;` for consistency with the existing `using namespace cuda::experimental::stf;` pattern in test files. Made-with: Cursor * disambiguate stf::reserved in C API after places namespace split Both stf::reserved and places::reserved exist, so unqualified 'reserved::' is ambiguous when both namespaces are imported. Made-with: Cursor * remove green_context.cuh backward compat block All 5 re-exported types (exec_place_green_ctx_impl, get_cuda_context_id, green_context_helper, green_ctx_data_place_impl, make_green_ctx_data_place) are either internal-only or test-convenience types. Tests that use green_context_helper now import cuda::experimental::places directly. Made-with: Cursor * remove unnecessary stf namespace import of exec_place_cuda_stream_impl Made-with: Cursor * remove unnecessary stf namespace imports from __places headers Made-with: Cursor * introduce independent places::hash template for places types Define places::hash<T> as a standalone primary template with specializations for data_place, exec_place, and green_ctx_view. Update all stf code and tests to use places::hash for places types instead of the former stf::hash specializations. Made-with: Cursor * use 'using namespace places' in test/example files instead of fully qualified places::hash Made-with: Cursor * clang-format * move green_ctx_view hash specialization to green_ctx_view.cuh Made-with: Cursor * clang-format * Fix places/STF namespace qualification in green context plumbing - Use places::green_context_helper in async_resources_handle (class lives in cuda::experimental::places; fixes cudax.test.places builds). - Qualify stf::mv and stf::hash_all in green_ctx_view.cuh; drop redundant using-declarations. Made-with: Cursor * Extra fixed for namespaces * use properly qualified names * Add missing using namespace in examples * add missing namespace * [STF] Disable stackable_loop_dispatch test The test is hitting a double free or corruption (fasttop) at runtime. Disabling it until the root cause is investigated. Made-with: Cursor * better message * cudax: re-export places into STF from STF-only headers; remove using namespace places from STF tests * cudax: fix STF places exports on CTK<12.4; extend places umbrella * cudax: re-export cyclic_partition into stf; include cyclic_shape in places umbrella * lookup fix * cudax(places): use cuda_try consistently; cuda_try<fn> for unambiguous APIs; update places tests * stf: export places names into stf; remove places:: call sites Split partition-related usings into stf_places_partition_into_stf.cuh to avoid include cycle with async_resources_handle. Extend stf_places_extended_exports with localized_array, stream ids, hash keys, and machine. * stf tests: include stf_places_partition_into_stf for place_partition APIs Fix CI after partition usings moved out of stf.cuh umbrella. * stf places bridges: self-contained headers, hash via extended_exports - stf_places_into_stf_core: include places.cuh for isolated header tests - stf_places_extended_exports: include hash.cuh so SortIncludes cannot break specialization order; drop redundant hash includes from callers - composite_slice: extended + into_stf_core only (sorted order safe) --------- Co-authored-by: Andrei Alexandrescu <andrei@erdani.com>

caugonnet self-assigned this Mar 30, 2026

caugonnet added stf Sequential Task Flow programming model places labels Mar 30, 2026

github-project-automation bot added this to CCCL Mar 30, 2026

github-project-automation bot moved this to Todo in CCCL Mar 30, 2026

cccl-authenticator-app bot moved this from Todo to In Progress in CCCL Mar 30, 2026

andralex reviewed Mar 30, 2026

View reviewed changes

cudax/include/cuda/experimental/__places/exec/cuda_stream.cuh Outdated Show resolved Hide resolved

andralex reviewed Mar 30, 2026

View reviewed changes

cudax/include/cuda/experimental/__places/exec/green_context.cuh Outdated Show resolved Hide resolved

caugonnet added 2 commits March 30, 2026 23:14

remove some redundant using

cd2ae89

caugonnet commented Mar 30, 2026

View reviewed changes

cudax/include/cuda/experimental/__places/places.cuh Outdated Show resolved Hide resolved

caugonnet commented Mar 30, 2026

View reviewed changes

cudax/include/cuda/experimental/__places/places.cuh Outdated Show resolved Hide resolved

caugonnet added 4 commits March 30, 2026 23:55

remove make_grid from places backward compat block

b04acbd

Made-with: Cursor

clang-format

49dd1cb

comments about a TODO

0ba31e2

caugonnet commented Mar 30, 2026

View reviewed changes

cudax/include/cuda/experimental/__stf/localization/composite_slice.cuh Show resolved Hide resolved

This comment has been minimized.

Sign in to view

caugonnet and others added 3 commits March 31, 2026 08:56

clang-format

f9b78eb

Merge branch 'main' into stf_places_namespaces

bf482a5

This comment has been minimized.

Sign in to view

stf: export places names into stf; remove places:: call sites

b3b4d1d

Split partition-related usings into stf_places_partition_into_stf.cuh to avoid include cycle with async_resources_handle. Extend stf_places_extended_exports with localized_array, stream ids, hash keys, and machine.

This comment has been minimized.

Sign in to view

stf tests: include stf_places_partition_into_stf for place_partition …

79b3ded

…APIs Fix CI after partition usings moved out of stf.cuh umbrella.

This comment has been minimized.

Sign in to view

andralex marked this pull request as ready for review April 2, 2026 05:32

andralex requested review from a team as code owners April 2, 2026 05:32

andralex requested review from andralex, gonidelis and wmaxey April 2, 2026 05:32

andralex enabled auto-merge (squash) April 2, 2026 05:32

cccl-authenticator-app bot moved this from In Progress to In Review in CCCL Apr 2, 2026

andralex approved these changes Apr 2, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

Merge branch 'main' into stf_places_namespaces

88b034b

miscco approved these changes Apr 2, 2026

View reviewed changes

NaderAlAwar approved these changes Apr 2, 2026

View reviewed changes

andralex merged commit 9832833 into NVIDIA:main Apr 2, 2026
72 of 74 checks passed

github-project-automation bot moved this from In Review to Done in CCCL Apr 2, 2026

Conversation

caugonnet commented Mar 30, 2026

Description

Checklist

Uh oh!

copy-pr-bot bot commented Mar 30, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

caugonnet Mar 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

caugonnet Mar 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

caugonnet commented Mar 30, 2026

Uh oh!

This comment has been minimized.

caugonnet commented Mar 31, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

andralex commented Apr 2, 2026

Uh oh!

This comment has been minimized.

andralex commented Apr 2, 2026

Uh oh!

This comment has been minimized.

andralex commented Apr 2, 2026

Uh oh!

andralex left a comment

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

andralex commented Apr 2, 2026

Uh oh!

copy-pr-bot bot commented Apr 2, 2026

Uh oh!

andralex commented Apr 2, 2026

Uh oh!

github-actions bot commented Apr 2, 2026

🥳 CI Workflow Results

🟩 Finished in 57m 18s: Pass: 100%/52 | Total: 22h 13m | Max: 43m 35s | Hits: 59%/26965

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

caugonnet Mar 30, 2026 •

edited

Loading

caugonnet Mar 30, 2026 •

edited

Loading