feat(cuda): accept a pre-linked CUBIN in load_module_bytes#7
Open
haixuanTao wants to merge 1 commit into
Open
Conversation
`load_module_bytes` assumed PTX text. Also accept a CUBIN, detected by the ELF magic (`\x7fELF`), loaded via `Ptx::from_binary`. This is required for modules that reference symbols the driver JIT cannot resolve on its own — e.g. libdevice `__nv_*` math — which a toolchain links into a self-contained binary ahead of time. PTX text continues to load unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
load_module_bytesassumed the input was PTX text. This makes it also accept a pre-linked CUBIN, detected by the ELF magic (\x7fELF) and loaded viacudarc::nvrtc::Ptx::from_binary. PTX text continues to load exactly as before.Why
A cubin is required when a module references symbols the CUDA driver JIT cannot resolve on its own — most commonly libdevice
__nv_*math (expf,fmaf, …). In that case a toolchain links libdevice into a self-contained cubin ahead of time, and the backend must load that binary rather than re-JIT the PTX. cudarc'sload_modulealready dispatches on the kind, so the change is just selectingfrom_binaryvsfrom_src.Small, backwards-compatible (text PTX path is untouched).
🤖 Generated with Claude Code