[llvm] [BOLT] Add support for safe-icf (PR #116275)

[Amir Ayupov via llvm-commits]

================
@@ -27,15 +27,49 @@ class IdenticalCodeFolding : public BinaryFunctionPass {
       return false;
     if (BF.hasSDTMarker())
       return false;
+    if (BF.hasAddressTaken())
+      return false;
     return BinaryFunctionPass::shouldOptimize(BF);
   }
 
 public:
+  enum class ICFLevel {
+    None, // No ICF. (Default)
+    Safe, // Safe ICF for all sections.
+    All,  // Aggressive ICF for code.
+  };
   explicit IdenticalCodeFolding(const cl::opt<bool> &PrintPass)
       : BinaryFunctionPass(PrintPass) {}
 
   const char *getName() const override { return "identical-code-folding"; }
   Error runOnFunctions(BinaryContext &BC) override;
+
+private:
+  /// Bit vector of memory addresses of vtables.
+  llvm::BitVector VtableBitVector;
----------------
aaupov wrote:

Yes, random access is O(N), but sequential is not:
> testing/setting bits in a SparseBitVector is O(distance away from last set bit).

Vtables should occupy a vanishingly small part of the binary, so sparse shouldn't be too inefficient. You can also drop an explicit size. Can you check the two and compare peak RSS (disabling debug info update) and pass wall time? 

Or let @maksfb weigh in.

https://github.com/llvm/llvm-project/pull/116275