[MIR2Vec] Refactor MIR vocabulary to use opcode-based indexing

[svkeerthy]

Refactor MIRVocabulary to improve opcode lookup and add Section enum for better organization. This is useful for embedder lookups (next patches)

(Tracking issue - #141817)

[svkeerthy]

• [MIR2Vec] Add MIR support to triplet generator script #164332
• [llvm-ir2vec][MIR2Vec] Supporting MIR mode in triplet and entity generation #164329
• [MLGO] Add MIR2Vec embedding documentation #164033
• [NFC][llvm-ir2vec] Standardize error message format using WithColor #164032
• [MIR2Vec][llvm-ir2vec] Add MIR2Vec support to llvm-ir2vec tool #164025
• [MIR2Vec] Handle Operands #163281
• [MIR2Vec] Add embedder for machine instructions #162161
• [MIR2Vec] Refactor MIR vocabulary to use opcode-based indexing #161713 👈 (View in Graphite)
• [IR2Vec] Initial infrastructure for MIR2Vec #161463
• [IR2Vec] Refactor vocabulary to use section-based storage #158376
• [IR2Vec] Add support for Cmp predicates in vocabulary and embeddings #156952
• [NFC][IR2Vec] Initialize Embedding vectors with zeros by default #155690
• [NFC][IR2Vec] Add usage documentation for obtaining embeddings #155516
• [IR2Vec] Refactor vocabulary to use canonical type IDs #155323 : 1 other dependent PR (#155700 )
• [NFC][IR2Vec] Add missed ptrtoaddr in vocab for tests #153094
• [IR2Vec] Make IR2VecCategory externally visible and reuse in llvm-ir2vec cl options #153089
• [IR2Vec][llvm-ir2vec] Supporting flow-aware embeddings #153087
• [IR2Vec] Add support for flow-aware embeddings #152613
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[svkeerthy]

Merge activity

• Oct 7, 11:43 PM UTC: A user started a stack merge that includes this pull request via Graphite.
• Oct 7, 11:44 PM UTC: @svkeerthy merged this pull request with Graphite.

[jyknight]

This introduces UB. In the error case, the nullptr in:

frame #1: 0x000055555ad9f2ee llc`::getMIR2VecVocabulary() at MIR2Vec.cpp:244:14
   241    if (StrVocabMap.empty()) {
   242      if (Error Err = readVocabulary()) {
   243        emitError(std::move(Err), M.getContext());
-> 244        return mir2vec::MIRVocabulary(std::move(StrVocabMap), nullptr);
   245      }
   246    }
   247 

is passed to the constructor which immediately dereferences it and stores it into a reference (invalid for nullptr):

   frame #0: 0x000055555ad9e01d llc`::MIRVocabulary() at MIR2Vec.cpp:0:1
   50  
   51   MIRVocabulary::MIRVocabulary(VocabMap &&OpcodeEntries,
   52                                const TargetInstrInfo *TII)
-> 53       : TII(*TII) {
   54     // Fixme: Use static factory methods for creating vocabularies instead of
   55     // public constructors
   56     // Early return for invalid inputs - creates empty/invalid vocabulary

Either the TII member needs to be a pointer, or else the constructor can't be called with nullptr.

[svkeerthy]

Thanks @jyknight. I will put up a patch shortly.