Implement the sqrt intrinsic

Lower this to LLVM's sqrt intrinsic, which on x64 turns into an SSE2 sqrt instruction.

include/Reader/reader.h

@@ -2894,7 +2894,13 @@ public:
   virtual void storeStaticField(CORINFO_RESOLVED_TOKEN *FieldToken,
                                 IRNode *ValueToStore, bool IsVolatile) = 0;
   virtual IRNode *stringGetChar(IRNode *Arg1, IRNode *Arg2) = 0;
-  virtual bool sqrt(IRNode *Arg1, IRNode **RetVal) = 0;
+
+  /// Optionally generate inline code for the \p sqrt operation
+  ///
+  /// \param Argument      input value for sqrt
+  /// \param Result [out]  resulting sqrt value, iff reader decided to expand
+  /// \returns             true iff Result represents the sqrt
+  virtual bool sqrt(IRNode *Argument, IRNode **Result) = 0;
 
   virtual bool interlockedIntrinsicBinOp(IRNode *Arg1, IRNode *Arg2,
                                          IRNode **RetVal,

include/Reader/readerir.h

@@ -458,10 +458,7 @@ public:
                         IRNode *ValueToStore, bool IsVolatile) override;
 
   IRNode *stringGetChar(IRNode *Arg1, IRNode *Arg2) override;
-
+  bool sqrt(IRNode *Argument, IRNode **Result) override;
-  bool sqrt(IRNode *Arg1, IRNode **RetVal) override {
-    throw NotYetImplementedException("sqrt");
-  };
 
   // The callTarget node is only required on IA64.
   bool interlockedIntrinsicBinOp(IRNode *Arg1, IRNode *Arg2, IRNode **RetVal,

lib/Reader/readerir.cpp

@@ -5479,6 +5479,22 @@ bool GenIR::abs(IRNode *Argument, IRNode **Result) {
   return false;
 }
 
+bool GenIR::sqrt(IRNode *Argument, IRNode **Result) {
+  Type *Ty = Argument->getType();
+
+  if (Ty->isFloatingPointTy()) {
+    Type *Types[] = {Ty};
+    Value *FSqrt = Intrinsic::getDeclaration(JitContext->CurrentModule,
+                                             Intrinsic::sqrt, Types);
+    bool MayThrow = false;
+    Value *Sqrt = makeCall(FSqrt, MayThrow, Argument).getInstruction();
+    *Result = (IRNode *)Sqrt;
+    return true;
+  }
+
+  return false;
+}
+
 IRNode *GenIR::localAlloc(IRNode *Arg, bool ZeroInit) {
   // Note that we've seen a localloc in this method, since it has repercussions
   // on other aspects of code generation.