[TableGen] Use std::move instead of swap. NFC.

[jayfoad]

Historically TableGen has used A.swap(B) to move containers without
the expense of copying them. Perhaps this predated rvalue references. In
any case A = std::move(B) seems like a more direct way to implement
this when only A is required after the operation.

[llvmbot]

@llvm/pr-subscribers-llvm-globalisel

Author: Jay Foad (jayfoad)

Changes

Historically TableGen has used A.swap(B) to move containers without
the expense of copying them. Perhaps this predated rvalue references. In
any case A = std::move(B) seems like a more direct way to implement
this when only A is required after the operation.

---

Full diff: https://github.com/llvm/llvm-project/pull/81606.diff

5 Files Affected:

• (modified) llvm/utils/TableGen/AsmMatcherEmitter.cpp (+4-5)
• (modified) llvm/utils/TableGen/CodeGenRegisters.cpp (+3-3)
• (modified) llvm/utils/TableGen/CodeGenSchedule.cpp (+1-1)
• (modified) llvm/utils/TableGen/GlobalISelMatchTable.cpp (+2-2)
• (modified) llvm/utils/TableGen/SubtargetEmitter.cpp (+1-1)

diff --git a/llvm/utils/TableGen/AsmMatcherEmitter.cpp b/llvm/utils/TableGen/AsmMatcherEmitter.cpp
index 9065885618069b..209df25c1fccb4 100644
--- a/llvm/utils/TableGen/AsmMatcherEmitter.cpp
+++ b/llvm/utils/TableGen/AsmMatcherEmitter.cpp
@@ -1270,11 +1270,10 @@ void AsmMatcherInfo::buildRegisterClasses(
       }
 
       RegisterSet Tmp;
-      std::swap(Tmp, ContainingSet);
-      std::insert_iterator<RegisterSet> II(ContainingSet,
-                                           ContainingSet.begin());
-      std::set_intersection(Tmp.begin(), Tmp.end(), RS.begin(), RS.end(), II,
-                            LessRecordByID());
+      std::insert_iterator<RegisterSet> II(Tmp, Tmp.begin());
+      std::set_intersection(ContainingSet.begin(), ContainingSet.end(),
+                            RS.begin(), RS.end(), II, LessRecordByID());
+      ContainingSet = std::move(Tmp);
     }
 
     if (!ContainingSet.empty()) {
diff --git a/llvm/utils/TableGen/CodeGenRegisters.cpp b/llvm/utils/TableGen/CodeGenRegisters.cpp
index 25f38648366300..0f0044957c9c6a 100644
--- a/llvm/utils/TableGen/CodeGenRegisters.cpp
+++ b/llvm/utils/TableGen/CodeGenRegisters.cpp
@@ -1964,9 +1964,9 @@ void CodeGenRegBank::pruneUnitSets() {
   for (unsigned i = 0, e = SuperSetIDs.size(); i != e; ++i) {
     unsigned SuperIdx = SuperSetIDs[i];
     PrunedUnitSets[i].Name = RegUnitSets[SuperIdx].Name;
-    PrunedUnitSets[i].Units.swap(RegUnitSets[SuperIdx].Units);
+    PrunedUnitSets[i].Units = std::move(RegUnitSets[SuperIdx].Units);
   }
-  RegUnitSets.swap(PrunedUnitSets);
+  RegUnitSets = std::move(PrunedUnitSets);
 }
 
 // Create a RegUnitSet for each RegClass that contains all units in the class
@@ -2140,7 +2140,7 @@ void CodeGenRegBank::computeRegUnitSets() {
     if (RCUnitSetsIdx == RegClassUnitSets.size()) {
       // Create a new list of UnitSets as a "fake" register class.
       RegClassUnitSets.resize(RCUnitSetsIdx + 1);
-      RegClassUnitSets[RCUnitSetsIdx].swap(RUSets);
+      RegClassUnitSets[RCUnitSetsIdx] = std::move(RUSets);
     }
   }
 }
diff --git a/llvm/utils/TableGen/CodeGenSchedule.cpp b/llvm/utils/TableGen/CodeGenSchedule.cpp
index 9cebc427dbdbc7..e56bf5bdee634b 100644
--- a/llvm/utils/TableGen/CodeGenSchedule.cpp
+++ b/llvm/utils/TableGen/CodeGenSchedule.cpp
@@ -1788,7 +1788,7 @@ void CodeGenSchedModels::inferFromRW(ArrayRef<unsigned> OperWrites,
     for (const PredTransition &Trans : LastTransitions)
       SubstitutedAny |= Transitions.substituteVariants(Trans);
     LLVM_DEBUG(Transitions.dump());
-    LastTransitions.swap(Transitions.TransVec);
+    LastTransitions = std::move(Transitions.TransVec);
   } while (SubstitutedAny);
 
   // WARNING: We are about to mutate the SchedClasses vector. Do not refer to
diff --git a/llvm/utils/TableGen/GlobalISelMatchTable.cpp b/llvm/utils/TableGen/GlobalISelMatchTable.cpp
index f7166ead9adc3d..d1bdc30849a7f6 100644
--- a/llvm/utils/TableGen/GlobalISelMatchTable.cpp
+++ b/llvm/utils/TableGen/GlobalISelMatchTable.cpp
@@ -545,8 +545,8 @@ void GroupMatcher::optimize() {
     if (T != E)
       F = ++T;
   }
-  optimizeRules<GroupMatcher>(Matchers, MatcherStorage).swap(Matchers);
-  optimizeRules<SwitchMatcher>(Matchers, MatcherStorage).swap(Matchers);
+  Matchers = optimizeRules<GroupMatcher>(Matchers, MatcherStorage);
+  Matchers = optimizeRules<SwitchMatcher>(Matchers, MatcherStorage);
 }
 
 //===- SwitchMatcher ------------------------------------------------------===//
diff --git a/llvm/utils/TableGen/SubtargetEmitter.cpp b/llvm/utils/TableGen/SubtargetEmitter.cpp
index b1502eaa20712a..ebe39167703c8c 100644
--- a/llvm/utils/TableGen/SubtargetEmitter.cpp
+++ b/llvm/utils/TableGen/SubtargetEmitter.cpp
@@ -1649,7 +1649,7 @@ static void collectProcessorIndices(const CodeGenSchedClass &SC,
     IdxVec PI;
     std::set_union(&T.ProcIndex, &T.ProcIndex + 1, ProcIndices.begin(),
                    ProcIndices.end(), std::back_inserter(PI));
-    ProcIndices.swap(PI);
+    ProcIndices = std::move(PI);
   }
 }
 

[jayfoad]

The motivation is just to write the code in a more obvious way. As a secondary effect this could conceivably reduce peak memory usage, since the memory for B could be freed slightly earlier; but I am not expecting that to be measurable.

[jayfoad]

No std::moves here since we would be moving from a temporary, and the compiler warns that this prevents copy elision.

[s-barannikov]

FWIW There is also llvm::set_intersect that could modify ContainingSet in-place. It does not accept a '<' comparator though, so migh not be suitable.

[dwblaikie]

Historically TableGen has used A.swap(B) to move containers without the expense of copying them. Perhaps this predated rvalue references. In any case A = std::move(B) seems like a more direct way to implement this when only A is required after the operation.

That's certainly a throwback indeed - nice cleanup!