Fixed heatmap mode

Created using spr 1.3.4

Author: aaupov

.github/new-prs-labeler.yml

@@ -702,6 +702,7 @@ mlgo:
   - llvm/unittests/CodeGen/ML*
   - llvm/test/CodeGen/MLRegAlloc/**
   - llvm/utils/mlgo-utils/**
+  - llvm/docs/MLGO.rst
 
 tools:llvm-exegesis:
   - llvm/tools/llvm-exegesis/**

bolt/docs/Heatmaps.md

@@ -89,7 +89,13 @@ For the generation, the default bucket size was used with a line size of 128.
 Some useful options are:
 
 ```
--line-size=<uint>   - number of entries per line (default 256)
+-block-size=<uint>  - heatmap bucket size in bytes (default 64)
+-line-size=<uint>   - number of entries per line (default 256).
+                      Use 128 if the heatmap doesn't fit screen horizontally.
 -max-address=<uint> - maximum address considered valid for heatmap (default 4GB)
 -print-mappings     - print mappings in the legend, between characters/blocks and text sections (default false)
+-heatmap-zoom-out=<uint>,... - print zoomed out heatmaps with given block sizes,
+                      must be multiples of block-size in ascending order.
+                      Suggested values: 4096 (default page size), 16384 (16k page),
+                      1048576 (1MB for XL workloads).
 ```

bolt/include/bolt/Profile/DataAggregator.h

@@ -370,10 +370,15 @@ class DataAggregator : public DataReader {
   /// memory.
   ///
   /// File format syntax:
-  /// {S|B|F|f|T} <start> [<end>] [<ft_end>] <count> [<mispred_count>]
+  /// E <event>
+  /// S <start> <count>
+  /// T <start> <end> <ft_end> <count>
+  /// B <start> <end> <count> <mispred_count>
+  /// [Ff] <start> <end> <count>
   ///
   /// where <start>, <end>, <ft_end> have the format [<id>:]<offset>
   ///
+  /// E - name of the sampling event used for subsequent entries
   /// S - indicates an aggregated basic sample at <start>
   /// B - indicates an aggregated branch from <start> to <end>
   /// F - an aggregated fall-through from <start> to <end>
@@ -390,23 +395,25 @@ class DataAggregator : public DataReader {
   /// <offset> - hex offset from the object base load address (0 for the
   /// main executable unless it's PIE) to the address.
   ///
-  /// <count> - total aggregated count of the branch or a fall-through.
+  /// <count> - total aggregated count.
   ///
   /// <mispred_count> - the number of times the branch was mispredicted.
-  /// Omitted for fall-throughs.
   ///
   /// Example:
   /// Basic samples profile:
+  /// E cycles
   /// S 41be50 3
+  /// E br_inst_retired.near_taken
+  /// S 41be60 6
   ///
-  /// Soft-deprecated branch profile with separate branches and fall-throughs:
+  /// Trace profile combining branches and fall-throughs:
+  /// T 4b196f 4b19e0 4b19ef 2
+  ///
+  /// Legacy branch profile with separate branches and fall-throughs:
   /// F 41be50 41be50 3
   /// F 41be90 41be90 4
   /// B 4b1942 39b57f0 3 0
   /// B 4b196f 4b19e0 2 0
-  ///
-  /// Recommended branch profile with pre-aggregated traces:
-  /// T 4b196f 4b19e0 4b19ef 2
   void parsePreAggregated();
 
   /// Parse the full output of pre-aggregated LBR samples generated by

bolt/lib/Profile/DataAggregator.cpp

@@ -1210,85 +1210,105 @@ ErrorOr<Location> DataAggregator::parseLocationOrOffset() {
 }
 
 std::error_code DataAggregator::parseAggregatedLBREntry() {
-  while (checkAndConsumeFS()) {
-  }
+  enum AggregatedLBREntry : char {
+    INVALID = 0,
+    EVENT_NAME,        // E
+    TRACE,             // T
+    SAMPLE,            // S
+    BRANCH,            // B
+    FT,                // F
+    FT_EXTERNAL_ORIGIN // f
+  } Type = INVALID;
+
+  // The number of fields to parse, set based on Type.
+  int AddrNum = 0;
+  int CounterNum = 0;
+  // Storage for parsed fields.
+  StringRef EventName;
+  std::optional<Location> Addr[3];
+  int64_t Counters[2];
 
