diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp
index 61f1c213431..32540d2a3b2 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp
@@ -4667,6 +4667,7 @@ void SelectionDAGLowering::visitInlineAsm(CallSite CS) {
         OpInfo.CallOperandVal = CS.getArgument(ArgNo++);
         break;
       }
+        
       // The return value of the call is this value.  As such, there is no
       // corresponding argument.
       assert(CS.getType() != Type::VoidTy && "Bad inline asm!");
@@ -4699,6 +4700,30 @@ void SelectionDAGLowering::visitInlineAsm(CallSite CS) {
     }
     
     OpInfo.ConstraintVT = OpVT;
+  }
+  
+  // Second pass over the constraints: compute which constraint option to use
+  // and assign registers to constraints that want a specific physreg.
+  for (unsigned i = 0, e = ConstraintInfos.size(); i != e; ++i) {
+    SDISelAsmOperandInfo &OpInfo = ConstraintOperands[i];
+    
+    // If this is an output operand with a matching input operand, look up the
+    // matching input.  It might have a different type (e.g. the output might be
+    // i32 and the input i64) and we need to pick the larger width to ensure we
+    // reserve the right number of registers.  
+    if (OpInfo.hasMatchingInput()) {
+      SDISelAsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput];
+      if (OpInfo.ConstraintVT != Input.ConstraintVT) {
+        assert(OpInfo.ConstraintVT.isInteger() &&
+               Input.ConstraintVT.isInteger() &&
+               "Asm constraints must be the same or different sized integers");
+        if (OpInfo.ConstraintVT.getSizeInBits() < 
+            Input.ConstraintVT.getSizeInBits())
+          OpInfo.ConstraintVT = Input.ConstraintVT;
+        else
+          Input.ConstraintVT = OpInfo.ConstraintVT;
+      }
+    }
     
     // Compute the constraint code and ConstraintType to use.
     TLI.ComputeConstraintToUse(OpInfo, OpInfo.CallOperand, hasMemory, &DAG);
@@ -4948,22 +4973,28 @@ void SelectionDAGLowering::visitInlineAsm(CallSite CS) {
   // and set it as the value of the call.
   if (!RetValRegs.Regs.empty()) {
     SDValue Val = RetValRegs.getCopyFromRegs(DAG, Chain, &Flag);
+    
+    // FIXME: Why don't we do this for inline asms with MRVs?
+    if (CS.getType()->isSingleValueType() && CS.getType()->isSized()) {
+      MVT ResultType = TLI.getValueType(CS.getType());
+    
+      // If any of the results of the inline asm is a vector, it may have the
+      // wrong width/num elts.  This can happen for register classes that can
+      // contain multiple different value types.  The preg or vreg allocated may
+      // not have the same VT as was expected.  Convert it to the right type
+      // with bit_convert.
+      if (ResultType != Val.getValueType() && Val.getValueType().isVector()) {
+        Val = DAG.getNode(ISD::BIT_CONVERT, ResultType, Val);
 
-    // If any of the results of the inline asm is a vector, it may have the
-    // wrong width/num elts.  This can happen for register classes that can
-    // contain multiple different value types.  The preg or vreg allocated may
-    // not have the same VT as was expected.  Convert it to the right type with
-    // bit_convert.
-    if (const StructType *ResSTy = dyn_cast<StructType>(CS.getType())) {
-      for (unsigned i = 0, e = ResSTy->getNumElements(); i != e; ++i) {
-        if (Val.getNode()->getValueType(i).isVector())
-          Val = DAG.getNode(ISD::BIT_CONVERT,
-                            TLI.getValueType(ResSTy->getElementType(i)), Val);
+      } else if (ResultType != Val.getValueType() && 
+                 ResultType.isInteger() && Val.getValueType().isInteger()) {
+        // If a result value was tied to an input value, the computed result may
+        // have a wider width than the expected result.  Extract the relevant
+        // portion.
+        Val = DAG.getNode(ISD::TRUNCATE, ResultType, Val);
       }
-    } else {
-      if (Val.getValueType().isVector())
-        Val = DAG.getNode(ISD::BIT_CONVERT, TLI.getValueType(CS.getType()),
-                          Val);
+    
+      assert(ResultType == Val.getValueType() && "Asm result value mismatch!");
     }
 
     setValue(CS.getInstruction(), Val);
@@ -5219,7 +5250,8 @@ TargetLowering::LowerCallTo(SDValue Chain, const Type *RetTy,
          Value != NumValues; ++Value) {
       MVT VT = ValueVTs[Value];
       const Type *ArgTy = VT.getTypeForMVT();
-      SDValue Op = SDValue(Args[i].Node.getNode(), Args[i].Node.getResNo() + Value);
+      SDValue Op = SDValue(Args[i].Node.getNode(),
+                           Args[i].Node.getResNo() + Value);
       ISD::ArgFlagsTy Flags;
       unsigned OriginalAlignment =
         getTargetData()->getABITypeAlignment(ArgTy);
diff --git a/test/CodeGen/PowerPC/2008-10-17-AsmMatchingOperands.ll b/test/CodeGen/PowerPC/2008-10-17-AsmMatchingOperands.ll
new file mode 100644
index 00000000000..eace40c82b4
--- /dev/null
+++ b/test/CodeGen/PowerPC/2008-10-17-AsmMatchingOperands.ll
@@ -0,0 +1,10 @@
+; RUN: llvm-as < %s | llc
+; PR2356
+target datalayout = "E-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f128:64:128"
+target triple = "powerpc-apple-darwin9"
+
+define i32 @test(i64 %x, i32* %p) nounwind {
+	%asmtmp = call i32 asm "", "=r,0"(i64 0) nounwind		; <i32> [#uses=0]
+	%y = add i32 %asmtmp, 1
+	ret i32 %y
+}