[llvm] [CodeGen][StaticDataPartitioning]Place local-linkage global variables in hot or unlikely prefixed sections based on profile information (PR #125756)

[David Li via llvm-commits]

================
@@ -0,0 +1,68 @@
+#ifndef LLVM_ANALYSIS_STATICDATAPROFILEINFO_H
+#define LLVM_ANALYSIS_STATICDATAPROFILEINFO_H
+
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/DenseSet.h"
+#include "llvm/IR/Constant.h"
+#include "llvm/Pass.h"
+
+namespace llvm {
+
+/// A class that holds the constants that represent static data and their
+/// profile information and provides methods to operate on them.
+class StaticDataProfileInfo {
+public:
+  /// Accummulate the profile count of a constant that will be lowered to static
+  /// data sections.
+  DenseMap<const Constant *, uint64_t> ConstantProfileCounts;
+
+  /// Keeps track of the constants that are seen at least once without profile
+  /// counts.
+  DenseSet<const Constant *> ConstantWithoutCounts;
+
+public:
+  StaticDataProfileInfo() = default;
+
+  /// If \p Count is not nullopt, add it to the profile count of the constant \p
+  /// C in a saturating way, and clamp the count to \p getInstrMaxCountValue if
+  /// the result exceeds it. Otherwise, mark the constant as having no profile
+  /// count.
+  void addConstantProfileCount(const Constant *C,
+                               std::optional<uint64_t> Count);
+
+  /// If \p C has a count, return it. Otherwise, return std::nullopt.
----------------
david-xl wrote:

When it returns nullopt, it does not mean the constant has unknown profile count -- it can just mean the profile count has not been set/aggregate for the constant.  When the analysis phase is done, returning nullopt seems the same has 'hasUnknownCount()' below. Perhaps clarify in the comment.

https://github.com/llvm/llvm-project/pull/125756