diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 38be3a82af658..2ce4fa51692b3 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -9987,19 +9987,29 @@ static bool IsElementEquivalent(int MaskSize, SDValue Op, SDValue ExpectedOp,
         MaskSize == (int)ExpectedOp.getNumOperands())
       return Op.getOperand(Idx) == ExpectedOp.getOperand(ExpectedIdx);
     break;
-  case ISD::BITCAST:
-    if (Op == ExpectedOp && (int)VT.getVectorNumElements() == MaskSize) {
-      SDValue Src = peekThroughBitcasts(Op);
-      EVT SrcVT = Src.getValueType();
-      if (SrcVT.isVector() &&
-          (SrcVT.getScalarSizeInBits() % VT.getScalarSizeInBits()) == 0) {
+  case ISD::BITCAST: {
+    SDValue Src = peekThroughBitcasts(Op);
+    EVT SrcVT = Src.getValueType();
+    if (Op == ExpectedOp && SrcVT.isVector() &&
+        (int)VT.getVectorNumElements() == MaskSize) {
+      if ((SrcVT.getScalarSizeInBits() % VT.getScalarSizeInBits()) == 0) {
         unsigned Scale = SrcVT.getScalarSizeInBits() / VT.getScalarSizeInBits();
         return (Idx % Scale) == (ExpectedIdx % Scale) &&
                IsElementEquivalent(SrcVT.getVectorNumElements(), Src, Src,
                                    Idx / Scale, ExpectedIdx / Scale);
       }
+      if ((VT.getScalarSizeInBits() % SrcVT.getScalarSizeInBits()) == 0) {
+        unsigned Scale = VT.getScalarSizeInBits() / SrcVT.getScalarSizeInBits();
+        for (unsigned I = 0; I != Scale; ++I)
+          if (!IsElementEquivalent(SrcVT.getVectorNumElements(), Src, Src,
+                                   (Idx * Scale) + I,
+                                   (ExpectedIdx * Scale) + I))
+            return false;
+        return true;
+      }
     }
     break;
+  }
   case ISD::VECTOR_SHUFFLE: {
     auto *SVN = cast<ShuffleVectorSDNode>(Op);
     return Op == ExpectedOp && (int)VT.getVectorNumElements() == MaskSize &&
diff --git a/llvm/test/CodeGen/X86/expand-vp-cast-intrinsics.ll b/llvm/test/CodeGen/X86/expand-vp-cast-intrinsics.ll
index 632f3c6c1e851..2609b06361af5 100644
--- a/llvm/test/CodeGen/X86/expand-vp-cast-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/expand-vp-cast-intrinsics.ll
@@ -532,10 +532,10 @@ define <2 x half> @vfptrunc_v2f16_v2f64(<2 x double> %a, <2 x i1> %m, i32 zeroex
 ; AVX512-NEXT:    vmovdqa %xmm0, (%rsp) # 16-byte Spill
 ; AVX512-NEXT:    callq __truncdfhf2@PLT
 ; AVX512-NEXT:    vpbroadcastw %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa (%rsp), %xmm1 # 16-byte Reload
-; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; AVX512-NEXT:    vpblendd $13, (%rsp), %xmm0, %xmm1 # 16-byte Folded Reload
+; AVX512-NEXT:    # xmm1 = mem[0],xmm0[1],mem[2,3]
 ; AVX512-NEXT:    vpbroadcastw %xmm0, %xmm0
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
 ; AVX512-NEXT:    addq $40, %rsp
 ; AVX512-NEXT:    .cfi_def_cfa_offset 8
 ; AVX512-NEXT:    retq
diff --git a/llvm/test/CodeGen/X86/horizontal-sum.ll b/llvm/test/CodeGen/X86/horizontal-sum.ll
index e2cc3ae0dca0a..443275e11459d 100644
--- a/llvm/test/CodeGen/X86/horizontal-sum.ll
+++ b/llvm/test/CodeGen/X86/horizontal-sum.ll
@@ -179,8 +179,8 @@ define <8 x float> @pair_sum_v8f32_v4f32(<4 x float> %0, <4 x float> %1, <4 x fl
