[llvm-profgen] Truncate the context with zero probe ID

[wlei]

Summary

Due to the debug info merging, there may have some contexts with zero probe id, we should truncate the context to avoid misleading pre-inliner.

[wlei]

Updating D114284: [llvm-profgen] Truncate the context with zero probe ID

[wlei]

Updating D114284: [llvm-profgen] Truncate the context with zero probe ID

[hoy]

lgtm, thanks.

llvm/tools/llvm-profgen/ProfiledBinary.h

459 ↗

(On Diff #388640)

nit: // clear the current context for an unknown probe

[wlei]

Updating D114284: [llvm-profgen] Truncate the context with zero probe ID

[wenlei]

llvm/tools/llvm-profgen/ProfileGenerator.cpp

775

Before this fix, when CalleeContextId only has one frame, what did we get for CallerProfile? A dummy root profile with empty name?

[wlei]

fix lint

llvm/tools/llvm-profgen/ProfileGenerator.cpp

775

Before this fix, CalleeContextId is guaranteed to have >1 frame, see there is a check if (InlinerDesc != nullptr) to make sure the callee always has the inliner which means we only infer caller's count for inlining.

llvm/tools/llvm-profgen/ProfileGenerator.cpp

for (auto PI : ProbeCounter) {
     // zero count probes of the frame later.
     FrameSamples[Probe->getInlineTreeNode()].insert(&FunctionProfile);
     FunctionProfile.addBodySamplesForProbe(Probe->getIndex(), Count);
     FunctionProfile.addTotalSamples(Count);
     if (Probe->isEntry()) {
       FunctionProfile.addHeadSamples(Count);
       // Look up for the caller's function profile
       const auto *InlinerDesc = Binary->getInlinerDescForProbe(Probe);
-      if (InlinerDesc != nullptr) {
+      SampleContextFrames CalleeContextId =
+          FunctionProfile.getContext().getContextFrames();
+      if (InlinerDesc != nullptr && CalleeContextId.size() > 1) {
         // Since the context id will be compressed, we have to use callee's
         // context id to infer caller's context id to ensure they share the
         // same context prefix.
-        SampleContextFrames CalleeContextId =
-            FunctionProfile.getContext().getContextFrames();
         SampleContextFrameVector CallerContextId;
         SampleContextFrame &&CallerLeafFrameLoc =
             getCallerContext(CalleeContextId, CallerContextId);
         uint64_t CallerIndex = CallerLeafFrameLoc.Location.LineOffset;
         assert(CallerIndex &&
                "Inferred caller's location index shouldn't be zero!");
         FunctionSamples &CallerProfile =
             getFunctionProfileForContext(CallerContextId);
         CallerProfile.setFunctionHash(InlinerDesc->FuncHash);
         CallerProfile.addBodySamples(CallerIndex, 0, Count);
         CallerProfile.addTotalSamples(Count);
         CallerProfile.addCalledTargetSamples(
             CallerIndex, 0, FunctionProfile.getContext().getName(), Count);
       }
     }

 
 FunctionSamples &CSProfileGenerator::getFunctionProfileForLeafProbe(
     SampleContextFrames ContextStack, const MCDecodedPseudoProbe *LeafProbe) {
 
   // Explicitly copy the context for appending the leaf context
   SampleContextFrameVector NewContextStack(ContextStack.begin(),
                                            ContextStack.end());
   Binary->getInlineContextForProbe(LeafProbe, NewContextStack, true);
+
+  int ZeroProbeIdx = -1;
+  for (uint32_t I = 0; I < NewContextStack.size() - 1; I++) {
+    auto &Callsite = NewContextStack[I];
+    // Ignore the context if the probe of the callsite is zero.
+    if (Callsite.Location.LineOffset == 0)
+      ZeroProbeIdx = I;
+  }
+  if (ZeroProbeIdx != -1) {
+    std::copy(NewContextStack.begin() + ZeroProbeIdx + 1, NewContextStack.end(),
+              NewContextStack.begin());
+    NewContextStack.resize(NewContextStack.size() - ZeroProbeIdx - 1);
+  }
+
   // For leaf inlined context with the top frame, we should strip off the top
   // frame's probe id, like:
   // Inlined stack: [foo:1, bar:2], the ContextId will be "foo:1 @ bar"
   auto LeafFrame = NewContextStack.back();
   LeafFrame.Location = LineLocation(0, 0);
   NewContextStack.pop_back();
   // Compress the context string except for the leaf frame
   CSProfileGenerator::compressRecursionContext(NewContextStack);