vendor/llvm/llvm-trunk-r307894

2 files changed, 25 insertions, 11 deletions

diff --git a/lib/Target/NVPTX/NVPTXISelLowering.cpp b/lib/Target/NVPTX/NVPTXISelLowering.cpp
index f26b9a7cb8dd..f800d91f4093 100644
--- a/lib/Target/NVPTX/NVPTXISelLowering.cpp
+++ b/lib/Target/NVPTX/NVPTXISelLowering.cpp
@@ -62,7 +62,6 @@
 #include <utility>

 #include <vector>

 

-#undef DEBUG_TYPE

 #define DEBUG_TYPE "nvptx-lower"

 

 using namespace llvm;

@@ -2456,7 +2455,7 @@ SDValue NVPTXTargetLowering::LowerFormalArguments(
             // v2f16 was loaded as an i32. Now we must bitcast it back.

             else if (EltVT == MVT::v2f16)

               Elt = DAG.getNode(ISD::BITCAST, dl, MVT::v2f16, Elt);

-            // Extend the element if necesary (e.g. an i8 is loaded

+            // Extend the element if necessary (e.g. an i8 is loaded

             // into an i16 register)

             if (Ins[InsIdx].VT.isInteger() &&

                 Ins[InsIdx].VT.getSizeInBits() > LoadVT.getSizeInBits()) {

diff --git a/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp b/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
index 3be291b48b8f..989f0a3aba2f 100644
--- a/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
+++ b/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
@@ -14,6 +14,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTXLowerAggrCopies.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/CodeGen/StackProtector.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DataLayout.h"
@@ -42,6 +43,7 @@ struct NVPTXLowerAggrCopies : public FunctionPass {
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.addPreserved<StackProtector>();
+    AU.addRequired<TargetTransformInfoWrapperPass>();
   }
 
   bool runOnFunction(Function &F) override;
@@ -61,6 +63,8 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
 
   const DataLayout &DL = F.getParent()->getDataLayout();
   LLVMContext &Context = F.getParent()->getContext();
+  const TargetTransformInfo &TTI =
+      getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
 
   // Collect all aggregate loads and mem* calls.
   for (Function::iterator BI = F.begin(), BE = F.end(); BI != BE; ++BI) {
@@ -104,15 +108,26 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
     Value *SrcAddr = LI->getOperand(0);
     Value *DstAddr = SI->getOperand(1);
     unsigned NumLoads = DL.getTypeStoreSize(LI->getType());
-    Value *CopyLen = ConstantInt::get(Type::getInt32Ty(Context), NumLoads);
+    ConstantInt *CopyLen =
+        ConstantInt::get(Type::getInt32Ty(Context), NumLoads);
 
-    createMemCpyLoop(/* ConvertedInst */ SI,
-                     /* SrcAddr */ SrcAddr, /* DstAddr */ DstAddr,
-                     /* CopyLen */ CopyLen,
-                     /* SrcAlign */ LI->getAlignment(),
-                     /* DestAlign */ SI->getAlignment(),
-                     /* SrcIsVolatile */ LI->isVolatile(),
-                     /* DstIsVolatile */ SI->isVolatile());
+    if (!TTI.useWideIRMemcpyLoopLowering()) {
+      createMemCpyLoop(/* ConvertedInst */ SI,
+                       /* SrcAddr */ SrcAddr, /* DstAddr */ DstAddr,
+                       /* CopyLen */ CopyLen,
+                       /* SrcAlign */ LI->getAlignment(),
+                       /* DestAlign */ SI->getAlignment(),
+                       /* SrcIsVolatile */ LI->isVolatile(),
+                       /* DstIsVolatile */ SI->isVolatile());
+    } else {
+      createMemCpyLoopKnownSize(/* ConvertedInst */ SI,
+                                /* SrcAddr */ SrcAddr, /* DstAddr */ DstAddr,
+                                /* CopyLen */ CopyLen,
+                                /* SrcAlign */ LI->getAlignment(),
+                                /* DestAlign */ SI->getAlignment(),
+                                /* SrcIsVolatile */ LI->isVolatile(),
+                                /* DstIsVolatile */ SI->isVolatile(), TTI);
+    }
 
     SI->eraseFromParent();
     LI->eraseFromParent();
@@ -121,7 +136,7 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
   // Transform mem* intrinsic calls.
   for (MemIntrinsic *MemCall : MemCalls) {
     if (MemCpyInst *Memcpy = dyn_cast<MemCpyInst>(MemCall)) {
-      expandMemCpyAsLoop(Memcpy);
+      expandMemCpyAsLoop(Memcpy, TTI);
     } else if (MemMoveInst *Memmove = dyn_cast<MemMoveInst>(MemCall)) {
       expandMemMoveAsLoop(Memmove);
     } else if (MemSetInst *Memset = dyn_cast<MemSetInst>(MemCall)) {