[llvm] [IR2Vec] Adding tests to check multiple invocations of `getFunctionVector()` and `getInstVecMap()` return same results (PR #162365)

[S. VenkataKeerthy via llvm-commits]

https://github.com/svkeerthy updated https://github.com/llvm/llvm-project/pull/162365

>From 2290ff5be630f8d8dd34f5b87cbc372243edfaae Mon Sep 17 00:00:00 2001
From: svkeerthy <venkatakeerthy at google.com>
Date: Mon, 6 Oct 2025 21:34:00 +0000
Subject: [PATCH 1/3] Reinitialize Function Vectors

---
 llvm/lib/Analysis/IR2Vec.cpp | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/llvm/lib/Analysis/IR2Vec.cpp b/llvm/lib/Analysis/IR2Vec.cpp
index 295b6d33525d9..688535161d4b9 100644
--- a/llvm/lib/Analysis/IR2Vec.cpp
+++ b/llvm/lib/Analysis/IR2Vec.cpp
@@ -200,6 +200,8 @@ void Embedder::computeEmbeddings() const {
   if (F.isDeclaration())
     return;
 
+  FuncVector = Embedding(Dimension, 0.0);
+
   // Consider only the basic blocks that are reachable from entry
   for (const BasicBlock *BB : depth_first(&F)) {
     computeEmbeddings(*BB);

>From 1739e5c93d88c917cb00c2cf1f04005082d1d002 Mon Sep 17 00:00:00 2001
From: svkeerthy <venkatakeerthy at google.com>
Date: Tue, 7 Oct 2025 20:19:02 +0000
Subject: [PATCH 2/3] Adding tests

---
 llvm/unittests/Analysis/IR2VecTest.cpp | 54 ++++++++++++++++++++++++++
 1 file changed, 54 insertions(+)

diff --git a/llvm/unittests/Analysis/IR2VecTest.cpp b/llvm/unittests/Analysis/IR2VecTest.cpp
index d136cb6a316b1..ee14596bcb8ca 100644
--- a/llvm/unittests/Analysis/IR2VecTest.cpp
+++ b/llvm/unittests/Analysis/IR2VecTest.cpp
@@ -430,6 +430,60 @@ TEST_F(IR2VecTestFixture, GetFunctionVector_FlowAware) {
   EXPECT_TRUE(FuncVec.approximatelyEquals(Embedding(2, 58.1)));
 }
 
+TEST_F(IR2VecTestFixture, MultipleComputeEmbeddingsConsistency_Symbolic) {
+  auto Emb = Embedder::create(IR2VecKind::Symbolic, *F, *V);
+  ASSERT_TRUE(static_cast<bool>(Emb));
+
+  // Get initial function vector
+  const auto &FuncVec1 = Emb->getFunctionVector();
+  
+  // Compute embeddings again by calling getFunctionVector multiple times
+  const auto &FuncVec2 = Emb->getFunctionVector();
+  const auto &FuncVec3 = Emb->getFunctionVector();
+
+  // All function vectors should be identical
+  EXPECT_TRUE(FuncVec1.approximatelyEquals(FuncVec2));
+  EXPECT_TRUE(FuncVec1.approximatelyEquals(FuncVec3));
+  EXPECT_TRUE(FuncVec2.approximatelyEquals(FuncVec3));
+
+  // Also check that instruction vectors remain consistent
+  const auto &InstMap1 = Emb->getInstVecMap();
+  const auto &InstMap2 = Emb->getInstVecMap();
+  
+  EXPECT_EQ(InstMap1.size(), InstMap2.size());
+  for (const auto &[Inst, Vec1] : InstMap1) {
+    ASSERT_TRUE(InstMap2.count(Inst));
+    EXPECT_TRUE(Vec1.approximatelyEquals(InstMap2.at(Inst)));
+  }
+}
+
+TEST_F(IR2VecTestFixture, MultipleComputeEmbeddingsConsistency_FlowAware) {
+  auto Emb = Embedder::create(IR2VecKind::FlowAware, *F, *V);
+  ASSERT_TRUE(static_cast<bool>(Emb));
+
+  // Get initial function vector
+  const auto &FuncVec1 = Emb->getFunctionVector();
+  
+  // Compute embeddings again by calling getFunctionVector multiple times
+  const auto &FuncVec2 = Emb->getFunctionVector();
+  const auto &FuncVec3 = Emb->getFunctionVector();
+
+  // All function vectors should be identical
+  EXPECT_TRUE(FuncVec1.approximatelyEquals(FuncVec2));
+  EXPECT_TRUE(FuncVec1.approximatelyEquals(FuncVec3));
+  EXPECT_TRUE(FuncVec2.approximatelyEquals(FuncVec3));
+
+  // Also check that instruction vectors remain consistent
+  const auto &InstMap1 = Emb->getInstVecMap();
+  const auto &InstMap2 = Emb->getInstVecMap();
+  
+  EXPECT_EQ(InstMap1.size(), InstMap2.size());
+  for (const auto &[Inst, Vec1] : InstMap1) {
+    ASSERT_TRUE(InstMap2.count(Inst));
+    EXPECT_TRUE(Vec1.approximatelyEquals(InstMap2.at(Inst)));
+  }
+}
+
 static constexpr unsigned MaxOpcodes = Vocabulary::MaxOpcodes;
 [[maybe_unused]]
 static constexpr unsigned MaxTypeIDs = Vocabulary::MaxTypeIDs;