 ; SSSE3-SLOW-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm2[0]
 ; SSSE3-SLOW-NEXT:    haddps %xmm7, %xmm6
 ; SSSE3-SLOW-NEXT:    haddps %xmm6, %xmm6
-; SSSE3-SLOW-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,3],xmm6[0,1]
-; SSSE3-SLOW-NEXT:    movaps %xmm2, %xmm1
+; SSSE3-SLOW-NEXT:    movhlps {{.*#+}} xmm6 = xmm2[1],xmm6[1]
+; SSSE3-SLOW-NEXT:    movaps %xmm6, %xmm1
 ; SSSE3-SLOW-NEXT:    retq
 ;
 ; SSSE3-FAST-LABEL: pair_sum_v8f32_v4f32:
@@ -345,8 +345,7 @@ define <8 x i32> @pair_sum_v8i32_v4i32(<4 x i32> %0, <4 x i32> %1, <4 x i32> %2,
 ; SSSE3-SLOW-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; SSSE3-SLOW-NEXT:    phaddd %xmm7, %xmm6
 ; SSSE3-SLOW-NEXT:    phaddd %xmm6, %xmm6
-; SSSE3-SLOW-NEXT:    palignr {{.*#+}} xmm6 = xmm1[8,9,10,11,12,13,14,15],xmm6[0,1,2,3,4,5,6,7]
-; SSSE3-SLOW-NEXT:    movdqa %xmm6, %xmm1
+; SSSE3-SLOW-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm6[1]
 ; SSSE3-SLOW-NEXT:    retq
 ;
 ; SSSE3-FAST-LABEL: pair_sum_v8i32_v4i32:
@@ -374,7 +373,7 @@ define <8 x i32> @pair_sum_v8i32_v4i32(<4 x i32> %0, <4 x i32> %1, <4 x i32> %2,
 ; AVX1-SLOW-NEXT:    vphaddd %xmm5, %xmm5, %xmm4
 ; AVX1-SLOW-NEXT:    vphaddd %xmm3, %xmm2, %xmm2
 ; AVX1-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,2,2,3]
-; AVX1-SLOW-NEXT:    vpunpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm1[0]
+; AVX1-SLOW-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm1[4,5,6,7]
 ; AVX1-SLOW-NEXT:    vpshufd {{.*#+}} xmm5 = xmm4[0,0,0,0]
 ; AVX1-SLOW-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5],xmm5[6,7]
 ; AVX1-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm2[1,3],xmm1[1,3]
@@ -397,7 +396,7 @@ define <8 x i32> @pair_sum_v8i32_v4i32(<4 x i32> %0, <4 x i32> %1, <4 x i32> %2,
 ; AVX1-FAST-NEXT:    vphaddd %xmm5, %xmm5, %xmm4
 ; AVX1-FAST-NEXT:    vphaddd %xmm3, %xmm2, %xmm2
 ; AVX1-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,2,2,3]
-; AVX1-FAST-NEXT:    vpunpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm1[0]
+; AVX1-FAST-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm1[4,5,6,7]
 ; AVX1-FAST-NEXT:    vpshufd {{.*#+}} xmm5 = xmm4[0,0,0,0]
 ; AVX1-FAST-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5],xmm5[6,7]
 ; AVX1-FAST-NEXT:    vshufps {{.*#+}} xmm1 = xmm2[1,3],xmm1[1,3]
@@ -422,7 +421,7 @@ define <8 x i32> @pair_sum_v8i32_v4i32(<4 x i32> %0, <4 x i32> %1, <4 x i32> %2,
 ; AVX2-SLOW-NEXT:    vphaddd %xmm5, %xmm5, %xmm4
 ; AVX2-SLOW-NEXT:    vphaddd %xmm3, %xmm2, %xmm2
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,2,2,3]
-; AVX2-SLOW-NEXT:    vpunpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm1[0]
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1],xmm1[2,3]
 ; AVX2-SLOW-NEXT:    vpbroadcastd %xmm4, %xmm5
 ; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1,2],xmm5[3]
 ; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm2[1,3],xmm1[1,1]
@@ -445,7 +444,7 @@ define <8 x i32> @pair_sum_v8i32_v4i32(<4 x i32> %0, <4 x i32> %1, <4 x i32> %2,
 ; AVX2-FAST-NEXT:    vphaddd %xmm5, %xmm5, %xmm4
 ; AVX2-FAST-NEXT:    vphaddd %xmm3, %xmm2, %xmm2
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,2,2,3]
-; AVX2-FAST-NEXT:    vpunpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm1[0]
