diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -2421,6 +2421,27 @@
       KnownUndef.setAllBits();
       return TLO.CombineTo(Op, TLO.DAG.getUNDEF(VT));
     }
+    SDValue ScalarSrc = Op.getOperand(0);
+    if (ScalarSrc.getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
+      SDValue Src = ScalarSrc.getOperand(0);
+      SDValue Idx = ScalarSrc.getOperand(1);
+      EVT SrcVT = Src.getValueType();
+
+      ElementCount SrcEltCnt = SrcVT.getVectorElementCount();
+
+      if (SrcEltCnt.isScalable())
+        return false;
+
+      unsigned NumSrcElts = SrcEltCnt.getFixedValue();
+      if (isNullConstant(Idx)) {
+        APInt SrcDemandedElts = APInt::getOneBitSet(NumSrcElts, 0);
+        APInt SrcUndef = KnownUndef.zextOrTrunc(NumSrcElts);
+        APInt SrcZero = KnownZero.zextOrTrunc(NumSrcElts);
+        if (SimplifyDemandedVectorElts(Src, SrcDemandedElts, SrcUndef, SrcZero,
+                                       TLO, Depth + 1))
+          return true;
+      }
+    }
     KnownUndef.setHighBits(NumElts - 1);
     break;
   }
diff --git a/llvm/test/CodeGen/X86/vector-fshl-256.ll b/llvm/test/CodeGen/X86/vector-fshl-256.ll
--- a/llvm/test/CodeGen/X86/vector-fshl-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-256.ll
@@ -943,7 +943,6 @@
 define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -964,7 +963,6 @@
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -978,7 +976,6 @@
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v8i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -992,7 +989,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -1006,7 +1002,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -1029,7 +1024,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -1049,7 +1043,6 @@
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [31,31,31,31]
 ; XOPAVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -1070,7 +1063,6 @@
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; XOPAVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -1089,8 +1081,6 @@
 define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,0,0,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1111,7 +1101,6 @@
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1125,7 +1114,6 @@
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v16i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1139,7 +1127,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1153,7 +1140,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1176,7 +1162,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1196,8 +1181,6 @@
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,0,0,4,5,6,7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; XOPAVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1218,7 +1201,6 @@
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; XOPAVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1237,31 +1219,30 @@
 define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm8, %xmm8, %xmm8
-; AVX1-NEXT:    vpshufb %xmm8, %xmm2, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX1-NEXT:    vpsrlw $1, %xmm4, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX1-NEXT:    vpandn %xmm6, %xmm2, %xmm7
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm7 = xmm7[0],zero,zero,zero,zero,zero,zero,zero,xmm7[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm7, %xmm4, %xmm4
-; AVX1-NEXT:    vpcmpeqd %xmm9, %xmm9, %xmm9
-; AVX1-NEXT:    vpsrlw %xmm7, %xmm9, %xmm3
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vpsrlw $1, %xmm3, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX1-NEXT:    vpandn %xmm8, %xmm2, %xmm6
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrlw %xmm6, %xmm3, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm7, %xmm7, %xmm7
+; AVX1-NEXT:    vpsrlw %xmm6, %xmm7, %xmm5
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw %xmm6, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw %xmm7, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
-; AVX1-NEXT:    vpand %xmm6, %xmm2, %xmm2
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vpand %xmm2, %xmm8, %xmm2
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm3, %xmm3
-; AVX1-NEXT:    vpsllw %xmm2, %xmm9, %xmm4
-; AVX1-NEXT:    vpshufb %xmm8, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllw %xmm2, %xmm7, %xmm4
+; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
@@ -1271,7 +1252,6 @@
 ;
 ; AVX2-LABEL: splatvar_funnnel_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1294,7 +1274,6 @@
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1317,7 +1296,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1339,7 +1317,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1357,7 +1334,6 @@
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1375,7 +1351,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1393,7 +1368,6 @@
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLVBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-fshl-512.ll b/llvm/test/CodeGen/X86/vector-fshl-512.ll
--- a/llvm/test/CodeGen/X86/vector-fshl-512.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-512.ll
@@ -597,7 +597,6 @@
 define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v16i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -611,7 +610,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -625,7 +623,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -645,7 +642,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -670,7 +666,6 @@
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -691,7 +686,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -712,7 +706,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -732,7 +725,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -757,7 +749,6 @@
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
 ; AVX512F-NEXT:    vpsrlw $1, %ymm3, %ymm3
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
@@ -790,7 +781,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
 ; AVX512VL-NEXT:    vpsrlw $1, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
@@ -823,7 +813,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -845,7 +834,6 @@
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -867,7 +855,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -889,7 +876,6 @@
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLVBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll b/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
