[SimplifyCFG] Simplify conditional branches on const icmp eq's

Author: yoni-lavi

llvm/include/llvm/Transforms/Utils/Local.h

@@ -392,7 +392,8 @@ Instruction *removeUnwindEdge(BasicBlock *BB, DomTreeUpdater *DTU = nullptr);
 ///
 /// Returns true if any basic block was removed.
 bool removeUnreachableBlocks(Function &F, DomTreeUpdater *DTU = nullptr,
-                             MemorySSAUpdater *MSSAU = nullptr);
+                             MemorySSAUpdater *MSSAU = nullptr,
+                             bool KeepOneInputPHIs = false);
 
 /// Combine the metadata of two instructions so that K can replace J. Some
 /// metadata kinds can only be kept if K does not move, meaning it dominated

llvm/lib/Transforms/Scalar/SimplifyCFGPass.cpp

@@ -26,6 +26,7 @@
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Analysis/AssumptionCache.h"
 #include "llvm/Analysis/CFG.h"
+#include "llvm/Analysis/ConstantFolding.h"
 #include "llvm/Analysis/DomTreeUpdater.h"
 #include "llvm/Analysis/GlobalsModRef.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
@@ -262,12 +263,52 @@ static bool iterativelySimplifyCFG(Function &F, const TargetTransformInfo &TTI,
   return Changed;
 }
 
+static void decayOneInputPHIs(BasicBlock &BB) {
+  if (BB.empty() || !isa<PHINode>(BB.begin()))
+    return;
+  unsigned NumPreds = cast<PHINode>(BB.front()).getNumIncomingValues();
+  if (NumPreds != 1)
+    return;
+  for (PHINode &Phi : make_early_inc_range(BB.phis())) {
+    if (Value *PhiConst = Phi.hasConstantValue()) {
+      Phi.replaceAllUsesWith(PhiConst);
+      Phi.eraseFromParent();
+      // Additionally, constant-fold conditional branch if there is one present,
+      // and the PHI has decayed into a constant, since it may make another CFG
+      // edge dead.
+      if (!isa<Constant>(PhiConst))
+        continue;
+      if (BranchInst *BI = dyn_cast<BranchInst>(BB.getTerminator()))
+        if (BI->isConditional())
+          if (auto *Cmp = dyn_cast<CmpInst>(BI->getCondition()))
+            if (auto *LHS = dyn_cast<Constant>(Cmp->getOperand(0)))
+              if (auto *RHS = dyn_cast<Constant>(Cmp->getOperand(1))) {
+                const DataLayout &DL = BB.getModule()->getDataLayout();
+                Constant *ConstCond = ConstantFoldCompareInstOperands(
+                    Cmp->getPredicate(), LHS, RHS, DL);
+                if (ConstCond)
+                  BI->setCondition(ConstCond);
+              }
+    }
+  }
+}
+
+static bool removeUnreachableBlocksAndSimplify(Function &F,
+                                               DomTreeUpdater *DTU) {
+  bool Changed = removeUnreachableBlocks(F, DTU, /*MSSAU=*/nullptr,
+                                         /*KeepOneInputPHIs=*/true);
+  if (Changed)
+    for (BasicBlock &BB : F)
+      decayOneInputPHIs(BB);
+  return Changed;
+}
+
 static bool simplifyFunctionCFGImpl(Function &F, const TargetTransformInfo &TTI,
                                     DominatorTree *DT,
                                     const SimplifyCFGOptions &Options) {
   DomTreeUpdater DTU(DT, DomTreeUpdater::UpdateStrategy::Eager);
 
-  bool EverChanged = removeUnreachableBlocks(F, DT ? &DTU : nullptr);
+  bool EverChanged = removeUnreachableBlocksAndSimplify(F, DT ? &DTU : nullptr);
   EverChanged |=
       tailMergeBlocksWithSimilarFunctionTerminators(F, DT ? &DTU : nullptr);
   EverChanged |= iterativelySimplifyCFG(F, TTI, DT ? &DTU : nullptr, Options);
@@ -285,7 +326,7 @@ static bool simplifyFunctionCFGImpl(Function &F, const TargetTransformInfo &TTI,
 
   do {
     EverChanged = iterativelySimplifyCFG(F, TTI, DT ? &DTU : nullptr, Options);
-    EverChanged |= removeUnreachableBlocks(F, DT ? &DTU : nullptr);
+    EverChanged |= removeUnreachableBlocksAndSimplify(F, DT ? &DTU : nullptr);
   } while (EverChanged);
 
