Addressed feedback in #76903

Created using spr 1.3.4

Author: aaupov

bolt/include/bolt/Profile/BoltAddressTranslation.h

@@ -79,21 +79,10 @@ class BoltAddressTranslation {
 
   BoltAddressTranslation() {}
 
-  /// Write the serialized address translation table for a function.
-  template <bool Cold>
-  void writeMaps(std::map<uint64_t, MapTy> &Maps, uint64_t &PrevAddress,
-                 raw_ostream &OS);
-
   /// Write the serialized address translation tables for each reordered
   /// function
   void write(const BinaryContext &BC, raw_ostream &OS);
 
-  /// Read the serialized address translation table for a function.
-  /// Return a parse error if failed.
-  template <bool Cold>
-  void parseMaps(std::vector<uint64_t> &HotFuncs, uint64_t &PrevAddress,
-                 DataExtractor &DE, uint64_t &Offset, Error &Err);
-
   /// Read the serialized address translation tables and load them internally
   /// in memory. Return a parse error if failed.
   std::error_code parse(StringRef Buf);
@@ -136,6 +125,17 @@ class BoltAddressTranslation {
   void writeEntriesForBB(MapTy &Map, const BinaryBasicBlock &BB,
                          uint64_t FuncAddress, uint64_t FuncInputAddress);
 
+  /// Write the serialized address translation table for a function.
+  template <bool Cold>
+  void writeMaps(std::map<uint64_t, MapTy> &Maps, uint64_t &PrevAddress,
+                 raw_ostream &OS);
+
+  /// Read the serialized address translation table for a function.
+  /// Return a parse error if failed.
+  template <bool Cold>
+  void parseMaps(std::vector<uint64_t> &HotFuncs, uint64_t &PrevAddress,
+                 DataExtractor &DE, uint64_t &Offset, Error &Err);
+
   /// Returns the bitmask with set bits corresponding to indices of BRANCHENTRY
   /// entries in function address translation map.
   APInt calculateBranchEntriesBitMask(MapTy &Map, size_t EqualElems);
@@ -145,7 +145,6 @@ class BoltAddressTranslation {
   size_t getNumEqualOffsets(const MapTy &Map) const;
 
   std::map<uint64_t, MapTy> Maps;
-  std::map<uint64_t, MapTy> ColdMaps;
 
   using BBHashMap = std::unordered_map<uint32_t, size_t>;
   std::unordered_map<uint64_t, std::pair<size_t, BBHashMap>> FuncHashes;

bolt/lib/Profile/BoltAddressTranslation.cpp

@@ -108,18 +108,17 @@ void BoltAddressTranslation::write(const BinaryContext &BC, raw_ostream &OS) {
       for (const BinaryBasicBlock *const BB : FF)
         writeEntriesForBB(Map, *BB, FF.getAddress(), Function.getAddress());
 
-      ColdMaps.emplace(FF.getAddress(), std::move(Map));
+      Maps.emplace(FF.getAddress(), std::move(Map));
       ColdPartSource.emplace(FF.getAddress(), Function.getOutputAddress());
     }
   }
 
   // Output addresses are delta-encoded
   uint64_t PrevAddress = 0;
   writeMaps</*Cold=*/false>(Maps, PrevAddress, OS);
-  writeMaps</*Cold=*/true>(ColdMaps, PrevAddress, OS);
+  writeMaps</*Cold=*/true>(Maps, PrevAddress, OS);
 
-  outs() << "BOLT-INFO: Wrote " << Maps.size() + ColdMaps.size()
-         << " BAT maps\n";
+  outs() << "BOLT-INFO: Wrote " << Maps.size() << " BAT maps\n";
   outs() << "BOLT-INFO: Wrote " << FuncHashes.size() << " function and "
          << std::accumulate(FuncHashes.begin(), FuncHashes.end(), 0ull,
                             [](size_t Acc, const auto &B) {
@@ -159,13 +158,19 @@ size_t BoltAddressTranslation::getNumEqualOffsets(const MapTy &Map) const {
 template <bool Cold>
 void BoltAddressTranslation::writeMaps(std::map<uint64_t, MapTy> &Maps,
                                        uint64_t &PrevAddress, raw_ostream &OS) {
-  const uint32_t NumFuncs = Maps.size();
+  const uint32_t NumFuncs =
+      llvm::count_if(llvm::make_first_range(Maps), [&](const uint64_t Address) {
+        return Cold == ColdPartSource.count(Address);
+      });
   encodeULEB128(NumFuncs, OS);
   LLVM_DEBUG(dbgs() << "Writing " << NumFuncs << (Cold ? " cold" : "")
                     << " functions for BAT.\n");
   size_t PrevIndex = 0;
   for (auto &MapEntry : Maps) {
     const uint64_t Address = MapEntry.first;
+    // Only process cold fragments in cold mode, and vice versa.
+    if (Cold != ColdPartSource.count(Address))
+      continue;
     const uint64_t HotInputAddress =
         ReverseMap[Cold ? ColdPartSource[Address] : Address];
     MapTy &Map = MapEntry.second;

