Repository navigation
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. |
github-actions
Bot
added
the
cuda.bindings
Contributor
Author
|
/ok to test |
Contributor
|
mdboom
marked this pull request as ready for review
October 26, 2026 14:01
|
No actionable comments were generated in the recent review.
 |
| Layer / File(s) | Summary |
|---|---|
Generated preamble imports cuda_bindings_13/cuda/bindings/_internal/*, cuda_bindings_13/cuda/bindings/cycufile.pxd, cuda_bindings_13/cuda/bindings/cydriver.pxd |
Generated bindings move existing Cython imports into preambles or reorder them. Generated content hashes change; the declarations and runtime behavior described in these files remain unchanged. |
cuFile structured-array wrappers cuda_bindings_13/cuda/bindings/cufile.pyx |
IOEvents, PerGpuStats, IOVec, Descr, and IOParams use structured-record arrays, named-field access, and buffer-based pointer access. Indexing returns owner-retaining views. |
Driver structured-array wrappers cuda_bindings_13/cuda/bindings/_v2/driver.pyx, cuda_bindings_13/docs/source/release/13.5.0-notes.rst |
Eight wrappers use structured-record arrays and owner-retaining views. Nested wrappers are created on first property access where applicable. The release note reports 5–30× faster struct-array operations in cufile, nvml, and _v2.driver. |
Suggested labels: enhancement
Suggested reviewers: rwgk
Priority:
Change: Feature
Merge Risk: 
58c77
-
This change speeds up wrappers for arrays of CUDA structs while preserving the existing API. No concrete merge-blocking issue remains after review.
Finishing Touches: 💡 1
-
🛠️ Fix failing CI checks: 💡
-
- Commit to this branch
- Create a new PR
-
- Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Comment @coderabbitai help to get the list of available commands.
This branch has not been deployed
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
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.
This brings in generator changes from !597.
The premise here is to stop using
numpy.recarray(which are implemented in Python, not C) for these arrays of structs. To avoid changing the API, certain operations lazily still returnrecarrays, but the most important operations avoid that for a /significantly/ faster implementation.Details
_dataas a plainnumpy.ndarraywith the samenumpy.recorddtype, so NumPy integer keys still returnnumpy.record.recarrayare stillrecarray: struct-typed fields and lists of field names are converted with.view(numpy.recarray)before they are returned.recarray._databy key (_data["field"]) instead of by attribute.ptr/int()get the address through the buffer protocol instead of.ctypes.data. The address is not cached, so it always reflects the current_data.