1 files changed, 38 insertions, 11 deletions
diff --git a/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp b/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
index f0c3663..6ab0fad 100644
--- a/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
+++ b/contrib/llvm/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
@@ -12,6 +12,8 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTXLowerAggrCopies.h"
+#include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/CodeGen/StackProtector.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/Function.h"
@@ -22,10 +24,33 @@
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/LLVMContext.h"
 #include "llvm/IR/Module.h"
+#include "llvm/Support/Debug.h"
+
+#define DEBUG_TYPE "nvptx"
 
 using namespace llvm;
 
-namespace llvm { FunctionPass *createLowerAggrCopies(); }
+namespace {
+// actual analysis class, which is a functionpass
+struct NVPTXLowerAggrCopies : public FunctionPass {
+  static char ID;
+
+  NVPTXLowerAggrCopies() : FunctionPass(ID) {}
+
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.addPreserved<MachineFunctionAnalysis>();
+    AU.addPreserved<StackProtector>();
+  }
+
+  bool runOnFunction(Function &F) override;
+
+  static const unsigned MaxAggrCopySize = 128;
+
+  const char *getPassName() const override {
+    return "Lower aggregate copies/intrinsics into loops";
+  }
+};
+} // namespace
 
 char NVPTXLowerAggrCopies::ID = 0;
 
@@ -45,8 +70,8 @@ static void convertTransferToLoop(
 
   // srcAddr and dstAddr are expected to be pointer types,
   // so no check is made here.
-  unsigned srcAS = dyn_cast<PointerType>(srcAddr->getType())->getAddressSpace();
-  unsigned dstAS = dyn_cast<PointerType>(dstAddr->getType())->getAddressSpace();
+  unsigned srcAS = cast<PointerType>(srcAddr->getType())->getAddressSpace();
+  unsigned dstAS = cast<PointerType>(dstAddr->getType())->getAddressSpace();
 
   // Cast pointers to (char *)
   srcAddr = builder.CreateBitCast(srcAddr, Type::getInt8PtrTy(Context, srcAS));
@@ -59,9 +84,11 @@ static void convertTransferToLoop(
   ind->addIncoming(ConstantInt::get(indType, 0), origBB);
 
   // load from srcAddr+ind
-  Value *val = loop.CreateLoad(loop.CreateGEP(srcAddr, ind), srcVolatile);
+  Value *val = loop.CreateLoad(loop.CreateGEP(loop.getInt8Ty(), srcAddr, ind),
+                               srcVolatile);
   // store at dstAddr+ind
-  loop.CreateStore(val, loop.CreateGEP(dstAddr, ind), dstVolatile);
+  loop.CreateStore(val, loop.CreateGEP(loop.getInt8Ty(), dstAddr, ind),
+                   dstVolatile);
 
   // The value for ind coming from backedge is (ind + 1)
   Value *newind = loop.CreateAdd(ind, ConstantInt::get(indType, 1));
@@ -81,7 +108,7 @@ static void convertMemSetToLoop(Instruction *splitAt, Value *dstAddr,
   origBB->getTerminator()->setSuccessor(0, loopBB);
   IRBuilder<> builder(origBB, origBB->getTerminator());
 
-  unsigned dstAS = dyn_cast<PointerType>(dstAddr->getType())->getAddressSpace();
+  unsigned dstAS = cast<PointerType>(dstAddr->getType())->getAddressSpace();
 
   // Cast pointer to the type of value getting stored
   dstAddr =
@@ -91,7 +118,7 @@ static void convertMemSetToLoop(Instruction *splitAt, Value *dstAddr,
   PHINode *ind = loop.CreatePHI(len->getType(), 0);
   ind->addIncoming(ConstantInt::get(len->getType(), 0), origBB);
 
-  loop.CreateStore(val, loop.CreateGEP(dstAddr, ind), false);
+  loop.CreateStore(val, loop.CreateGEP(val->getType(), dstAddr, ind), false);
 
   Value *newind = loop.CreateAdd(ind, ConstantInt::get(len->getType(), 1));
   ind->addIncoming(newind, loopBB);
@@ -104,7 +131,7 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
   SmallVector<MemTransferInst *, 4> aggrMemcpys;
   SmallVector<MemSetInst *, 4> aggrMemsets;
 
-  const DataLayout *DL = &getAnalysis<DataLayoutPass>().getDataLayout();
+  const DataLayout &DL = F.getParent()->getDataLayout();
   LLVMContext &Context = F.getParent()->getContext();
 
   //
@@ -117,10 +144,10 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
          ++II) {
       if (LoadInst *load = dyn_cast<LoadInst>(II)) {
 
-        if (load->hasOneUse() == false)
+        if (!load->hasOneUse())
           continue;
 
-        if (DL->getTypeStoreSize(load->getType()) < MaxAggrCopySize)
+        if (DL.getTypeStoreSize(load->getType()) < MaxAggrCopySize)
           continue;
 
         User *use = load->user_back();
@@ -166,7 +193,7 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
     StoreInst *store = dyn_cast<StoreInst>(*load->user_begin());
     Value *srcAddr = load->getOperand(0);
     Value *dstAddr = store->getOperand(1);
-    unsigned numLoads = DL->getTypeStoreSize(load->getType());
+    unsigned numLoads = DL.getTypeStoreSize(load->getType());
     Value *len = ConstantInt::get(Type::getInt32Ty(Context), numLoads);
 
     convertTransferToLoop(store, srcAddr, dstAddr, len, load->isVolatile(),