   return true;

llvm/lib/Transforms/Utils/Local.cpp

@@ -3080,7 +3080,8 @@ Instruction *llvm::removeUnwindEdge(BasicBlock *BB, DomTreeUpdater *DTU) {
 /// if they are in a dead cycle.  Return true if a change was made, false
 /// otherwise.
 bool llvm::removeUnreachableBlocks(Function &F, DomTreeUpdater *DTU,
-                                   MemorySSAUpdater *MSSAU) {
+                                   MemorySSAUpdater *MSSAU,
+                                   bool KeepOneInputPHIs) {
   SmallPtrSet<BasicBlock *, 16> Reachable;
   bool Changed = markAliveBlocks(F, Reachable, DTU);
 
@@ -3111,7 +3112,7 @@ bool llvm::removeUnreachableBlocks(Function &F, DomTreeUpdater *DTU,
   if (MSSAU)
     MSSAU->removeBlocks(BlocksToRemove);
 
-  DeleteDeadBlocks(BlocksToRemove.takeVector(), DTU);
+  DeleteDeadBlocks(BlocksToRemove.takeVector(), DTU, KeepOneInputPHIs);
 
   return Changed;
 }

llvm/test/Analysis/MemorySSA/nondeterminism.ll

@@ -1,7 +1,7 @@
 ; RUN: opt -passes=simplifycfg -S --preserve-ll-uselistorder %s | FileCheck %s
 ; REQUIRES: x86-registered-target
 ; CHECK-LABEL: @n
-; CHECK: uselistorder i16 0, { 3, 2, 4, 1, 5, 0, 6 }
+; CHECK: uselistorder i16 0, { 0, 1, 4, 3, 5, 2, 6 }
 
 ; Note: test was added in an effort to ensure determinism when updating memoryssa. See PR42574.
 ; If the uselistorder check becomes no longer relevant, the test can be disabled or removed.

llvm/test/Transforms/Coroutines/coro-alloca-01.ll

@@ -42,7 +42,7 @@ suspend:
 ; CHECK-LABEL: @f(
 ; CHECK:         %x.reload.addr = getelementptr inbounds %f.Frame, ptr %hdl, i32 0, i32 2
 ; CHECK:         %y.reload.addr = getelementptr inbounds %f.Frame, ptr %hdl, i32 0, i32 3
-; CHECK:         %alias_phi = phi ptr [ %y.reload.addr, %merge.from.flag_false ], [ %x.reload.addr, %entry ]
+; CHECK:         %alias_phi = select i1 %n, ptr %x.reload.addr, ptr %y.reload.addr
 ; CHECK:         %alias_phi.spill.addr = getelementptr inbounds %f.Frame, ptr %hdl, i32 0, i32 4
 ; CHECK:         store ptr %alias_phi, ptr %alias_phi.spill.addr, align 8
 

llvm/test/Transforms/Coroutines/coro-alloca-07.ll

@@ -70,11 +70,7 @@ declare void @free(ptr)
 ; CHECK-NEXT:    store ptr @f.destroy, ptr [[DESTROY_ADDR]], align 8
 ; CHECK-NEXT:    [[X_RELOAD_ADDR:%.*]] = getelementptr inbounds [[F_FRAME]], ptr [[HDL]], i32 0, i32 2
 ; CHECK-NEXT:    [[Y_RELOAD_ADDR:%.*]] = getelementptr inbounds [[F_FRAME]], ptr [[HDL]], i32 0, i32 3
-; CHECK-NEXT:    br i1 [[N:%.*]], label [[MERGE:%.*]], label [[MERGE_FROM_FLAG_FALSE:%.*]]
-; CHECK:       merge.from.flag_false:
-; CHECK-NEXT:    br label [[MERGE:%.*]]
-; CHECK:       merge:
-; CHECK-NEXT:    [[ALIAS_PHI:%.*]] = phi ptr [ [[Y_RELOAD_ADDR]], [[MERGE_FROM_FLAG_FALSE]] ], [ [[X_RELOAD_ADDR]], [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[ALIAS_PHI:%.*]] = select i1 [[N:%.*]], ptr [[X_RELOAD_ADDR]], ptr [[Y_RELOAD_ADDR]]
 ; CHECK-NEXT:    [[ALIAS_PHI_SPILL_ADDR:%.*]] = getelementptr inbounds [[F_FRAME]], ptr [[HDL]], i32 0, i32 4
 ; CHECK-NEXT:    store ptr [[ALIAS_PHI]], ptr [[ALIAS_PHI_SPILL_ADDR]], align 8
 ; CHECK-NEXT:    store i8 1, ptr [[ALIAS_PHI]], align 1