--- a/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
@@ -995,7 +995,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -1012,7 +1011,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -1029,7 +1027,6 @@
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VBMI2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -1046,7 +1043,6 @@
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLVBMI2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-fshl-rot-512.ll b/llvm/test/CodeGen/X86/vector-fshl-rot-512.ll
--- a/llvm/test/CodeGen/X86/vector-fshl-rot-512.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-rot-512.ll
@@ -454,7 +454,6 @@
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512F-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -478,7 +477,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512VL-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -502,7 +500,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -517,7 +514,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -549,97 +545,94 @@
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm1, %zmm2
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX512F-NEXT:    vpsubb %ymm2, %ymm5, %ymm5
-; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm5, %ymm5
-; AVX512F-NEXT:    vpsllw $5, %ymm5, %ymm5
-; AVX512F-NEXT:    vpblendvb %ymm5, %ymm3, %ymm1, %ymm3
-; AVX512F-NEXT:    vpsrlw $2, %ymm3, %ymm6
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; AVX512F-NEXT:    vpsrlw $4, %ymm2, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpxor %xmm6, %xmm6, %xmm6
+; AVX512F-NEXT:    vpsubb %ymm3, %ymm6, %ymm3
+; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
+; AVX512F-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm4, %ymm2, %ymm4
+; AVX512F-NEXT:    vpsrlw $2, %ymm4, %ymm6
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm7 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
 ; AVX512F-NEXT:    vpand %ymm7, %ymm6, %ymm6
-; AVX512F-NEXT:    vpaddb %ymm5, %ymm5, %ymm8
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm3, %ymm3
-; AVX512F-NEXT:    vpsrlw $1, %ymm3, %ymm6
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm8
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsrlw $1, %ymm4, %ymm6
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm9 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; AVX512F-NEXT:    vpand %ymm6, %ymm9, %ymm6
 ; AVX512F-NEXT:    vpaddb %ymm8, %ymm8, %ymm10
-; AVX512F-NEXT:    vpblendvb %ymm10, %ymm6, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm10, %ymm6, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm6
-; AVX512F-NEXT:    vpand %ymm4, %ymm6, %ymm4
-; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm0, %ymm4
-; AVX512F-NEXT:    vpsrlw $2, %ymm4, %ymm5
+; AVX512F-NEXT:    vpand %ymm5, %ymm6, %ymm5
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm0, %ymm3
+; AVX512F-NEXT:    vpsrlw $2, %ymm3, %ymm5
 ; AVX512F-NEXT:    vpand %ymm7, %ymm5, %ymm5
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw $1, %ymm4, %ymm5
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $1, %ymm3, %ymm5
 ; AVX512F-NEXT:    vpand %ymm5, %ymm9, %ymm5
-; AVX512F-NEXT:    vpblendvb %ymm10, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
-; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm1
+; AVX512F-NEXT:    vpblendvb %ymm10, %ymm5, %ymm3, %ymm3
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
+; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT:    vpsllw %xmm1, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllw %xmm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm2
 ; AVX512F-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512F-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsllw %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpbroadcastb %xmm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
-; AVX512F-NEXT:    vpternlogq $236, %zmm1, %zmm3, %zmm0
+; AVX512F-NEXT:    vpternlogq $236, %zmm2, %zmm3, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm1, %ymm1
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm1, %zmm2
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpsubb %ymm2, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm5, %ymm5
-; AVX512VL-NEXT:    vpsllw $5, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm3, %ymm1, %ymm3
-; AVX512VL-NEXT:    vpsrlw $2, %ymm3, %ymm6
+; AVX512VL-NEXT:    vpbroadcastb %xmm1, %ymm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; AVX512VL-NEXT:    vpsrlw $4, %ymm2, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpxor %xmm6, %xmm6, %xmm6
+; AVX512VL-NEXT:    vpsubb %ymm3, %ymm6, %ymm3
+; AVX512VL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm2, %ymm4
+; AVX512VL-NEXT:    vpsrlw $2, %ymm4, %ymm6
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm7 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
 ; AVX512VL-NEXT:    vpand %ymm7, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpaddb %ymm5, %ymm5, %ymm8
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpsrlw $1, %ymm3, %ymm6
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm8
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlw $1, %ymm4, %ymm6
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm9 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm9, %ymm6
 ; AVX512VL-NEXT:    vpaddb %ymm8, %ymm8, %ymm10
-; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm6, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm6, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm6
-; AVX512VL-NEXT:    vpand %ymm4, %ymm6, %ymm4
-; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpsrlw $2, %ymm4, %ymm5
+; AVX512VL-NEXT:    vpand %ymm5, %ymm6, %ymm5
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpsrlw $2, %ymm3, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm7, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw $1, %ymm4, %ymm5
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $1, %ymm3, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm9, %ymm5
-; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
-; AVX512VL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm2, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm1
+; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
