diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index eadfa1f..2be31c8 100644
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -364,6 +364,13 @@
   
   unsigned NumVectorRegs = 1;
   
+  // FIXME: We don't support non-power-of-2-sized vectors for now.  Ideally we 
+  // could break down into LHS/RHS like LegalizeDAG does.
+  if (!isPowerOf2_32(NumElts)) {
+    NumVectorRegs = NumElts;
+    NumElts = 1;
+  }
+  
   // Divide the input until we get to a supported size.  This will always
   // end with a scalar if the target doesn't support vectors.
   while (NumElts > 1 &&
diff --git a/lib/Target/X86/X86CallingConv.td b/lib/Target/X86/X86CallingConv.td
index bb054e8..aacd5a4 100644
--- a/lib/Target/X86/X86CallingConv.td
+++ b/lib/Target/X86/X86CallingConv.td
@@ -52,10 +52,12 @@
 
 // X86-32 FastCC return-value convention.
 def RetCC_X86_32_Fast : CallingConv<[
-  // The X86-32 fastcc returns FP values in XMM0 if the target has SSE2,
-  // otherwise it is the the C calling conventions.
-  CCIfType<[f32], CCIfSubtarget<"hasSSE2()", CCAssignToReg<[XMM0]>>>,
-  CCIfType<[f64], CCIfSubtarget<"hasSSE2()", CCAssignToReg<[XMM0]>>>,
+  // The X86-32 fastcc returns 1, 2, or 3 FP values in XMM0-2 if the target has
+  // SSE2, otherwise it is the the C calling conventions.
+  // This can happen when a float, 2 x float, or 3 x float vector is split by
+  // target lowering, and is returned in 1-3 sse regs.
+  CCIfType<[f32], CCIfSubtarget<"hasSSE2()", CCAssignToReg<[XMM0,XMM1,XMM2]>>>,
+  CCIfType<[f64], CCIfSubtarget<"hasSSE2()", CCAssignToReg<[XMM0,XMM1,XMM2]>>>,
   CCDelegateTo<RetCC_X86Common>
 ]>;