[DebugInfo][DWARF] Utilize DW_AT_LLVM_stmt_sequence attr in line table lookups

llvm/include/llvm/DebugInfo/DWARF/DWARFDebugLine.h

@@ -209,6 +209,9 @@ class DWARFDebugLine {
     unsigned LastRowIndex;
     bool Empty;
 
+    /// The offset into the line table where this sequence begins
+    uint64_t StmtSeqOffset = UINT64_MAX;
+
     void reset();
 
     static bool orderByHighPC(const Sequence &LHS, const Sequence &RHS) {
@@ -243,8 +246,20 @@ class DWARFDebugLine {
     uint32_t lookupAddress(object::SectionedAddress Address,
                            bool *IsApproximateLine = nullptr) const;
 
-    bool lookupAddressRange(object::SectionedAddress Address, uint64_t Size,
-                            std::vector<uint32_t> &Result) const;
+    /// Fills the Result argument with the indices of the rows that correspond
+    /// to the address range specified by \p Address and \p Size.
+    ///
+    /// \param Address - The starting address of the range.
+    /// \param Size - The size of the address range.
+    /// \param Result - The vector to fill with row indices.
+    /// \param StmtSequenceOffset - if provided, only rows from the sequence
+    /// starting at the matching offset will be added to the result.
+    ///
+    /// Returns true if any rows were found.
+    bool lookupAddressRange(
+        object::SectionedAddress Address, uint64_t Size,
+        std::vector<uint32_t> &Result,
+        std::optional<uint64_t> StmtSequenceOffset = std::nullopt) const;
 
     bool hasFileAtIndex(uint64_t FileIndex) const {
       return Prologue.hasFileAtIndex(FileIndex);
@@ -305,8 +320,20 @@ class DWARFDebugLine {
     uint32_t lookupAddressImpl(object::SectionedAddress Address,
                                bool *IsApproximateLine = nullptr) const;
 
-    bool lookupAddressRangeImpl(object::SectionedAddress Address, uint64_t Size,
-                                std::vector<uint32_t> &Result) const;
+    /// Fills the Result argument with the indices of the rows that correspond
+    /// to the address range specified by \p Address and \p Size.
+    ///
+    /// \param Address - The starting address of the range.
+    /// \param Size - The size of the address range.
+    /// \param Result - The vector to fill with row indices.
+    /// \param StmtSequenceOffset - if provided, only rows from the sequence
+    /// starting at the matching offset will be added to the result.
+    ///
+    /// Returns true if any rows were found.
+    bool
+    lookupAddressRangeImpl(object::SectionedAddress Address, uint64_t Size,
+                           std::vector<uint32_t> &Result,
+                           std::optional<uint64_t> StmtSequenceOffset) const;
   };
 
   const LineTable *getLineTable(uint64_t Offset) const;
@@ -376,7 +403,7 @@ class DWARFDebugLine {
     ParsingState(struct LineTable *LT, uint64_t TableOffset,
                  function_ref<void(Error)> ErrorHandler);
 
-    void resetRowAndSequence();
+    void resetRowAndSequence(uint64_t Offset);
     void appendRowToMatrix();
 
     struct AddrOpIndexDelta {

llvm/lib/DebugInfo/DWARF/DWARFDebugLine.cpp

@@ -531,6 +531,7 @@ void DWARFDebugLine::Sequence::reset() {
   FirstRowIndex = 0;
   LastRowIndex = 0;
   Empty = true;
+  StmtSeqOffset = UINT64_MAX;
 }
 
 DWARFDebugLine::LineTable::LineTable() { clear(); }
@@ -561,13 +562,12 @@ void DWARFDebugLine::LineTable::clear() {
 DWARFDebugLine::ParsingState::ParsingState(
     struct LineTable *LT, uint64_t TableOffset,
     function_ref<void(Error)> ErrorHandler)
-    : LineTable(LT), LineTableOffset(TableOffset), ErrorHandler(ErrorHandler) {
-  resetRowAndSequence();
-}
+    : LineTable(LT), LineTableOffset(TableOffset), ErrorHandler(ErrorHandler) {}
 
