Remove REMAT_DEBUG and break ties in score

Author: lucas-rami

llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp

@@ -1105,16 +1105,16 @@ bool ClusteredLowOccStage::initGCNSchedStage() {
 void PreRARematStage::printTargetRegions(bool PrintAll) const {
   if (PrintAll) {
     for (auto [I, Target] : enumerate(RPTargets))
-      REMAT_DEBUG(dbgs() << "  [" << I << "] " << Target << '\n');
+      dbgs() << REMAT_PREFIX << "  [" << I << "] " << Target << '\n';
     return;
   }
   if (TargetRegions.none()) {
-    REMAT_DEBUG(dbgs() << "No target regions\n");
+    dbgs() << REMAT_PREFIX << "No target regions\n";
     return;
   }
-  REMAT_DEBUG(dbgs() << "Target regions:\n");
+  dbgs() << REMAT_PREFIX << "Target regions:\n";
   for (unsigned I : TargetRegions.set_bits())
-    REMAT_DEBUG(dbgs() << "  [" << I << "] " << RPTargets[I] << '\n');
+    dbgs() << REMAT_PREFIX << "  [" << I << "] " << RPTargets[I] << '\n';
 }
 
 void PreRARematStage::RematReg::print(
@@ -1230,8 +1230,8 @@ bool PreRARematStage::initGCNSchedStage() {
   }
   unsetSatisifedRPTargets(RescheduleRegions);
 
+  LLVM_DEBUG(printTargetRegions());
 #ifndef NDEBUG
-  printTargetRegions();
   unsigned RoundNum = 0;
 #endif
 
@@ -1242,7 +1242,7 @@ bool PreRARematStage::initGCNSchedStage() {
     // (Re-)Score and (re-)sort all remats in increasing score order.
     for (ScoredRemat &Remat : ScoredRemats)
       Remat.update(TargetRegions, RPTargets, RegionFreq, !TargetOcc);
-    stable_sort(ScoredRemats);
+    sort(ScoredRemats);
 
     REMAT_DEBUG({
       dbgs() << "==== ROUND " << RoundNum << " ====\n";
@@ -1294,8 +1294,8 @@ bool PreRARematStage::initGCNSchedStage() {
       unsetSatisifedRPTargets(Remat.Live);
     }
 
+    LLVM_DEBUG(printTargetRegions());
 #ifndef NDEBUG
-    printTargetRegions();
     ++RoundNum;
 #endif
 

llvm/lib/Target/AMDGPU/GCNSchedStrategy.h

@@ -530,8 +530,14 @@ class PreRARematStage : public GCNSchedStage {
 
     int getScore() const { return Score; }
 
-    bool operator<(const ScoredRemat &O) const { return Score < O.Score; }
-    bool operator==(const ScoredRemat &O) const { return Score == O.Score; }
+    bool operator<(const ScoredRemat &O) const {
+      // Break ties using pointer to rematerializable register. Since
+      // rematerializations are collected in instruction order, registers
+      // appearing earlier have a "higher score" than those appearing later.
+      if (Score == O.Score)
+        return Remat > O.Remat;
+      return Score < O.Score;
+    }
 
   private:
     /// Estimated save/restore latency costs for spilling a register to stack.