Index: llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
+++ llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
@@ -702,17 +702,6 @@
   } else
     Promoted = DAG.getNode(NewOpc, dl, NVT, Node->getOperand(0));
 
-  // Assert that the converted value fits in the original type.  If it doesn't
-  // (eg: because the value being converted is too big), then the result of the
-  // original operation was undefined anyway, so the assert is still correct.
-  if (Node->getOpcode() == ISD::FP_TO_UINT ||
-      Node->getOpcode() == ISD::STRICT_FP_TO_UINT)
-    NewOpc = ISD::AssertZext;
-  else
-    NewOpc = ISD::AssertSext;
-
-  Promoted = DAG.getNode(NewOpc, dl, NVT, Promoted,
-                         DAG.getValueType(VT.getScalarType()));
   Promoted = DAG.getNode(ISD::TRUNCATE, dl, VT, Promoted);
   Results.push_back(Promoted);
   if (IsStrict)
Index: llvm/test/CodeGen/X86/avx-cvt-2.ll
===================================================================
--- llvm/test/CodeGen/X86/avx-cvt-2.ll
+++ llvm/test/CodeGen/X86/avx-cvt-2.ll
@@ -12,7 +12,10 @@
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; CHECK-NEXT:    vmovdqa %xmm0, (%rdi)
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
@@ -26,7 +29,10 @@
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; CHECK-NEXT:    vmovdqa %xmm0, (%rdi)
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
@@ -40,7 +46,10 @@
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; CHECK-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; CHECK-NEXT:    vmovq %xmm0, (%rdi)
 ; CHECK-NEXT:    vzeroupper
@@ -55,7 +64,10 @@
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; CHECK-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; CHECK-NEXT:    vmovq %xmm0, (%rdi)
 ; CHECK-NEXT:    vzeroupper
Index: llvm/test/CodeGen/X86/avx512-cvt.ll
===================================================================
--- llvm/test/CodeGen/X86/avx512-cvt.ll
+++ llvm/test/CodeGen/X86/avx512-cvt.ll
@@ -502,14 +502,16 @@
 ; NOVL-LABEL: f64to8uc:
 ; NOVL:       # %bb.0:
 ; NOVL-NEXT:    vcvttpd2dq %zmm0, %ymm0
-; NOVL-NEXT:    vpmovdb %zmm0, %xmm0
+; NOVL-NEXT:    vpmovdw %zmm0, %ymm0
+; NOVL-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; NOVL-NEXT:    vzeroupper
 ; NOVL-NEXT:    retq
 ;
 ; VL-LABEL: f64to8uc:
 ; VL:       # %bb.0:
 ; VL-NEXT:    vcvttpd2dq %zmm0, %ymm0
-; VL-NEXT:    vpmovdb %ymm0, %xmm0
+; VL-NEXT:    vpmovdw %ymm0, %xmm0
+; VL-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; VL-NEXT:    vzeroupper
 ; VL-NEXT:    retq
   %res = fptoui <8 x double> %f to <8 x i8>
@@ -645,14 +647,16 @@
 ; NOVL-LABEL: f64to8sc:
 ; NOVL:       # %bb.0:
 ; NOVL-NEXT:    vcvttpd2dq %zmm0, %ymm0
-; NOVL-NEXT:    vpmovdb %zmm0, %xmm0
+; NOVL-NEXT:    vpmovdw %zmm0, %ymm0
+; NOVL-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; NOVL-NEXT:    vzeroupper
 ; NOVL-NEXT:    retq
 ;
 ; VL-LABEL: f64to8sc:
 ; VL:       # %bb.0:
 ; VL-NEXT:    vcvttpd2dq %zmm0, %ymm0
-; VL-NEXT:    vpmovdb %ymm0, %xmm0
+; VL-NEXT:    vpmovdw %ymm0, %xmm0
+; VL-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; VL-NEXT:    vzeroupper
 ; VL-NEXT:    retq
   %res = fptosi <8 x double> %f to <8 x i8>
@@ -2356,6 +2360,7 @@
 ; NOVLDQ:       # %bb.0:
 ; NOVLDQ-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; NOVLDQ-NEXT:    vcvttpd2dq %ymm0, %xmm0
+; NOVLDQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; NOVLDQ-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; NOVLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; NOVLDQ-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
@@ -2364,6 +2369,7 @@
 ; VLDQ-LABEL: test_4f64tosb:
 ; VLDQ:       # %bb.0:
 ; VLDQ-NEXT:    vcvttpd2dq %ymm0, %xmm0
+; VLDQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; VLDQ-NEXT:    vpmovd2m %xmm0, %k1
 ; VLDQ-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
 ; VLDQ-NEXT:    retq
@@ -2371,6 +2377,7 @@
 ; VLNODQ-LABEL: test_4f64tosb:
 ; VLNODQ:       # %bb.0:
 ; VLNODQ-NEXT:    vcvttpd2dq %ymm0, %xmm0
+; VLNODQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; VLNODQ-NEXT:    vptestmd %xmm0, %xmm0, %k1
 ; VLNODQ-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
 ; VLNODQ-NEXT:    retq
@@ -2379,6 +2386,7 @@
 ; DQNOVL:       # %bb.0:
 ; DQNOVL-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; DQNOVL-NEXT:    vcvttpd2dq %ymm0, %xmm0
+; DQNOVL-NEXT:    vpslld $31, %xmm0, %xmm0
 ; DQNOVL-NEXT:    vpmovd2m %zmm0, %k1
 ; DQNOVL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; DQNOVL-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
@@ -2392,6 +2400,7 @@
 ; NOVLDQ-LABEL: test_8f64tosb:
 ; NOVLDQ:       # %bb.0:
 ; NOVLDQ-NEXT:    vcvttpd2dq %zmm0, %ymm0
+; NOVLDQ-NEXT:    vpslld $31, %ymm0, %ymm0
 ; NOVLDQ-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; NOVLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; NOVLDQ-NEXT:    retq
@@ -2399,6 +2408,7 @@
 ; VLDQ-LABEL: test_8f64tosb:
 ; VLDQ:       # %bb.0:
 ; VLDQ-NEXT:    vcvttpd2dq %zmm0, %ymm0
+; VLDQ-NEXT:    vpslld $31, %ymm0, %ymm0
 ; VLDQ-NEXT:    vpmovd2m %ymm0, %k1
 ; VLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; VLDQ-NEXT:    retq
@@ -2406,6 +2416,7 @@
 ; VLNODQ-LABEL: test_8f64tosb:
 ; VLNODQ:       # %bb.0:
 ; VLNODQ-NEXT:    vcvttpd2dq %zmm0, %ymm0
+; VLNODQ-NEXT:    vpslld $31, %ymm0, %ymm0
 ; VLNODQ-NEXT:    vptestmd %ymm0, %ymm0, %k1
 ; VLNODQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; VLNODQ-NEXT:    retq
@@ -2413,6 +2424,7 @@
 ; DQNOVL-LABEL: test_8f64tosb:
 ; DQNOVL:       # %bb.0:
 ; DQNOVL-NEXT:    vcvttpd2dq %zmm0, %ymm0
+; DQNOVL-NEXT:    vpslld $31, %ymm0, %ymm0
 ; DQNOVL-NEXT:    vpmovd2m %zmm0, %k1
 ; DQNOVL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; DQNOVL-NEXT:    retq
@@ -2426,6 +2438,7 @@
 ; NOVLDQ:       # %bb.0:
 ; NOVLDQ-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; NOVLDQ-NEXT:    vcvttps2dq %xmm0, %xmm0
+; NOVLDQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; NOVLDQ-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; NOVLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; NOVLDQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
@@ -2435,6 +2448,7 @@
 ; VLDQ-LABEL: test_2f32tosb:
 ; VLDQ:       # %bb.0:
 ; VLDQ-NEXT:    vcvttps2dq %xmm0, %xmm0
+; VLDQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; VLDQ-NEXT:    vpmovd2m %xmm0, %k1
 ; VLDQ-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} {z}
 ; VLDQ-NEXT:    retq
@@ -2442,6 +2456,7 @@
 ; VLNODQ-LABEL: test_2f32tosb:
 ; VLNODQ:       # %bb.0:
 ; VLNODQ-NEXT:    vcvttps2dq %xmm0, %xmm0
+; VLNODQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; VLNODQ-NEXT:    vptestmd %xmm0, %xmm0, %k1
 ; VLNODQ-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} {z}
 ; VLNODQ-NEXT:    retq
@@ -2450,6 +2465,7 @@
 ; DQNOVL:       # %bb.0:
 ; DQNOVL-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; DQNOVL-NEXT:    vcvttps2dq %xmm0, %xmm0
+; DQNOVL-NEXT:    vpslld $31, %xmm0, %xmm0
 ; DQNOVL-NEXT:    vpmovd2m %zmm0, %k1
 ; DQNOVL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; DQNOVL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
@@ -2465,6 +2481,7 @@
 ; NOVLDQ:       # %bb.0:
 ; NOVLDQ-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; NOVLDQ-NEXT:    vcvttps2dq %xmm0, %xmm0
+; NOVLDQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; NOVLDQ-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; NOVLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; NOVLDQ-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
@@ -2473,6 +2490,7 @@
 ; VLDQ-LABEL: test_4f32tosb:
 ; VLDQ:       # %bb.0:
 ; VLDQ-NEXT:    vcvttps2dq %xmm0, %xmm0
+; VLDQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; VLDQ-NEXT:    vpmovd2m %xmm0, %k1
 ; VLDQ-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
 ; VLDQ-NEXT:    retq
@@ -2480,6 +2498,7 @@
 ; VLNODQ-LABEL: test_4f32tosb:
 ; VLNODQ:       # %bb.0:
 ; VLNODQ-NEXT:    vcvttps2dq %xmm0, %xmm0
+; VLNODQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; VLNODQ-NEXT:    vptestmd %xmm0, %xmm0, %k1
 ; VLNODQ-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
 ; VLNODQ-NEXT:    retq
@@ -2488,6 +2507,7 @@
 ; DQNOVL:       # %bb.0:
 ; DQNOVL-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; DQNOVL-NEXT:    vcvttps2dq %xmm0, %xmm0
+; DQNOVL-NEXT:    vpslld $31, %xmm0, %xmm0
 ; DQNOVL-NEXT:    vpmovd2m %zmm0, %k1
 ; DQNOVL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; DQNOVL-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
@@ -2501,6 +2521,7 @@
 ; NOVLDQ-LABEL: test_8f32tosb:
 ; NOVLDQ:       # %bb.0:
 ; NOVLDQ-NEXT:    vcvttps2dq %ymm0, %ymm0
+; NOVLDQ-NEXT:    vpslld $31, %ymm0, %ymm0
 ; NOVLDQ-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; NOVLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; NOVLDQ-NEXT:    retq
@@ -2508,6 +2529,7 @@
 ; VLDQ-LABEL: test_8f32tosb:
 ; VLDQ:       # %bb.0:
 ; VLDQ-NEXT:    vcvttps2dq %ymm0, %ymm0
+; VLDQ-NEXT:    vpslld $31, %ymm0, %ymm0
 ; VLDQ-NEXT:    vpmovd2m %ymm0, %k1
 ; VLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; VLDQ-NEXT:    retq
@@ -2515,6 +2537,7 @@
 ; VLNODQ-LABEL: test_8f32tosb:
 ; VLNODQ:       # %bb.0:
 ; VLNODQ-NEXT:    vcvttps2dq %ymm0, %ymm0
+; VLNODQ-NEXT:    vpslld $31, %ymm0, %ymm0
 ; VLNODQ-NEXT:    vptestmd %ymm0, %ymm0, %k1
 ; VLNODQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; VLNODQ-NEXT:    retq
@@ -2522,6 +2545,7 @@
 ; DQNOVL-LABEL: test_8f32tosb:
 ; DQNOVL:       # %bb.0:
 ; DQNOVL-NEXT:    vcvttps2dq %ymm0, %ymm0
+; DQNOVL-NEXT:    vpslld $31, %ymm0, %ymm0
 ; DQNOVL-NEXT:    vpmovd2m %zmm0, %k1
 ; DQNOVL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
 ; DQNOVL-NEXT:    retq
@@ -2534,6 +2558,7 @@
 ; NODQ-LABEL: test_16f32tosb:
 ; NODQ:       # %bb.0:
 ; NODQ-NEXT:    vcvttps2dq %zmm0, %zmm0
+; NODQ-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NODQ-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; NODQ-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
 ; NODQ-NEXT:    retq
@@ -2541,6 +2566,7 @@
 ; VLDQ-LABEL: test_16f32tosb:
 ; VLDQ:       # %bb.0:
 ; VLDQ-NEXT:    vcvttps2dq %zmm0, %zmm0
+; VLDQ-NEXT:    vpslld $31, %zmm0, %zmm0
 ; VLDQ-NEXT:    vpmovd2m %zmm0, %k1
 ; VLDQ-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
 ; VLDQ-NEXT:    retq
@@ -2548,6 +2574,7 @@
 ; DQNOVL-LABEL: test_16f32tosb:
 ; DQNOVL:       # %bb.0:
 ; DQNOVL-NEXT:    vcvttps2dq %zmm0, %zmm0
+; DQNOVL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; DQNOVL-NEXT:    vpmovd2m %zmm0, %k1
 ; DQNOVL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
 ; DQNOVL-NEXT:    retq
Index: llvm/test/CodeGen/X86/min-legal-vector-width.ll
===================================================================
--- llvm/test/CodeGen/X86/min-legal-vector-width.ll
+++ llvm/test/CodeGen/X86/min-legal-vector-width.ll
@@ -627,8 +627,10 @@
 ; CHECK-LABEL: test_16f32tosb_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vcvttps2dq (%rdi), %ymm1
+; CHECK-NEXT:    vpslld $31, %ymm1, %ymm1
 ; CHECK-NEXT:    vpmovd2m %ymm1, %k0
 ; CHECK-NEXT:    vcvttps2dq 32(%rdi), %ymm1
+; CHECK-NEXT:    vpslld $31, %ymm1, %ymm1
 ; CHECK-NEXT:    vpmovd2m %ymm1, %k1
 ; CHECK-NEXT:    kunpckbw %k0, %k1, %k1
 ; CHECK-NEXT:    vmovdqu16 %ymm0, %ymm0 {%k1} {z}
@@ -643,6 +645,7 @@
 ; CHECK-LABEL: test_16f32tosb_512:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vcvttps2dq (%rdi), %zmm1
+; CHECK-NEXT:    vpslld $31, %zmm1, %zmm1
 ; CHECK-NEXT:    vpmovd2m %zmm1, %k1
 ; CHECK-NEXT:    vmovdqu16 %ymm0, %ymm0 {%k1} {z}
 ; CHECK-NEXT:    retq
Index: llvm/test/CodeGen/X86/vec-strict-fptoint-128.ll
===================================================================
--- llvm/test/CodeGen/X86/vec-strict-fptoint-128.ll
+++ llvm/test/CodeGen/X86/vec-strict-fptoint-128.ll
@@ -3229,6 +3229,7 @@
 ; AVX512F-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcvttps2dq %xmm0, %xmm0
+; AVX512F-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