-void DWARFDebugLine::ParsingState::resetRowAndSequence() {
+void DWARFDebugLine::ParsingState::resetRowAndSequence(uint64_t Offset) {
   Row.reset(LineTable->Prologue.DefaultIsStmt);
   Sequence.reset();
+  Sequence.StmtSeqOffset = Offset;
 }
 
 void DWARFDebugLine::ParsingState::appendRowToMatrix() {
@@ -848,6 +848,10 @@ Error DWARFDebugLine::LineTable::parse(
     *OS << '\n';
     Row::dumpTableHeader(*OS, /*Indent=*/Verbose ? 12 : 0);
   }
+  // *OffsetPtr points to the end of the prologue - i.e. the start of the first
+  // sequence. So initialize the first sequence offset accordingly.
+  State.resetRowAndSequence(*OffsetPtr);
+
   bool TombstonedAddress = false;
   auto EmitRow = [&] {
     if (!TombstonedAddress) {
@@ -912,7 +916,9 @@ Error DWARFDebugLine::LineTable::parse(
         // into this code path - if it were invalid, the default case would be
         // followed.
         EmitRow();
-        State.resetRowAndSequence();
+        // Cursor now points to right after the end_sequence opcode - so points
+        // to the start of the next sequence - if one exists.
+        State.resetRowAndSequence(Cursor.tell());
         break;
 
       case DW_LNE_set_address:
@@ -1364,66 +1370,111 @@ DWARFDebugLine::LineTable::lookupAddressImpl(object::SectionedAddress Address,
 
 bool DWARFDebugLine::LineTable::lookupAddressRange(
     object::SectionedAddress Address, uint64_t Size,
-    std::vector<uint32_t> &Result) const {
+    std::vector<uint32_t> &Result,
+    std::optional<uint64_t> StmtSequenceOffset) const {
 
   // Search for relocatable addresses
-  if (lookupAddressRangeImpl(Address, Size, Result))
+  if (lookupAddressRangeImpl(Address, Size, Result, StmtSequenceOffset))
     return true;
 
   if (Address.SectionIndex == object::SectionedAddress::UndefSection)
     return false;
 
   // Search for absolute addresses
   Address.SectionIndex = object::SectionedAddress::UndefSection;
-  return lookupAddressRangeImpl(Address, Size, Result);
+  return lookupAddressRangeImpl(Address, Size, Result, StmtSequenceOffset);
 }
 
 bool DWARFDebugLine::LineTable::lookupAddressRangeImpl(
     object::SectionedAddress Address, uint64_t Size,
-    std::vector<uint32_t> &Result) const {
+    std::vector<uint32_t> &Result,
+    std::optional<uint64_t> StmtSequenceOffset) const {
   if (Sequences.empty())
     return false;
-  uint64_t EndAddr = Address.Address + Size;
-  // First, find an instruction sequence containing the given address.
-  DWARFDebugLine::Sequence Sequence;
-  Sequence.SectionIndex = Address.SectionIndex;
-  Sequence.HighPC = Address.Address;
-  SequenceIter LastSeq = Sequences.end();
-  SequenceIter SeqPos = llvm::upper_bound(
-      Sequences, Sequence, DWARFDebugLine::Sequence::orderByHighPC);
-  if (SeqPos == LastSeq || !SeqPos->containsPC(Address))
-    return false;
 
-  SequenceIter StartPos = SeqPos;
+  const uint64_t EndAddr = Address.Address + Size;
+
+  // Helper: Given a sequence and a starting address, find the subset
+  // of rows within [Address, EndAddr) and append them to `Result`.
+  auto addRowsFromSequence = [&](const Sequence &Seq,
+                                 object::SectionedAddress StartAddr,
+                                 uint64_t EndAddress, bool IsFirstSequence) {
+    // If this sequence does not intersect our [StartAddr, EndAddress) range,
+    // do nothing.
+    if (Seq.HighPC <= StartAddr.Address || Seq.LowPC >= EndAddress)
+      return;
 
-  // Add the rows from the first sequence to the vector, starting with the
-  // index we just calculated
+    // For the "first" sequence in the search, we must figure out which row
+    // actually starts within our address range.
+    uint32_t FirstRowIndex = Seq.FirstRowIndex;
+    if (IsFirstSequence)
+      FirstRowIndex = findRowInSeq(Seq, StartAddr);
 
-  while (SeqPos != LastSeq && SeqPos->LowPC < EndAddr) {
-    const DWARFDebugLine::Sequence &CurSeq = *SeqPos;
-    // For the first sequence, we need to find which row in the sequence is the
-    // first in our range.
-    uint32_t FirstRowIndex = CurSeq.FirstRowIndex;
-    if (SeqPos == StartPos)
-      FirstRowIndex = findRowInSeq(CurSeq, Address);
-
-    // Figure out the last row in the range.
+    // Similarly, compute the last row that is within [StartAddr, EndAddress).
     uint32_t LastRowIndex =
-        findRowInSeq(CurSeq, {EndAddr - 1, Address.SectionIndex});
+        findRowInSeq(Seq, {EndAddress - 1, StartAddr.SectionIndex});
     if (LastRowIndex == UnknownRowIndex)
-      LastRowIndex = CurSeq.LastRowIndex - 1;
+      LastRowIndex = Seq.LastRowIndex - 1;
 
     assert(FirstRowIndex != UnknownRowIndex);
     assert(LastRowIndex != UnknownRowIndex);
 
