; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vextracti64x4 $1, %zmm0, %ymm3
-; CHECK-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3,4,5],ymm0[6,7]
+; CHECK-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3],ymm3[4,5],ymm0[6,7]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqq %ymm3, %ymm2, %k1
; CHECK-NEXT: vpermq {{.*#+}} ymm1 {%k1} = ymm0[2,3,2,3]
; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vextracti64x4 $1, %zmm0, %ymm2
-; CHECK-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3,4,5],ymm0[6,7]
+; CHECK-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3],ymm2[4,5],ymm0[6,7]
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %ymm2, %ymm1, %k1
; CHECK-NEXT: vpermq {{.*#+}} ymm0 {%k1} {z} = ymm0[2,3,2,3]
; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mask4:
; CHECK: # %bb.0:
; CHECK-NEXT: vextracti64x4 $1, %zmm0, %ymm3
-; CHECK-NEXT: vmovdqa {{.*#+}} ymm4 = [7,3,3,1]
-; CHECK-NEXT: vpermi2q %ymm0, %ymm3, %ymm4
-; CHECK-NEXT: vpxor %xmm0, %xmm0, %xmm0
-; CHECK-NEXT: vpcmpeqq %ymm0, %ymm2, %k1
-; CHECK-NEXT: vpblendmq %ymm4, %ymm1, %ymm0 {%k1}
+; CHECK-NEXT: vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm3[1],ymm0[3],ymm3[3]
+; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
+; CHECK-NEXT: vpcmpeqq %ymm3, %ymm2, %k1
+; CHECK-NEXT: vpermq {{.*#+}} ymm1 {%k1} = ymm0[2,3,3,1]
+; CHECK-NEXT: vmovdqa %ymm1, %ymm0
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 3, i32 7, i32 7, i32 5>
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mask4(<8 x i64> %vec, <4 x i64> %mask) {
; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask4:
; CHECK: # %bb.0:
-; CHECK-NEXT: vextracti64x4 $1, %zmm0, %ymm3
-; CHECK-NEXT: vmovdqa {{.*#+}} ymm2 = [7,3,3,1]
-; CHECK-NEXT: vpxor %xmm4, %xmm4, %xmm4
-; CHECK-NEXT: vpcmpeqq %ymm4, %ymm1, %k1
-; CHECK-NEXT: vpermi2q %ymm0, %ymm3, %ymm2 {%k1} {z}
-; CHECK-NEXT: vmovdqa %ymm2, %ymm0
+; CHECK-NEXT: vextracti64x4 $1, %zmm0, %ymm2
+; CHECK-NEXT: vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm2[1],ymm0[3],ymm2[3]
+; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
+; CHECK-NEXT: vpcmpeqq %ymm2, %ymm1, %k1
+; CHECK-NEXT: vpermq {{.*#+}} ymm0 {%k1} {z} = ymm0[2,3,3,1]
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 3, i32 7, i32 7, i32 5>
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
; CHECK: # %bb.0:
; CHECK-NEXT: vmovdqa64 (%rdi), %zmm2
; CHECK-NEXT: vextracti64x4 $1, %zmm2, %ymm3
-; CHECK-NEXT: vmovdqa {{.*#+}} ymm4 = [7,1,1,5]
-; CHECK-NEXT: vpermi2q %ymm3, %ymm2, %ymm4
-; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; CHECK-NEXT: vpcmpeqq %ymm2, %ymm1, %k1
-; CHECK-NEXT: vmovdqa64 %ymm4, %ymm0 {%k1}
+; CHECK-NEXT: vpunpckhqdq {{.*#+}} ymm2 = ymm3[1],ymm2[1],ymm3[3],ymm2[3]
+; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
+; CHECK-NEXT: vpcmpeqq %ymm3, %ymm1, %k1
+; CHECK-NEXT: vpermq {{.*#+}} ymm0 {%k1} = ymm2[2,1,1,0]
; CHECK-NEXT: retq
%vec = load <8 x i64>, <8 x i64>* %vp
%shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 1, i32 1, i32 5>
define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mem_mask2(<8 x i64>* %vp, <4 x i64> %mask) {
; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask2:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmovdqa64 (%rdi), %zmm2
-; CHECK-NEXT: vextracti64x4 $1, %zmm2, %ymm3
-; CHECK-NEXT: vmovdqa {{.*#+}} ymm1 = [7,1,1,5]
-; CHECK-NEXT: vpxor %xmm4, %xmm4, %xmm4
-; CHECK-NEXT: vpcmpeqq %ymm4, %ymm0, %k1
-; CHECK-NEXT: vpermi2q %ymm3, %ymm2, %ymm1 {%k1} {z}
-; CHECK-NEXT: vmovdqa %ymm1, %ymm0
+; CHECK-NEXT: vmovdqa64 (%rdi), %zmm1
+; CHECK-NEXT: vextracti64x4 $1, %zmm1, %ymm2
+; CHECK-NEXT: vpunpckhqdq {{.*#+}} ymm1 = ymm2[1],ymm1[1],ymm2[3],ymm1[3]
+; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
+; CHECK-NEXT: vpcmpeqq %ymm2, %ymm0, %k1
+; CHECK-NEXT: vpermq {{.*#+}} ymm0 {%k1} {z} = ymm1[2,1,1,0]
; CHECK-NEXT: retq
%vec = load <8 x i64>, <8 x i64>* %vp
%shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 1, i32 1, i32 5>
define <4 x i64> @test_8xi64_to_4xi64_perm_mem_mask3(<8 x i64>* %vp) {
; CHECK-LABEL: test_8xi64_to_4xi64_perm_mem_mask3:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmovaps (%rdi), %zmm0
-; CHECK-NEXT: vextractf64x4 $1, %zmm0, %ymm1
-; CHECK-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
