Merged master:8260db752c91 into amd-gfx:362b06229d79

Local branch amd-gfx 362b062 Merged master:7ba3293691be into amd-gfx:f2bcfb477c22
Remote branch master 8260db7 [mlir][Python] Return and accept OpView for all functions.

llvm/include/llvm/Analysis/MustExecute.h

@@ -27,6 +27,8 @@
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/Analysis/EHPersonalities.h"
 #include "llvm/Analysis/InstructionPrecedenceTracking.h"
+#include "llvm/IR/PassManager.h"
+#include "llvm/Support/raw_ostream.h"
 
 namespace llvm {
 
@@ -541,6 +543,23 @@ struct MustBeExecutedContextExplorer {
   MustBeExecutedIterator EndIterator;
 };
 
+class MustExecutePrinterPass : public PassInfoMixin<MustExecutePrinterPass> {
+  raw_ostream &OS;
+
+public:
+  MustExecutePrinterPass(raw_ostream &OS) : OS(OS) {}
+  PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
+};
+
+class MustBeExecutedContextPrinterPass
+    : public PassInfoMixin<MustBeExecutedContextPrinterPass> {
+  raw_ostream &OS;
+
+public:
+  MustBeExecutedContextPrinterPass(raw_ostream &OS) : OS(OS) {}
+  PreservedAnalyses run(Module &M, ModuleAnalysisManager &AM);
+};
+
 } // namespace llvm
 
 #endif

llvm/lib/Analysis/MustExecute.cpp

@@ -16,9 +16,11 @@
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/IR/AssemblyAnnotationWriter.h"
 #include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Dominators.h"
 #include "llvm/IR/InstIterator.h"
 #include "llvm/IR/LLVMContext.h"
 #include "llvm/IR/Module.h"
+#include "llvm/IR/PassManager.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormattedStream.h"
@@ -300,30 +302,31 @@ bool ICFLoopSafetyInfo::doesNotWriteMemoryBefore(const Instruction &I,
 }
 
 namespace {
-  struct MustExecutePrinter : public FunctionPass {
+struct MustExecutePrinter : public FunctionPass {
 
-    static char ID; // Pass identification, replacement for typeid
-    MustExecutePrinter() : FunctionPass(ID) {
-      initializeMustExecutePrinterPass(*PassRegistry::getPassRegistry());
-    }
-    void getAnalysisUsage(AnalysisUsage &AU) const override {
-      AU.setPreservesAll();
-      AU.addRequired<DominatorTreeWrapperPass>();
-      AU.addRequired<LoopInfoWrapperPass>();
-    }
-    bool runOnFunction(Function &F) override;
-  };
-  struct MustBeExecutedContextPrinter : public ModulePass {
-    static char ID;
+  static char ID; // Pass identification, replacement for typeid
+  MustExecutePrinter() : FunctionPass(ID) {
+    initializeMustExecutePrinterPass(*PassRegistry::getPassRegistry());
+  }
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.setPreservesAll();
+    AU.addRequired<DominatorTreeWrapperPass>();
+    AU.addRequired<LoopInfoWrapperPass>();
+  }
+  bool runOnFunction(Function &F) override;
+};
+struct MustBeExecutedContextPrinter : public ModulePass {
+  static char ID;
 
-    MustBeExecutedContextPrinter() : ModulePass(ID) {
-      initializeMustBeExecutedContextPrinterPass(*PassRegistry::getPassRegistry());
-    }
-    void getAnalysisUsage(AnalysisUsage &AU) const override {
-      AU.setPreservesAll();
-    }
-    bool runOnModule(Module &M) override;
-  };
+  MustBeExecutedContextPrinter() : ModulePass(ID) {
+    initializeMustBeExecutedContextPrinterPass(
+        *PassRegistry::getPassRegistry());
+  }
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.setPreservesAll();
+  }
+  bool runOnModule(Module &M) override;
+};
 }
 
 char MustExecutePrinter::ID = 0;
@@ -339,15 +342,16 @@ FunctionPass *llvm::createMustExecutePrinter() {
 }
 
