[BOLT] Support pre-aggregated basic sample profile (#140196)

Define a pre-aggregated basic sample format:
```
E <event name>
S <location> <count>
```

`-nl` flag is required to use parsed basic samples.

Test Plan: update pre-aggregated-perf.test

Author: aaupov

bolt/include/bolt/Profile/DataAggregator.h

@@ -370,33 +370,46 @@ class DataAggregator : public DataReader {
   /// memory.
   ///
   /// File format syntax:
-  /// {B|F|f|T} [<start_id>:]<start_offset> [<end_id>:]<end_offset> [<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>
   ///
-  /// B - indicates an aggregated branch
-  /// F - an aggregated fall-through
+  /// 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>
   /// f - an aggregated fall-through with external origin - used to disambiguate
   ///       between a return hitting a basic block head and a regular internal
   ///       jump to the block
-  /// T - an aggregated trace: branch with a fall-through (from, to, ft_end)
-  ///
-  /// <start_id> - build id of the object containing the start address. We can
-  /// skip it for the main binary and use "X" for an unknown object. This will
-  /// save some space and facilitate human parsing.
-  ///
-  /// <start_offset> - hex offset from the object base load address (0 for the
-  /// main executable unless it's PIE) to the start address.
+  /// T - an aggregated trace: branch from <start> to <end> with a fall-through
+  ///       to <ft_end>
   ///
-  /// <end_id>, <end_offset> - same for the end address.
+  /// <id> - build id of the object containing the address. We can skip it for
+  /// the main binary and use "X" for an unknown object. This will save some
+  /// space and facilitate human parsing.
   ///
-  /// <ft_end> - same for the fallthrough_end address.
+  /// <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
+  ///
+  /// 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

bolt/lib/Profile/DataAggregator.cpp

@@ -1204,77 +1204,106 @@ 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];
+
+  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();
 
-  ErrorOr<StringRef> TypeOrErr = parseString(FieldSeparator);
-  if (std::error_code EC = TypeOrErr.getError())
-    return EC;
-  enum AggregatedLBREntry { TRACE, BRANCH, FT, FT_EXTERNAL_ORIGIN, INVALID };
-  auto Type = StringSwitch<AggregatedLBREntry>(TypeOrErr.get())
-                  .Case("T", TRACE)
-                  .Case("B", BRANCH)
-                  .Case("F", FT)
-                  .Case("f", FT_EXTERNAL_ORIGIN)
-                  .Default(INVALID);
-  if (Type == INVALID) {
-    reportError("expected T, B, F or f");
-    return make_error_code(llvm::errc::io_error);
-  }
+    if (Type == EVENT_NAME) {
+      EventName = Str;
+      break;
+    }
 
-  while (checkAndConsumeFS()) {
-  }
-  ErrorOr<Location> From = parseLocationOrOffset();
-  if (std::error_code EC = From.getError())
-    return EC;
+    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);
+    }
 
-  while (checkAndConsumeFS()) {
+    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);
   }
-  ErrorOr<Location> To = parseLocationOrOffset();
-  if (std::error_code EC = To.getError())
-    return EC;
 
-  ErrorOr<Location> TraceFtEnd = std::error_code();
-  if (Type == AggregatedLBREntry::TRACE) {
+  for (int I = 0; I < AddrNum; ++I) {
     while (checkAndConsumeFS()) {
     }
-    TraceFtEnd = parseLocationOrOffset();
-    if (std::error_code EC = TraceFtEnd.getError())
+    ErrorOr<Location> AddrOrErr = parseLocationOrOffset();
+    if (std::error_code EC = AddrOrErr.getError())
       return EC;
+    Addr[I] = AddrOrErr.get();
   }
 
