[mlir][OpenMP] Add the tgt_entry_info attribute #78223
Draft
Conversation
This file contains 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
…entry manager This patch adds an optional field in the create target method to store the offload entry in a custom location and not register the entry in the entry manager. This change is required to enable JIT compilation in MLIR for OpenMP target offload ops, as arrays of entries are handled differently for standalone MLIR compilation.
This patch adds the `OffloadHandler` utility class for creating LLVM offload entries. LLVM offload entries hold information on offload symbols; for example, for a GPU kernel, this includes its host address to identify the kernel and the kernel identifier in the binary. Arrays of offload entries can be used to register functions within the CUDA/HIP runtime. Libomptarget also uses these entries to register OMP target offload kernels and variables. This patch is 1/4 on introducing the `OffloadEmbeddingAttr` GPU translation attribute.
This patch adds the `omp.tgt_entry_info` attribute. This attribute provides information to identify offload entries uniquely and partially reflects the information in the `llvm::TargetRegionEntryInfo` struct. An `info` parameter was added to `TargetOp` to specify the offload entry information from the operation explicitly. Both the host and device versions must have the same `info` attribute; otherwise, the constructs won't correctly map between each other. This patch is required to enable JIT compilation for the OMP dialect, as the entry array has to be fully constructed in the IR instead of using sections to implicitly construct it. Note: Ignore the base commits.
fabianmcg
added a commit
to fabianmcg/llvm-project
that referenced
this pull request
Jan 16, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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 patch adds the
omp.tgt_entry_infoattribute. This attribute providesinformation to identify offload entries uniquely and partially reflects the
information in the
llvm::TargetRegionEntryInfostruct.An
infoparameter was added toTargetOpto specify the offload entryinformation from the operation explicitly. Both the host and device versions
must have the same
infoattribute; otherwise, the constructs won't correctlymap between each other.
This patch is required to enable JIT compilation for the OMP dialect, as
the entry array has to be fully constructed in the IR instead of using
sections to implicitly construct it.
Note: Ignore the base commits.