[llvm] [BOLT] Fix debug line emission for functions in multiple compilation units (PR #151230)

[Paschalis Mpeis via llvm-commits]

================
@@ -852,6 +832,100 @@ class DwarfLineTable {
   // Returns DWARF Version for this line table.
   uint16_t getDwarfVersion() const { return DwarfVersion; }
 };
+
+/// ClusteredRows represents a collection of debug line table row references.
+/// Since a Binary function can belong to multiple compilation units (CUs),
+/// a single MCInst can have multiple debug line table rows associated with it
+/// from different CUs. This class manages such clustered row references.
+///
+/// MEMORY LAYOUT AND DESIGN:
+/// This class uses a flexible array member pattern to store all
+/// DebugLineTableRowRef elements in a single contiguous memory allocation.
+/// The memory layout is:
+///
+/// +------------------+
+/// | ClusteredRows    |  <- Object header (Size + first element)
+/// | - Size           |
+/// | - Raws (element) |  <- First DebugLineTableRowRef element
+/// +------------------+
+/// | element[1]       |  <- Additional DebugLineTableRowRef elements
+/// | element[2]       |     stored immediately after the object
+/// | ...              |
+/// | element[Size-1]  |
+/// +------------------+
+///
+/// PERFORMANCE BENEFITS:
+/// - Single memory allocation: All elements are stored in one contiguous block,
+///   eliminating the need for separate heap allocations for the array.
+/// - No extra dereferencing: Elements are accessed directly via pointer
+///   arithmetic (beginPtr() + offset) rather than through an additional
+///   pointer indirection.
+/// - Cache locality: All elements are guaranteed to be adjacent in memory,
+///   improving cache performance during iteration.
+/// - Memory efficiency: No overhead from separate pointer storage or
+///   fragmented allocations.
----------------
paschalis-mpeis wrote:

Could keep maybe in the commit message and add a short 1-2 lines summary here?

https://github.com/llvm/llvm-project/pull/151230