 char MustBeExecutedContextPrinter::ID = 0;
-INITIALIZE_PASS_BEGIN(
-    MustBeExecutedContextPrinter, "print-must-be-executed-contexts",
-    "print the must-be-executed-contexed for all instructions", false, true)
+INITIALIZE_PASS_BEGIN(MustBeExecutedContextPrinter,
+                      "print-must-be-executed-contexts",
+                      "print the must-be-executed-context for all instructions",
+                      false, true)
 INITIALIZE_PASS_DEPENDENCY(PostDominatorTreeWrapperPass)
 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
 INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)
 INITIALIZE_PASS_END(MustBeExecutedContextPrinter,
                     "print-must-be-executed-contexts",
-                    "print the must-be-executed-contexed for all instructions",
+                    "print the must-be-executed-context for all instructions",
                     false, true)
 
 ModulePass *llvm::createMustBeExecutedContextPrinter() {
@@ -835,3 +839,42 @@ const Instruction *MustBeExecutedIterator::advance() {
   Tail = nullptr;
   return nullptr;
 }
+
+PreservedAnalyses MustExecutePrinterPass::run(Function &F,
+                                              FunctionAnalysisManager &AM) {
+  auto &LI = AM.getResult<LoopAnalysis>(F);
+  auto &DT = AM.getResult<DominatorTreeAnalysis>(F);
+
+  MustExecuteAnnotatedWriter Writer(F, DT, LI);
+  F.print(OS, &Writer);
+  return PreservedAnalyses::all();
+}
+
+PreservedAnalyses
+MustBeExecutedContextPrinterPass::run(Module &M, ModuleAnalysisManager &AM) {
+  FunctionAnalysisManager &FAM =
+      AM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
+  GetterTy<const LoopInfo> LIGetter = [&](const Function &F) {
+    return &FAM.getResult<LoopAnalysis>(const_cast<Function &>(F));
+  };
+  GetterTy<const DominatorTree> DTGetter = [&](const Function &F) {
+    return &FAM.getResult<DominatorTreeAnalysis>(const_cast<Function &>(F));
+  };
+  GetterTy<const PostDominatorTree> PDTGetter = [&](const Function &F) {
+    return &FAM.getResult<PostDominatorTreeAnalysis>(const_cast<Function &>(F));
+  };
+
+  MustBeExecutedContextExplorer Explorer(
+      /* ExploreInterBlock */ true,
+      /* ExploreCFGForward */ true,
+      /* ExploreCFGBackward */ true, LIGetter, DTGetter, PDTGetter);
+
+  for (Function &F : M) {
+    for (Instruction &I : instructions(F)) {
+      OS << "-- Explore context of: " << I << "\n";
+      for (const Instruction *CI : Explorer.range(&I))
+        OS << "  [F: " << CI->getFunction()->getName() << "] " << *CI << "\n";
+    }
+  }
+  return PreservedAnalyses::all();
+}

llvm/lib/Passes/PassBuilder.cpp

@@ -51,6 +51,7 @@
 #include "llvm/Analysis/MemorySSA.h"
 #include "llvm/Analysis/ModuleDebugInfoPrinter.h"
 #include "llvm/Analysis/ModuleSummaryAnalysis.h"
+#include "llvm/Analysis/MustExecute.h"
 #include "llvm/Analysis/ObjCARCAliasAnalysis.h"
 #include "llvm/Analysis/OptimizationRemarkEmitter.h"
 #include "llvm/Analysis/PhiValues.h"

