[NFC][IR2Vec] Initialize Embedding vectors with zeros by default

[svkeerthy]

Initialize Embedding vectors with zeros by default when only size is provided.

[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
• [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 👈 (View in Graphite)
• [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.

[llvmbot]

@llvm/pr-subscribers-llvm-analysis

Author: S. VenkataKeerthy (svkeerthy)

Changes

Initialize Embedding vectors with zeros by default when only size is provided.

---

Full diff: https://github.com/llvm/llvm-project/pull/155690.diff

2 Files Affected:

• (modified) llvm/include/llvm/Analysis/IR2Vec.h (+1-1)
• (modified) llvm/lib/Analysis/IR2Vec.cpp (+4-4)

diff --git a/llvm/include/llvm/Analysis/IR2Vec.h b/llvm/include/llvm/Analysis/IR2Vec.h
index 44932a3385e16..6fb8f736da092 100644
--- a/llvm/include/llvm/Analysis/IR2Vec.h
+++ b/llvm/include/llvm/Analysis/IR2Vec.h
@@ -92,7 +92,7 @@ struct Embedding {
   Embedding(std::vector<double> &&V) : Data(std::move(V)) {}
   Embedding(std::initializer_list<double> IL) : Data(IL) {}
 
-  explicit Embedding(size_t Size) : Data(Size) {}
+  explicit Embedding(size_t Size) : Data(Size, 0.0) {}
   Embedding(size_t Size, double InitialValue) : Data(Size, InitialValue) {}
 
   size_t size() const { return Data.size(); }
diff --git a/llvm/lib/Analysis/IR2Vec.cpp b/llvm/lib/Analysis/IR2Vec.cpp
index 565ec2a6287b7..6b90f1aabacfa 100644
--- a/llvm/lib/Analysis/IR2Vec.cpp
+++ b/llvm/lib/Analysis/IR2Vec.cpp
@@ -155,7 +155,7 @@ void Embedding::print(raw_ostream &OS) const {
 Embedder::Embedder(const Function &F, const Vocabulary &Vocab)
     : F(F), Vocab(Vocab), Dimension(Vocab.getDimension()),
       OpcWeight(::OpcWeight), TypeWeight(::TypeWeight), ArgWeight(::ArgWeight),
-      FuncVector(Embedding(Dimension, 0)) {}
+      FuncVector(Embedding(Dimension)) {}
 
 std::unique_ptr<Embedder> Embedder::create(IR2VecKind Mode, const Function &F,
                                            const Vocabulary &Vocab) {
@@ -472,7 +472,7 @@ void IR2VecVocabAnalysis::generateNumMappedVocab() {
 
   // Handle Opcodes
   std::vector<Embedding> NumericOpcodeEmbeddings(Vocabulary::MaxOpcodes,
-                                                 Embedding(Dim, 0));
+                                                 Embedding(Dim));
   NumericOpcodeEmbeddings.reserve(Vocabulary::MaxOpcodes);
   for (unsigned Opcode : seq(0u, Vocabulary::MaxOpcodes)) {
     StringRef VocabKey = Vocabulary::getVocabKeyForOpcode(Opcode + 1);
@@ -487,7 +487,7 @@ void IR2VecVocabAnalysis::generateNumMappedVocab() {
 
   // Handle Types - only canonical types are present in vocabulary
   std::vector<Embedding> NumericTypeEmbeddings(Vocabulary::MaxCanonicalTypeIDs,
-                                               Embedding(Dim, 0));
+                                               Embedding(Dim));
   NumericTypeEmbeddings.reserve(Vocabulary::MaxCanonicalTypeIDs);
   for (unsigned CTypeID : seq(0u, Vocabulary::MaxCanonicalTypeIDs)) {
     StringRef VocabKey = Vocabulary::getVocabKeyForCanonicalTypeID(
@@ -503,7 +503,7 @@ void IR2VecVocabAnalysis::generateNumMappedVocab() {
 
   // Handle Arguments/Operands
   std::vector<Embedding> NumericArgEmbeddings(Vocabulary::MaxOperandKinds,
-                                              Embedding(Dim, 0));
+                                              Embedding(Dim));
   NumericArgEmbeddings.reserve(Vocabulary::MaxOperandKinds);
   for (unsigned OpKind : seq(0u, Vocabulary::MaxOperandKinds)) {
     Vocabulary::OperandKind Kind = static_cast<Vocabulary::OperandKind>(OpKind);

[llvmbot]

@llvm/pr-subscribers-mlgo

Author: S. VenkataKeerthy (svkeerthy)

Changes

Initialize Embedding vectors with zeros by default when only size is provided.