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1],xmm1[2,3]
 ; AVX2-FAST-NEXT:    vpbroadcastd %xmm4, %xmm5
 ; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1,2],xmm5[3]
 ; AVX2-FAST-NEXT:    vshufps {{.*#+}} xmm1 = xmm2[1,3],xmm1[1,1]
diff --git a/llvm/test/CodeGen/X86/vector-half-conversions.ll b/llvm/test/CodeGen/X86/vector-half-conversions.ll
index 8510b1031d717..1bbf92e45fc6c 100644
--- a/llvm/test/CodeGen/X86/vector-half-conversions.ll
+++ b/llvm/test/CodeGen/X86/vector-half-conversions.ll
@@ -3138,10 +3138,10 @@ define <2 x i16> @cvt_2f64_to_2i16(<2 x double> %a0) nounwind {
 ; AVX512-NEXT:    vmovdqa %xmm0, (%rsp) # 16-byte Spill
 ; AVX512-NEXT:    callq __truncdfhf2@PLT
 ; AVX512-NEXT:    vpbroadcastw %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa (%rsp), %xmm1 # 16-byte Reload
-; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; AVX512-NEXT:    vpblendd $13, (%rsp), %xmm0, %xmm1 # 16-byte Folded Reload
+; AVX512-NEXT:    # xmm1 = mem[0],xmm0[1],mem[2,3]
 ; AVX512-NEXT:    vpbroadcastw %xmm0, %xmm0
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
 ; AVX512-NEXT:    addq $40, %rsp
 ; AVX512-NEXT:    retq
   %1 = fptrunc <2 x double> %a0 to <2 x half>
@@ -3272,8 +3272,8 @@ define <4 x i16> @cvt_4f64_to_4i16(<4 x double> %a0) nounwind {
 ; AVX512-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX512-NEXT:    callq __truncdfhf2@PLT
 ; AVX512-NEXT:    vpbroadcastw %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512-NEXT:    vpblendd $3, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512-NEXT:    # xmm0 = mem[0,1],xmm0[2,3]
 ; AVX512-NEXT:    addq $72, %rsp
 ; AVX512-NEXT:    retq
   %1 = fptrunc <4 x double> %a0 to <4 x half>
@@ -3404,8 +3404,8 @@ define <8 x i16> @cvt_4f64_to_8i16_undef(<4 x double> %a0) nounwind {
 ; AVX512-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX512-NEXT:    callq __truncdfhf2@PLT
 ; AVX512-NEXT:    vpbroadcastw %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512-NEXT:    vpblendd $3, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512-NEXT:    # xmm0 = mem[0,1],xmm0[2,3]
 ; AVX512-NEXT:    addq $72, %rsp
 ; AVX512-NEXT:    retq
   %1 = fptrunc <4 x double> %a0 to <4 x half>
@@ -4107,8 +4107,8 @@ define void @store_cvt_4f64_to_8i16_undef(<4 x double> %a0, ptr %a1) nounwind {
 ; AVX512-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX512-NEXT:    callq __truncdfhf2@PLT
 ; AVX512-NEXT:    vpbroadcastw %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512-NEXT:    vpblendd $3, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512-NEXT:    # xmm0 = mem[0,1],xmm0[2,3]
 ; AVX512-NEXT:    vmovdqa %xmm0, (%rbx)
 ; AVX512-NEXT:    addq $64, %rsp
 ; AVX512-NEXT:    popq %rbx
diff --git a/llvm/test/CodeGen/X86/vector-mul.ll b/llvm/test/CodeGen/X86/vector-mul.ll
index a166bebae721c..98b5bab98c4f9 100644
--- a/llvm/test/CodeGen/X86/vector-mul.ll
+++ b/llvm/test/CodeGen/X86/vector-mul.ll
@@ -1569,11 +1569,12 @@ define <2 x i64> @mul_v2i64_neg_17_65(<2 x i64> %a0) nounwind {
 }
 
 define <2 x i64> @mul_v2i64_0_1(<2 x i64> %a0) nounwind {
-; X86-SSE2-LABEL: mul_v2i64_0_1:
-; X86-SSE2:       # %bb.0:
-; X86-SSE2-NEXT:    xorpd %xmm1, %xmm1
