1 files changed, 97 insertions, 51 deletions

diff --git a/contrib/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp b/contrib/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
index e623054..31ee920 100644
--- a/contrib/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
+++ b/contrib/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
@@ -14,16 +14,32 @@
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPU.h"
-#include "AMDGPUIntrinsicInfo.h"
 #include "AMDGPUSubtarget.h"
 #include "AMDGPUTargetMachine.h"
-
+#include "llvm/ADT/StringRef.h"
 #include "llvm/Analysis/DivergenceAnalysis.h"
 #include "llvm/CodeGen/Passes.h"
-#include "llvm/IR/InstVisitor.h"
+#include "llvm/CodeGen/TargetPassConfig.h"
+#include "llvm/IR/Attributes.h"
+#include "llvm/IR/BasicBlock.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
 #include "llvm/IR/IRBuilder.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/raw_ostream.h"
+#include "llvm/IR/InstVisitor.h"
+#include "llvm/IR/InstrTypes.h"
+#include "llvm/IR/Instruction.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Operator.h"
+#include "llvm/IR/Type.h"
+#include "llvm/IR/Value.h"
+#include "llvm/Pass.h"
+#include "llvm/Support/Casting.h"
+#include <cassert>
+#include <iterator>
 
 #define DEBUG_TYPE "amdgpu-codegenprepare"
 
@@ -33,18 +49,15 @@ namespace {
 
 class AMDGPUCodeGenPrepare : public FunctionPass,
                              public InstVisitor<AMDGPUCodeGenPrepare, bool> {
-  const GCNTargetMachine *TM;
-  const SISubtarget *ST;
-  DivergenceAnalysis *DA;
-  Module *Mod;
-  bool HasUnsafeFPMath;
+  const SISubtarget *ST = nullptr;
+  DivergenceAnalysis *DA = nullptr;
+  Module *Mod = nullptr;
+  bool HasUnsafeFPMath = false;
 
   /// \brief Copies exact/nsw/nuw flags (if any) from binary operation \p I to
   /// binary operation \p V.
   ///
   /// \returns Binary operation \p V.
-  Value *copyFlags(const BinaryOperator &I, Value *V) const;
-
   /// \returns \p T's base element bit width.
   unsigned getBaseElementBitWidth(const Type *T) const;
 
@@ -113,13 +126,8 @@ class AMDGPUCodeGenPrepare : public FunctionPass,
 
 public:
   static char ID;
-  AMDGPUCodeGenPrepare(const TargetMachine *TM = nullptr) :
-    FunctionPass(ID),
-    TM(static_cast<const GCNTargetMachine *>(TM)),
-    ST(nullptr),
-    DA(nullptr),
-    Mod(nullptr),
-    HasUnsafeFPMath(false) { }
+
+  AMDGPUCodeGenPrepare() : FunctionPass(ID) {}
 
   bool visitFDiv(BinaryOperator &I);
 
@@ -142,22 +150,7 @@ public:
  }
 };
 
-} // End anonymous namespace
-
-Value *AMDGPUCodeGenPrepare::copyFlags(
-    const BinaryOperator &I, Value *V) const {
-  BinaryOperator *BinOp = dyn_cast<BinaryOperator>(V);
-  if (!BinOp) // Possibly constant expression.
-    return V;
-
-  if (isa<OverflowingBinaryOperator>(BinOp)) {
-    BinOp->setHasNoSignedWrap(I.hasNoSignedWrap());
-    BinOp->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
-  } else if (isa<PossiblyExactOperator>(BinOp))
-    BinOp->setIsExact(I.isExact());
-
-  return V;
-}
+} // end anonymous namespace
 
 unsigned AMDGPUCodeGenPrepare::getBaseElementBitWidth(const Type *T) const {
   assert(needsPromotionToI32(T) && "T does not need promotion to i32");
@@ -186,12 +179,48 @@ bool AMDGPUCodeGenPrepare::isSigned(const SelectInst &I) const {
 }
 
 bool AMDGPUCodeGenPrepare::needsPromotionToI32(const Type *T) const {
-  if (T->isIntegerTy() && T->getIntegerBitWidth() > 1 &&
-      T->getIntegerBitWidth() <= 16)
+  const IntegerType *IntTy = dyn_cast<IntegerType>(T);
+  if (IntTy && IntTy->getBitWidth() > 1 && IntTy->getBitWidth() <= 16)
+    return true;
+
+  if (const VectorType *VT = dyn_cast<VectorType>(T)) {
+    // TODO: The set of packed operations is more limited, so may want to
+    // promote some anyway.
+    if (ST->hasVOP3PInsts())
+      return false;
+
+    return needsPromotionToI32(VT->getElementType());
+  }
+
+  return false;
+}
+
+// Return true if the op promoted to i32 should have nsw set.
+static bool promotedOpIsNSW(const Instruction &I) {
+  switch (I.getOpcode()) {
+  case Instruction::Shl:
+  case Instruction::Add:
+  case Instruction::Sub:
     return true;
-  if (!T->isVectorTy())
+  case Instruction::Mul:
+    return I.hasNoUnsignedWrap();
+  default:
+    return false;
+  }
+}
+
+// Return true if the op promoted to i32 should have nuw set.
+static bool promotedOpIsNUW(const Instruction &I) {
+  switch (I.getOpcode()) {
+  case Instruction::Shl:
+  case Instruction::Add:
+  case Instruction::Mul:
+    return true;
+  case Instruction::Sub:
+    return I.hasNoUnsignedWrap();
+  default:
     return false;
-  return needsPromotionToI32(cast<VectorType>(T)->getElementType());
+  }
 }
 