---

Full diff: https://github.com/llvm/llvm-project/pull/155690.diff

2 Files Affected:

• (modified) llvm/include/llvm/Analysis/IR2Vec.h (+1-1)
• (modified) llvm/lib/Analysis/IR2Vec.cpp (+4-4)

diff --git a/llvm/include/llvm/Analysis/IR2Vec.h b/llvm/include/llvm/Analysis/IR2Vec.h
index 44932a3385e16..6fb8f736da092 100644
--- a/llvm/include/llvm/Analysis/IR2Vec.h
+++ b/llvm/include/llvm/Analysis/IR2Vec.h
@@ -92,7 +92,7 @@ struct Embedding {
   Embedding(std::vector<double> &&V) : Data(std::move(V)) {}
   Embedding(std::initializer_list<double> IL) : Data(IL) {}
 
-  explicit Embedding(size_t Size) : Data(Size) {}
+  explicit Embedding(size_t Size) : Data(Size, 0.0) {}
   Embedding(size_t Size, double InitialValue) : Data(Size, InitialValue) {}
 
   size_t size() const { return Data.size(); }
diff --git a/llvm/lib/Analysis/IR2Vec.cpp b/llvm/lib/Analysis/IR2Vec.cpp
index 565ec2a6287b7..6b90f1aabacfa 100644
--- a/llvm/lib/Analysis/IR2Vec.cpp
+++ b/llvm/lib/Analysis/IR2Vec.cpp
@@ -155,7 +155,7 @@ void Embedding::print(raw_ostream &OS) const {
 Embedder::Embedder(const Function &F, const Vocabulary &Vocab)
     : F(F), Vocab(Vocab), Dimension(Vocab.getDimension()),
       OpcWeight(::OpcWeight), TypeWeight(::TypeWeight), ArgWeight(::ArgWeight),
-      FuncVector(Embedding(Dimension, 0)) {}
+      FuncVector(Embedding(Dimension)) {}
 
 std::unique_ptr<Embedder> Embedder::create(IR2VecKind Mode, const Function &F,
                                            const Vocabulary &Vocab) {
@@ -472,7 +472,7 @@ void IR2VecVocabAnalysis::generateNumMappedVocab() {
 
   // Handle Opcodes
   std::vector<Embedding> NumericOpcodeEmbeddings(Vocabulary::MaxOpcodes,
-                                                 Embedding(Dim, 0));
+                                                 Embedding(Dim));
   NumericOpcodeEmbeddings.reserve(Vocabulary::MaxOpcodes);
   for (unsigned Opcode : seq(0u, Vocabulary::MaxOpcodes)) {
     StringRef VocabKey = Vocabulary::getVocabKeyForOpcode(Opcode + 1);
@@ -487,7 +487,7 @@ void IR2VecVocabAnalysis::generateNumMappedVocab() {
 
   // Handle Types - only canonical types are present in vocabulary
   std::vector<Embedding> NumericTypeEmbeddings(Vocabulary::MaxCanonicalTypeIDs,
-                                               Embedding(Dim, 0));
+                                               Embedding(Dim));
   NumericTypeEmbeddings.reserve(Vocabulary::MaxCanonicalTypeIDs);
   for (unsigned CTypeID : seq(0u, Vocabulary::MaxCanonicalTypeIDs)) {
     StringRef VocabKey = Vocabulary::getVocabKeyForCanonicalTypeID(
@@ -503,7 +503,7 @@ void IR2VecVocabAnalysis::generateNumMappedVocab() {
 
   // Handle Arguments/Operands
   std::vector<Embedding> NumericArgEmbeddings(Vocabulary::MaxOperandKinds,
-                                              Embedding(Dim, 0));
+                                              Embedding(Dim));
   NumericArgEmbeddings.reserve(Vocabulary::MaxOperandKinds);
   for (unsigned OpKind : seq(0u, Vocabulary::MaxOperandKinds)) {
     Vocabulary::OperandKind Kind = static_cast<Vocabulary::OperandKind>(OpKind);

[svkeerthy]

Merge activity

• Sep 4, 8:29 PM UTC: A user started a stack merge that includes this pull request via Graphite.
• Sep 4, 8:31 PM UTC: @svkeerthy merged this pull request with Graphite.