-  ErrorOr<StringRef> TypeOrErr = parseString(FieldSeparator);
-  if (std::error_code EC = TypeOrErr.getError())
-    return EC;
-  enum AggregatedLBREntry {
-    TRACE,
-    SAMPLE,
-    BRANCH,
-    FT,
-    FT_EXTERNAL_ORIGIN,
-    INVALID
-  };
-  auto Type = StringSwitch<AggregatedLBREntry>(TypeOrErr.get())
-                  .Case("T", TRACE)
-                  .Case("S", SAMPLE)
-                  .Case("B", BRANCH)
-                  .Case("F", FT)
-                  .Case("f", FT_EXTERNAL_ORIGIN)
-                  .Default(INVALID);
-  if (Type == INVALID) {
-    reportError("expected T, S, B, F or f");
-    return make_error_code(llvm::errc::io_error);
-  }
+  while (Type == INVALID || Type == EVENT_NAME) {
+    while (checkAndConsumeFS()) {
+    }
+    ErrorOr<StringRef> StrOrErr =
+        parseString(FieldSeparator, Type == EVENT_NAME);
+    if (std::error_code EC = StrOrErr.getError())
+      return EC;
+    StringRef Str = StrOrErr.get();
 
-  std::optional<Location> Addrs[3];
-  int AddrNum = 2;
-  if (Type == TRACE)
-    AddrNum = 3;
-  else if (Type == SAMPLE)
-    AddrNum = 1;
+    if (Type == EVENT_NAME) {
+      EventName = Str;
+      break;
+    }
 
-  int64_t Counters[2];
-  int CounterNum = 1;
-  if (Type == BRANCH)
-    CounterNum = 2;
+    Type = StringSwitch<AggregatedLBREntry>(Str)
+               .Case("T", TRACE)
+               .Case("S", SAMPLE)
+               .Case("E", EVENT_NAME)
+               .Case("B", BRANCH)
+               .Case("F", FT)
+               .Case("f", FT_EXTERNAL_ORIGIN)
+               .Default(INVALID);
+
+    if (Type == INVALID) {
+      reportError("expected T, S, E, B, F or f");
+      return make_error_code(llvm::errc::io_error);
+    }
+
+    using SSI = StringSwitch<int>;
+    AddrNum = SSI(Str).Case("T", 3).Case("S", 1).Case("E", 0).Default(2);
+    CounterNum = SSI(Str).Case("B", 2).Case("E", 0).Default(1);
+  }
 
   for (int I = 0; I < AddrNum; ++I) {
     while (checkAndConsumeFS()) {
     }
-    if (ErrorOr<Location> Addr = parseLocationOrOffset())
-      Addrs[I] = Addr.get();
-    else
-      return Addr.getError();
+    ErrorOr<Location> AddrOrErr = parseLocationOrOffset();
+    if (std::error_code EC = AddrOrErr.getError())
+      return EC;
+    Addr[I] = AddrOrErr.get();
   }
 
   for (int I = 0; I < CounterNum; ++I) {
     while (checkAndConsumeFS()) {
     }
-    if (ErrorOr<int64_t> Count = parseNumberField(FieldSeparator, I + 1 == CounterNum))
-      Counters[I] = Count.get();
-    else
-      return Count.getError();
+    ErrorOr<int64_t> CountOrErr =
+        parseNumberField(FieldSeparator, I + 1 == CounterNum);
+    if (std::error_code EC = CountOrErr.getError())
+      return EC;
+    Counters[I] = CountOrErr.get();
   }
 
   if (!checkAndConsumeNewLine()) {
     reportError("expected end of line");
     return make_error_code(llvm::errc::io_error);
   }
 
-  const uint64_t FromOffset = Addrs[0]->Offset;
+  if (Type == EVENT_NAME) {
+    EventNames.insert(EventName);
+    return std::error_code();
+  }
+
+  const uint64_t FromOffset = Addr[0]->Offset;
   BinaryFunction *FromFunc = getBinaryFunctionContainingAddress(FromOffset);
   if (FromFunc)
     FromFunc->setHasProfileAvailable();
 
+  int64_t Count = Counters[0];
+  int64_t Mispreds = Counters[1];
+
   if (Type == SAMPLE) {
-    BasicSamples[FromOffset] += Counters[0];
+    BasicSamples[FromOffset] += Count;
+    NumTotalSamples += Count;
     return std::error_code();
   }
 
-  const uint64_t ToOffset = Addrs[1]->Offset;
+  const uint64_t ToOffset = Addr[1]->Offset;
   BinaryFunction *ToFunc = getBinaryFunctionContainingAddress(ToOffset);
   if (ToFunc)
     ToFunc->setHasProfileAvailable();
 