llvm/test/Transforms/Coroutines/coro-alloca-outside-frame.ll

@@ -43,7 +43,7 @@ suspend:
 ; CHECK:         %x = alloca i64, align 8, !coro.outside.frame !0
 ; CHECK-NOT:     %x.reload.addr = getelementptr inbounds %f.Frame, ptr %hdl, i32 0, i32 2
 ; CHECK:         %y.reload.addr = getelementptr inbounds %f.Frame, ptr %hdl, i32 0, i32 2
-; CHECK:         %alias_phi = phi ptr [ %y.reload.addr, %merge.from.flag_false ], [ %x, %entry ]
+; CHECK:         %alias_phi = select i1 %n, ptr %x, ptr %y.reload.addr
 
 declare ptr @llvm.coro.free(token, ptr)
 declare i32 @llvm.coro.size.i32()

llvm/test/Transforms/Coroutines/coro-spill-defs-before-corobegin.ll

@@ -51,15 +51,15 @@ lpad:
 ; CHECK-LABEL: @f(
 ; CHECK:       %alloc = call ptr @malloc(i32 32)
 ; CHECK-NEXT:  %flag = call i1 @check(ptr %alloc)
-; CHECK-NEXT:  %spec.select = select i1 %flag, i32 0, i32 1
+; CHECK-NEXT:  %value_phi = select i1 %flag, i32 0, i32 1
 ; CHECK-NEXT:  %value_invoke = call i32 @calc()
 ; CHECK-NEXT:  %hdl = call noalias nonnull ptr @llvm.coro.begin(token %id, ptr %alloc)
 
 ; CHECK:       store ptr @f.destroy, ptr %destroy.addr
 ; CHECK-NEXT:  %value_invoke.spill.addr = getelementptr inbounds %f.Frame, ptr %hdl, i32 0, i32 3
 ; CHECK-NEXT:  store i32 %value_invoke, ptr %value_invoke.spill.addr
 ; CHECK-NEXT:  %value_phi.spill.addr = getelementptr inbounds %f.Frame, ptr %hdl, i32 0, i32 2
-; CHECK-NEXT:  store i32 %spec.select, ptr %value_phi.spill.addr
+; CHECK-NEXT:  store i32 %value_phi, ptr %value_phi.spill.addr
 
 declare ptr @llvm.coro.free(token, ptr)
 declare i32 @llvm.coro.size.i32()

llvm/test/Transforms/SimplifyCFG/intersection-block-with-dead-predecessor.ll

@@ -0,0 +1,52 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt < %s -passes=simplifycfg -simplifycfg-require-and-preserve-domtree=1 -S | FileCheck %s
+
+; A pre-existing foldable pattern should remain intact
+define void @const_valued_cond_br(ptr %P) {
+; CHECK-LABEL: define void @const_valued_cond_br(
+; CHECK-SAME: ptr [[P:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 42, 42
+; CHECK-NEXT:    br i1 [[COND]], label [[A:%.*]], label [[B:%.*]]
+; CHECK:       a:
+; CHECK-NEXT:    store i32 123, ptr [[P]], align 4
+; CHECK-NEXT:    br label [[B]]
+; CHECK:       b:
+; CHECK-NEXT:    ret void
+;
+entry:
+  %cond = icmp eq i32 42, 42
+  br i1 %cond, label %a, label %b
+a:
+  store i32 123, ptr %P
+  br label %b
+b:
+  ret void
+}
+
+; When the phi decays to a constant, the terminator of `b` gets constant-folded,
+; enabling further simplification.
+define void @intersection_block_with_dead_predecessor(ptr %P) {
+; CHECK-LABEL: define void @intersection_block_with_dead_predecessor(
+; CHECK-SAME: ptr [[P:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 1, 1
+; CHECK-NEXT:    store i32 321, ptr [[P]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %b
+b:
+  %x = phi i32 [1, %entry], [2, %a]
+  switch i32 %x, label %c [
+  i32 1, label %d
+  ]
+c:
+  store i32 123, ptr %P
+  ret void
+d:
+  store i32 321, ptr %P
+  ret void
+a: ; unreachable
+  br label %b
+}