clang/cmake/caches/Fuchsia.cmake

@@ -38,9 +38,25 @@ set(_FUCHSIA_BOOTSTRAP_PASSTHROUGH
   CURL_ROOT
   OpenSSL_ROOT
   httplib_ROOT
+
+  # Deprecated
   CursesAndPanel_ROOT
+
+  CURSES_INCLUDE_DIRS
+  CURSES_LIBRARIES
+  PANEL_LIBRARIES
+
+  # Deprecated
   Terminfo_ROOT
+
+  Terminfo_LIBRARIES
+
+  # Deprecated
   LibEdit_ROOT
+
+  LibEdit_INCLUDE_DIRS
+  LibEdit_LIBRARIES
+
   FUCHSIA_ENABLE_LLDB
   LLDB_ENABLE_CURSES
   LLDB_ENABLE_LIBEDIT

flang/include/flang/Common/Fortran.h

@@ -87,6 +87,10 @@ ENUM_CLASS(CUDASubprogramAttrs, Host, Device, HostDevice, Global, Grid_Global)
 // CUDA data attributes; mutually exclusive
 ENUM_CLASS(CUDADataAttr, Constant, Device, Managed, Pinned, Shared, Texture)
 
+// OpenACC device types
+ENUM_CLASS(
+    OpenACCDeviceType, Star, Default, Nvidia, Radeon, Host, Multicore, None)
+
 // OpenMP atomic_default_mem_order clause allowed values
 ENUM_CLASS(OmpAtomicDefaultMemOrderType, SeqCst, AcqRel, Relaxed)
 

flang/include/flang/Parser/dump-parse-tree.h

@@ -48,6 +48,7 @@ class ParseTreeDumper {
   NODE(std, uint64_t)
   NODE_ENUM(common, CUDADataAttr)
   NODE_ENUM(common, CUDASubprogramAttrs)
+  NODE_ENUM(common, OpenACCDeviceType)
   NODE(format, ControlEditDesc)
   NODE(format::ControlEditDesc, Kind)
   NODE(format, DerivedTypeDataEditDesc)
@@ -101,7 +102,7 @@ class ParseTreeDumper {
   NODE(parser, AccSelfClause)
   NODE(parser, AccStandaloneDirective)
   NODE(parser, AccDeviceTypeExpr)
-  NODE_ENUM(parser::AccDeviceTypeExpr, Device)
+
   NODE(parser, AccDeviceTypeExprList)
   NODE(parser, AccTileExpr)
   NODE(parser, AccTileExprList)

flang/include/flang/Parser/parse-tree.h

@@ -4072,8 +4072,8 @@ struct AccWaitArgument {
 };
 
 struct AccDeviceTypeExpr {
-  ENUM_CLASS(Device, Star, Default, Nvidia, Radeon, Host, Multicore)
-  WRAPPER_CLASS_BOILERPLATE(AccDeviceTypeExpr, Device);
+  WRAPPER_CLASS_BOILERPLATE(
+      AccDeviceTypeExpr, Fortran::common::OpenACCDeviceType);
   CharBlock source;
 };
 

flang/include/flang/Semantics/symbol.h

@@ -112,7 +112,8 @@ class WithBindName {
   bool isExplicitBindName_{false};
 };
 
-class OpenACCRoutineInfo {
+// Device type specific OpenACC routine information
+class OpenACCRoutineDeviceTypeInfo {
 public:
   bool isSeq() const { return isSeq_; }
   void set_isSeq(bool value = true) { isSeq_ = value; }
@@ -124,21 +125,43 @@ class OpenACCRoutineInfo {
   void set_isGang(bool value = true) { isGang_ = value; }
   unsigned gangDim() const { return gangDim_; }
   void set_gangDim(unsigned value) { gangDim_ = value; }
-  bool isNohost() const { return isNohost_; }
-  void set_isNohost(bool value = true) { isNohost_ = value; }
   const std::string *bindName() const {
     return bindName_ ? &*bindName_ : nullptr;
   }
   void set_bindName(std::string &&name) { bindName_ = std::move(name); }
+  void set_dType(Fortran::common::OpenACCDeviceType dType) {
+    deviceType_ = dType;
+  }
+  Fortran::common::OpenACCDeviceType dType() const { return deviceType_; }
 