llvm/lib/Passes/PassRegistry.def

@@ -86,6 +86,7 @@ MODULE_PASS("print-callgraph", CallGraphPrinterPass(dbgs()))
 MODULE_PASS("print", PrintModulePass(dbgs()))
 MODULE_PASS("print-lcg", LazyCallGraphPrinterPass(dbgs()))
 MODULE_PASS("print-lcg-dot", LazyCallGraphDOTPrinterPass(dbgs()))
+MODULE_PASS("print-must-be-executed-contexts", MustBeExecutedContextPrinterPass(dbgs()))
 MODULE_PASS("print-stack-safety", StackSafetyGlobalPrinterPass(dbgs()))
 MODULE_PASS("print<module-debuginfo>", ModuleDebugInfoPrinterPass(dbgs()))
 MODULE_PASS("rewrite-statepoints-for-gc", RewriteStatepointsForGC())
@@ -276,9 +277,10 @@ FUNCTION_PASS("print<phi-values>", PhiValuesPrinterPass(dbgs()))
 FUNCTION_PASS("print<regions>", RegionInfoPrinterPass(dbgs()))
 FUNCTION_PASS("print<scalar-evolution>", ScalarEvolutionPrinterPass(dbgs()))
 FUNCTION_PASS("print<stack-safety-local>", StackSafetyPrinterPass(dbgs()))
-// TODO: rename to print<alias-sets> after NPM switch
+// TODO: rename to print<foo> after NPM switch
 FUNCTION_PASS("print-alias-sets", AliasSetsPrinterPass(dbgs()))
 FUNCTION_PASS("print-predicateinfo", PredicateInfoPrinterPass(dbgs()))
+FUNCTION_PASS("print-mustexecute", MustExecutePrinterPass(dbgs()))
 FUNCTION_PASS("reassociate", ReassociatePass())
 FUNCTION_PASS("scalarizer", ScalarizerPass())
 FUNCTION_PASS("sccp", SCCPPass())

llvm/lib/Transforms/AggressiveInstCombine/AggressiveInstCombine.cpp

@@ -39,8 +39,6 @@ using namespace PatternMatch;
 STATISTIC(NumAnyOrAllBitsSet, "Number of any/all-bits-set patterns folded");
 STATISTIC(NumGuardedRotates,
           "Number of guarded rotates transformed into funnel shifts");
-STATISTIC(NumGuardedFunnelShifts,
-          "Number of guarded funnel shifts transformed into funnel shifts");
 STATISTIC(NumPopCountRecognized, "Number of popcount idioms recognized");
 
 namespace {
@@ -69,17 +67,17 @@ class AggressiveInstCombinerLegacyPass : public FunctionPass {
 };
 } // namespace
 
-/// Match a pattern for a bitwise funnel/rotate operation that partially guards
-/// against undefined behavior by branching around the funnel-shift/rotation
-/// when the shift amount is 0.
-static bool foldGuardedFunnelShift(Instruction &I) {
+/// Match a pattern for a bitwise rotate operation that partially guards
+/// against undefined behavior by branching around the rotation when the shift
+/// amount is 0.
+static bool foldGuardedRotateToFunnelShift(Instruction &I) {
   if (I.getOpcode() != Instruction::PHI || I.getNumOperands() != 2)
     return false;
 
   // As with the one-use checks below, this is not strictly necessary, but we
   // are being cautious to avoid potential perf regressions on targets that
-  // do not actually have a funnel/rotate instruction (where the funnel shift
-  // would be expanded back into math/shift/logic ops).
+  // do not actually have a rotate instruction (where the funnel shift would be
+  // expanded back into math/shift/logic ops).
   if (!isPowerOf2_32(I.getType()->getScalarSizeInBits()))
     return false;
 
@@ -113,33 +111,27 @@ static bool foldGuardedFunnelShift(Instruction &I) {
     return Intrinsic::not_intrinsic;
   };
 