+    // Append all row indices in [FirstRowIndex, LastRowIndex].
     for (uint32_t I = FirstRowIndex; I <= LastRowIndex; ++I) {
       Result.push_back(I);
     }
+  };
 
+  // If a stmt_sequence_offset was provided, do a binary search to find the
+  // single sequence that matches that offset. Then add only its relevant rows.
+  if (StmtSequenceOffset) {
+    // Binary-search the Sequences by their StmtSeqOffset:
+    auto It = llvm::lower_bound(Sequences, *StmtSequenceOffset,
+                                [](const Sequence &Seq, uint64_t OffsetVal) {
+                                  return Seq.StmtSeqOffset < OffsetVal;
+                                });
+
+    // If not found or mismatched, there’s no match to return.
+    if (It == Sequences.end() || It->StmtSeqOffset != *StmtSequenceOffset)
+      return false;
+
+    // Now add rows from the discovered sequence if it intersects [Address,
+    // EndAddr).
+    addRowsFromSequence(*It, Address, EndAddr, /*IsFirstSequence=*/true);
+    return !Result.empty();
+  }
+
+  // Otherwise, fall back to logic of walking sequences by Address.
+  // We first find a sequence containing `Address` (via upper_bound on HighPC),
+  // then proceed forward through overlapping sequences in ascending order.
+
+  // Construct a dummy Sequence to find where `Address` fits by HighPC.
+  DWARFDebugLine::Sequence SearchSeq;
+  SearchSeq.SectionIndex = Address.SectionIndex;
+  SearchSeq.HighPC = Address.Address;
+
+  auto LastSeq = Sequences.end();
+  auto SeqPos = llvm::upper_bound(Sequences, SearchSeq,
+                                  DWARFDebugLine::Sequence::orderByHighPC);
+  if (SeqPos == LastSeq || !SeqPos->containsPC(Address))
+    return false;
+
+  // This marks the first sequence we found that might contain Address.
+  const auto StartPos = SeqPos;
+
+  // Walk forward until sequences no longer intersect [Address, EndAddr).
+  while (SeqPos != LastSeq && SeqPos->LowPC < EndAddr) {
+    // Add rows only if the sequence is in the same section:
+    if (SeqPos->SectionIndex == Address.SectionIndex) {
+      // For the very first sequence, we need the row that lines up with
+      // `Address`.
+      bool IsFirst = (SeqPos == StartPos);
+      addRowsFromSequence(*SeqPos, Address, EndAddr, IsFirst);
+    }
     ++SeqPos;
   }
 
-  return true;
+  return !Result.empty();
 }
 
 std::optional<StringRef>

llvm/unittests/DebugInfo/DWARF/DWARFDebugLineTest.cpp

@@ -6,10 +6,10 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "llvm/DebugInfo/DWARF/DWARFDebugLine.h"
 #include "DwarfGenerator.h"
 #include "DwarfUtils.h"
 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
-#include "llvm/DebugInfo/DWARF/DWARFDebugLine.h"
 #include "llvm/Object/ObjectFile.h"
 #include "llvm/Testing/Support/Error.h"
 #include "gtest/gtest.h"
@@ -2035,4 +2035,117 @@ TEST_F(DebugLineBasicFixture, PrintPathsProperly) {
   EXPECT_THAT(Result.c_str(), MatchesRegex("a dir.b dir.b file"));
 }
 