-  int64_t Count = Counters[0];
-  int64_t Mispreds = Counters[1];
-
   Trace Trace(FromOffset, ToOffset);
   // Taken trace
   if (Type == TRACE || Type == BRANCH) {
@@ -1300,7 +1320,7 @@ std::error_code DataAggregator::parseAggregatedLBREntry() {
   }
   // Construct fallthrough part of the trace
   if (Type == TRACE) {
-    const uint64_t TraceFtEndOffset = Addrs[2]->Offset;
+    const uint64_t TraceFtEndOffset = Addr[2]->Offset;
     Trace.From = ToOffset;
     Trace.To = TraceFtEndOffset;
     Type = FromFunc == ToFunc ? FT : FT_EXTERNAL_ORIGIN;

bolt/test/X86/Inputs/pre-aggregated-basic.txt

@@ -1,3 +1,4 @@
+E cycles
 S 4005f0 1
 S 4005f0 1
 S 400610 1

bolt/test/X86/pre-aggregated-perf.test

@@ -60,10 +60,12 @@ RUN: cat %t.bolt.yaml | FileCheck %s -check-prefix=NEWFORMAT
 ## Test pre-aggregated basic profile
 RUN: perf2bolt %t.exe -o %t --pa -p %p/Inputs/pre-aggregated-basic.txt -o %t.ba \
 RUN:   2>&1 | FileCheck %s --check-prefix=BASIC-ERROR
-RUN: perf2bolt %t.exe -o %t --pa -p %p/Inputs/pre-aggregated-basic.txt -o %t.ba \
+RUN: perf2bolt %t.exe -o %t --pa -p %p/Inputs/pre-aggregated-basic.txt -o %t.ba.nl \
 RUN:   -nl 2>&1 | FileCheck %s --check-prefix=BASIC-SUCCESS
+RUN: FileCheck %s --input-file %t.ba.nl --check-prefix CHECK-BASIC-NL
 BASIC-ERROR: BOLT-INFO: 0 out of 7 functions in the binary (0.0%) have non-empty execution profile
 BASIC-SUCCESS: BOLT-INFO: 4 out of 7 functions in the binary (57.1%) have non-empty execution profile
+CHECK-BASIC-NL: no_lbr cycles
 
 PERF2BOLT: 0 [unknown] 7f36d18d60c0 1 main 53c 0 2
 PERF2BOLT: 1 main 451 1 SolveCubic 0 0 2

bolt/test/link_fdata.py

@@ -36,7 +36,7 @@
 fdata_pat = re.compile(r"([01].*) (?P<exec>\d+) (?P<mispred>\d+)")
 
 # Pre-aggregated profile:
-# {T|S|B|F|f} <start> [<end>] [<ft_end>] <count> [<mispred_count>]
+# {T|S|E|B|F|f} <start> [<end>] [<ft_end>] <count> [<mispred_count>]
 # <loc>: [<id>:]<offset>
 preagg_pat = re.compile(r"(?P<type>[TSBFf]) (?P<offsets_count>.*)")
 

clang/docs/ReleaseNotes.rst

@@ -615,6 +615,8 @@ Bug Fixes in This Version
   argument which contains a pragma. (#GH113722)
 - Fixed assertion failures when generating name lookup table in modules. (#GH61065, #GH134739)
 - Fixed an assertion failure in constant compound literal statements. (#GH139160)
+- Fix crash due to unknown references and pointer implementation and handling of
+  base classes. (GH139452)
 
 Bug Fixes to Compiler Builtins
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

clang/include/clang/Basic/Attr.td

@@ -4790,6 +4790,7 @@ def HLSLResourceBinding: InheritableAttr {
       RegisterType RegType;
       std::optional<unsigned> SlotNumber;
       unsigned SpaceNumber;
+      std::optional<unsigned> ImplicitBindingOrderID;
 
   public:
       void setBinding(RegisterType RT, std::optional<unsigned> SlotNum, unsigned SpaceNum) {
@@ -4811,6 +4812,16 @@ def HLSLResourceBinding: InheritableAttr {
       unsigned getSpaceNumber() const {
         return SpaceNumber;
       }
+      void setImplicitBindingOrderID(uint32_t Value) {
+        ImplicitBindingOrderID = Value;
+      }
+      bool hasImplicitBindingOrderID() const { 
+        return ImplicitBindingOrderID.has_value();
+      }
+      uint32_t getImplicitBindingOrderID() const {
+        assert(hasImplicitBindingOrderID() && "attribute does not have implicit binding order id");
+        return ImplicitBindingOrderID.value();
+      }
   }];
 }
 

clang/lib/AST/ExprConstant.cpp

@@ -3314,7 +3314,11 @@ static bool HandleLValueBase(EvalInfo &Info, const Expr *E, LValue &Obj,
     return false;
 
   // Extract most-derived object and corresponding type.
-  DerivedDecl = D.MostDerivedType->getAsCXXRecordDecl();
+  // FIXME: After implementing P2280R4 it became possible to get references
+  // here. We do MostDerivedType->getAsCXXRecordDecl() in several other
+  // locations and if we see crashes in those locations in the future
+  // it may make more sense to move this fix into Lvalue::set.
+  DerivedDecl = D.MostDerivedType.getNonReferenceType()->getAsCXXRecordDecl();
   if (!CastToDerivedClass(Info, E, Obj, DerivedDecl, D.MostDerivedPathLength))
     return false;