 private:
   bool isSeq_{false};
   bool isVector_{false};
   bool isWorker_{false};
   bool isGang_{false};
   unsigned gangDim_{0};
-  bool isNohost_{false};
   std::optional<std::string> bindName_;
+  Fortran::common::OpenACCDeviceType deviceType_{
+      Fortran::common::OpenACCDeviceType::None};
+};
+
+// OpenACC routine information. Device independent info are stored on the
+// OpenACCRoutineInfo instance while device dependent info are stored
+// in as objects in the OpenACCRoutineDeviceTypeInfo list.
+class OpenACCRoutineInfo : public OpenACCRoutineDeviceTypeInfo {
+public:
+  bool isNohost() const { return isNohost_; }
+  void set_isNohost(bool value = true) { isNohost_ = value; }
+  std::list<OpenACCRoutineDeviceTypeInfo> &deviceTypeInfos() {
+    return deviceTypeInfos_;
+  }
+  void add_deviceTypeInfo(OpenACCRoutineDeviceTypeInfo &info) {
+    deviceTypeInfos_.push_back(info);
+  }
+
+private:
+  std::list<OpenACCRoutineDeviceTypeInfo> deviceTypeInfos_;
+  bool isNohost_{false};
 };
 
 // A subroutine or function definition, or a subprogram interface defined

flang/lib/Lower/OpenACC.cpp

@@ -1483,20 +1483,22 @@ genAsyncClause(Fortran::lower::AbstractConverter &converter,
 }
 
 static mlir::acc::DeviceType
-getDeviceType(Fortran::parser::AccDeviceTypeExpr::Device device) {
+getDeviceType(Fortran::common::OpenACCDeviceType device) {
   switch (device) {
-  case Fortran::parser::AccDeviceTypeExpr::Device::Star:
+  case Fortran::common::OpenACCDeviceType::Star:
     return mlir::acc::DeviceType::Star;
-  case Fortran::parser::AccDeviceTypeExpr::Device::Default:
+  case Fortran::common::OpenACCDeviceType::Default:
     return mlir::acc::DeviceType::Default;
-  case Fortran::parser::AccDeviceTypeExpr::Device::Nvidia:
+  case Fortran::common::OpenACCDeviceType::Nvidia:
     return mlir::acc::DeviceType::Nvidia;
-  case Fortran::parser::AccDeviceTypeExpr::Device::Radeon:
+  case Fortran::common::OpenACCDeviceType::Radeon:
     return mlir::acc::DeviceType::Radeon;
-  case Fortran::parser::AccDeviceTypeExpr::Device::Host:
+  case Fortran::common::OpenACCDeviceType::Host:
     return mlir::acc::DeviceType::Host;
-  case Fortran::parser::AccDeviceTypeExpr::Device::Multicore:
+  case Fortran::common::OpenACCDeviceType::Multicore:
     return mlir::acc::DeviceType::Multicore;
+  case Fortran::common::OpenACCDeviceType::None:
+    return mlir::acc::DeviceType::None;
   }
   return mlir::acc::DeviceType::None;
 }

flang/lib/Parser/CMakeLists.txt

@@ -30,7 +30,6 @@ add_flang_library(FortranParser
   LINK_COMPONENTS
   Support
   FrontendOpenACC
-  FrontendOpenMP
 
   DEPENDS
   omp_gen

flang/lib/Parser/openacc-parsers.cpp

@@ -54,13 +54,13 @@ TYPE_PARSER(construct<AccSizeExpr>(scalarIntExpr) ||
 TYPE_PARSER(construct<AccSizeExprList>(nonemptyList(Parser<AccSizeExpr>{})))
 
 TYPE_PARSER(sourced(construct<AccDeviceTypeExpr>(
-    first("*" >> pure(AccDeviceTypeExpr::Device::Star),
-        "DEFAULT" >> pure(AccDeviceTypeExpr::Device::Default),
-        "NVIDIA" >> pure(AccDeviceTypeExpr::Device::Nvidia),
-        "ACC_DEVICE_NVIDIA" >> pure(AccDeviceTypeExpr::Device::Nvidia),
-        "RADEON" >> pure(AccDeviceTypeExpr::Device::Radeon),
-        "HOST" >> pure(AccDeviceTypeExpr::Device::Host),
-        "MULTICORE" >> pure(AccDeviceTypeExpr::Device::Multicore)))))
+    first("*" >> pure(Fortran::common::OpenACCDeviceType::Star),
+        "DEFAULT" >> pure(Fortran::common::OpenACCDeviceType::Default),
+        "NVIDIA" >> pure(Fortran::common::OpenACCDeviceType::Nvidia),
+        "ACC_DEVICE_NVIDIA" >> pure(Fortran::common::OpenACCDeviceType::Nvidia),
+        "RADEON" >> pure(Fortran::common::OpenACCDeviceType::Radeon),
+        "HOST" >> pure(Fortran::common::OpenACCDeviceType::Host),
+        "MULTICORE" >> pure(Fortran::common::OpenACCDeviceType::Multicore)))))
 
 TYPE_PARSER(
     construct<AccDeviceTypeExprList>(nonemptyList(Parser<AccDeviceTypeExpr>{})))