[llvm] [StaticDataLayout][PGO] Add profile format for static data layout, and the classes to operate on the profiles. (PR #138170)

[Mingming Liu via llvm-commits]

================
@@ -0,0 +1,246 @@
+#include "llvm/ProfileData/DataAccessProf.h"
+#include "llvm/ADT/DenseMapInfoVariant.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ProfileData/InstrProf.h"
+#include "llvm/Support/Compression.h"
+#include "llvm/Support/Endian.h"
+#include "llvm/Support/Errc.h"
+#include "llvm/Support/Error.h"
+#include "llvm/Support/StringSaver.h"
+#include "llvm/Support/raw_ostream.h"
+#include <sys/types.h>
+
+namespace llvm {
+namespace data_access_prof {
+
+// If `Map` has an entry keyed by `Str`, returns the entry iterator. Otherwise,
+// creates an owned copy of `Str`, adds a map entry for it and returns the
+// iterator.
+static MapVector<StringRef, uint64_t>::iterator
+saveStringToMap(MapVector<StringRef, uint64_t> &Map,
+                llvm::UniqueStringSaver &saver, StringRef Str) {
+  auto [Iter, Inserted] = Map.try_emplace(saver.save(Str), Map.size());
+  return Iter;
+}
+
+const DataAccessProfRecord *
+DataAccessProfData::getProfileRecord(const SymbolID SymbolID) const {
+  auto Key = SymbolID;
+  if (std::holds_alternative<StringRef>(SymbolID))
+    Key = InstrProfSymtab::getCanonicalName(std::get<StringRef>(SymbolID));
+
+  auto It = Records.find(Key);
+  if (It != Records.end())
+    return &It->second;
+
+  return nullptr;
+}
+
+bool DataAccessProfData::isKnownColdSymbol(const SymbolID SymID) const {
+  if (std::holds_alternative<uint64_t>(SymID))
+    return KnownColdHashes.count(std::get<uint64_t>(SymID));
+  return KnownColdSymbols.count(std::get<StringRef>(SymID));
+}
+
+Error DataAccessProfData::addSymbolizedDataAccessProfile(SymbolID Symbol,
+                                                         uint64_t AccessCount) {
+  uint64_t RecordID = -1;
+  bool IsStringLiteral = false;
+  SymbolID Key;
+  if (std::holds_alternative<uint64_t>(Symbol)) {
+    RecordID = std::get<uint64_t>(Symbol);
+    Key = RecordID;
+    IsStringLiteral = true;
+  } else {
+    StringRef SymbolName = std::get<StringRef>(Symbol);
+    if (SymbolName.empty())
+      return make_error<StringError>("Empty symbol name",
+                                     llvm::errc::invalid_argument);
+
+    StringRef CanonicalName = InstrProfSymtab::getCanonicalName(SymbolName);
+    Key = CanonicalName;
+    RecordID = saveStringToMap(StrToIndexMap, saver, CanonicalName)->second;
+    IsStringLiteral = false;
+  }
+
+  auto [Iter, Inserted] = Records.try_emplace(
+      Key, DataAccessProfRecord{RecordID, AccessCount, IsStringLiteral});
----------------
mingmingl-llvm wrote:

The updated PR adds a constructor for `DataAccessProfRecord` and does `Records.try_emplace(Key, RecordID, AccessCount, IsStringLiteral);` as suggested. I also re-configured my cmake to set `-DCMAKE_C_COMPILER` and `-DCMAKE_CXX_COMPILER` to clang in the environment. Now I no longer see `missing field 'Locations' initializer`.

In the future when C++20 is available for LLVM developer, we can possibly just  use designated initializer here.

https://github.com/llvm/llvm-project/pull/138170