[profcheck] Require `unknown` metadata have an origin parameter (#157594)

Rather than passes using `!prof = !{!”unknown”}`​for cases where don’t have enough information to emit profile values, this patch captures the pass (or some other information) that can help diagnostics - i.e. `!{!”unknown”, !”some-pass-name”}`​.

For example, suppose we emitted a `select`​ with the unknown metadata, and, later, end up needing to lower that to a conditional branch. If we observe (via sample profiling, for example) that the branch is biased and would have benefitted from a valid profile, the extra information can help speed up debugging.

We can also (in a subsequent pass) generate optimization remarks about such lowered selects, with a similar aim - identify patterns lowering to `select`​ that may be worth some extra investment in extracting a more precise profile.

1 files changed, 18 insertions, 7 deletions

diff --git a/llvm/lib/IR/Verifier.cpp b/llvm/lib/IR/Verifier.cpp
index 81a5372..7d362ce 100644
--- a/llvm/lib/IR/Verifier.cpp
+++ b/llvm/lib/IR/Verifier.cpp
@@ -631,7 +631,7 @@ private:
   void verifyFunctionAttrs(FunctionType *FT, AttributeList Attrs,
                            const Value *V, bool IsIntrinsic, bool IsInlineAsm);
   void verifyFunctionMetadata(ArrayRef<std::pair<unsigned, MDNode *>> MDs);
-
+  void verifyUnknownProfileMetadata(MDNode *MD);
   void visitConstantExprsRecursively(const Constant *EntryC);
   void visitConstantExpr(const ConstantExpr *CE);
   void visitConstantPtrAuth(const ConstantPtrAuth *CPA);
@@ -2515,19 +2515,31 @@ void Verifier::verifyFunctionAttrs(FunctionType *FT, AttributeList Attrs,
                   V);
   }
 }
+void Verifier::verifyUnknownProfileMetadata(MDNode *MD) {
+  Check(MD->getNumOperands() == 2,
+        "'unknown' !prof should have a single additional operand", MD);
+  auto *PassName = dyn_cast<MDString>(MD->getOperand(1));
+  Check(PassName != nullptr,
+        "'unknown' !prof should have an additional operand of type "
+        "string");
+  Check(!PassName->getString().empty(),
+        "the 'unknown' !prof operand should not be an empty string");
+}
 
 void Verifier::verifyFunctionMetadata(
     ArrayRef<std::pair<unsigned, MDNode *>> MDs) {
   for (const auto &Pair : MDs) {
     if (Pair.first == LLVMContext::MD_prof) {
       MDNode *MD = Pair.second;
+      Check(MD->getNumOperands() >= 2,
+            "!prof annotations should have no less than 2 operands", MD);
       // We may have functions that are synthesized by the compiler, e.g. in
       // WPD, that we can't currently determine the entry count.
-      if (isExplicitlyUnknownProfileMetadata(*MD))
+      if (MD->getOperand(0).equalsStr(
+              MDProfLabels::UnknownBranchWeightsMarker)) {
+        verifyUnknownProfileMetadata(MD);
         continue;
-
-      Check(MD->getNumOperands() >= 2,
-            "!prof annotations should have no less than 2 operands", MD);
+      }
 
       // Check first operand.
       Check(MD->getOperand(0) != nullptr, "first operand should not be null",
@@ -5054,8 +5066,7 @@ void Verifier::visitProfMetadata(Instruction &I, MDNode *MD) {
           "'unknown' !prof should only appear on instructions on which "
           "'branch_weights' would",
           MD);
-    Check(MD->getNumOperands() == 1,
-          "'unknown' !prof should have no additional operands", MD);
+    verifyUnknownProfileMetadata(MD);
     return;
   }