+; AVX512VL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VL-NEXT:    vpsllw %xmm1, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllw %xmm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm2
 ; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpsllw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpbroadcastb %xmm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpternlogq $236, %zmm1, %zmm3, %zmm0
+; AVX512VL-NEXT:    vpternlogq $236, %zmm2, %zmm3, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -661,7 +654,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -683,7 +675,6 @@
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VBMI2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -705,7 +696,6 @@
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLVBMI2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-fshr-256.ll b/llvm/test/CodeGen/X86/vector-fshr-256.ll
--- a/llvm/test/CodeGen/X86/vector-fshr-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-256.ll
@@ -933,7 +933,6 @@
 define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -954,7 +953,6 @@
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -968,7 +966,6 @@
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v8i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -982,7 +979,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -996,7 +992,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -1019,7 +1014,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -1040,7 +1034,6 @@
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [31,31,31,31]
 ; XOPAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -1061,7 +1054,6 @@
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -1080,8 +1072,6 @@
 define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,0,0,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1102,7 +1092,6 @@
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1116,7 +1105,6 @@
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v16i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1130,7 +1118,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1144,7 +1131,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1167,7 +1153,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1188,8 +1173,6 @@
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,0,0,4,5,6,7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; XOPAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1210,7 +1193,6 @@
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -1229,27 +1211,26 @@
 define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm8, %xmm8, %xmm8
-; AVX1-NEXT:    vpshufb %xmm8, %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm5
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,zero,zero,zero,zero,xmm5[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
-; AVX1-NEXT:    vpsrlw %xmm5, %xmm6, %xmm6
-; AVX1-NEXT:    vpcmpeqd %xmm7, %xmm7, %xmm7
-; AVX1-NEXT:    vpsrlw %xmm5, %xmm7, %xmm3
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpand %xmm3, %xmm6, %xmm6
-; AVX1-NEXT:    vpsrlw %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
-; AVX1-NEXT:    vpandn %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
+; AVX1-NEXT:    vpcmpeqd %xmm6, %xmm6, %xmm6
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm6, %xmm7
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm7, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
+; AVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm3, %xmm3
-; AVX1-NEXT:    vpsllw %xmm2, %xmm7, %xmm4
-; AVX1-NEXT:    vpshufb %xmm8, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllw %xmm2, %xmm6, %xmm4
+; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
@@ -1260,7 +1241,6 @@
 ;
 ; AVX2-LABEL: splatvar_funnnel_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1282,7 +1262,6 @@
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1304,7 +1283,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1325,7 +1303,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1342,7 +1319,6 @@
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1359,7 +1335,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -1376,7 +1351,6 @@
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLVBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-fshr-512.ll b/llvm/test/CodeGen/X86/vector-fshr-512.ll
--- a/llvm/test/CodeGen/X86/vector-fshr-512.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-512.ll
@@ -593,7 +593,6 @@
 define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v16i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -607,7 +606,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -621,7 +619,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -642,7 +639,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
@@ -668,7 +664,6 @@
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -689,7 +684,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -710,7 +704,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -731,7 +724,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
 ; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
@@ -757,7 +749,6 @@
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -787,7 +778,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -817,7 +807,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -838,7 +827,6 @@
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -859,7 +847,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
@@ -880,7 +867,6 @@
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLVBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll b/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
--- a/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
@@ -1076,7 +1076,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -1093,7 +1092,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -1110,7 +1108,6 @@
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VBMI2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -1127,7 +1124,6 @@
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLVBMI2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-fshr-rot-512.ll b/llvm/test/CodeGen/X86/vector-fshr-rot-512.ll
--- a/llvm/test/CodeGen/X86/vector-fshr-rot-512.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-rot-512.ll
@@ -454,7 +454,6 @@
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512F-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -478,7 +477,6 @@