+/// Test that lookupAddressRange correctly filters rows based on
+/// a statement-sequence offset (simulating DW_AT_LLVM_stmt_sequence).
+///
+/// This test verifies that:
+/// 1. When a statement-sequence offset is provided, lookupAddressRange
+///    only returns rows from the sequence starting at that offset.
+/// 2. When an invalid statement-sequence offset is provided, no rows
+///    are returned.
+/// 3. When no statement-sequence offset is provided, all matching rows
+///    in the table are returned.
+///
+/// We build a line table with two sequences at the same address range
+/// but different line numbers. Then we try lookups with various statement-
+/// sequence offsets to check the filtering logic.
+TEST_F(DebugLineBasicFixture, LookupAddressRangeWithStmtSequenceOffset) {
+  if (!setupGenerator())
+    GTEST_SKIP();
+
+  // Create a line table that has two sequences covering [0x1000, 0x1004).
+  // Each sequence has two rows: addresses at 0x1000 and 0x1004, but
+  // they differ by line numbers (100 vs. 200, etc.).
+  //
+  // We'll pretend the first sequence starts at offset 0x2e in the line table,
+  // the second at 0x42, and we'll also test an invalid offset 0x66.
+
+  LineTable &LT = Gen->addLineTable();
+
+  // First sequence at offset 0x2e: addresses 0x1000(Ln=100), 0x1004(Ln=101)
+  LT.addExtendedOpcode(9, DW_LNE_set_address, {{0x1000U, LineTable::Quad}});
+  LT.addStandardOpcode(DW_LNS_set_prologue_end, {});
+  // Advance the line register by 99 (so line=100) and copy.
+  LT.addStandardOpcode(DW_LNS_advance_line, {{99, LineTable::SLEB}});
+  LT.addStandardOpcode(DW_LNS_copy, {});
+  // 0x4b is a special opcode: address += 4, line += 1 (so line=101).
+  LT.addByte(0x4b);
+  // End this sequence.
+  LT.addExtendedOpcode(1, DW_LNE_end_sequence, {});
+
+  // Second sequence at offset 0x42: addresses 0x1000(Ln=200), 0x1004(Ln=201)
+  LT.addExtendedOpcode(9, DW_LNE_set_address, {{0x1000U, LineTable::Quad}});
+  LT.addStandardOpcode(DW_LNS_set_prologue_end, {});
+  LT.addStandardOpcode(DW_LNS_advance_line, {{199, LineTable::SLEB}});
+  LT.addStandardOpcode(DW_LNS_copy, {});
+  // 0x4b again: address += 4, line += 1 (so line=201).
+  LT.addByte(0x4b);
+  // End this second sequence.
+  LT.addExtendedOpcode(1, DW_LNE_end_sequence, {});
+
+  // Generate the DWARF data.
+  generate();
+
+  // Parse the line table.
+  auto ExpectedLineTable =
+      Line.getOrParseLineTable(LineData, /*Offset=*/0, *Context,
+                               /*DwarfUnit=*/nullptr, RecordRecoverable);
+  ASSERT_THAT_EXPECTED(ExpectedLineTable, Succeeded());
+  const auto *Table = *ExpectedLineTable;
+
+  // The table should have two sequences, each starting at our chosen offsets.
+  ASSERT_EQ(Table->Sequences.size(), 2u);
+
+  // 1) Try looking up with an invalid offset (simulating an invalid
+  //    DW_AT_LLVM_stmt_sequence). We expect no rows.
+  {
+    std::vector<uint32_t> Rows;
+    bool Found = Table->lookupAddressRange(
+        {0x1000, object::SectionedAddress::UndefSection}, /*Size=*/1, Rows,
+        /*StmtSequenceOffset=*/0x66); // invalid offset
+    EXPECT_FALSE(Found);
+    EXPECT_TRUE(Rows.empty());
+  }
+
+  // 2) Look up using the offset 0x2e (our first sequence). We expect
+  //    to get the rows from that sequence only (which for 0x1000 is row #0).
+  {
+    std::vector<uint32_t> Rows;
+    bool Found = Table->lookupAddressRange(
+        {0x1000, object::SectionedAddress::UndefSection}, /*Size=*/1, Rows,
+        /*StmtSequenceOffset=*/0x2e);
+    EXPECT_TRUE(Found);
+    ASSERT_EQ(Rows.size(), 1u);
+    // The first sequence's first row is index 0.
+    EXPECT_EQ(Rows[0], 0u);
+  }
+
+  // 3) Look up using the offset 0x42 (second sequence). For address 0x1000
+  //    in that second sequence, we should see row #2.
+  {
+    std::vector<uint32_t> Rows;
+    bool Found = Table->lookupAddressRange(
+        {0x1000, object::SectionedAddress::UndefSection}, /*Size=*/1, Rows,
+        /*StmtSequenceOffset=*/0x42);
+    EXPECT_TRUE(Found);
+    ASSERT_EQ(Rows.size(), 1u);
+    // The second sequence's first row is index 2 in the table.
+    EXPECT_EQ(Rows[0], 3u);
+  }
+
+  // 4) Look up with no statement-sequence offset specified.
+  //    We should get rows from both sequences for address 0x1000.
+  {
+    std::vector<uint32_t> Rows;
+    bool Found = Table->lookupAddressRange(
+        {0x1000, object::SectionedAddress::UndefSection}, /*Size=*/1, Rows,
+        std::nullopt /* no filter */);
+    EXPECT_TRUE(Found);
+    // The first sequence's row is #0, second's row is #2, so both should
+    // appear.
+    ASSERT_EQ(Rows.size(), 2u);
+    EXPECT_EQ(Rows[0], 0u);
+    EXPECT_EQ(Rows[1], 3u);
+  }
+}
 } // end anonymous namespace