| blob | 4668d7b6870ef27c42c8187f9d21848ee1a7cc5a |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck --check-prefixes=ALL,KNL %s
3 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx | FileCheck --check-prefixes=ALL,SKX %s
5 target triple = "x86_64-unknown-unknown"
7 define <32 x i16> @shuffle_v32i16(<32 x i16> %a) {
8 ; KNL-LABEL: shuffle_v32i16:
9 ; KNL: ## %bb.0:
10 ; KNL-NEXT: vpbroadcastw %xmm0, %ymm0
11 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
12 ; KNL-NEXT: retq
13 ;
14 ; SKX-LABEL: shuffle_v32i16:
15 ; SKX: ## %bb.0:
16 ; SKX-NEXT: vpbroadcastw %xmm0, %zmm0
17 ; SKX-NEXT: retq
18 %c = shufflevector <32 x i16> %a, <32 x i16> undef, <32 x i32> zeroinitializer
19 ret <32 x i16> %c
20 }
22 define <32 x i16> @shuffle_v32i16_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08(<32 x i16> %a) {
23 ; KNL-LABEL: shuffle_v32i16_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08:
24 ; KNL: ## %bb.0:
25 ; KNL-NEXT: vextracti128 $1, %ymm0, %xmm0
26 ; KNL-NEXT: vpbroadcastw %xmm0, %ymm0
27 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
28 ; KNL-NEXT: retq
29 ;
30 ; SKX-LABEL: shuffle_v32i16_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08:
31 ; SKX: ## %bb.0:
32 ; SKX-NEXT: vpbroadcastw {{.*#+}} zmm1 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
33 ; SKX-NEXT: vpermw %zmm0, %zmm1, %zmm0
34 ; SKX-NEXT: retq
35 %c = shufflevector <32 x i16> %a, <32 x i16> undef, <32 x i32> <i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
36 ret <32 x i16> %c
37 }
39 define <32 x i16> @shuffle_v32i16_02_05_u_u_07_u_0a_01_00_05_u_04_07_u_0a_01_02_05_u_u_07_u_0a_01_00_05_u_04_07_u_0a_1f(<32 x i16> %a) {
40 ; KNL-LABEL: shuffle_v32i16_02_05_u_u_07_u_0a_01_00_05_u_04_07_u_0a_01_02_05_u_u_07_u_0a_01_00_05_u_04_07_u_0a_1f:
41 ; KNL: ## %bb.0:
42 ; KNL-NEXT: vpshufb {{.*#+}} ymm1 = ymm0[4,5,10,11,4,5,6,7,14,15,2,3,4,5,2,3,20,21,26,27,20,21,22,23,30,31,18,19,20,21,18,19]
43 ; KNL-NEXT: vpermq {{.*#+}} ymm2 = ymm0[2,3,0,1]
44 ; KNL-NEXT: vpshufb {{.*#+}} ymm3 = ymm2[0,1,10,11,8,9,8,9,14,15,6,7,4,5,14,15,16,17,26,27,24,25,24,25,30,31,22,23,20,21,30,31]
45 ; KNL-NEXT: vmovdqa {{.*#+}} ymm4 = <255,255,255,255,u,u,u,u,255,255,u,u,0,0,255,255,0,0,0,0,u,u,0,0,0,0,u,u,255,255,u,u>
46 ; KNL-NEXT: vpblendvb %ymm4, %ymm1, %ymm3, %ymm3
47 ; KNL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
48 ; KNL-NEXT: vpblendw {{.*#+}} ymm0 = ymm3[0,1,2,3,4,5,6],ymm0[7],ymm3[8,9,10,11,12,13,14],ymm0[15]
49 ; KNL-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
50 ; KNL-NEXT: vpshufb {{.*#+}} ymm2 = ymm2[0,1,10,11,8,9,8,9,14,15,2,3,4,5,2,3,16,17,26,27,24,25,24,25,30,31,18,19,20,21,18,19]
51 ; KNL-NEXT: vmovdqa {{.*#+}} ymm3 = <0,0,0,0,u,u,u,u,0,0,u,u,255,255,0,0,255,255,255,255,u,u,255,255,255,255,u,u,0,0,255,255>
