cuda.coop: execute all Block Load and Store algorithms - #11200
Draft
tpn wants to merge 4 commits into
Draft
Conversation
Contributor
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
This was referenced Sep 4, 2026
tpn
force-pushed
the
codex/cuda-coop-numba-block-load-store-algorithms
branch
from
September 4, 2026 16:52
8ecfb54 to
39755f7
Compare
Route all six selectors through declarative provider metadata. Use separate factories for storage-free and storage-bearing forms. Preserve Store payloads for exchange-based algorithms. Keep invalid partial Load values through a family-owned wrapper. Recover keyword-form local array extents in the generic planner. Signed-off-by: Trent Nelson <trent@trent.me>
Cover all layouts with independent host oracles. Exercise portable and qualified core and backend routes. Check provider identity and real NVRTC compilation. Check whole-function storage rewriting and RTX runtime behavior. Signed-off-by: Trent Nelson <trent@trent.me>
Describe every algorithm layout and storage contract. Call out Store payload preservation and temporary storage reuse. Signed-off-by: Trent Nelson <trent@trent.me>
Keep portable and qualified algorithm selection on one normalized lowercase string contract. Reject enum-like objects and integers before provider materialization, including the former ordinal value zero. Cover planner and provider paths and document the runtime contract. Signed-off-by: Trent Nelson <trent@trent.me>
tpn
force-pushed
the
codex/cuda-coop-numba-block-load-store-algorithms
branch
from
September 4, 2026 17:54
39755f7 to
a4d8965
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why this is needed
PR #11182 establishes portable group planning and the Numba-CUDA-MLIR provider
infrastructure, but deliberately limits Block Load and Store to DIRECT. This
stacked PR completes the CUB Block Load/Store algorithm matrix while keeping
activation, dispatch, caching, phase orchestration, and shared-storage
machinery family-independent.
Reviewer-visible behavior
All six lowercase selectors execute through the portable and qualified APIs:
direct,striped,vectorizetranspose,warp_transpose,warp_transpose_timeslicedexchange scratch space.
oob_defaultpreserves each invalid payloadslot through a family-owned CUB wrapper.
and reuse validation supplied by Add Numba-CUDA-MLIR Block Load and Store to cuda.coop #11182.
cuda.local.array(shape=...).synchronization remain declarative family metadata.
Validation
The exact final-stack head
9b299e894f, which includes this layer, passed:tests outside the source tree.
Stack and scope
This draft is stacked on #11182. It does not add Warp groups, Reduce, Scan,
Exchange, Shuffle, CUDAX dispatch, callbacks, or Python operator compilation.
It inherits the guarded Numba-CUDA-MLIR compatibility shim debt from the
foundation PR.