You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
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
This PR allows us to optionally compile in dynamic runtime support for nvMatmulHeuristics. Users will still need to pip install nvidia-matmul-heuristics and then set their LD_LIBRARY_PATH to the installed location (on my machine it is at $HOME/.local/lib/python3.12/site-packages/nvidia/nvMatmulHeuristics/lib). If the user doesn't do this, we gracefully just print a warning and use the default CutlassParams, avoiding a dynamic linker error if we had done a compile-time link against libnvMatmulHeuristics.so.
I have not yet verified that this improves performance. This PR is so far just an exploration of the mechanics of calling nvmmh dynamically.
NOTE: CMakeLists.txt changes are not yet pushed...
The PR uses TORCH_WARN_ONCE for version mismatch between compiled and runtime nvMatmulHeuristics, but continues execution despite requiring exactly matching versions. This may lead to undefined behavior if the versions are incompatible.
TORCH_WARN_ONCE(
"nvFuser was compiled against nvMatmulHeuristics version ",
NVMMH_VERSION_MAJOR,
".",
NVMMH_VERSION_MINOR,
" but found nvMatmulHeuristics shared library version ",
lib_major,
".",
lib_minor,
". Exactly matching versions are required");
}
Hardcoded problem dimensions (M=N=K=8192) are used when querying nvMatmulHeuristics, which may result in suboptimal or invalid kernel configurations for actual runtime tensor sizes.
constexpruint32_t M = 8192, N = 8192, K = 8192, Batch = 1;
nvmmhMatmulProblem_t problem{M, N, K, Batch, layout};
If nvMatmulHeuristicsCreate succeeds but a subsequent call fails, the created handle is not destroyed before returning, potentially causing resource leaks.
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 PR allows us to optionally compile in dynamic runtime support for nvMatmulHeuristics. Users will still need to
pip install nvidia-matmul-heuristicsand then set theirLD_LIBRARY_PATHto the installed location (on my machine it is at$HOME/.local/lib/python3.12/site-packages/nvidia/nvMatmulHeuristics/lib). If the user doesn't do this, we gracefully just print a warning and use the defaultCutlassParams, avoiding a dynamic linker error if we had done a compile-time link againstlibnvMatmulHeuristics.so.I have not yet verified that this improves performance. This PR is so far just an exploration of the mechanics of calling nvmmh dynamically.
NOTE: CMakeLists.txt changes are not yet pushed...