From 821d7471f91858ce91de86df940cedfab4166543 Mon Sep 17 00:00:00 2001
From: Justin Bogner <mail@justinbogner.com>
Date: Fri, 1 Aug 2014 22:50:07 +0000
Subject: InstrProf: Allow multiple functions with the same name

This updates the instrumentation based profiling format so that when
we have multiple functions with the same name (but different function
hashes) we keep all of them instead of rejecting the later ones.

There are a number of scenarios where this can come up where it's more
useful to keep multiple function profiles:

* Name collisions in unrelated libraries that are profiled together.
* Multiple "main" functions from multiple tools built against a common
  library.
* Combining profiles from different build configurations (ie, asserts
  and no-asserts)

The profile format now stores the number of counters between the hash
and the counts themselves, so that multiple sets of counts can be
stored. Since this is backwards incompatible, I've bumped the format
version and added some trivial logic to skip this when reading the old
format.

llvm-svn: 214585
---
 llvm/lib/ProfileData/InstrProfReader.cpp | 63 ++++++++++++++++++++++++--------
 1 file changed, 48 insertions(+), 15 deletions(-)

(limited to 'llvm/lib/ProfileData/InstrProfReader.cpp')

diff --git a/llvm/lib/ProfileData/InstrProfReader.cpp b/llvm/lib/ProfileData/InstrProfReader.cpp
index e8f64614df5..5c1993766aa 100644
--- a/llvm/lib/ProfileData/InstrProfReader.cpp
+++ b/llvm/lib/ProfileData/InstrProfReader.cpp
@@ -307,8 +307,8 @@ std::error_code IndexedInstrProfReader::readHeader() {
     return error(instrprof_error::bad_magic);
 
   // Read the version.
-  uint64_t Version = endian::readNext<uint64_t, little, unaligned>(Cur);
-  if (Version != IndexedInstrProf::Version)
+  FormatVersion = endian::readNext<uint64_t, little, unaligned>(Cur);
+  if (FormatVersion > IndexedInstrProf::Version)
     return error(instrprof_error::unsupported_version);
 
   // Read the maximal function count.
@@ -331,18 +331,31 @@ std::error_code IndexedInstrProfReader::readHeader() {
 }
 
 std::error_code IndexedInstrProfReader::getFunctionCounts(
-    StringRef FuncName, uint64_t &FuncHash, std::vector<uint64_t> &Counts) {
-  const auto &Iter = Index->find(FuncName);
+    StringRef FuncName, uint64_t FuncHash, std::vector<uint64_t> &Counts) {
+  auto Iter = Index->find(FuncName);
   if (Iter == Index->end())
     return error(instrprof_error::unknown_function);
 
-  // Found it. Make sure it's valid before giving back a result.
-  const InstrProfRecord &Record = *Iter;
-  if (Record.Name.empty())
-    return error(instrprof_error::malformed);
-  FuncHash = Record.Hash;
-  Counts = Record.Counts;
-  return success();
+  // Found it. Look for counters with the right hash.
+  ArrayRef<uint64_t> Data = (*Iter).Data;
+  uint64_t NumCounts;
+  for (uint64_t I = 0, E = Data.size(); I != E; I += NumCounts) {
+    // The function hash comes first.
+    uint64_t FoundHash = Data[I++];
+    // In v1, we have at least one count. Later, we have the number of counts.
+    if (I == E)
+      return error(instrprof_error::malformed);
+    NumCounts = FormatVersion == 1 ? E - I : Data[I++];
+    // If we have more counts than data, this is bogus.
+    if (I + NumCounts > E)
+      return error(instrprof_error::malformed);
+    // Check for a match and fill the vector if there is one.
+    if (FoundHash == FuncHash) {
+      Counts = Data.slice(I, NumCounts);
+      return success();
+    }
+  }
+  return error(instrprof_error::hash_mismatch);
 }
 
 std::error_code
@@ -351,10 +364,30 @@ IndexedInstrProfReader::readNextRecord(InstrProfRecord &Record) {
   if (RecordIterator == Index->data_end())
     return error(instrprof_error::eof);
 
-  // Read the next one.
-  Record = *RecordIterator;
-  ++RecordIterator;
-  if (Record.Name.empty())
+  // Record the current function name.
+  Record.Name = (*RecordIterator).Name;
+
+  ArrayRef<uint64_t> Data = (*RecordIterator).Data;
+  // Valid data starts with a hash and either a count or the number of counts.
+  if (CurrentOffset + 1 > Data.size())
+    return error(instrprof_error::malformed);
+  // First we have a function hash.
+  Record.Hash = Data[CurrentOffset++];
+  // In version 1 we knew the number of counters implicitly, but in newer
+  // versions we store the number of counters next.
+  uint64_t NumCounts =
+      FormatVersion == 1 ? Data.size() - CurrentOffset : Data[CurrentOffset++];
+  if (CurrentOffset + NumCounts > Data.size())
     return error(instrprof_error::malformed);
+  // And finally the counts themselves.
+  Record.Counts = Data.slice(CurrentOffset, NumCounts);
+
+  // If we've exhausted this function's data, increment the record.
+  CurrentOffset += NumCounts;
+  if (CurrentOffset == Data.size()) {
+    ++RecordIterator;
+    CurrentOffset = 0;
+  }
+
   return success();
 }
-- 
cgit v1.2.3