Index: lib/Target/X86/X86InstrSSE.td
===================================================================
--- lib/Target/X86/X86InstrSSE.td
+++ lib/Target/X86/X86InstrSSE.td
@@ -2282,6 +2282,9 @@
 
 let Predicates = [HasAVX, NoVLX] in {
   // Match fpround and fpextend for 128/256-bit conversions
+  def : Pat<(v4f32 (bitconvert (X86vzmovl (v2f64 (bitconvert
+                               (v4f32 (X86vfpround (v2f64 VR128:$src)))))))),
+            (VCVTPD2PSrr VR128:$src)>;
   def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
             (VCVTPD2PSrr VR128:$src)>;
   def : Pat<(v4f32 (X86vfpround (loadv2f64 addr:$src))),
@@ -2301,6 +2304,9 @@
 
 let Predicates = [UseSSE2] in {
   // Match fpround and fpextend for 128 conversions
+  def : Pat<(v4f32 (bitconvert (X86vzmovl (v2f64 (bitconvert
+                               (v4f32 (X86vfpround (v2f64 VR128:$src)))))))),
+            (CVTPD2PSrr VR128:$src)>;
   def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
             (CVTPD2PSrr VR128:$src)>;
   def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
Index: lib/Target/X86/X86IntrinsicsInfo.h
===================================================================
--- lib/Target/X86/X86IntrinsicsInfo.h
+++ lib/Target/X86/X86IntrinsicsInfo.h
@@ -1884,6 +1884,7 @@
   X86_INTRINSIC_DATA(sse2_comile_sd,    COMI, X86ISD::COMI, ISD::SETLE),
   X86_INTRINSIC_DATA(sse2_comilt_sd,    COMI, X86ISD::COMI, ISD::SETLT),
   X86_INTRINSIC_DATA(sse2_comineq_sd,   COMI, X86ISD::COMI, ISD::SETNE),
+  X86_INTRINSIC_DATA(sse2_cvtpd2ps,     INTR_TYPE_1OP, X86ISD::VFPROUND, 0),
   X86_INTRINSIC_DATA(sse2_max_pd,       INTR_TYPE_2OP, X86ISD::FMAX, 0),
   X86_INTRINSIC_DATA(sse2_min_pd,       INTR_TYPE_2OP, X86ISD::FMIN, 0),
   X86_INTRINSIC_DATA(sse2_movmsk_pd,    INTR_TYPE_1OP, X86ISD::MOVMSK, 0),
Index: test/CodeGen/X86/sse2-intrinsics-x86.ll
===================================================================
--- test/CodeGen/X86/sse2-intrinsics-x86.ll
+++ test/CodeGen/X86/sse2-intrinsics-x86.ll
@@ -229,13 +229,11 @@
 ; SSE-LABEL: test_x86_sse2_cvtpd2ps_zext:
 ; SSE:       ## BB#0:
 ; SSE-NEXT:    cvtpd2ps %xmm0, %xmm0
-; SSE-NEXT:    movq {{.*#+}} xmm0 = xmm0[0],zero
 ; SSE-NEXT:    retl
 ;
 ; KNL-LABEL: test_x86_sse2_cvtpd2ps_zext:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vcvtpd2ps %xmm0, %xmm0
-; KNL-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
 ; KNL-NEXT:    retl
   %cvt = call <4 x float> @llvm.x86.sse2.cvtpd2ps(<2 x double> %a0)
   %res = shufflevector <4 x float> %cvt, <4 x float> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
Index: test/CodeGen/X86/vec_fptrunc.ll
===================================================================
--- test/CodeGen/X86/vec_fptrunc.ll
+++ test/CodeGen/X86/vec_fptrunc.ll
@@ -138,26 +138,22 @@
 ; X32-SSE:       # BB#0:
 ; X32-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-SSE-NEXT:    cvtpd2ps (%eax), %xmm0
-; X32-SSE-NEXT:    movq {{.*#+}} xmm0 = xmm0[0],zero
 ; X32-SSE-NEXT:    retl
 ;
 ; X32-AVX-LABEL: fptrunc_frommem2_zext:
 ; X32-AVX:       # BB#0:
 ; X32-AVX-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-AVX-NEXT:    vcvtpd2psx (%eax), %xmm0
-; X32-AVX-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
 ; X32-AVX-NEXT:    retl
 ;
 ; X64-SSE-LABEL: fptrunc_frommem2_zext:
 ; X64-SSE:       # BB#0:
 ; X64-SSE-NEXT:    cvtpd2ps (%rdi), %xmm0
-; X64-SSE-NEXT:    movq {{.*#+}} xmm0 = xmm0[0],zero
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX-LABEL: fptrunc_frommem2_zext:
 ; X64-AVX:       # BB#0:
 ; X64-AVX-NEXT:    vcvtpd2psx (%rdi), %xmm0
-; X64-AVX-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
 ; X64-AVX-NEXT:    retq
   %arg = load <2 x double>, <2 x double> * %ld, align 16
   %cvt = fptrunc <2 x double> %arg to <2 x float>
@@ -169,25 +165,21 @@
 ; X32-SSE-LABEL: fptrunc_fromreg2_zext:
 ; X32-SSE:       # BB#0:
 ; X32-SSE-NEXT:    cvtpd2ps %xmm0, %xmm0
-; X32-SSE-NEXT:    movq {{.*#+}} xmm0 = xmm0[0],zero
 ; X32-SSE-NEXT:    retl
 ;
 ; X32-AVX-LABEL: fptrunc_fromreg2_zext:
 ; X32-AVX:       # BB#0:
 ; X32-AVX-NEXT:    vcvtpd2ps %xmm0, %xmm0
-; X32-AVX-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
 ; X32-AVX-NEXT:    retl
 ;
 ; X64-SSE-LABEL: fptrunc_fromreg2_zext:
 ; X64-SSE:       # BB#0:
 ; X64-SSE-NEXT:    cvtpd2ps %xmm0, %xmm0
-; X64-SSE-NEXT:    movq {{.*#+}} xmm0 = xmm0[0],zero
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX-LABEL: fptrunc_fromreg2_zext:
 ; X64-AVX:       # BB#0:
 ; X64-AVX-NEXT:    vcvtpd2ps %xmm0, %xmm0
-; X64-AVX-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
 ; X64-AVX-NEXT:    retq
   %cvt = fptrunc <2 x double> %arg to <2 x float>
   %ret = shufflevector <2 x float> %cvt, <2 x float> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 2>