commit e29bb9a038245320164c5890d1a75843e4a664ef
author S. VenkataKeerthy <31350914+svkeerthy@users.noreply.github.com> Tue Jun 17 10:57:52 2025 -0700
committer GitHub <noreply@github.com> Tue Jun 17 10:57:52 2025 -0700
tree 12ef077166016675fe598cfd2f3491bf99f0dffe
parent 0cfc59ff51720ee60a71dd34077fc161886a3701

[IR2Vec] Consider only reachable BBs and non-debug instructions (#143476)

Changes to consider BBs that are reachable from the entry block. Similarly we skip debug instruction while computing the embeddings.

(Tracking issue - #141817)

llvm/lib/Analysis/IR2Vec.cpp

diff --git a/llvm/lib/Analysis/IR2Vec.cpp b/llvm/lib/Analysis/IR2Vec.cpp
index 0f7303c..fa38c35 100644
--- a/llvm/lib/Analysis/IR2Vec.cpp
+++ b/llvm/lib/Analysis/IR2Vec.cpp
@@ -13,7 +13,9 @@
 
 #include "llvm/Analysis/IR2Vec.h"
 
+#include "llvm/ADT/DepthFirstIterator.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/IR/CFG.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/PassManager.h"
 #include "llvm/Support/Debug.h"
@@ -190,7 +192,8 @@
 void SymbolicEmbedder::computeEmbeddings(const BasicBlock &BB) const {
   Embedding BBVector(Dimension, 0);
 
-  for (const auto &I : BB) {
+  // We consider only the non-debug and non-pseudo instructions
+  for (const auto &I : BB.instructionsWithoutDebug()) {
     Embedding InstVector(Dimension, 0);
 
     const auto OpcVec = lookupVocab(I.getOpcodeName());
@@ -215,9 +218,11 @@
 void SymbolicEmbedder::computeEmbeddings() const {
   if (F.isDeclaration())
     return;
-  for (const auto &BB : F) {
-    computeEmbeddings(BB);
-    FuncVector += BBVecMap[&BB];
+
+  // Consider only the basic blocks that are reachable from entry
+  for (const BasicBlock *BB : depth_first(&F)) {
+    computeEmbeddings(*BB);
+    FuncVector += BBVecMap[BB];
   }
 }