-; CHECK-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,0,0,2]
+; CHECK-NEXT: vmovdqa64 (%rdi), %zmm0
+; CHECK-NEXT: vextracti64x4 $1, %zmm0, %ymm1
+; CHECK-NEXT: vpalignr {{.*#+}} ymm0 = ymm1[8,9,10,11,12,13,14,15],ymm0[0,1,2,3,4,5,6,7],ymm1[24,25,26,27,28,29,30,31],ymm0[16,17,18,19,20,21,22,23]
+; CHECK-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,1,1,3]
; CHECK-NEXT: retq
%vec = load <8 x i64>, <8 x i64>* %vp
%res = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 0, i32 0, i32 2>
; CHECK: # %bb.0:
; CHECK-NEXT: vmovdqa64 (%rdi), %zmm2
; CHECK-NEXT: vextracti64x4 $1, %zmm2, %ymm3
-; CHECK-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6,7]
+; CHECK-NEXT: vpalignr {{.*#+}} ymm2 = ymm3[8,9,10,11,12,13,14,15],ymm2[0,1,2,3,4,5,6,7],ymm3[24,25,26,27,28,29,30,31],ymm2[16,17,18,19,20,21,22,23]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqq %ymm3, %ymm1, %k1
-; CHECK-NEXT: vpermq {{.*#+}} ymm0 {%k1} = ymm2[3,0,0,2]
+; CHECK-NEXT: vpermq {{.*#+}} ymm0 {%k1} = ymm2[2,1,1,3]
; CHECK-NEXT: retq
%vec = load <8 x i64>, <8 x i64>* %vp
%shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 0, i32 0, i32 2>
; CHECK: # %bb.0:
; CHECK-NEXT: vmovdqa64 (%rdi), %zmm1
; CHECK-NEXT: vextracti64x4 $1, %zmm1, %ymm2
-; CHECK-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; CHECK-NEXT: vpalignr {{.*#+}} ymm1 = ymm2[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm2[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %ymm2, %ymm0, %k1
-; CHECK-NEXT: vpermq {{.*#+}} ymm0 {%k1} {z} = ymm1[3,0,0,2]
+; CHECK-NEXT: vpermq {{.*#+}} ymm0 {%k1} {z} = ymm1[2,1,1,3]
; CHECK-NEXT: retq
%vec = load <8 x i64>, <8 x i64>* %vp
%shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 0, i32 0, i32 2>
; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask7:
; CHECK: # %bb.0:
; CHECK-NEXT: vmovdqa64 (%rdi), %zmm2
-; CHECK-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm3
-; CHECK-NEXT: vextracti64x4 $1, %zmm2, %ymm2
-; CHECK-NEXT: vpermq {{.*#+}} ymm2 = ymm2[3,3,1,3]
-; CHECK-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6,7]
+; CHECK-NEXT: vextracti64x4 $1, %zmm2, %ymm3
+; CHECK-NEXT: vpunpckhqdq {{.*#+}} ymm2 = ymm3[1],ymm2[1],ymm3[3],ymm2[3]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqq %ymm3, %ymm1, %k1
-; CHECK-NEXT: vmovdqa64 %ymm2, %ymm0 {%k1}
+; CHECK-NEXT: vpermq {{.*#+}} ymm0 {%k1} = ymm2[2,2,0,1]
; CHECK-NEXT: retq
%vec = load <8 x i64>, <8 x i64>* %vp
%shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 7, i32 5, i32 1>
; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask7:
; CHECK: # %bb.0:
; CHECK-NEXT: vmovdqa64 (%rdi), %zmm1
-; CHECK-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm2
-; CHECK-NEXT: vextracti64x4 $1, %zmm1, %ymm1
-; CHECK-NEXT: vpermq {{.*#+}} ymm1 = ymm1[3,3,1,3]
-; CHECK-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; CHECK-NEXT: vextracti64x4 $1, %zmm1, %ymm2
+; CHECK-NEXT: vpunpckhqdq {{.*#+}} ymm1 = ymm2[1],ymm1[1],ymm2[3],ymm1[3]
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %ymm2, %ymm0, %k1
-; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 {%k1} {z}
+; CHECK-NEXT: vpermq {{.*#+}} ymm0 {%k1} {z} = ymm1[2,2,0,1]
; CHECK-NEXT: retq
%vec = load <8 x i64>, <8 x i64>* %vp
%shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 7, i32 5, i32 1>
; CHECK: # %bb.0:
; CHECK-NEXT: vmovdqa64 (%rdi), %zmm2
; CHECK-NEXT: vextracti64x4 $1, %zmm2, %ymm3
-; CHECK-NEXT: vextracti128 $1, %ymm3, %xmm3
-; CHECK-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; CHECK-NEXT: vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; CHECK-NEXT: vpunpcklqdq {{.*#+}} ymm2 = ymm3[0],ymm2[0],ymm3[2],ymm2[2]
+; CHECK-NEXT: vextracti128 $1, %ymm2, %xmm2
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqq %xmm3, %xmm1, %k1
; CHECK-NEXT: vmovdqa64 %xmm2, %xmm0 {%k1}
; CHECK: # %bb.0:
; CHECK-NEXT: vmovdqa64 (%rdi), %zmm1
; CHECK-NEXT: vextracti64x4 $1, %zmm1, %ymm2
-; CHECK-NEXT: vextracti128 $1, %ymm2, %xmm2
-; CHECK-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; CHECK-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3]
+; CHECK-NEXT: vpunpcklqdq {{.*#+}} ymm1 = ymm2[0],ymm1[0],ymm2[2],ymm1[2]
+; CHECK-NEXT: vextracti128 $1, %ymm1, %xmm1
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %xmm2, %xmm0, %k1
; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 {%k1} {z}