52 ; KNL-NEXT: vpblendvb %ymm3, %ymm2, %ymm1, %ymm1
53 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
54 ; KNL-NEXT: retq
55 ;
56 ; SKX-LABEL: shuffle_v32i16_02_05_u_u_07_u_0a_01_00_05_u_04_07_u_0a_01_02_05_u_u_07_u_0a_01_00_05_u_04_07_u_0a_1f:
57 ; SKX: ## %bb.0:
58 ; SKX-NEXT: vmovdqa64 {{.*#+}} zmm1 = <2,5,u,u,7,u,10,1,0,5,u,4,7,u,10,1,2,5,u,u,7,u,10,1,0,5,u,4,7,u,10,31>
59 ; SKX-NEXT: vpermw %zmm0, %zmm1, %zmm0
60 ; SKX-NEXT: retq
61 %c = shufflevector <32 x i16> %a, <32 x i16> undef, <32 x i32> <i32 2, i32 5, i32 undef, i32 undef, i32 7, i32 undef, i32 10, i32 1, i32 0, i32 5, i32 undef, i32 4, i32 7, i32 undef, i32 10, i32 1, i32 2, i32 5, i32 undef, i32 undef, i32 7, i32 undef, i32 10, i32 1, i32 0, i32 5, i32 undef, i32 4, i32 7, i32 undef, i32 10, i32 31>
62 ret <32 x i16> %c
63 }
65 define <32 x i16> @shuffle_v32i16_0f_1f_0e_16_0d_1d_04_1e_0b_1b_0a_1a_09_19_08_18_0f_1f_0e_16_0d_1d_04_1e_0b_1b_0a_1a_09_19_08_38(<32 x i16> %a, <32 x i16> %b) {
66 ; KNL-LABEL: shuffle_v32i16_0f_1f_0e_16_0d_1d_04_1e_0b_1b_0a_1a_09_19_08_18_0f_1f_0e_16_0d_1d_04_1e_0b_1b_0a_1a_09_19_08_38:
67 ; KNL: ## %bb.0:
68 ; KNL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
69 ; KNL-NEXT: vpermq {{.*#+}} ymm2 = ymm2[3,1,2,3]
70 ; KNL-NEXT: vpshufb {{.*#+}} ymm3 = ymm2[6,7,12,13,2,3,0,1,u,u,u,u,u,u,u,u,22,23,20,21,18,19,u,u,u,u,u,u,u,u,u,u]
71 ; KNL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,1,2,3]
72 ; KNL-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[6,7,4,5,2,3,8,9,u,u,u,u,u,u,u,u,22,23,20,21,18,19,16,17,u,u,u,u,u,u,u,u]
73 ; KNL-NEXT: vpunpcklwd {{.*#+}} ymm3 = ymm0[0],ymm3[0],ymm0[1],ymm3[1],ymm0[2],ymm3[2],ymm0[3],ymm3[3],ymm0[8],ymm3[8],ymm0[9],ymm3[9],ymm0[10],ymm3[10],ymm0[11],ymm3[11]
74 ; KNL-NEXT: vextracti32x4 $3, %zmm1, %xmm1
75 ; KNL-NEXT: vpbroadcastw %xmm1, %ymm1
76 ; KNL-NEXT: vpblendw {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5,6],ymm1[7],ymm3[8,9,10,11,12,13,14],ymm1[15]
77 ; KNL-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
78 ; KNL-NEXT: vpshufb {{.*#+}} ymm2 = ymm2[6,7,12,13,2,3,0,1,u,u,u,u,u,u,u,u,22,23,20,21,18,19,16,17,u,u,u,u,u,u,u,u]
79 ; KNL-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm0[0],ymm2[0],ymm0[1],ymm2[1],ymm0[2],ymm2[2],ymm0[3],ymm2[3],ymm0[8],ymm2[8],ymm0[9],ymm2[9],ymm0[10],ymm2[10],ymm0[11],ymm2[11]
80 ; KNL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
81 ; KNL-NEXT: retq
82 ;
83 ; SKX-LABEL: shuffle_v32i16_0f_1f_0e_16_0d_1d_04_1e_0b_1b_0a_1a_09_19_08_18_0f_1f_0e_16_0d_1d_04_1e_0b_1b_0a_1a_09_19_08_38:
84 ; SKX: ## %bb.0:
85 ; SKX-NEXT: vmovdqa64 {{.*#+}} zmm2 = [15,31,14,22,13,29,4,28,11,27,10,26,9,25,8,24,15,31,14,22,13,29,4,28,11,27,10,26,9,25,8,56]