 bool AMDGPUCodeGenPrepare::promoteUniformOpToI32(BinaryOperator &I) const {
@@ -218,7 +247,19 @@ bool AMDGPUCodeGenPrepare::promoteUniformOpToI32(BinaryOperator &I) const {
     ExtOp0 = Builder.CreateZExt(I.getOperand(0), I32Ty);
     ExtOp1 = Builder.CreateZExt(I.getOperand(1), I32Ty);
   }
-  ExtRes = copyFlags(I, Builder.CreateBinOp(I.getOpcode(), ExtOp0, ExtOp1));
+
+  ExtRes = Builder.CreateBinOp(I.getOpcode(), ExtOp0, ExtOp1);
+  if (Instruction *Inst = dyn_cast<Instruction>(ExtRes)) {
+    if (promotedOpIsNSW(cast<Instruction>(I)))
+      Inst->setHasNoSignedWrap();
+
+    if (promotedOpIsNUW(cast<Instruction>(I)))
+      Inst->setHasNoUnsignedWrap();
+
+    if (const auto *ExactOp = dyn_cast<PossiblyExactOperator>(&I))
+      Inst->setIsExact(ExactOp->isExact());
+  }
+
   TruncRes = Builder.CreateTrunc(ExtRes, I.getType());
 
   I.replaceAllUsesWith(TruncRes);
@@ -339,16 +380,16 @@ bool AMDGPUCodeGenPrepare::visitFDiv(BinaryOperator &FDiv) {
   FastMathFlags FMF = FPOp->getFastMathFlags();
   bool UnsafeDiv = HasUnsafeFPMath || FMF.unsafeAlgebra() ||
                                       FMF.allowReciprocal();
-  if (ST->hasFP32Denormals() && !UnsafeDiv)
+
+  // With UnsafeDiv node will be optimized to just rcp and mul.
+  if (ST->hasFP32Denormals() || UnsafeDiv)
     return false;
 
   IRBuilder<> Builder(FDiv.getParent(), std::next(FDiv.getIterator()), FPMath);
   Builder.setFastMathFlags(FMF);
   Builder.SetCurrentDebugLocation(FDiv.getDebugLoc());
 
-  const AMDGPUIntrinsicInfo *II = TM->getIntrinsicInfo();
-  Function *Decl
-    = II->getDeclaration(Mod, AMDGPUIntrinsic::amdgcn_fdiv_fast, {});
+  Function *Decl = Intrinsic::getDeclaration(Mod, Intrinsic::amdgcn_fdiv_fast);
 
   Value *Num = FDiv.getOperand(0);
   Value *Den = FDiv.getOperand(1);
@@ -447,10 +488,15 @@ bool AMDGPUCodeGenPrepare::doInitialization(Module &M) {
 }
 
 bool AMDGPUCodeGenPrepare::runOnFunction(Function &F) {
-  if (!TM || skipFunction(F))
+  if (skipFunction(F))
+    return false;
+
+  auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
+  if (!TPC)
     return false;
 
-  ST = &TM->getSubtarget<SISubtarget>(F);
+  const TargetMachine &TM = TPC->getTM<TargetMachine>();
+  ST = &TM.getSubtarget<SISubtarget>(F);
   DA = &getAnalysis<DivergenceAnalysis>();
   HasUnsafeFPMath = hasUnsafeFPMath(F);
 
@@ -467,14 +513,14 @@ bool AMDGPUCodeGenPrepare::runOnFunction(Function &F) {
   return MadeChange;
 }
 
-INITIALIZE_TM_PASS_BEGIN(AMDGPUCodeGenPrepare, DEBUG_TYPE,
+INITIALIZE_PASS_BEGIN(AMDGPUCodeGenPrepare, DEBUG_TYPE,
                       "AMDGPU IR optimizations", false, false)
 INITIALIZE_PASS_DEPENDENCY(DivergenceAnalysis)
-INITIALIZE_TM_PASS_END(AMDGPUCodeGenPrepare, DEBUG_TYPE,
-                       "AMDGPU IR optimizations", false, false)
+INITIALIZE_PASS_END(AMDGPUCodeGenPrepare, DEBUG_TYPE, "AMDGPU IR optimizations",
+                    false, false)
 
 char AMDGPUCodeGenPrepare::ID = 0;
 
-FunctionPass *llvm::createAMDGPUCodeGenPreparePass(const GCNTargetMachine *TM) {
-  return new AMDGPUCodeGenPrepare(TM);
+FunctionPass *llvm::createAMDGPUCodeGenPreparePass() {
+  return new AMDGPUCodeGenPrepare();
 }