reapply Sanjiv's patch to genericize memcpy/memset/memmove to take an

arbitrary integer width for the count.

llvm-svn: 59823

llvm/docs/LangRef.html

@@ -5070,7 +5070,13 @@ for more efficient code generation.
 <div class="doc_text">
 
 <h5>Syntax:</h5>
+<p>This is an overloaded intrinsic. You can use llvm.memcpy on any integer bit
+width. Not all targets support all bit widths however.</p>
 <pre>
+  declare void @llvm.memcpy.i8(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
+                                i8 &lt;len&gt;, i32 &lt;align&gt;)
+  declare void @llvm.memcpy.i16(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
+                                i16 &lt;len&gt;, i32 &lt;align&gt;)
   declare void @llvm.memcpy.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
                                 i32 &lt;len&gt;, i32 &lt;align&gt;)
   declare void @llvm.memcpy.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
@@ -5124,7 +5130,13 @@ be set to 0 or 1.
 <div class="doc_text">
 
 <h5>Syntax:</h5>
+<p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
+width. Not all targets support all bit widths however.</p>
 <pre>
+  declare void @llvm.memmove.i8(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
+                                 i8 &lt;len&gt;, i32 &lt;align&gt;)
+  declare void @llvm.memmove.i16(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
+                                 i16 &lt;len&gt;, i32 &lt;align&gt;)
   declare void @llvm.memmove.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
                                  i32 &lt;len&gt;, i32 &lt;align&gt;)
   declare void @llvm.memmove.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
@@ -5179,7 +5191,13 @@ be set to 0 or 1.
 <div class="doc_text">
 
 <h5>Syntax:</h5>
+<p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
+width. Not all targets support all bit widths however.</p>
 <pre>
+  declare void @llvm.memset.i8(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
+                                i8 &lt;len&gt;, i32 &lt;align&gt;)
+  declare void @llvm.memset.i16(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
+                                i16 &lt;len&gt;, i32 &lt;align&gt;)
   declare void @llvm.memset.i32(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
                                 i32 &lt;len&gt;, i32 &lt;align&gt;)
   declare void @llvm.memset.i64(i8 * &lt;dest&gt;, i8 &lt;val&gt;,

llvm/examples/BrainF/BrainF.cpp

@@ -53,7 +53,9 @@ void BrainF::header() {
   //Function prototypes
 
   //declare void @llvm.memset.i32(i8 *, i8, i32, i32)
-  Function *memset_func = Intrinsic::getDeclaration(module, Intrinsic::memset_i32);
+  const Type *Tys[] = { Type::Int32Ty };
+  Function *memset_func = Intrinsic::getDeclaration(module, Intrinsic::memset,
+                                                    Tys, 1);
 
   //declare i32 @getchar()
   getchar_func = cast<Function>(module->

llvm/include/llvm/IntrinsicInst.h

@@ -208,12 +208,9 @@ namespace llvm {
     static inline bool classof(const MemIntrinsic *) { return true; }
     static inline bool classof(const IntrinsicInst *I) {
       switch (I->getIntrinsicID()) {
-      case Intrinsic::memcpy_i32:
+      case Intrinsic::memcpy:
-      case Intrinsic::memcpy_i64:
+      case Intrinsic::memmove:
-      case Intrinsic::memmove_i32:
+      case Intrinsic::memset:
-      case Intrinsic::memmove_i64:
-      case Intrinsic::memset_i32:
-      case Intrinsic::memset_i64:
         return true;
       default: return false;
       }
@@ -246,8 +243,7 @@ namespace llvm {
     // Methods for support type inquiry through isa, cast, and dyn_cast:
     static inline bool classof(const MemCpyInst *) { return true; }
     static inline bool classof(const IntrinsicInst *I) {
-      return I->getIntrinsicID() == Intrinsic::memcpy_i32 ||
+      return I->getIntrinsicID() == Intrinsic::memcpy;
-             I->getIntrinsicID() == Intrinsic::memcpy_i64;
     }
     static inline bool classof(const Value *V) {
       return isa<IntrinsicInst>(V) && classof(cast<IntrinsicInst>(V));
@@ -275,8 +271,7 @@ namespace llvm {
     // Methods for support type inquiry through isa, cast, and dyn_cast:
     static inline bool classof(const MemMoveInst *) { return true; }
     static inline bool classof(const IntrinsicInst *I) {
-      return I->getIntrinsicID() == Intrinsic::memmove_i32 ||
+      return I->getIntrinsicID() == Intrinsic::memmove;
-             I->getIntrinsicID() == Intrinsic::memmove_i64;
     }
     static inline bool classof(const Value *V) {
       return isa<IntrinsicInst>(V) && classof(cast<IntrinsicInst>(V));
@@ -299,8 +294,7 @@ namespace llvm {
     // Methods for support type inquiry through isa, cast, and dyn_cast:
     static inline bool classof(const MemSetInst *) { return true; }
     static inline bool classof(const IntrinsicInst *I) {
-      return I->getIntrinsicID() == Intrinsic::memset_i32 ||
+      return I->getIntrinsicID() == Intrinsic::memset;
-             I->getIntrinsicID() == Intrinsic::memset_i64;
     }
     static inline bool classof(const Value *V) {
       return isa<IntrinsicInst>(V) && classof(cast<IntrinsicInst>(V));