-; X86-SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
-; X86-SSE2-NEXT:    retl
+; SSE2-LABEL: mul_v2i64_0_1:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    xorps %xmm1, %xmm1
+; SSE2-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
+; SSE2-NEXT:    movaps %xmm1, %xmm0
+; SSE2-NEXT:    ret{{[l|q]}}
 ;
 ; SSE4-LABEL: mul_v2i64_0_1:
 ; SSE4:       # %bb.0:
@@ -1581,13 +1582,6 @@ define <2 x i64> @mul_v2i64_0_1(<2 x i64> %a0) nounwind {
 ; SSE4-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
 ; SSE4-NEXT:    ret{{[l|q]}}
 ;
-; X64-SSE2-LABEL: mul_v2i64_0_1:
-; X64-SSE2:       # %bb.0:
-; X64-SSE2-NEXT:    xorps %xmm1, %xmm1
-; X64-SSE2-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
-; X64-SSE2-NEXT:    movaps %xmm1, %xmm0
-; X64-SSE2-NEXT:    retq
-;
 ; X64-AVX-LABEL: mul_v2i64_0_1:
 ; X64-AVX:       # %bb.0:
 ; X64-AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
diff --git a/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll b/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll
index fed72ff5c2b39..7fbb211b69ccf 100644
--- a/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll
+++ b/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll
@@ -7065,17 +7065,29 @@ define void @vec512_i64_widen_to_i256_factor4_broadcast_to_v2i256_factor2(ptr %i
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
-; AVX512BW-LABEL: vec512_i64_widen_to_i256_factor4_broadcast_to_v2i256_factor2:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX512BW-NEXT:    vpaddb (%rsi), %xmm0, %xmm0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpmovsxbq {{.*#+}} zmm2 = [0,9,2,11,0,13,2,15]
-; AVX512BW-NEXT:    vpermi2q %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpaddb (%rdx), %zmm2, %zmm0
-; AVX512BW-NEXT:    vmovdqa64 %zmm0, (%rcx)
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
+; AVX512BW-SLOW-LABEL: vec512_i64_widen_to_i256_factor4_broadcast_to_v2i256_factor2:
+; AVX512BW-SLOW:       # %bb.0:
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX512BW-SLOW-NEXT:    vpaddb (%rsi), %xmm0, %xmm0
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
+; AVX512BW-SLOW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-SLOW-NEXT:    vpunpcklqdq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
+; AVX512BW-SLOW-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm0, (%rcx)
+; AVX512BW-SLOW-NEXT:    vzeroupper
+; AVX512BW-SLOW-NEXT:    retq
+;
+; AVX512BW-FAST-LABEL: vec512_i64_widen_to_i256_factor4_broadcast_to_v2i256_factor2:
+; AVX512BW-FAST:       # %bb.0:
+; AVX512BW-FAST-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX512BW-FAST-NEXT:    vpaddb (%rsi), %xmm0, %xmm0
+; AVX512BW-FAST-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-FAST-NEXT:    vpmovsxbq {{.*#+}} zmm2 = [0,9,2,11,0,13,2,15]
+; AVX512BW-FAST-NEXT:    vpermi2q %zmm1, %zmm0, %zmm2
+; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm2, %zmm0
+; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm0, (%rcx)
+; AVX512BW-FAST-NEXT:    vzeroupper
+; AVX512BW-FAST-NEXT:    retq
   %in.vec.base = load <64 x i8>, ptr %in.vec.base.ptr, align 64
   %in.vec.bias = load <64 x i8>, ptr %in.vec.bias.ptr, align 64
   %in.vec = add <64 x i8> %in.vec.base, %in.vec.bias