-  // One phi operand must be a funnel/rotate operation, and the other phi
-  // operand must be the source value of that funnel/rotate operation:
+  // One phi operand must be a rotate operation, and the other phi operand must
+  // be the source value of that rotate operation:
   // phi [ rotate(RotSrc, ShAmt), FunnelBB ], [ RotSrc, GuardBB ]
-  // phi [ fshl(ShlVal0, ShlVal1, ShAmt), FunnelBB ], [ ShlVal0, GuardBB ]
-  // phi [ fshr(ShlVal0, ShlVal1, ShAmt), FunnelBB ], [ ShlVal1, GuardBB ]
   PHINode &Phi = cast<PHINode>(I);
   unsigned FunnelOp = 0, GuardOp = 1;
   Value *P0 = Phi.getOperand(0), *P1 = Phi.getOperand(1);
   Value *ShVal0, *ShVal1, *ShAmt;
   Intrinsic::ID IID = matchFunnelShift(P0, ShVal0, ShVal1, ShAmt);
-  if (IID == Intrinsic::not_intrinsic ||
-      (IID == Intrinsic::fshl && ShVal0 != P1) ||
-      (IID == Intrinsic::fshr && ShVal1 != P1)) {
+  if (IID == Intrinsic::not_intrinsic || ShVal0 != ShVal1 || ShVal0 != P1) {
     IID = matchFunnelShift(P1, ShVal0, ShVal1, ShAmt);
-    if (IID == Intrinsic::not_intrinsic ||
-        (IID == Intrinsic::fshl && ShVal0 != P0) ||
-        (IID == Intrinsic::fshr && ShVal1 != P0))
+    if (IID == Intrinsic::not_intrinsic || ShVal0 != ShVal1 || ShVal0 != P0)
       return false;
     assert((IID == Intrinsic::fshl || IID == Intrinsic::fshr) &&
            "Pattern must match funnel shift left or right");
     std::swap(FunnelOp, GuardOp);
   }
+  assert(ShVal0 == ShVal1 && "Rotation funnel shift pattern expected");
 
   // The incoming block with our source operand must be the "guard" block.
-  // That must contain a cmp+branch to avoid the funnel/rotate when the shift
-  // amount is equal to 0. The other incoming block is the block with the
-  // funnel/rotate.
+  // That must contain a cmp+branch to avoid the rotate when the shift amount
+  // is equal to 0. The other incoming block is the block with the rotate.
   BasicBlock *GuardBB = Phi.getIncomingBlock(GuardOp);
   BasicBlock *FunnelBB = Phi.getIncomingBlock(FunnelOp);
   Instruction *TermI = GuardBB->getTerminator();
@@ -158,21 +150,18 @@ static bool foldGuardedFunnelShift(Instruction &I) {
   //   br i1 %cmp, label %PhiBB, label %FunnelBB
   // FunnelBB:
   //   %sub = sub i32 32, %ShAmt
-  //   %shr = lshr i32 %ShVal1, %sub
-  //   %shl = shl i32 %ShVal0, %ShAmt
-  //   %fsh = or i32 %shr, %shl
+  //   %shr = lshr i32 %RotSrc, %sub
+  //   %shl = shl i32 %RotSrc, %ShAmt
+  //   %rot = or i32 %shr, %shl
   //   br label %PhiBB
   // PhiBB:
-  //   %cond = phi i32 [ %fsh, %FunnelBB ], [ %ShVal0, %GuardBB ]
+  //   %cond = phi i32 [ %RotSrc, %FunnelBB ], [ %RotSrc, %GuardBB ]
   // -->
-  // llvm.fshl.i32(i32 %ShVal0, i32 %ShVal1, i32 %ShAmt)
+  // llvm.fshl.i32(i32 %RotSrc, i32 %RotSrc, i32 %ShAmt)
   IRBuilder<> Builder(PhiBB, PhiBB->getFirstInsertionPt());
   Function *F = Intrinsic::getDeclaration(Phi.getModule(), IID, Phi.getType());
   Phi.replaceAllUsesWith(Builder.CreateCall(F, {ShVal0, ShVal1, ShAmt}));
-  if (ShVal0 == ShVal1)
-    ++NumGuardedRotates;
-  else
-    ++NumGuardedFunnelShifts;
+  ++NumGuardedRotates;
   return true;
 }
 