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512VL-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -502,7 +500,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -517,7 +514,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -549,93 +545,90 @@
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm1, %zmm1
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
-; AVX512F-NEXT:    vpsllw $4, %ymm2, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX512F-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX512F-NEXT:    vpsubb %ymm1, %ymm5, %ymm5
-; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm5, %ymm5
-; AVX512F-NEXT:    vpsllw $5, %ymm5, %ymm5
-; AVX512F-NEXT:    vpblendvb %ymm5, %ymm3, %ymm2, %ymm3
-; AVX512F-NEXT:    vpsllw $2, %ymm3, %ymm6
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm2
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512F-NEXT:    vpsllw $4, %ymm3, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX512F-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpxor %xmm6, %xmm6, %xmm6
+; AVX512F-NEXT:    vpsubb %ymm2, %ymm6, %ymm2
+; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm4
+; AVX512F-NEXT:    vpsllw $2, %ymm4, %ymm6
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
 ; AVX512F-NEXT:    vpand %ymm7, %ymm6, %ymm6
-; AVX512F-NEXT:    vpaddb %ymm5, %ymm5, %ymm8
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm3, %ymm3
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm6
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm8
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm6
 ; AVX512F-NEXT:    vpaddb %ymm8, %ymm8, %ymm9
-; AVX512F-NEXT:    vpblendvb %ymm9, %ymm6, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm9, %ymm6, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm6
-; AVX512F-NEXT:    vpand %ymm4, %ymm6, %ymm4
-; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm0, %ymm4
-; AVX512F-NEXT:    vpsllw $2, %ymm4, %ymm5
+; AVX512F-NEXT:    vpand %ymm5, %ymm6, %ymm5
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm5, %ymm0, %ymm2
+; AVX512F-NEXT:    vpsllw $2, %ymm2, %ymm5
 ; AVX512F-NEXT:    vpand %ymm7, %ymm5, %ymm5
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
-; AVX512F-NEXT:    vpblendvb %ymm9, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm5
+; AVX512F-NEXT:    vpblendvb %ymm9, %ymm5, %ymm2, %ymm2
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm2
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm1, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsrlw %xmm1, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm2
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm3
 ; AVX512F-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpbroadcastb %xmm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
-; AVX512F-NEXT:    vpternlogq $236, %zmm2, %zmm3, %zmm0
+; AVX512F-NEXT:    vpternlogq $236, %zmm3, %zmm2, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm1, %ymm1
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm1, %zmm1
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
-; AVX512VL-NEXT:    vpsllw $4, %ymm2, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpsubb %ymm1, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm5, %ymm5
-; AVX512VL-NEXT:    vpsllw $5, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm3, %ymm2, %ymm3
-; AVX512VL-NEXT:    vpsllw $2, %ymm3, %ymm6
+; AVX512VL-NEXT:    vpbroadcastb %xmm1, %ymm2
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512VL-NEXT:    vpsllw $4, %ymm3, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX512VL-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpxor %xmm6, %xmm6, %xmm6
+; AVX512VL-NEXT:    vpsubb %ymm2, %ymm6, %ymm2
+; AVX512VL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm4
+; AVX512VL-NEXT:    vpsllw $2, %ymm4, %ymm6
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
 ; AVX512VL-NEXT:    vpand %ymm7, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpaddb %ymm5, %ymm5, %ymm8
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm6
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm8
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm6
 ; AVX512VL-NEXT:    vpaddb %ymm8, %ymm8, %ymm9
-; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm6, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm6, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm6
-; AVX512VL-NEXT:    vpand %ymm4, %ymm6, %ymm4
-; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpsllw $2, %ymm4, %ymm5
+; AVX512VL-NEXT:    vpand %ymm5, %ymm6, %ymm5
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm5, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm7, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
-; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm5
+; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm2
 ; AVX512VL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm1, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsrlw %xmm1, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm2
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm3
 ; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpbroadcastb %xmm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpternlogq $236, %zmm2, %zmm3, %zmm0
+; AVX512VL-NEXT:    vpternlogq $236, %zmm3, %zmm2, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -657,7 +650,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -679,7 +671,6 @@
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VBMI2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -701,7 +692,6 @@
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLVBMI2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-rotate-256.ll b/llvm/test/CodeGen/X86/vector-rotate-256.ll
--- a/llvm/test/CodeGen/X86/vector-rotate-256.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-256.ll
@@ -979,7 +979,6 @@
 ;
 ; AVX512BW-LABEL: splatvar_rotate_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm2, %zmm0, %zmm2
@@ -993,7 +992,6 @@
 ;
 ; AVX512VLBW-LABEL: splatvar_rotate_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm2, %zmm0, %zmm2
@@ -1007,7 +1005,6 @@
 ;
 ; AVX512VBMI2-LABEL: splatvar_rotate_v32i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsllw %xmm2, %zmm0, %zmm2
@@ -1021,7 +1018,6 @@
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_rotate_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLVBMI2-NEXT:    vpsllw %xmm2, %zmm0, %zmm2