86 ; SKX-NEXT: vpermt2w %zmm1, %zmm2, %zmm0
87 ; SKX-NEXT: retq
88 %c = shufflevector <32 x i16> %a, <32 x i16> %b, <32 x i32> <i32 15, i32 31, i32 14, i32 22, i32 13, i32 29, i32 4, i32 28, i32 11, i32 27, i32 10, i32 26, i32 9, i32 25, i32 8, i32 24, i32 15, i32 31, i32 14, i32 22, i32 13, i32 29, i32 4, i32 28, i32 11, i32 27, i32 10, i32 26, i32 9, i32 25, i32 8, i32 56>
89 ret <32 x i16> %c
90 }
92 define <32 x i16> @shuffle_v16i32_0_32_1_33_2_34_3_35_8_40_9_41_u_u_u_u(<32 x i16> %a, <32 x i16> %b) {
93 ; ALL-LABEL: shuffle_v16i32_0_32_1_33_2_34_3_35_8_40_9_41_u_u_u_u:
94 ; ALL: ## %bb.0:
95 ; ALL-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
96 ; ALL-NEXT: retq
97 %c = shufflevector <32 x i16> %a, <32 x i16> %b, <32 x i32> <i32 0, i32 32, i32 1, i32 33, i32 2, i32 34, i32 3, i32 35, i32 8, i32 40, i32 9, i32 41, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
98 ret <32 x i16> %c
99 }
101 define <32 x i16> @shuffle_v16i32_4_36_5_37_6_38_7_39_12_44_13_45_u_u_u_u(<32 x i16> %a, <32 x i16> %b) {
102 ; ALL-LABEL: shuffle_v16i32_4_36_5_37_6_38_7_39_12_44_13_45_u_u_u_u:
103 ; ALL: ## %bb.0:
104 ; ALL-NEXT: vpunpckhwd {{.*#+}} ymm0 = ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15]
105 ; ALL-NEXT: retq
106 %c = shufflevector <32 x i16> %a, <32 x i16> %b, <32 x i32> <i32 4, i32 36, i32 5, i32 37, i32 6, i32 38, i32 7, i32 39, i32 12, i32 44, i32 13, i32 45, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
107 ret <32 x i16> %c
108 }
110 define <32 x i16> @shuffle_v32i16_1_z_3_z_5_z_7_z_9_z_11_z_13_z_15_z_17_z_19_z_21_z_23_z_25_z_27_z_29_z_31_z(<32 x i16> %a, <32 x i16> %b) {
111 ; ALL-LABEL: shuffle_v32i16_1_z_3_z_5_z_7_z_9_z_11_z_13_z_15_z_17_z_19_z_21_z_23_z_25_z_27_z_29_z_31_z:
112 ; ALL: ## %bb.0:
113 ; ALL-NEXT: vpsrld $16, %zmm0, %zmm0
114 ; ALL-NEXT: retq
115 %c = shufflevector <32 x i16> %a, <32 x i16> zeroinitializer, <32 x i32> <i32 1, i32 34, i32 3, i32 34, i32 5, i32 34, i32 7, i32 34, i32 9, i32 34, i32 11, i32 34, i32 13, i32 34, i32 15, i32 34, i32 17, i32 34, i32 19, i32 34, i32 21, i32 34, i32 23, i32 34, i32 25, i32 34, i32 27, i32 34, i32 29, i32 34, i32 31, i32 34>
116 ret <32 x i16> %c
117 }
119 define <32 x i16> @shuffle_v32i16_z_0_z_2_z_4_z_6_z_8_z_10_z_12_z_14_z_16_z_18_z_20_z_22_z_24_z_26_z_28_z_30(<32 x i16> %a, <32 x i16> %b) {
120 ; ALL-LABEL: shuffle_v32i16_z_0_z_2_z_4_z_6_z_8_z_10_z_12_z_14_z_16_z_18_z_20_z_22_z_24_z_26_z_28_z_30:
121 ; ALL: ## %bb.0:
122 ; ALL-NEXT: vpslld $16, %zmm0, %zmm0
123 ; ALL-NEXT: retq
124 %c = shufflevector <32 x i16> %a, <32 x i16> zeroinitializer, <32 x i32> <i32 34, i32 0, i32 34, i32 2, i32 34, i32 4, i32 34, i32 6, i32 34, i32 8, i32 34, i32 10, i32 34, i32 12, i32 34, i32 14, i32 34, i32 16, i32 34, i32 18, i32 34, i32 20, i32 34, i32 22, i32 34, i32 24, i32 34, i32 26, i32 34, i32 28, i32 34, i32 30>
