diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index d8eaee71c89..69acf1a0181 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -3833,11 +3833,13 @@ X86TargetLowering::LowerEXTRACT_VECTOR_ELT_SSE4(SDOperand Op,
   } else if (VT == MVT::f32) {
     // EXTRACTPS outputs to a GPR32 register which will require a movd to copy
     // the result back to FR32 register. It's only worth matching if the
-    // result has a single use which is a store.
+    // result has a single use which is a store or a bitcast to i32.
     if (!Op.hasOneUse())
       return SDOperand();
     SDNode *User = Op.Val->use_begin()->getUser();
-    if (User->getOpcode() != ISD::STORE)
+    if (User->getOpcode() != ISD::STORE &&
+        (User->getOpcode() != ISD::BIT_CONVERT ||
+         User->getValueType(0) != MVT::i32))
       return SDOperand();
     SDOperand Extract = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, MVT::i32,
                     DAG.getNode(ISD::BIT_CONVERT, MVT::v4i32, Op.getOperand(0)),
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 442847cda8e..982b0dc6df6 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -3387,13 +3387,12 @@ defm PEXTRD      : SS41I_extract32<0x16, "pextrd">;
 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
 /// destination
 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
-  // Not worth matching to rr form of extractps since the result is in GPR32.
   def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
                  (ins VR128:$src1, i32i8imm:$src2),
                  !strconcat(OpcodeStr, 
                   "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-                 [/*(set GR32:$dst,
-                  (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))*/]>,
+                 [(set GR32:$dst,
+                    (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
            OpSize;
   def mr : SS4AIi8<opc, MRMDestMem, (outs), 
                  (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
diff --git a/test/CodeGen/X86/sse41-extractps-bitcast-0.ll b/test/CodeGen/X86/sse41-extractps-bitcast-0.ll
new file mode 100644
index 00000000000..bcfaa7a0e5a
--- /dev/null
+++ b/test/CodeGen/X86/sse41-extractps-bitcast-0.ll
@@ -0,0 +1,12 @@
+; RUN: llvm-as < %s | llc -march=x86 -mattr=sse41 | grep extractps | count 2
+
+define i32 @foo(<4 x float> %v) {
+  %s = extractelement <4 x float> %v, i32 3
+  %i = bitcast float %s to i32
+  ret i32 %i
+}
+define i32 @boo(<4 x float> %v) {
+  %t = bitcast <4 x float> %v to <4 x i32>
+  %s = extractelement <4 x i32> %t, i32 3
+  ret i32 %s
+}
diff --git a/test/CodeGen/X86/sse41-extractps-bitcast-1.ll b/test/CodeGen/X86/sse41-extractps-bitcast-1.ll
new file mode 100644
index 00000000000..fc0df060623
--- /dev/null
+++ b/test/CodeGen/X86/sse41-extractps-bitcast-1.ll
@@ -0,0 +1,19 @@
+; RUN: llvm-as < %s | llc -march=x86 -mattr=sse41 | not grep extractps
+
+; The non-store form of extractps puts its result into a GPR.
+; This makes it suitable for an extract from a <4 x float> that
+; is bitcasted to i32, but unsuitable for much of anything else.
+
+define float @bar(<4 x float> %v) {
+  %s = extractelement <4 x float> %v, i32 3
+  %t = add float %s, 1.0
+  ret float %t
+}
+define float @baz(<4 x float> %v) {
+  %s = extractelement <4 x float> %v, i32 3
+  ret float %s
+}
+define i32 @qux(<4 x i32> %v) {
+  %i = extractelement <4 x i32> %v, i32 3
+  ret i32 %i
+}