llvm/include/llvm/Intrinsics.td

@@ -193,33 +193,15 @@ def int_stackprotector : Intrinsic<[llvm_void_ty],
 //
 
 let Properties = [IntrWriteArgMem] in {
-  def int_memcpy_i16  : Intrinsic<[llvm_void_ty],
+  def int_memcpy  : Intrinsic<[llvm_void_ty],
+                               [llvm_ptr_ty, llvm_ptr_ty,
+                                llvm_anyint_ty, llvm_i32_ty]>;
+  def int_memmove : Intrinsic<[llvm_void_ty],
                                   [llvm_ptr_ty, llvm_ptr_ty,
-                                   llvm_i16_ty, llvm_i16_ty]>;
+                                   llvm_anyint_ty, llvm_i32_ty]>;
-  def int_memcpy_i32  : Intrinsic<[llvm_void_ty],
+  def int_memset  : Intrinsic<[llvm_void_ty],
-                                  [llvm_ptr_ty, llvm_ptr_ty,
+                               [llvm_ptr_ty, llvm_i8_ty,
-                                   llvm_i32_ty, llvm_i32_ty]>;
+                                llvm_anyint_ty, llvm_i32_ty]>;
-  def int_memcpy_i64  : Intrinsic<[llvm_void_ty],
-                                  [llvm_ptr_ty, llvm_ptr_ty,
-                                   llvm_i64_ty, llvm_i32_ty]>;
-  def int_memmove_i16 : Intrinsic<[llvm_void_ty],
-                                  [llvm_ptr_ty, llvm_ptr_ty,
-                                   llvm_i16_ty, llvm_i16_ty]>;
-  def int_memmove_i32 : Intrinsic<[llvm_void_ty],
-                                  [llvm_ptr_ty, llvm_ptr_ty,
-                                   llvm_i32_ty, llvm_i32_ty]>;
-  def int_memmove_i64 : Intrinsic<[llvm_void_ty],
-                                  [llvm_ptr_ty, llvm_ptr_ty,
-                                   llvm_i64_ty, llvm_i32_ty]>;
-  def int_memset_i16  : Intrinsic<[llvm_void_ty],
-                                  [llvm_ptr_ty, llvm_i8_ty,
-                                   llvm_i16_ty, llvm_i16_ty]>;
-  def int_memset_i32  : Intrinsic<[llvm_void_ty],
-                                  [llvm_ptr_ty, llvm_i8_ty,
-                                   llvm_i32_ty, llvm_i32_ty]>;
-  def int_memset_i64  : Intrinsic<[llvm_void_ty],
-                                  [llvm_ptr_ty, llvm_i8_ty,
-                                   llvm_i64_ty, llvm_i32_ty]>;
 }
 
 // These functions do not actually read memory, but they are sensitive to the

llvm/lib/Analysis/IPA/Andersens.cpp

@@ -903,8 +903,7 @@ bool Andersens::AddConstraintsForExternalCall(CallSite CS, Function *F) {
       F->getName() == "atol" || F->getName() == "atoll" ||
       F->getName() == "remove" || F->getName() == "unlink" ||
       F->getName() == "rename" || F->getName() == "memcmp" ||
-      F->getName() == "llvm.memset.i32" ||
+      F->getName() == "llvm.memset" ||
-      F->getName() == "llvm.memset.i64" ||
       F->getName() == "strcmp" || F->getName() == "strncmp" ||
       F->getName() == "execl" || F->getName() == "execlp" ||
       F->getName() == "execle" || F->getName() == "execv" ||
@@ -942,8 +941,8 @@ bool Andersens::AddConstraintsForExternalCall(CallSite CS, Function *F) {
 
 
   // These functions do induce points-to edges.
-  if (F->getName() == "llvm.memcpy.i32" || F->getName() == "llvm.memcpy.i64" ||
+  if (F->getName() == "llvm.memcpy" ||
-      F->getName() == "llvm.memmove.i32" ||F->getName() == "llvm.memmove.i64" ||
+      F->getName() == "llvm.memmove" ||
       F->getName() == "memmove") {
 
