diff --git a/llvm/lib/CodeGen/SelectOptimize.cpp b/llvm/lib/CodeGen/SelectOptimize.cpp
index ca46e8aa7d424f..b5a8a80193adbe 100644
--- a/llvm/lib/CodeGen/SelectOptimize.cpp
+++ b/llvm/lib/CodeGen/SelectOptimize.cpp
@@ -555,6 +555,12 @@ void SelectOptimize::findProfitableSIGroupsBase(SelectGroups &SIGroups,
   }
 }
 
+static void EmitAndPrintRemark(OptimizationRemarkEmitter *ORE,
+                               DiagnosticInfoOptimizationBase &Rem) {
+  LLVM_DEBUG(dbgs() << Rem.getMsg() << "\n");
+  ORE->emit(Rem);
+}
+
 void SelectOptimize::findProfitableSIGroupsInnerLoops(
     const Loop *L, SelectGroups &SIGroups, SelectGroups &ProfSIGroups) {
   NumSelectOptAnalyzed += SIGroups.size();
@@ -589,7 +595,7 @@ void SelectOptimize::findProfitableSIGroupsInnerLoops(
       OR << "Profitable to convert to branch (loop analysis). BranchCost="
          << BranchCost.toString() << ", SelectCost=" << SelectCost.toString()
          << ". ";
-      ORE->emit(OR);
+      EmitAndPrintRemark(ORE, OR);
       ++NumSelectConvertedLoop;
       ProfSIGroups.push_back(ASI);
     } else {
@@ -597,7 +603,7 @@ void SelectOptimize::findProfitableSIGroupsInnerLoops(
       ORmiss << "Select is more profitable (loop analysis). BranchCost="
              << BranchCost.toString()
              << ", SelectCost=" << SelectCost.toString() << ". ";
-      ORE->emit(ORmiss);
+      EmitAndPrintRemark(ORE, ORmiss);
     }
   }
 }
@@ -605,6 +611,7 @@ void SelectOptimize::findProfitableSIGroupsInnerLoops(
 bool SelectOptimize::isConvertToBranchProfitableBase(
     const SmallVector<SelectInst *, 2> &ASI) {
   SelectInst *SI = ASI.front();
+  LLVM_DEBUG(dbgs() << "Analyzing select group containing " << *SI << "\n");
   OptimizationRemark OR(DEBUG_TYPE, "SelectOpti", SI);
   OptimizationRemarkMissed ORmiss(DEBUG_TYPE, "SelectOpti", SI);
 
@@ -612,7 +619,7 @@ bool SelectOptimize::isConvertToBranchProfitableBase(
   if (PSI->isColdBlock(SI->getParent(), BFI.get())) {
     ++NumSelectColdBB;
     ORmiss << "Not converted to branch because of cold basic block. ";
-    ORE->emit(ORmiss);
+    EmitAndPrintRemark(ORE, ORmiss);
     return false;
   }
 
@@ -620,7 +627,7 @@ bool SelectOptimize::isConvertToBranchProfitableBase(
   if (SI->getMetadata(LLVMContext::MD_unpredictable)) {
     ++NumSelectUnPred;
     ORmiss << "Not converted to branch because of unpredictable branch. ";
-    ORE->emit(ORmiss);
+    EmitAndPrintRemark(ORE, ORmiss);
     return false;
   }
 
@@ -629,7 +636,7 @@ bool SelectOptimize::isConvertToBranchProfitableBase(
   if (isSelectHighlyPredictable(SI) && TLI->isPredictableSelectExpensive()) {
     ++NumSelectConvertedHighPred;
     OR << "Converted to branch because of highly predictable branch. ";
-    ORE->emit(OR);
+    EmitAndPrintRemark(ORE, OR);
     return true;
   }
 
@@ -638,12 +645,12 @@ bool SelectOptimize::isConvertToBranchProfitableBase(
   if (hasExpensiveColdOperand(ASI)) {
     ++NumSelectConvertedExpColdOperand;
     OR << "Converted to branch because of expensive cold operand.";
-    ORE->emit(OR);
+    EmitAndPrintRemark(ORE, OR);
     return true;
   }
 
