[CUDA] refactor fatbinary to use --image3#21054
Merged
KornevNikita merged 1 commit intointel:syclfrom Jan 29, 2026
Merged
Conversation
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
Contributor
Author
|
@intel/dpcpp-clang-driver-reviewers @intel/dpcpp-tools-reviewers fails on BMG are not related with my change |
bratpiorka
force-pushed
the
rrudnick_cuda_fatbin
branch
from
2016b55 to
ced9034
Compare
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
| // CHK-CMDS-AOT-NV-NEXT: clang{{.*}} -o [[CLANGOUT:.*]] -dumpdir a.out.nvptx64.sm_50.img. --target=nvptx64-nvidia-cuda -march={{.*}} | ||
| // CHK-CMDS-AOT-NV-NEXT: ptxas{{.*}} --output-file [[PTXASOUT:.*]] [[CLANGOUT]] | ||
| // CHK-CMDS-AOT-NV-NEXT: fatbinary{{.*}} --create [[FATBINOUT:.*]] --image=profile={{.*}},file=[[CLANGOUT]] --image=profile={{.*}},file=[[PTXASOUT]] | ||
| // CHK-CMDS-AOT-NV-NEXT: fatbinary{{.*}} --create [[FATBINOUT:.*]] --image3=kind=ptx,sm={{(compute_)?50}},file=[[CLANGOUT]] --image3=kind=elf,sm=50,file=[[PTXASOUT]] |
Contributor
There was a problem hiding this comment.
My understanding is, by specifying compute_*, you can compile code for a range of NVidia GPUs and produce PTX files, while specifying sm_* will generate machine code for the exact GPU specified.
Could you please explain this check? sm={{(compute_)?50}}
This will match sm=50 or sm=compute_50
It makes sense to set the kind to PTX only when sm=compute_50 but the regex here would set it for both sm=50 and sm=compute_50.
Please correct if this is not the case.
Contributor
Author
There was a problem hiding this comment.
I simplified the patterns, and now I’m only looking for sm=50, which works well - I checked both CUDA 12.6 and CUDA 13 toolkits
bratpiorka
force-pushed
the
rrudnick_cuda_fatbin
branch
from
ced9034 to
68c0735
Compare
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
srividya-sundaram
approved these changes
Jan 16, 2026
Contributor
srividya-sundaram
left a comment
srividya-sundaram
left a comment
There was a problem hiding this comment.
Driver specific changes LGTM
hchilama
approved these changes
Jan 16, 2026
YuriPlyakhin
approved these changes
Jan 28, 2026
bratpiorka
force-pushed
the
rrudnick_cuda_fatbin
branch
from
68c0735 to
b41feee
Compare
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
bratpiorka
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
Contributor
Author
|
@intel/llvm-gatekeepers please merge |
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
5 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
This change updates clang-linker-wrapper to invoke NVIDIA fatbinary using the --image3=... image specification format (kind + SM + file) instead of the legacy --image=profile=... format, which is no longer supported in CUDA Toolkit version 13 and later.
Since newer CUDA toolkits have removed the legacy --image option, clang-linker-wrapper must use --image3 to maintain compatibility.
Changes
NOTE: upstream already has this changed implemented in: llvm/llvm-project@79d8a26