@@ -361,7 +350,7 @@ static bool foldUnusualPatterns(Function &F, DominatorTree &DT) {
     // iteratively in this loop rather than waiting until the end.
     for (Instruction &I : make_range(BB.rbegin(), BB.rend())) {
       MadeChange |= foldAnyOrAllBitsSet(I);
-      MadeChange |= foldGuardedFunnelShift(I);
+      MadeChange |= foldGuardedRotateToFunnelShift(I);
       MadeChange |= tryToRecognizePopCount(I); 
     }
   }

llvm/test/Analysis/MustExecute/const-cond.ll

@@ -1,46 +1,47 @@
-; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: opt -disable-output -print-mustexecute %s 2>&1 | FileCheck %s
-
-; In general the CFG below is easily simplified but this is useful for
-; pass ordering issue elimination.
-define i1 @const_cond(i32 %high) {
-; CHECK-LABEL: @const_cond(
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    br label [[LOOP:%.*]]
-; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[BACKEDGE:%.*]] ] ; (mustexec in: loop)
-; CHECK-NEXT:    br i1 true, label [[NEXT:%.*]], label [[NEVER1:%.*]] ; (mustexec in: loop)
-; CHECK:       next:
-; CHECK-NEXT:    br i1 false, label [[NEVER2:%.*]], label [[BACKEDGE]] ; (mustexec in: loop)
-; CHECK:       backedge:
-; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1 ; (mustexec in: loop)
-; CHECK-NEXT:    [[EXIT_TEST:%.*]] = icmp slt i32 [[IV]], [[HIGH:%.*]] ; (mustexec in: loop)
-; CHECK-NEXT:    br i1 [[EXIT_TEST]], label [[LOOP]], label [[EXIT:%.*]] ; (mustexec in: loop)
-; CHECK:       exit:
-; CHECK-NEXT:    ret i1 false
-; CHECK:       never1:
-; CHECK-NEXT:    unreachable
-; CHECK:       never2:
-; CHECK-NEXT:    unreachable
-;
-entry:
-  br label %loop
-
-loop:
-  %iv = phi i32 [0, %entry], [%iv.next, %backedge]
-  br i1 true, label %next, label %never1
-next:
-  br i1 false, label %never2, label %backedge
-backedge:
-  %iv.next = add nsw nuw i32 %iv, 1
-  %exit.test = icmp slt i32 %iv, %high
-  br i1 %exit.test, label %loop, label %exit
-
-exit:
-  ret i1 false
-never1:
-  unreachable
-never2:
-  unreachable
-}
-
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -disable-output -print-mustexecute %s 2>&1 | FileCheck %s
+; RUN: opt -disable-output -passes=print-mustexecute %s 2>&1 | FileCheck %s
+
+; In general the CFG below is easily simplified but this is useful for
+; pass ordering issue elimination.
+define i1 @const_cond(i32 %high) {
+; CHECK-LABEL: @const_cond(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[BACKEDGE:%.*]] ] ; (mustexec in: loop)
+; CHECK-NEXT:    br i1 true, label [[NEXT:%.*]], label [[NEVER1:%.*]] ; (mustexec in: loop)
+; CHECK:       next:
+; CHECK-NEXT:    br i1 false, label [[NEVER2:%.*]], label [[BACKEDGE]] ; (mustexec in: loop)
+; CHECK:       backedge:
+; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1 ; (mustexec in: loop)
+; CHECK-NEXT:    [[EXIT_TEST:%.*]] = icmp slt i32 [[IV]], [[HIGH:%.*]] ; (mustexec in: loop)
+; CHECK-NEXT:    br i1 [[EXIT_TEST]], label [[LOOP]], label [[EXIT:%.*]] ; (mustexec in: loop)
+; CHECK:       exit:
+; CHECK-NEXT:    ret i1 false
+; CHECK:       never1:
+; CHECK-NEXT:    unreachable
+; CHECK:       never2:
+; CHECK-NEXT:    unreachable
+;
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i32 [0, %entry], [%iv.next, %backedge]
+  br i1 true, label %next, label %never1
+next:
+  br i1 false, label %never2, label %backedge
+backedge:
+  %iv.next = add nsw nuw i32 %iv, 1
+  %exit.test = icmp slt i32 %iv, %high
+  br i1 %exit.test, label %loop, label %exit
+
+exit:
+  ret i1 false
+never1:
+  unreachable
+never2:
+  unreachable
+}
+