Index: lib/Target/X86/X86InstrSSE.td
===================================================================
--- lib/Target/X86/X86InstrSSE.td
+++ lib/Target/X86/X86InstrSSE.td
@@ -1307,9 +1307,7 @@
                                  (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    "movhpd\t{$src, $dst|$dst, $src}",
-                   [(store (f64 (vector_extract
-                                 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
-                                 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
+                   [], IIC_SSE_MOV_LH>, VEX;
 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    "movhps\t{$src, $dst|$dst, $src}",
                    [(store (f64 (vector_extract
@@ -1318,9 +1316,7 @@
                                  (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    "movhpd\t{$src, $dst|$dst, $src}",
-                   [(store (f64 (vector_extract
-                                 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
-                                 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
+                   [], IIC_SSE_MOV_LH>;
 } // SchedRW
 
 let Predicates = [HasAVX] in {
@@ -1332,6 +1328,8 @@
                  (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
             (VMOVHPSrm VR128:$src1, addr:$src2)>;
 
+  // VMOVHPD patterns
+  
   // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
   // is during lowering, where it's not possible to recognize the load fold
   // cause it has two uses through a bitcast. One use disappears at isel time
@@ -1344,6 +1342,16 @@
   def : Pat<(v2f64 (X86Unpckl VR128:$src1,
                       (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src2)))))),
             (VMOVHPDrm VR128:$src1, addr:$src2)>;
+
+  def : Pat<(store (f64 (vector_extract
+                          (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
+                          (iPTR 0))), addr:$dst),
+            (VMOVHPDmr addr:$dst, VR128:$src)>;
+            
+  def : Pat<(store (f64 (vector_extract
+                          (v2f64 (X86VPermilpi VR128:$src, (i8 1))),
+                          (iPTR 0))), addr:$dst),
+            (VMOVHPDmr addr:$dst, VR128:$src)>;
 }
 
 let Predicates = [UseSSE1] in {
@@ -1369,6 +1377,16 @@
   def : Pat<(v2f64 (X86Unpckl VR128:$src1,
                       (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src2)))))),
             (MOVHPDrm VR128:$src1, addr:$src2)>;
+
+  def : Pat<(store (f64 (vector_extract
+                          (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
+                          (iPTR 0))), addr:$dst),
+            (MOVHPDmr addr:$dst, VR128:$src)>;
+
+  def : Pat<(store (f64 (vector_extract
+                          (v2f64 (X86Shufp VR128:$src, VR128:$src, (i8 1))),
+                          (iPTR 0))), addr:$dst),
+            (MOVHPDmr addr:$dst, VR128:$src)>;
 }
 
 //===----------------------------------------------------------------------===//
Index: test/CodeGen/X86/extractelement-load.ll
===================================================================
--- test/CodeGen/X86/extractelement-load.ll
+++ test/CodeGen/X86/extractelement-load.ll
@@ -1,5 +1,6 @@
 ; RUN: llc < %s -march=x86 -mattr=+sse2 -mcpu=yonah | FileCheck %s
 ; RUN: llc < %s -march=x86-64 -mattr=+sse2 -mcpu=core2 | FileCheck %s
+; RUN: llc < %s -march=x86-64 -mattr=+avx -mcpu=btver2 | FileCheck %s
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 
@@ -29,16 +30,15 @@
 ; This case could easily end up inf-looping in the DAG combiner due to an
 ; low alignment load of the vector which prevents us from reliably forming a
 ; narrow load.
-; FIXME: It would be nice to detect whether the target has fast and legal
-; unaligned loads and use them here.
+
+; The expected codegen is identical for the AVX case except
+; load/store instructions will have a leading 'v', so we don't
+; need to special-case the checks.
+
 define void @t3() {
 ; CHECK-LABEL: t3:
-;
-; This movs the entire vector, shuffling the high double down. If we fixed the
-; FIXME above it would just move the high double directly.
 ; CHECK: movupd
-; CHECK: shufpd
-; CHECK: movlpd
+; CHECK: movhpd
 
 bb:
   %tmp13 = load <2 x double>* undef, align 1