Fix mixed-IR liveness for inline overload DCE#795
Open
cpcloud wants to merge 5 commits intoNVIDIA:mainfrom
Open
Fix mixed-IR liveness for inline overload DCE#795cpcloud wants to merge 5 commits intoNVIDIA:mainfrom
cpcloud wants to merge 5 commits intoNVIDIA:mainfrom
Conversation
Contributor
|
Automatic reviews are disabled for this repository. |
|
Auto-sync is disabled for ready for review pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
Contributor
Author
|
/ok to test |
1 similar comment
Contributor
Author
|
/ok to test |
Contributor
Author
|
/ok to test |
brandon-b-miller
added a commit
that referenced
this pull request
Feb 20, 2026
## Summary - align CUDA `np.dtype` overload parameters with NumPy (`dtype`, `align`, `copy`) while avoiding unsupported `**kwargs` in typing templates - align CUDA `np.dot` and `np.vdot` overload parameter names with NumPy (`a`, `b`, `out`) so signature-compatibility checks pass across NumPy variants - preserve existing lowering behavior for supported dtype and BLAS-backed dot/vdot paths Extracted from #795 to keep the issue-718 DCE fix focused. --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: brandon-b-miller <brmiller@nvidia.com>
Contributor
Author
|
/ok to test |
Issue NVIDIA#718 mixed IR node classes after inline overload expansion, which made use/def and liveness tracking miss live vars and allowed DCE to drop required expressions. Use compatibility var collection in analysis/DCE and add a user-level regression that exercises inline="always" with cuda.local.array. Co-authored-by: Cursor <cursoragent@cursor.com>
Use a direct list_vars call with AttributeError fallback in compat_list_vars_node instead of pre-checking attributes, which keeps the compatibility path straightforward while preserving mixed-IR behavior. Co-authored-by: Cursor <cursoragent@cursor.com>
Update local package records in pixi.lock so all matrix variants of numba-cuda reflect version 0.27.0 during pixi-managed test/build workflows. Co-authored-by: Cursor <cursoragent@cursor.com>
Match the CUDA np.dtype overload parameters to NumPy's public signature so signature-compatibility checks pass on environments where dtype exposes align/copy/kwargs parameters. Co-authored-by: Cursor <cursoragent@cursor.com>
This reverts commit af57423.
Contributor
Author
|
/ok to test |
cpcloud
force-pushed
the
fix/issue-718-inline-overload-dce
branch
from
3c93c51 to
ccf93f2
Compare
gmarkall
modified the milestones:
numba-cuda backlog,
numba-cuda v0.28.0,
numba-cuda v0.29.0
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
3 participants
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.
Summary
numba.coreand vendored CUDA IR nodes created byinline="always"overload inliningstatic_getitemfeeding the return value in issue [BUG]KeyErrorwithinline="always"and some interaction with DCE #718cuda.local.arraypixi.locklocal package entries sonumba-cudavariants resolve to0.27.0Related: the separate
np.dtypesignature compatibility fix was split into #797.Fixes #718