@@ -3238,6 +3239,7 @@
 ; AVX512VL-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vcvttps2dq %xmm0, %xmm0
+; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
 ; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
@@ -3246,6 +3248,7 @@
 ; AVX512DQ-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vcvttps2dq %xmm0, %xmm0
+; AVX512DQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpmovd2m %zmm0, %k0
 ; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
@@ -3255,6 +3258,7 @@
 ; AVX512VLDQ-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
 ; AVX512VLDQ:       # %bb.0:
 ; AVX512VLDQ-NEXT:    vcvttps2dq %xmm0, %xmm0
+; AVX512VLDQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512VLDQ-NEXT:    vpmovd2m %xmm0, %k0
 ; AVX512VLDQ-NEXT:    vpmovm2d %k0, %xmm0
 ; AVX512VLDQ-NEXT:    ret{{[l|q]}}
Index: llvm/test/CodeGen/X86/vec-strict-fptoint-256.ll
===================================================================
--- llvm/test/CodeGen/X86/vec-strict-fptoint-256.ll
+++ llvm/test/CodeGen/X86/vec-strict-fptoint-256.ll
@@ -1280,6 +1280,7 @@
 ; AVX512F-LABEL: strict_vector_fptosi_v4f64_to_v4i1:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcvttpd2dq %ymm0, %xmm0
+; AVX512F-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
@@ -1289,6 +1290,7 @@
 ; AVX512VL-LABEL: strict_vector_fptosi_v4f64_to_v4i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vcvttpd2dq %ymm0, %xmm0
+; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
 ; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
@@ -1298,6 +1300,7 @@
 ; AVX512DQ-LABEL: strict_vector_fptosi_v4f64_to_v4i1:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vcvttpd2dq %ymm0, %xmm0
+; AVX512DQ-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpmovd2m %zmm0, %k0
 ; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
@@ -1307,6 +1310,7 @@
 ; AVX512DQVL-LABEL: strict_vector_fptosi_v4f64_to_v4i1:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vcvttpd2dq %ymm0, %xmm0
+; AVX512DQVL-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpmovd2m %xmm0, %k0
 ; AVX512DQVL-NEXT:    vpmovm2d %k0, %xmm0
 ; AVX512DQVL-NEXT:    vzeroupper
@@ -1423,7 +1427,10 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; AVX-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    ret{{[l|q]}}
 ;
@@ -1466,7 +1473,10 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; AVX-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    ret{{[l|q]}}
 ;
@@ -1509,7 +1519,10 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; AVX-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    ret{{[l|q]}}
@@ -1517,28 +1530,32 @@
 ; AVX512F-LABEL: strict_vector_fptosi_v8f32_to_v8i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcvttps2dq %ymm0, %ymm0
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512VL-LABEL: strict_vector_fptosi_v8f32_to_v8i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vcvttps2dq %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmovdb %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512DQ-LABEL: strict_vector_fptosi_v8f32_to_v8i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vcvttps2dq %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512DQ-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512DQ-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512DQVL-LABEL: strict_vector_fptosi_v8f32_to_v8i8:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vcvttps2dq %ymm0, %ymm0
-; AVX512DQVL-NEXT:    vpmovdb %ymm0, %xmm0
+; AVX512DQVL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512DQVL-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vzeroupper
 ; AVX512DQVL-NEXT:    ret{{[l|q]}}
   %ret = call <8 x i8> @llvm.experimental.constrained.fptosi.v8i8.v8f32(<8 x float> %a,
@@ -1551,7 +1568,10 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; AVX-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    ret{{[l|q]}}
@@ -1559,28 +1579,32 @@
 ; AVX512F-LABEL: strict_vector_fptoui_v8f32_to_v8i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcvttps2dq %ymm0, %ymm0
