[llvm] [LLVM-Reduce] - Distinct Metadata Reduction (PR #104624)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,147 @@
+//===- ReduceDistinctMetadata.cpp - Specialized Delta Pass ------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===------------------------------------------------------------------------------===//
+//
+// This file implements two functions used by the Generic Delta Debugging
+// Algorithm, which are used to reduce unnamed distinct metadata nodes.
+//
+//===------------------------------------------------------------------------------===//
+
+#include "ReduceDistinctMetadata.h"
+#include "Delta.h"
+#include "llvm/ADT/Sequence.h"
+#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/IR/InstIterator.h"
+#include <algorithm>
+#include <queue>
+
+using namespace llvm;
+
+// Traverse the graph breadth-first and try to remove unnamed metadata nodes
+void reduceNodes(MDNode *Root,
+                 SetVector<std::pair<unsigned int, MDNode *>> &NodesToDelete,
+                 MDNode *TemporaryNode, Oracle &O, Module &Program) {
+  std::queue<MDNode *> NodesToTraverse{};
+  // Keep track of visited nodes not to get into loops
+  SetVector<MDNode *> VisitedNodes{};
+  NodesToTraverse.push(Root);
+
+  while (!NodesToTraverse.empty()) {
+    MDNode *CurrentNode = NodesToTraverse.front();
+    NodesToTraverse.pop();
+
+    // Mark the nodes for removal
+    for (unsigned int I = 0; I < CurrentNode->getNumOperands(); ++I) {
+      if (MDNode *Operand =
+              dyn_cast<MDNode>(CurrentNode->getOperand(I).get())) {
+        // Check whether node has been visited
+        if (!VisitedNodes.contains(Operand)) {
+          NodesToTraverse.push(Operand);
+          VisitedNodes.insert(Operand);
+        }
+        // Delete the node only if it is distinct
+        if (Operand->isDistinct())
+          // Add to removal list
+          NodesToDelete.insert(std::make_pair(I, CurrentNode));
+      }
+    }
+
+    // Remove the nodes
+    for (auto [PositionToReplace, Node] : NodesToDelete) {
+      if (!O.shouldKeep())
+        Node->replaceOperandWith(PositionToReplace, TemporaryNode);
+    }
+    NodesToDelete.clear();
+  }
+}
+
+// After reducing metadata, we need to remove references to the temporary node,
+// this is also done with BFS
+void cleanUpTemporaries(NamedMDNode &NamedNode, MDTuple *TemporaryTuple,
+                        Module &Program) {
+  std::queue<MDTuple *> NodesToTraverse{};
+  SetVector<MDTuple *> VisitedNodes{};
+
+  // Push all first level operands of the named node to the queue
+  for (auto I = NamedNode.op_begin(); I != NamedNode.op_end(); ++I) {
+    // If the node hasn't been traversed yet, add it to the queue of nodes to
+    // traverse.
+    if (MDTuple *TupleI = dyn_cast<MDTuple>((*I))) {
+      if (!VisitedNodes.contains(TupleI)) {
+        NodesToTraverse.push(TupleI);
+        VisitedNodes.insert(TupleI);
+      }
+    }
+  }
+
+  while (!NodesToTraverse.empty()) {
+    MDTuple *CurrentTuple = NodesToTraverse.front();
+    NodesToTraverse.pop();
+
+    // Shift all of the interesting elements to the left, pop remaining
+    // afterwards
+    if (CurrentTuple
+            ->isDistinct()) { // Do resizing and cleaning operations only if
+                              // the node is distinct, as resizing is not
+                              // supported for unique nodes and is redundant.
----------------
arsenm wrote:

Move the comment inside the condition? The wrapping is weird 

https://github.com/llvm/llvm-project/pull/104624