[LoopUnroll] Add RuntimeUnrollMultiExit to loop unroll options (NFC) (#124462)

Add an extra knob to RuntimeUnrollMultiExit to let backends control
whether to allow multi-exit unrolling on a per-loop basis.

This gives backends more fine-grained control on deciding if multi-exit
unrolling is profitable for a given loop and uarch. Similar to
4226e0a.

PR: #124462

Author: fhahn

llvm/include/llvm/Analysis/TargetTransformInfo.h

@@ -622,6 +622,11 @@ class TargetTransformInfo {
     /// Don't allow runtime unrolling if expanding the trip count takes more
     /// than SCEVExpansionBudget.
     unsigned SCEVExpansionBudget;
+    /// Allow runtime unrolling multi-exit loops. Should only be set if the
+    /// target determined that multi-exit unrolling is profitable for the loop.
+    /// Fall back to the generic logic to determine whether multi-exit unrolling
+    /// is profitable if set to false.
+    bool RuntimeUnrollMultiExit;
   };
 
   /// Get target-customized preferences for the generic loop unrolling

llvm/include/llvm/Transforms/Utils/UnrollLoop.h

@@ -76,6 +76,7 @@ struct UnrollLoopOptions {
   bool ForgetAllSCEV;
   const Instruction *Heart = nullptr;
   unsigned SCEVExpansionBudget;
+  bool RuntimeUnrollMultiExit = false;
 };
 
 LoopUnrollResult UnrollLoop(Loop *L, UnrollLoopOptions ULO, LoopInfo *LI,
@@ -91,7 +92,8 @@ bool UnrollRuntimeLoopRemainder(
     bool UseEpilogRemainder, bool UnrollRemainder, bool ForgetAllSCEV,
     LoopInfo *LI, ScalarEvolution *SE, DominatorTree *DT, AssumptionCache *AC,
     const TargetTransformInfo *TTI, bool PreserveLCSSA,
-    unsigned SCEVExpansionBudget, Loop **ResultLoop = nullptr);
+    unsigned SCEVExpansionBudget, bool RuntimeUnrollMultiExit,
+    Loop **ResultLoop = nullptr);
 
 LoopUnrollResult UnrollAndJamLoop(Loop *L, unsigned Count, unsigned TripCount,
                                   unsigned TripMultiple, bool UnrollRemainder,

llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp

@@ -220,6 +220,7 @@ TargetTransformInfo::UnrollingPreferences llvm::gatherUnrollingPreferences(
   UP.UnrollAndJamInnerLoopThreshold = 60;
   UP.MaxIterationsCountToAnalyze = UnrollMaxIterationsCountToAnalyze;
   UP.SCEVExpansionBudget = SCEVCheapExpansionBudget;
+  UP.RuntimeUnrollMultiExit = false;
 
   // Override with any target specific settings
   TTI.getUnrollingPreferences(L, SE, UP, &ORE);
@@ -1352,6 +1353,7 @@ tryToUnrollLoop(Loop *L, DominatorTree &DT, LoopInfo *LI, ScalarEvolution &SE,
   ULO.ForgetAllSCEV = ForgetAllSCEV;
   ULO.Heart = getLoopConvergenceHeart(L);
   ULO.SCEVExpansionBudget = UP.SCEVExpansionBudget;
+  ULO.RuntimeUnrollMultiExit = UP.RuntimeUnrollMultiExit;
   LoopUnrollResult UnrollResult = UnrollLoop(
       L, ULO, LI, &SE, &DT, &AC, &TTI, &ORE, PreserveLCSSA, &RemainderLoop, AA);
   if (UnrollResult == LoopUnrollResult::Unmodified)

llvm/lib/Transforms/Utils/LoopUnroll.cpp

@@ -590,10 +590,11 @@ llvm::UnrollLoop(Loop *L, UnrollLoopOptions ULO, LoopInfo *LI,
                                               : isEpilogProfitable(L);
 
   if (ULO.Runtime &&
-      !UnrollRuntimeLoopRemainder(
-          L, ULO.Count, ULO.AllowExpensiveTripCount, EpilogProfitability,
-          ULO.UnrollRemainder, ULO.ForgetAllSCEV, LI, SE, DT, AC, TTI,
-          PreserveLCSSA, ULO.SCEVExpansionBudget, RemainderLoop)) {
+      !UnrollRuntimeLoopRemainder(L, ULO.Count, ULO.AllowExpensiveTripCount,
+                                  EpilogProfitability, ULO.UnrollRemainder,
+                                  ULO.ForgetAllSCEV, LI, SE, DT, AC, TTI,
+                                  PreserveLCSSA, ULO.SCEVExpansionBudget,
+                                  ULO.RuntimeUnrollMultiExit, RemainderLoop)) {
     if (ULO.Force)
       ULO.Runtime = false;
     else {

llvm/lib/Transforms/Utils/LoopUnrollRuntime.cpp

@@ -461,7 +461,7 @@ CloneLoopBlocks(Loop *L, Value *NewIter, const bool UseEpilogRemainder,
 
 /// Returns true if we can profitably unroll the multi-exit loop L. Currently,
 /// we return true only if UnrollRuntimeMultiExit is set to true.
-static bool canProfitablyUnrollMultiExitLoop(
+static bool canProfitablyRuntimeUnrollMultiExitLoop(
     Loop *L, SmallVectorImpl<BasicBlock *> &OtherExits, BasicBlock *LatchExit,
     bool UseEpilogRemainder) {
 
@@ -583,7 +583,8 @@ bool llvm::UnrollRuntimeLoopRemainder(
     bool UseEpilogRemainder, bool UnrollRemainder, bool ForgetAllSCEV,
     LoopInfo *LI, ScalarEvolution *SE, DominatorTree *DT, AssumptionCache *AC,
     const TargetTransformInfo *TTI, bool PreserveLCSSA,
-    unsigned SCEVExpansionBudget, Loop **ResultLoop) {
+    unsigned SCEVExpansionBudget, bool RuntimeUnrollMultiExit,
+    Loop **ResultLoop) {
   LLVM_DEBUG(dbgs() << "Trying runtime unrolling on Loop: \n");
   LLVM_DEBUG(L->dump());
   LLVM_DEBUG(UseEpilogRemainder ? dbgs() << "Using epilog remainder.\n"
@@ -632,8 +633,9 @@ bool llvm::UnrollRuntimeLoopRemainder(
     if (!PreserveLCSSA)
       return false;
 
-    if (!canProfitablyUnrollMultiExitLoop(L, OtherExits, LatchExit,
-                                          UseEpilogRemainder)) {
+    if (!RuntimeUnrollMultiExit &&
+        !canProfitablyRuntimeUnrollMultiExitLoop(L, OtherExits, LatchExit,
+                                                 UseEpilogRemainder)) {
       LLVM_DEBUG(
           dbgs()
           << "Multiple exit/exiting blocks in loop and multi-exit unrolling not "

llvm/unittests/Transforms/Utils/UnrollLoopTest.cpp

@@ -73,6 +73,6 @@ while.end:                                        ; preds = %while.cond
 
   bool ret =
       UnrollRuntimeLoopRemainder(L, 4, true, false, false, false, &LI, &SE, &DT,
-                                 &AC, /*TTI=*/nullptr, PreserveLCSSA, 4);
+                                 &AC, /*TTI=*/nullptr, PreserveLCSSA, 4, false);
   EXPECT_FALSE(ret);
 }