>From a32b12d4e843bbfedc11fcd506aef4598fb88c6f Mon Sep 17 00:00:00 2001
From: svkeerthy <venkatakeerthy at google.com>
Date: Tue, 7 Oct 2025 20:41:10 +0000
Subject: [PATCH 3/3] Fix formatting issues

---
 llvm/unittests/Analysis/IR2VecTest.cpp | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/llvm/unittests/Analysis/IR2VecTest.cpp b/llvm/unittests/Analysis/IR2VecTest.cpp
index ee14596bcb8ca..40b4aa21f2b46 100644
--- a/llvm/unittests/Analysis/IR2VecTest.cpp
+++ b/llvm/unittests/Analysis/IR2VecTest.cpp
@@ -436,7 +436,7 @@ TEST_F(IR2VecTestFixture, MultipleComputeEmbeddingsConsistency_Symbolic) {
 
   // Get initial function vector
   const auto &FuncVec1 = Emb->getFunctionVector();
-  
+
   // Compute embeddings again by calling getFunctionVector multiple times
   const auto &FuncVec2 = Emb->getFunctionVector();
   const auto &FuncVec3 = Emb->getFunctionVector();
@@ -449,7 +449,7 @@ TEST_F(IR2VecTestFixture, MultipleComputeEmbeddingsConsistency_Symbolic) {
   // Also check that instruction vectors remain consistent
   const auto &InstMap1 = Emb->getInstVecMap();
   const auto &InstMap2 = Emb->getInstVecMap();
-  
+
   EXPECT_EQ(InstMap1.size(), InstMap2.size());
   for (const auto &[Inst, Vec1] : InstMap1) {
     ASSERT_TRUE(InstMap2.count(Inst));
@@ -463,7 +463,7 @@ TEST_F(IR2VecTestFixture, MultipleComputeEmbeddingsConsistency_FlowAware) {
 
   // Get initial function vector
   const auto &FuncVec1 = Emb->getFunctionVector();
-  
+
   // Compute embeddings again by calling getFunctionVector multiple times
   const auto &FuncVec2 = Emb->getFunctionVector();
   const auto &FuncVec3 = Emb->getFunctionVector();
@@ -476,7 +476,7 @@ TEST_F(IR2VecTestFixture, MultipleComputeEmbeddingsConsistency_FlowAware) {
   // Also check that instruction vectors remain consistent
   const auto &InstMap1 = Emb->getInstVecMap();
   const auto &InstMap2 = Emb->getInstVecMap();
-  
+
   EXPECT_EQ(InstMap1.size(), InstMap2.size());
   for (const auto &[Inst, Vec1] : InstMap1) {
     ASSERT_TRUE(InstMap2.count(Inst));