   ORmiss << "Not profitable to convert to branch (base heuristic).";
-  ORE->emit(ORmiss);
+  EmitAndPrintRemark(ORE, ORmiss);
   return false;
 }
 
@@ -665,7 +672,7 @@ bool SelectOptimize::hasExpensiveColdOperand(
     OptimizationRemarkMissed ORmiss(DEBUG_TYPE, "SelectOpti", ASI.front());
     ORmiss << "Profile data available but missing branch-weights metadata for "
               "select instruction. ";
-    ORE->emit(ORmiss);
+    EmitAndPrintRemark(ORE, ORmiss);
   }
   if (!ColdOperand)
     return false;
@@ -801,7 +808,7 @@ bool SelectOptimize::checkLoopHeuristics(const Loop *L,
       LoopCost[1].NonPredCost >= LoopCost[1].PredCost) {
     ORmissL << "No select conversion in the loop due to no reduction of loop's "
                "critical path. ";
-    ORE->emit(ORmissL);
+    EmitAndPrintRemark(ORE, ORmissL);
     return false;
   }
 
@@ -818,7 +825,7 @@ bool SelectOptimize::checkLoopHeuristics(const Loop *L,
                "loop's critical path. Gain="
             << Gain[1].toString()
             << ", RelativeGain=" << RelativeGain.toString() << "%. ";
-    ORE->emit(ORmissL);
+    EmitAndPrintRemark(ORE, ORmissL);
     return false;
   }
 
@@ -834,7 +841,7 @@ bool SelectOptimize::checkLoopHeuristics(const Loop *L,
       ORmissL << "No select conversion in the loop due to small gradient gain. "
                  "GradientGain="
               << GradientGain.toString() << "%. ";
-      ORE->emit(ORmissL);
+      EmitAndPrintRemark(ORE, ORmissL);
       return false;
     }
   }
@@ -842,7 +849,7 @@ bool SelectOptimize::checkLoopHeuristics(const Loop *L,
   else if (Gain[1] < Gain[0]) {
     ORmissL
         << "No select conversion in the loop due to negative gradient gain. ";
-    ORE->emit(ORmissL);
+    EmitAndPrintRemark(ORE, ORmissL);
     return false;
   }
 
@@ -858,6 +865,8 @@ bool SelectOptimize::checkLoopHeuristics(const Loop *L,
 bool SelectOptimize::computeLoopCosts(
     const Loop *L, const SelectGroups &SIGroups,
     DenseMap<const Instruction *, CostInfo> &InstCostMap, CostInfo *LoopCost) {
+  LLVM_DEBUG(dbgs() << "Calculating Latency / IPredCost / INonPredCost of loop "
+                    << L->getHeader()->getName() << "\n");
   const auto &SIset = getSIset(SIGroups);
   // Compute instruction and loop-critical-path costs across two iterations for
   // both predicated and non-predicated version.
@@ -891,7 +900,7 @@ bool SelectOptimize::computeLoopCosts(
           ORmissL << "Invalid instruction cost preventing analysis and "
                      "optimization of the inner-most loop containing this "
                      "instruction. ";
-          ORE->emit(ORmissL);
+          EmitAndPrintRemark(ORE, ORmissL);
           return false;
         }
         IPredCost += Scaled64::get(ILatency.value());
@@ -925,12 +934,17 @@ bool SelectOptimize::computeLoopCosts(
 
           INonPredCost = PredictedPathCost + MispredictCost;
         }
+        LLVM_DEBUG(dbgs() << " " << ILatency << "/" << IPredCost << "/"
+                          << INonPredCost << " for " << I << "\n");
 
         InstCostMap[&I] = {IPredCost, INonPredCost};
         MaxCost.PredCost = std::max(MaxCost.PredCost, IPredCost);
         MaxCost.NonPredCost = std::max(MaxCost.NonPredCost, INonPredCost);
       }
     }
+    LLVM_DEBUG(dbgs() << "Iteration " << Iter + 1
+                      << " MaxCost = " << MaxCost.PredCost << " "
+                      << MaxCost.NonPredCost << "\n");
   }
   return true;
 }