-  while (checkAndConsumeFS()) {
-  }
-  ErrorOr<int64_t> Frequency =
-      parseNumberField(FieldSeparator, Type != AggregatedLBREntry::BRANCH);
-  if (std::error_code EC = Frequency.getError())
-    return EC;
-
-  uint64_t Mispreds = 0;
-  if (Type == AggregatedLBREntry::BRANCH) {
+  for (int I = 0; I < CounterNum; ++I) {
     while (checkAndConsumeFS()) {
     }
-    ErrorOr<int64_t> MispredsOrErr = parseNumberField(FieldSeparator, true);
-    if (std::error_code EC = MispredsOrErr.getError())
+    ErrorOr<int64_t> CountOrErr =
+        parseNumberField(FieldSeparator, I + 1 == CounterNum);
+    if (std::error_code EC = CountOrErr.getError())
       return EC;
-    Mispreds = static_cast<uint64_t>(MispredsOrErr.get());
+    Counters[I] = CountOrErr.get();
   }
 
   if (!checkAndConsumeNewLine()) {
     reportError("expected end of line");
     return make_error_code(llvm::errc::io_error);
   }
 
-  BinaryFunction *FromFunc = getBinaryFunctionContainingAddress(From->Offset);
-  BinaryFunction *ToFunc = getBinaryFunctionContainingAddress(To->Offset);
+  if (Type == EVENT_NAME) {
+    EventNames.insert(EventName);
+    return std::error_code();
+  }
 
-  for (BinaryFunction *BF : {FromFunc, ToFunc})
-    if (BF)
-      BF->setHasProfileAvailable();
+  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] += Count;
+    NumTotalSamples += Count;
+    return std::error_code();
+  }
 
-  uint64_t Count = static_cast<uint64_t>(Frequency.get());
+  const uint64_t ToOffset = Addr[1]->Offset;
+  BinaryFunction *ToFunc = getBinaryFunctionContainingAddress(ToOffset);
+  if (ToFunc)
+    ToFunc->setHasProfileAvailable();
 
-  Trace Trace(From->Offset, To->Offset);
+  Trace Trace(FromOffset, ToOffset);
   // Taken trace
   if (Type == TRACE || Type == BRANCH) {
     TakenBranchInfo &Info = BranchLBRs[Trace];
@@ -1285,8 +1314,9 @@ std::error_code DataAggregator::parseAggregatedLBREntry() {
   }
   // Construct fallthrough part of the trace
   if (Type == TRACE) {
-    Trace.From = To->Offset;
-    Trace.To = TraceFtEnd->Offset;
+    const uint64_t TraceFtEndOffset = Addr[2]->Offset;
+    Trace.From = ToOffset;
+    Trace.To = TraceFtEndOffset;
     Type = FromFunc == ToFunc ? FT : FT_EXTERNAL_ORIGIN;
   }
   // Add fallthrough trace

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

@@ -0,0 +1,19 @@
+E cycles
+S 4005f0 1
+S 4005f0 1
+S 400610 1
+S 400ad1 2
+S 400b10 1
+S 400bb7 1
+S 400bbc 2
+S 400d90 1
+S 400dae 1
+S 400e00 2
+S 401170 22
+S 401180 58
+S 4011a0 33
+S 4011a9 33
+S 4011ad 58
+S 4011b2 22
+S X:7f36d18d60c0 2
+S X:7f36d18f2ce0 1

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

@@ -57,6 +57,16 @@ RUN: llvm-bolt %t.exe -o %t.bolt.yaml --pa -p %p/Inputs/pre-aggregated.txt \
 RUN:   --aggregate-only --profile-format=yaml --profile-use-dfs
 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.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
 PERF2BOLT: 1 main 490 0 [unknown] 4005f0 0 1

bolt/test/link_fdata.py

@@ -36,9 +36,9 @@
 fdata_pat = re.compile(r"([01].*) (?P<mispred>\d+) (?P<exec>\d+)")
 
 # Pre-aggregated profile:
-# {T|B|F|f} [<start_id>:]<start_offset> [<end_id>:]<end_offset> [<ft_end>]
-# <count> [<mispred_count>]
-preagg_pat = re.compile(r"(?P<type>[TBFf]) (?P<offsets_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>.*)")
 
 # No-LBR profile:
 # <is symbol?> <closest elf symbol or DSO name> <relative address> <count>