Misc optimizer+codegen work for 'cmpxchg' and 'atomicrmw'. They appear to be

working on x86 (at least for trivial testcases); other architectures will
need more work so that they actually emit the appropriate instructions for
orderings stricter than 'monotonic'. (As far as I can tell, the ARM, PPC,
Mips, and Alpha backends need such changes.)



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@136457 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 90 insertions, 16 deletions

diff --git a/lib/Transforms/IPO/MergeFunctions.cpp b/lib/Transforms/IPO/MergeFunctions.cpp
index f3d7e46ad2..e46375b955 100644
--- a/lib/Transforms/IPO/MergeFunctions.cpp
+++ b/lib/Transforms/IPO/MergeFunctions.cpp
@@ -317,22 +317,22 @@ bool FunctionComparator::isEquivalentOperation(const Instruction *I1,
   if (const InvokeInst *CI = dyn_cast<InvokeInst>(I1))
     return CI->getCallingConv() == cast<InvokeInst>(I2)->getCallingConv() &&
            CI->getAttributes() == cast<InvokeInst>(I2)->getAttributes();
-  if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(I1)) {
-    if (IVI->getNumIndices() != cast<InsertValueInst>(I2)->getNumIndices())
-      return false;
-    for (unsigned i = 0, e = IVI->getNumIndices(); i != e; ++i)
-      if (IVI->idx_begin()[i] != cast<InsertValueInst>(I2)->idx_begin()[i])
-        return false;
-    return true;
-  }
-  if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(I1)) {
-    if (EVI->getNumIndices() != cast<ExtractValueInst>(I2)->getNumIndices())
-      return false;
-    for (unsigned i = 0, e = EVI->getNumIndices(); i != e; ++i)
-      if (EVI->idx_begin()[i] != cast<ExtractValueInst>(I2)->idx_begin()[i])
-        return false;
-    return true;
-  }
+  if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(I1))
+    return IVI->getIndices() == cast<InsertValueInst>(I2)->getIndices();
+  if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(I1))
+    return EVI->getIndices() == cast<ExtractValueInst>(I2)->getIndices();
+  if (const FenceInst *FI = dyn_cast<FenceInst>(I1))
+    return FI->getOrdering() == cast<FenceInst>(I2)->getOrdering() &&
+           FI->getSynchScope() == cast<FenceInst>(I2)->getSynchScope();
+  if (const AtomicCmpXchgInst *CXI = dyn_cast<AtomicCmpXchgInst>(I1))
+    return CXI->isVolatile() == cast<AtomicCmpXchgInst>(I2)->isVolatile() &&
+           CXI->getOrdering() == cast<AtomicCmpXchgInst>(I2)->getOrdering() &&
+           CXI->getSynchScope() == cast<AtomicCmpXchgInst>(I2)->getSynchScope();
+  if (const AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(I1))
+    return RMWI->getOperation() == cast<AtomicRMWInst>(I2)->getOperation() &&
+           RMWI->isVolatile() == cast<AtomicRMWInst>(I2)->isVolatile() &&
+           RMWI->getOrdering() == cast<AtomicRMWInst>(I2)->getOrdering() &&
+           RMWI->getSynchScope() == cast<AtomicRMWInst>(I2)->getSynchScope();
 
   return true;
 }
diff --git a/lib/Transforms/Scalar/LowerAtomic.cpp b/lib/Transforms/Scalar/LowerAtomic.cpp
index 0954f77f3b..5f5ed03bd8 100644
--- a/lib/Transforms/Scalar/LowerAtomic.cpp
+++ b/lib/Transforms/Scalar/LowerAtomic.cpp
@@ -115,6 +115,76 @@ static bool LowerAtomicIntrinsic(IntrinsicInst *II) {
   return true;
 }
 
+bool LowerAtomicCmpXchgInst(AtomicCmpXchgInst *CXI) {
+  IRBuilder<> Builder(CXI->getParent(), CXI);
+  Value *Ptr = CXI->getPointerOperand();
+  Value *Cmp = CXI->getCompareOperand();
+  Value *Val = CXI->getNewValOperand();
+ 
+  LoadInst *Orig = Builder.CreateLoad(Ptr);
+  Value *Equal = Builder.CreateICmpEQ(Orig, Cmp);
+  Value *Res = Builder.CreateSelect(Equal, Val, Orig);
+  Builder.CreateStore(Res, Ptr);
+ 
+  CXI->replaceAllUsesWith(Orig);
+  CXI->eraseFromParent();
+  return true;
+}
+
+bool LowerAtomicRMWInst(AtomicRMWInst *RMWI) {
+  IRBuilder<> Builder(RMWI->getParent(), RMWI);
+  Value *Ptr = RMWI->getPointerOperand();
+  Value *Val = RMWI->getValOperand();
+
+  LoadInst *Orig = Builder.CreateLoad(Ptr);
+  Value *Res = NULL;
+
+  switch (RMWI->getOperation()) {
+  default: llvm_unreachable("Unexpected RMW operation");
+  case AtomicRMWInst::Xchg:
+    Res = Val;
+    break;
+  case AtomicRMWInst::Add:
+    Res = Builder.CreateAdd(Orig, Val);
+    break;
+  case AtomicRMWInst::Sub:
+    Res = Builder.CreateSub(Orig, Val);
+    break;
+  case AtomicRMWInst::And:
+    Res = Builder.CreateAnd(Orig, Val);
+    break;
+  case AtomicRMWInst::Nand:
+    Res = Builder.CreateNot(Builder.CreateAnd(Orig, Val));
+    break;
+  case AtomicRMWInst::Or:
+    Res = Builder.CreateOr(Orig, Val);
+    break;
+  case AtomicRMWInst::Xor:
+    Res = Builder.CreateXor(Orig, Val);
+    break;
+  case AtomicRMWInst::Max:
+    Res = Builder.CreateSelect(Builder.CreateICmpSLT(Orig, Val),
+                               Val, Orig);
+    break;
+  case AtomicRMWInst::Min:
+    Res = Builder.CreateSelect(Builder.CreateICmpSLT(Orig, Val),
+                               Orig, Val);
+    break;
+  case AtomicRMWInst::UMax:
+    Res = Builder.CreateSelect(Builder.CreateICmpULT(Orig, Val),
+                               Val, Orig);
+    break;
+  case AtomicRMWInst::UMin:
+    Res = Builder.CreateSelect(Builder.CreateICmpULT(Orig, Val),
+                               Orig, Val);
+    break;
+  }
+  Builder.CreateStore(Res, Ptr);
+  RMWI->replaceAllUsesWith(Orig);
+  RMWI->eraseFromParent();
+  return true;
+}
+
 static bool LowerFenceInst(FenceInst *FI) {
   FI->eraseFromParent();
   return true;
@@ -134,6 +204,10 @@ namespace {
           Changed |= LowerAtomicIntrinsic(II);
         else if (FenceInst *FI = dyn_cast<FenceInst>(Inst))
           Changed |= LowerFenceInst(FI);
+        else if (AtomicCmpXchgInst *CXI = dyn_cast<AtomicCmpXchgInst>(Inst))
+          Changed |= LowerAtomicCmpXchgInst(CXI);
+        else if (AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(Inst))
+          Changed |= LowerAtomicRMWInst(RMWI);
       }
       return Changed;
     }