Fix linking of external code from callees #137
Merged
Conversation
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
The original linking implementation for linkable code in device declarations did not consider calls inside callees; this change recurses through the typing to find all calls requiring linkable code.
isVoid
reviewed
Mar 2, 2025
numba_cuda/numba/cuda/dispatcher.py
Outdated
Comment on lines
51
to
68
| # The typemap of the function includes calls, so we can traverse it to find | ||
| # the references we need. | ||
| for name, v in cres.fndesc.typemap.items(): | ||
|
|
||
| # CUDADispatchers represent a call to a device function, so we need to | ||
| # look up the linkable code for those recursively. | ||
| if isinstance(v, cuda_types.CUDADispatcher): | ||
| # We need to locate the signature of the call so we can find the | ||
| # correct overload. | ||
| for call, sig in cres.fndesc.calltypes.items(): | ||
| if isinstance(call, ir.Expr) and call.op == 'call': | ||
| # There will likely be multiple calls in the typemap; we | ||
| # can uniquely identify the relevant one using its SSA | ||
| # name. | ||
| if call.func.name == name: | ||
| called_cres = v.dispatcher.overloads[sig.args] | ||
| called_link_objects = get_cres_link_objects(called_cres) | ||
| link_objects.update(called_link_objects) |
Contributor
There was a problem hiding this comment.
This is cool. I learnt a few things by reading through this section. Do you think the below simplifies the code and reduces the code complexity for a little bit?
I made a PR here:
gmarkall#4
I think this reduces the size of the list for both of the nested for-loop. This is proportional to O(num_calls^2), not O(num_typings^2)
Contributor
Author
There was a problem hiding this comment.
Thanks - I've incorporated your changes. I don't think there's much of a performance impact (the number of calls and typings won't be that large), but I think your modifications improved the readability of the code.
gmarkall
added
4 - Waiting on reviewer
gmarkall
added a commit
to gmarkall/numba-cuda
that referenced
this pull request
Mar 6, 2025
- Fix linking of external code from callees (NVIDIA#137) - Try using a newer branch workflow (NVIDIA#148) - Move publish step out of `wheels-build.yaml` (NVIDIA#147) - Upload wheels to PyPI from GitHub-hosted runner (NVIDIA#142) - Add paddle to interoperability chapter (NVIDIA#144) - Fix the debug info of GridGroup type (NVIDIA#131) - Remove dead `prepare_cuda_kernel()` (NVIDIA#130) - Add a CUDA DI Builder (NVIDIA#104) - dont launch extra kernels when stats counting is disabled (NVIDIA#127) - Fixup debug metadata in kernel fixup (NVIDIA#97) - Implement debuginfo bool name fix (numba/numba#9888) in numba-cuda (NVIDIA#106)
Merged
gmarkall
added a commit
that referenced
this pull request
Mar 6, 2025
- Fix linking of external code from callees (#137) - Try using a newer branch workflow (#148) - Move publish step out of `wheels-build.yaml` (#147) - Upload wheels to PyPI from GitHub-hosted runner (#142) - Add paddle to interoperability chapter (#144) - Fix the debug info of GridGroup type (#131) - Remove dead `prepare_cuda_kernel()` (#130) - Add a CUDA DI Builder (#104) - dont launch extra kernels when stats counting is disabled (#127) - Fixup debug metadata in kernel fixup (#97) - Implement debuginfo bool name fix (numba/numba#9888) in numba-cuda (#106)
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.
The original linking implementation for linkable code in device declarations did not consider calls inside callees; this change recurses through the typing to find all calls requiring linkable code.