release/20.x: Reduce memory usage in AST parent map generation by lazily checking if nodes have been seen (#129934)

[llvmbot]

Backport 8c7f0ea

Requested by: @higher-performance

[llvmbot]

@higher-performance What do you think about merging this PR to the release branch?

[llvmbot]

@llvm/pr-subscribers-clang

Author: None (llvmbot)

Changes

Backport 8c7f0ea

Requested by: @higher-performance

---

Full diff: https://github.com/llvm/llvm-project/pull/131209.diff

1 Files Affected:

• (modified) clang/lib/AST/ParentMapContext.cpp (+11-6)

diff --git a/clang/lib/AST/ParentMapContext.cpp b/clang/lib/AST/ParentMapContext.cpp
index 7ff492443031d..d8dd352c42d6b 100644
--- a/clang/lib/AST/ParentMapContext.cpp
+++ b/clang/lib/AST/ParentMapContext.cpp
@@ -12,10 +12,11 @@
 //===----------------------------------------------------------------------===//
 
 #include "clang/AST/ParentMapContext.h"
-#include "clang/AST/RecursiveASTVisitor.h"
 #include "clang/AST/Decl.h"
 #include "clang/AST/Expr.h"
+#include "clang/AST/RecursiveASTVisitor.h"
 #include "clang/AST/TemplateBase.h"
+#include "llvm/ADT/SmallPtrSet.h"
 
 using namespace clang;
 
@@ -69,17 +70,21 @@ class ParentMapContext::ParentMap {
       for (; N > 0; --N)
         push_back(Value);
     }
-    bool contains(const DynTypedNode &Value) {
-      return Seen.contains(Value);
+    bool contains(const DynTypedNode &Value) const {
+      const void *Identity = Value.getMemoizationData();
+      assert(Identity);
+      return Dedup.contains(Identity);
     }
     void push_back(const DynTypedNode &Value) {
-      if (!Value.getMemoizationData() || Seen.insert(Value).second)
+      const void *Identity = Value.getMemoizationData();
+      if (!Identity || Dedup.insert(Identity).second) {
         Items.push_back(Value);
+      }
     }
     llvm::ArrayRef<DynTypedNode> view() const { return Items; }
   private:
-    llvm::SmallVector<DynTypedNode, 2> Items;
-    llvm::SmallDenseSet<DynTypedNode, 2> Seen;
+    llvm::SmallVector<DynTypedNode, 1> Items;
+    llvm::SmallPtrSet<const void *, 2> Dedup;
   };
 
   /// Maps from a node to its parents. This is used for nodes that have

[higher-performance]

LGTM. @erichkeane?

[erichkeane]

FWIW: I'm in favor of this. This is quite low risk/a pretty trivial change that has an incredibly big impact on the memory pressure of our compiler in certain situations. I think this is a good candidate to back port.

[github-actions]

@higher-performance (or anyone else). If you would like to add a note about this fix in the release notes (completely optional). Please reply to this comment with a one or two sentence description of the fix. When you are done, please add the release:note label to this PR.