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512VL-LABEL: strict_vector_fptoui_v8f32_to_v8i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vcvttps2dq %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmovdb %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512DQ-LABEL: strict_vector_fptoui_v8f32_to_v8i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vcvttps2dq %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512DQ-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512DQ-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512DQVL-LABEL: strict_vector_fptoui_v8f32_to_v8i8:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vcvttps2dq %ymm0, %ymm0
-; AVX512DQVL-NEXT:    vpmovdb %ymm0, %xmm0
+; AVX512DQVL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512DQVL-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vzeroupper
 ; AVX512DQVL-NEXT:    ret{{[l|q]}}
   %ret = call <8 x i8> @llvm.experimental.constrained.fptoui.v8i8.v8f32(<8 x float> %a,
@@ -1593,13 +1617,17 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; AVX-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512F-LABEL: strict_vector_fptosi_v8f32_to_v8i1:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcvttps2dq %ymm0, %ymm0
+; AVX512F-NEXT:    vpslld $31, %ymm0, %ymm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
@@ -1610,6 +1638,7 @@
 ; AVX512VL-LABEL: strict_vector_fptosi_v8f32_to_v8i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vcvttps2dq %ymm0, %ymm0
+; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k1
 ; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
@@ -1620,6 +1649,7 @@
 ; AVX512DQ-LABEL: strict_vector_fptosi_v8f32_to_v8i1:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vcvttps2dq %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpslld $31, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpmovd2m %zmm0, %k0
 ; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
 ; AVX512DQ-NEXT:    vpmovdw %zmm0, %ymm0
@@ -1630,6 +1660,7 @@
 ; AVX512DQVL-LABEL: strict_vector_fptosi_v8f32_to_v8i1:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vcvttps2dq %ymm0, %ymm0
+; AVX512DQVL-NEXT:    vpslld $31, %ymm0, %ymm0
 ; AVX512DQVL-NEXT:    vpmovd2m %ymm0, %k0
 ; AVX512DQVL-NEXT:    vpmovm2d %k0, %ymm0
 ; AVX512DQVL-NEXT:    vpmovdw %ymm0, %xmm0
@@ -1645,7 +1676,10 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; AVX-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    ret{{[l|q]}}
 ;
Index: llvm/test/CodeGen/X86/vec-strict-fptoint-512.ll
===================================================================
--- llvm/test/CodeGen/X86/vec-strict-fptoint-512.ll
+++ llvm/test/CodeGen/X86/vec-strict-fptoint-512.ll
@@ -678,14 +678,16 @@
 ; AVX512VL-LABEL: strict_vector_fptosi_v8f64_to_v8i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vcvttpd2dq %zmm0, %ymm0
-; AVX512VL-NEXT:    vpmovdb %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512DQ-LABEL: strict_vector_fptosi_v8f64_to_v8i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vcvttpd2dq %zmm0, %ymm0
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512DQ-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512DQ-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    ret{{[l|q]}}
   %ret = call <8 x i8> @llvm.experimental.constrained.fptosi.v8i8.v8f64(<8 x double> %a,
@@ -697,14 +699,16 @@
 ; AVX512VL-LABEL: strict_vector_fptoui_v8f64_to_v8i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vcvttpd2dq %zmm0, %ymm0
-; AVX512VL-NEXT:    vpmovdb %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512DQ-LABEL: strict_vector_fptoui_v8f64_to_v8i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vcvttpd2dq %zmm0, %ymm0
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512DQ-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512DQ-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    ret{{[l|q]}}
