[llvm-profgen][NFC] Detect pre-aggregated format

[aaupov]

Distinguish the following profgen input formats:

1. perf script with brstack only,
2. perf script with call stack and brstack (hybrid),
3. pre-aggregated input with count, call stack and brstack.

Pre-aggregated input lacks mmap information, so to enable using single input file for optimizing the main binary and shared libraries, the addresses will be augmented with buildid information in a follow-up (#190863).

[llvmbot]

@llvm/pr-subscribers-pgo

Author: Amir Ayupov (aaupov)

Changes

Distinguish the following profgen input formats:

1. perf script with brstack only,
2. perf script with call stack and brstack (hybrid),
3. pre-aggregated input with count, call stack and brstack.

Pre-aggregated input lacks mmap information, so to enable using single input file for optimizing the main binary and shared libraries, the addresses will be augmented with buildid information in a follow-up (#190863).

---

Full diff: https://github.com/llvm/llvm-project/pull/191593.diff

2 Files Affected:

• (modified) llvm/tools/llvm-profgen/PerfReader.cpp (+9-6)
• (modified) llvm/tools/llvm-profgen/PerfReader.h (+7-2)

diff --git a/llvm/tools/llvm-profgen/PerfReader.cpp b/llvm/tools/llvm-profgen/PerfReader.cpp
index 1dc59321fd91f..bbfde1256f2cc 100644
--- a/llvm/tools/llvm-profgen/PerfReader.cpp
+++ b/llvm/tools/llvm-profgen/PerfReader.cpp
@@ -365,9 +365,11 @@ PerfReaderBase::create(ProfiledBinary *Binary, PerfInputFile &PerfInput,
 
   PerfInput.Content =
       PerfScriptReader::checkPerfScriptType(PerfInput.InputFile);
-  if (PerfInput.Content == PerfContent::LBRStack) {
-    PerfReader.reset(
-        new HybridPerfReader(Binary, PerfInput.InputFile, PIDFilter));
+  if (PerfInput.Content == PerfContent::LBRStack ||
+      PerfInput.Content == PerfContent::AggLBRStack) {
+    auto *Reader = new HybridPerfReader(Binary, PerfInput.InputFile, PIDFilter);
+    Reader->setIsPreAggregated(PerfInput.Content == PerfContent::AggLBRStack);
+    PerfReader.reset(Reader);
   } else if (PerfInput.Content == PerfContent::LBR) {
     PerfReader.reset(new LBRPerfReader(Binary, PerfInput.InputFile, PIDFilter));
   } else {
@@ -1191,8 +1193,9 @@ PerfContent PerfScriptReader::checkPerfScriptType(StringRef FileName) {
   TraceStream TraceIt(FileName);
   uint64_t FrameAddr = 0;
   while (!TraceIt.isAtEoF()) {
-    // Skip the aggregated count
-    if (!TraceIt.getCurrentLine().getAsInteger(10, FrameAddr))
+    // Skip the aggregated count and detect pre-aggregated input.
+    bool HasAggCount = !TraceIt.getCurrentLine().getAsInteger(10, FrameAddr);
+    if (HasAggCount)
       TraceIt.advance();
 
     // Detect sample with call stack
@@ -1205,7 +1208,7 @@ PerfContent PerfScriptReader::checkPerfScriptType(StringRef FileName) {
     if (!TraceIt.isAtEoF()) {
       if (isLBRSample(TraceIt.getCurrentLine())) {
         if (Count > 0)
-          return PerfContent::LBRStack;
+          return HasAggCount ? PerfContent::AggLBRStack : PerfContent::LBRStack;
         else
           return PerfContent::LBR;
       }
diff --git a/llvm/tools/llvm-profgen/PerfReader.h b/llvm/tools/llvm-profgen/PerfReader.h
index 2a4c7594d3a93..83c4fb0447c5c 100644
--- a/llvm/tools/llvm-profgen/PerfReader.h
+++ b/llvm/tools/llvm-profgen/PerfReader.h
@@ -69,8 +69,9 @@ enum PerfFormat {
 // The type of perfscript content.
 enum PerfContent {
   UnknownContent = 0,
-  LBR = 1,      // Only LBR sample.
-  LBRStack = 2, // Hybrid sample including call stack and LBR stack.
+  LBR = 1,         // Only LBR sample.
+  LBRStack = 2,    // Hybrid sample including call stack and LBR stack.
+  AggLBRStack = 3, // Pre-aggregated hybrid sample.
 };
 
 struct PerfInputFile {
@@ -631,6 +632,8 @@ class PerfScriptReader : public PerfReaderBase {
   // receiving signals.
   static SmallVector<CleanupInstaller, 2> TempFileCleanups;
 
+  void setIsPreAggregated(bool V) { IsPreAggregated = V; }
+
 protected:
   // Check whether a given line is LBR sample
   static bool isLBRSample(StringRef Line);
@@ -676,6 +679,8 @@ class PerfScriptReader : public PerfReaderBase {
   std::set<uint64_t> InvalidReturnAddresses;
   // PID for the process of interest
   std::optional<int32_t> PIDFilter;
+  // Whether the input is pre-aggregated
+  bool IsPreAggregated = false;
 };
 
 /*

[HighW4y2H3ll]

What is the pre-aggregated input? Is this the same as putting 1 as the aggregated count for each sample? Could you explain why we need the new format? And does this new format only applicable to LBRStack samples?

[aaupov]

What is the pre-aggregated input? Is this the same as putting 1 as the aggregated count for each sample? Could you explain why we need the new format? And does this new format only applicable to LBRStack samples?

Pre-aggregated means there's a count line preceding call/branch stacks. It's worth distinguishing it as a separate format to allow buildid prefix only for it – because non-aggregated input comes directly from perf script and includes mmap events which serves as a main mechanism to distinguish addresses from the binary/DSOs.

[HighW4y2H3ll]

Why do we need to restrict buildID annotation only for LBRStack samples? It looks like the current implementation is compatible with the pre-aggregated input and if there's no aggregated count the default will be 1:

llvm-project/llvm/tools/llvm-profgen/PerfReader.cpp

Lines 1048 to 1054 in b9998b1

	uint64_t PerfScriptReader::parseAggregatedCount(TraceStream &TraceIt) {
	// The aggregated count is optional, so do not skip the line and return 1 if
	// it's unmatched
	uint64_t Count = 1;
	if (!TraceIt.getCurrentLine().getAsInteger(10, Count))
	TraceIt.advance();
	return Count;

[aaupov]

Why do we need to restrict buildID annotation only for LBRStack samples?

We can add buildid prefix support in two ways:

1. narrow: pre-aggregated only where it's needed,
2. uniform: pre-aggregated and perf script input where it's not needed.

I chose option 1 to keep perf script parsing strict.