     // *Dest = *Src, which requires an artificial graph node to represent the

llvm/lib/CodeGen/IntrinsicLowering.cpp

@@ -98,22 +98,19 @@ void IntrinsicLowering::AddPrototypes(Module &M) {
         EnsureFunctionExists(M, "abort", I->arg_end(), I->arg_end(),
                              Type::VoidTy);
         break;
-      case Intrinsic::memcpy_i32:
+      case Intrinsic::memcpy:
-      case Intrinsic::memcpy_i64:
         M.getOrInsertFunction("memcpy", PointerType::getUnqual(Type::Int8Ty),
                               PointerType::getUnqual(Type::Int8Ty), 
                               PointerType::getUnqual(Type::Int8Ty), 
                               TD.getIntPtrType(), (Type *)0);
         break;
-      case Intrinsic::memmove_i32:
+      case Intrinsic::memmove:
-      case Intrinsic::memmove_i64:
         M.getOrInsertFunction("memmove", PointerType::getUnqual(Type::Int8Ty),
                               PointerType::getUnqual(Type::Int8Ty), 
                               PointerType::getUnqual(Type::Int8Ty), 
                               TD.getIntPtrType(), (Type *)0);
         break;
-      case Intrinsic::memset_i32:
+      case Intrinsic::memset:
-      case Intrinsic::memset_i64:
         M.getOrInsertFunction("memset", PointerType::getUnqual(Type::Int8Ty),
                               PointerType::getUnqual(Type::Int8Ty), 
                               Type::Int32Ty, 
@@ -784,8 +781,7 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
   case Intrinsic::var_annotation:
     break;   // Strip out annotate intrinsic
     
-  case Intrinsic::memcpy_i32:
+  case Intrinsic::memcpy: {
-  case Intrinsic::memcpy_i64: {
     static Constant *MemcpyFCache = 0;
     Value *Size = CI->getOperand(3);
     const Type *IntPtr = TD.getIntPtrType();
@@ -803,8 +799,7 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
                     MemcpyFCache);
     break;
   }
-  case Intrinsic::memmove_i32: 
+  case Intrinsic::memmove: {
-  case Intrinsic::memmove_i64: {
     static Constant *MemmoveFCache = 0;
     Value *Size = CI->getOperand(3);
     const Type *IntPtr = TD.getIntPtrType();
@@ -822,8 +817,7 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
                     MemmoveFCache);
     break;
   }
-  case Intrinsic::memset_i32:
+  case Intrinsic::memset: {
-  case Intrinsic::memset_i64: {
     static Constant *MemsetFCache = 0;
     Value *Size = CI->getOperand(3);
     const Type *IntPtr = TD.getIntPtrType();

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp

@@ -3663,8 +3663,7 @@ SelectionDAGLowering::visitIntrinsicCall(CallInst &I, unsigned Intrinsic) {
   case Intrinsic::longjmp:
     return "_longjmp"+!TLI.usesUnderscoreLongJmp();
     break;
-  case Intrinsic::memcpy_i32:
+  case Intrinsic::memcpy: {
-  case Intrinsic::memcpy_i64: {
     SDValue Op1 = getValue(I.getOperand(1));
     SDValue Op2 = getValue(I.getOperand(2));
     SDValue Op3 = getValue(I.getOperand(3));
@@ -3673,8 +3672,7 @@ SelectionDAGLowering::visitIntrinsicCall(CallInst &I, unsigned Intrinsic) {
                               I.getOperand(1), 0, I.getOperand(2), 0));
     return 0;
   }
-  case Intrinsic::memset_i32:
+  case Intrinsic::memset: {
-  case Intrinsic::memset_i64: {
     SDValue Op1 = getValue(I.getOperand(1));
     SDValue Op2 = getValue(I.getOperand(2));
     SDValue Op3 = getValue(I.getOperand(3));
@@ -3683,8 +3681,7 @@ SelectionDAGLowering::visitIntrinsicCall(CallInst &I, unsigned Intrinsic) {
                               I.getOperand(1), 0));
     return 0;
   }
-  case Intrinsic::memmove_i32:
+  case Intrinsic::memmove: {
-  case Intrinsic::memmove_i64: {
     SDValue Op1 = getValue(I.getOperand(1));
     SDValue Op2 = getValue(I.getOperand(2));
     SDValue Op3 = getValue(I.getOperand(3));