   %ret = call <8 x i8> @llvm.experimental.constrained.fptoui.v8i8.v8f64(<8 x double> %a,
@@ -716,6 +720,7 @@
 ; AVX512VL-LABEL: strict_vector_fptosi_v8f64_to_v8i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vcvttpd2dq %zmm0, %ymm0
+; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k1
 ; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
@@ -726,6 +731,7 @@
 ; AVX512DQ-LABEL: strict_vector_fptosi_v8f64_to_v8i1:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vcvttpd2dq %zmm0, %ymm0
+; AVX512DQ-NEXT:    vpslld $31, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpmovd2m %zmm0, %k0
 ; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
 ; AVX512DQ-NEXT:    vpmovdw %zmm0, %ymm0
@@ -834,6 +840,7 @@
 ; AVX512VL-LABEL: strict_vector_fptosi_v16f32_to_v16i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vcvttps2dq %zmm0, %zmm0
+; AVX512VL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; AVX512VL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
@@ -843,6 +850,7 @@
 ; AVX512DQ-LABEL: strict_vector_fptosi_v16f32_to_v16i1:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vcvttps2dq %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpslld $31, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vpmovd2m %zmm0, %k0
 ; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
 ; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
Index: llvm/test/CodeGen/X86/vec_cast2.ll
===================================================================
--- llvm/test/CodeGen/X86/vec_cast2.ll
+++ llvm/test/CodeGen/X86/vec_cast2.ll
@@ -98,7 +98,10 @@
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; CHECK-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retl
@@ -111,7 +114,10 @@
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retl
   %res = fptosi <8 x float> %src to <8 x i16>
@@ -144,7 +150,10 @@
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; CHECK-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retl
@@ -157,7 +166,10 @@
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retl
   %res = fptoui <8 x float> %src to <8 x i16>
Index: llvm/test/CodeGen/X86/vec_fp_to_int.ll
===================================================================
--- llvm/test/CodeGen/X86/vec_fp_to_int.ll
+++ llvm/test/CodeGen/X86/vec_fp_to_int.ll
@@ -2661,10 +2661,16 @@
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vcvttps2dq %ymm1, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX1-NEXT:    vpshufb %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
 ; AVX1-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpshufb %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
 ; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
@@ -2672,11 +2678,12 @@
 ; AVX2-LABEL: fptosi_16f32_to_16i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vcvttps2dq %ymm1, %ymm1
-; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
+; AVX2-NEXT:    vpshufb %ymm2, %ymm1, %ymm1
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
 ; AVX2-NEXT:    vcvttps2dq %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpshufb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
@@ -2707,10 +2714,16 @@
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vcvttps2dq %ymm1, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = <0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u>
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX1-NEXT:    vpshufb %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
 ; AVX1-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpshufb %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
@@ -2718,11 +2731,12 @@
 ; AVX2-LABEL: fptoui_16f32_to_16i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vcvttps2dq %ymm1, %ymm1
-; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
+; AVX2-NEXT:    vpshufb %ymm2, %ymm1, %ymm1
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
 ; AVX2-NEXT:    vcvttps2dq %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpshufb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq