Introduce CounterExpressionBuilder::replace(C, Map)

[chapuni]

This return a counter for each term in the expression replaced by ReplaceMap.

At the moment, this doesn't update the Map, so Map is marked as const.

[llvmbot]

@llvm/pr-subscribers-pgo

Author: NAKAMURA Takumi (chapuni)

Changes

This return a counter for each term in the expression replaced by ReplaceMap.

At the moment, this doesn't update the Map, so Map is marked as const.

---

Full diff: https://github.com/llvm/llvm-project/pull/112698.diff

2 Files Affected:

• (modified) llvm/include/llvm/ProfileData/Coverage/CoverageMapping.h (+6)
• (modified) llvm/lib/ProfileData/Coverage/CoverageMapping.cpp (+32)

diff --git a/llvm/include/llvm/ProfileData/Coverage/CoverageMapping.h b/llvm/include/llvm/ProfileData/Coverage/CoverageMapping.h
index fa07b3a9e8b14f..d6528bd407ef34 100644
--- a/llvm/include/llvm/ProfileData/Coverage/CoverageMapping.h
+++ b/llvm/include/llvm/ProfileData/Coverage/CoverageMapping.h
@@ -34,6 +34,7 @@
 #include <cassert>
 #include <cstdint>
 #include <iterator>
+#include <map>
 #include <memory>
 #include <sstream>
 #include <string>
@@ -213,6 +214,11 @@ class CounterExpressionBuilder {
   /// Return a counter that represents the expression that subtracts RHS from
   /// LHS.
   Counter subtract(Counter LHS, Counter RHS, bool Simplify = true);
+
+  using ReplaceMap = std::map<Counter, Counter>;
+
+  /// Return a counter for each term in the expression replaced by ReplaceMap.
+  Counter replace(Counter C, const ReplaceMap &Map);
 };
 
 using LineColPair = std::pair<unsigned, unsigned>;
diff --git a/llvm/lib/ProfileData/Coverage/CoverageMapping.cpp b/llvm/lib/ProfileData/Coverage/CoverageMapping.cpp
index c713371da81e40..b50f025d261e13 100644
--- a/llvm/lib/ProfileData/Coverage/CoverageMapping.cpp
+++ b/llvm/lib/ProfileData/Coverage/CoverageMapping.cpp
@@ -135,6 +135,38 @@ Counter CounterExpressionBuilder::subtract(Counter LHS, Counter RHS,
   return Simplify ? simplify(Cnt) : Cnt;
 }
 
+Counter CounterExpressionBuilder::replace(Counter C, const ReplaceMap &Map) {
+  auto I = Map.find(C);
+
+  // Replace C with the Map even if C is Expression.
+  if (I != Map.end())
+    return I->second;
+
+  // Traverse only Expression.
+  if (!C.isExpression())
+    return C;
+
+  auto CE = Expressions[C.getExpressionID()];
+  auto NewLHS = replace(CE.LHS, Map);
+  auto NewRHS = replace(CE.RHS, Map);
+
+  // Reconstruct Expression with induced subexpressions.
+  switch (CE.Kind) {
+  case CounterExpression::Add:
+    C = add(NewLHS, NewRHS);
+    break;
+  case CounterExpression::Subtract:
+    C = subtract(NewLHS, NewRHS);
+    break;
+  }
+
+  // Reconfirm if the reconstructed expression would hit the Map.
+  if ((I = Map.find(C)) != Map.end())
+    return I->second;
+
+  return C;
+}
+
 void CounterMappingContext::dump(const Counter &C, raw_ostream &OS) const {
   switch (C.getKind()) {
   case Counter::Zero:

[ornata]

why std::map instead of e.g. DenseMap?

https://llvm.org/docs/ProgrammersManual.html#map

std::map has similar characteristics to std::set: it uses a single allocation per pair inserted into the map, it offers log(n) lookup with an extremely large constant factor, imposes a space penalty of 3 pointers per pair in the map, etc.

std::map is most useful when your keys or values are very large, if you need to iterate over the collection in sorted order, or if you need stable iterators into the map (i.e. they don’t get invalidated if an insertion or deletion of another element takes place).

[chapuni]

I am so lazy to add hash helper into llvm::Counter. I have taken std::map since Counter has provided enough operators.

The map may grow up to hundreds of keys. Could we make Counter ready for hash maps?

[ornata]

Replace C with the Map...

"replace C with the value found in Map..."?

[ornata]

Why?

[chapuni]

Because others (Zero or CounterRef) are not traversable :)

Is this line too naive to you? I could eliminate.

[ornata]

Can this have a more descriptive name? This is replacing each term with an existing term, right?

[chapuni]

I thought its name may be simple here, since it is declared in the class.

"This replaces each term (or possibly recognizable expression) with an existing expression."

[ornata]

How deep can this recurse? Should there be a depth limit?

[chapuni]

I suppose at most the number of terms, unless the expression is made buggy with loops. (Then, other CounterExpression stuff would hang or crash, I guess) Would we really check if the expr would not be buggy?

If we should introduce such a check, it may be not depth limit but the set of Seen.

[chapuni]

Thanks for comments!

[chapuni]

I am so lazy to add hash helper into llvm::Counter. I have taken std::map since Counter has provided enough operators.

The map may grow up to hundreds of keys. Could we make Counter ready for hash maps?

[chapuni]

I thought its name may be simple here, since it is declared in the class.

"This replaces each term (or possibly recognizable expression) with an existing expression."

[chapuni]

Because others (Zero or CounterRef) are not traversable :)

Is this line too naive to you? I could eliminate.

[chapuni]

I suppose at most the number of terms, unless the expression is made buggy with loops. (Then, other CounterExpression stuff would hang or crash, I guess) Would we really check if the expr would not be buggy?

If we should introduce such a check, it may be not depth limit but the set of Seen.