llvm/lib/Transforms/Scalar/InstructionCombining.cpp

@@ -9200,12 +9200,11 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
       if (GlobalVariable *GVSrc = dyn_cast<GlobalVariable>(MMI->getSource()))
         if (GVSrc->isConstant()) {
           Module *M = CI.getParent()->getParent()->getParent();
-          Intrinsic::ID MemCpyID;
+          Intrinsic::ID MemCpyID = Intrinsic::memcpy;
-          if (CI.getOperand(3)->getType() == Type::Int32Ty)
+          const Type *Tys[1];
-            MemCpyID = Intrinsic::memcpy_i32;
+          Tys[0] = CI.getOperand(3)->getType();
-          else
+          CI.setOperand(0, 
-            MemCpyID = Intrinsic::memcpy_i64;
+                        Intrinsic::getDeclaration(M, MemCpyID, Tys, 1));
-          CI.setOperand(0, Intrinsic::getDeclaration(M, MemCpyID));
           Changed = true;
         }
 

llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp

@@ -427,9 +427,12 @@ bool MemCpyOpt::processStore(StoreInst *SI, BasicBlock::iterator& BBI) {
     // instruction needed by the start of the block.
     BasicBlock::iterator InsertPt = BI;
   
-    if (MemSetF == 0)
+    if (MemSetF == 0) {
+      const Type *Tys[] = {Type::Int64Ty};
       MemSetF = Intrinsic::getDeclaration(SI->getParent()->getParent()
-                                          ->getParent(), Intrinsic::memset_i64);
+                                          ->getParent(), Intrinsic::memset,
+                                          Tys, 1);
+   }
     
     // Get the starting pointer of the block.
     StartPtr = Range.StartPtr;
@@ -671,9 +674,11 @@ bool MemCpyOpt::processMemCpy(MemCpyInst* M) {
     return false;
   
   // If all checks passed, then we can transform these memcpy's
+  const Type *Tys[1];
+  Tys[0] = M->getLength()->getType();
   Function* MemCpyFun = Intrinsic::getDeclaration(
                                  M->getParent()->getParent()->getParent(),
-                                 M->getIntrinsicID());
+                                 M->getIntrinsicID(), Tys, 1);
     
   std::vector<Value*> args;
   args.push_back(M->getRawDest());

llvm/lib/Transforms/Scalar/SimplifyLibCalls.cpp

@@ -130,9 +130,10 @@ Value *LibCallOptimization::EmitStrLen(Value *Ptr, IRBuilder<> &B) {
 Value *LibCallOptimization::EmitMemCpy(Value *Dst, Value *Src, Value *Len,
                                        unsigned Align, IRBuilder<> &B) {
   Module *M = Caller->getParent();
-  Intrinsic::ID IID = Len->getType() == Type::Int32Ty ?
+  Intrinsic::ID IID = Intrinsic::memcpy;
-                           Intrinsic::memcpy_i32 : Intrinsic::memcpy_i64;
+  const Type *Tys[1];
-  Value *MemCpy = Intrinsic::getDeclaration(M, IID);
+  Tys[0] = Len->getType();
+  Value *MemCpy = Intrinsic::getDeclaration(M, IID, Tys, 1);
   return B.CreateCall4(MemCpy, CastToCStr(Dst, B), CastToCStr(Src, B), Len,
                        ConstantInt::get(Type::Int32Ty, Align));
 }

llvm/lib/Transforms/Utils/InlineFunction.cpp

@@ -257,8 +257,10 @@ bool llvm::InlineFunction(CallSite CS, CallGraph *CG, const TargetData *TD) {
         Value *NewAlloca = new AllocaInst(AggTy, 0, Align, I->getName(),
                                           Caller->begin()->begin());
         // Emit a memcpy.
+        const Type *Tys[] = { Type::Int64Ty };
         Function *MemCpyFn = Intrinsic::getDeclaration(Caller->getParent(),
-                                                       Intrinsic::memcpy_i64);
+                                                       Intrinsic::memcpy, 
+                                                       Tys, 1);
         Value *DestCast = new BitCastInst(NewAlloca, VoidPtrTy, "tmp", TheCall);
         Value *SrcCast = new BitCastInst(*AI, VoidPtrTy, "tmp", TheCall);
 

llvm/lib/VMCore/Verifier.cpp

@@ -1336,12 +1336,9 @@ void Verifier::visitIntrinsicFunctionCall(Intrinsic::ID ID, CallInst &CI) {
   switch (ID) {
   default:
     break;
-  case Intrinsic::memcpy_i32:
+  case Intrinsic::memcpy:
-  case Intrinsic::memcpy_i64:
+  case Intrinsic::memmove:
-  case Intrinsic::memmove_i32:
+  case Intrinsic::memset:
-  case Intrinsic::memmove_i64:
-  case Intrinsic::memset_i32:
-  case Intrinsic::memset_i64:
     Assert1(isa<ConstantInt>(CI.getOperand(4)),
             "alignment argument of memory intrinsics must be a constant int",
             &CI);