125 ret <32 x i16> %c
126 }
128 define <32 x i16> @shuffle_v32i16_1_1_0_0_4_5_6_7_9_9_8_8_12_13_14_15_17_17_16_16_20_21_22_23_25_25_24_24_28_29_30_31(<32 x i16> %a, <32 x i16> %b) {
129 ; KNL-LABEL: shuffle_v32i16_1_1_0_0_4_5_6_7_9_9_8_8_12_13_14_15_17_17_16_16_20_21_22_23_25_25_24_24_28_29_30_31:
130 ; KNL: ## %bb.0:
131 ; KNL-NEXT: vpshuflw {{.*#+}} ymm1 = ymm0[1,1,0,0,4,5,6,7,9,9,8,8,12,13,14,15]
132 ; KNL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
133 ; KNL-NEXT: vpshuflw {{.*#+}} ymm0 = ymm0[1,1,0,0,4,5,6,7,9,9,8,8,12,13,14,15]
134 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
135 ; KNL-NEXT: retq
136 ;
137 ; SKX-LABEL: shuffle_v32i16_1_1_0_0_4_5_6_7_9_9_8_8_12_13_14_15_17_17_16_16_20_21_22_23_25_25_24_24_28_29_30_31:
138 ; SKX: ## %bb.0:
139 ; SKX-NEXT: vpshuflw {{.*#+}} zmm0 = zmm0[1,1,0,0,4,5,6,7,9,9,8,8,12,13,14,15,17,17,16,16,20,21,22,23,25,25,24,24,28,29,30,31]
140 ; SKX-NEXT: retq
141 %c = shufflevector <32 x i16> %a, <32 x i16> zeroinitializer, <32 x i32> <i32 1, i32 1, i32 0, i32 0, i32 4, i32 5, i32 6, i32 7, i32 9, i32 9, i32 8, i32 8, i32 12, i32 13, i32 14, i32 15, i32 17, i32 17, i32 16, i32 16, i32 20, i32 21, i32 22, i32 23, i32 25, i32 25, i32 24, i32 24, i32 28, i32 29, i32 30, i32 31>
142 ret <32 x i16> %c
143 }
145 define <32 x i16> @shuffle_v32i16_0_1_2_3_5_5_4_4_8_9_10_11_13_13_12_12_16_17_18_19_21_21_20_20_24_25_26_27_29_29_28_28(<32 x i16> %a, <32 x i16> %b) {
146 ; KNL-LABEL: shuffle_v32i16_0_1_2_3_5_5_4_4_8_9_10_11_13_13_12_12_16_17_18_19_21_21_20_20_24_25_26_27_29_29_28_28:
147 ; KNL: ## %bb.0:
148 ; KNL-NEXT: vpshufhw {{.*#+}} ymm1 = ymm0[0,1,2,3,5,5,4,4,8,9,10,11,13,13,12,12]
149 ; KNL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
150 ; KNL-NEXT: vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,5,5,4,4,8,9,10,11,13,13,12,12]
151 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
152 ; KNL-NEXT: retq
153 ;
154 ; SKX-LABEL: shuffle_v32i16_0_1_2_3_5_5_4_4_8_9_10_11_13_13_12_12_16_17_18_19_21_21_20_20_24_25_26_27_29_29_28_28:
155 ; SKX: ## %bb.0:
156 ; SKX-NEXT: vpshufhw {{.*#+}} zmm0 = zmm0[0,1,2,3,5,5,4,4,8,9,10,11,13,13,12,12,16,17,18,19,21,21,20,20,24,25,26,27,29,29,28,28]
157 ; SKX-NEXT: retq
158 %c = shufflevector <32 x i16> %a, <32 x i16> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 5, i32 4, i32 4, i32 8, i32 9, i32 10, i32 11, i32 13, i32 13, i32 12, i32 12, i32 16, i32 17, i32 18, i32 19, i32 21, i32 21, i32 20, i32 20, i32 24, i32 25, i32 26, i32 27, i32 29, i32 29, i32 28, i32 28>
159 ret <32 x i16> %c
160 }
162 define <32 x i16> @shuffle_v32i16_1_1_0_0_5_5_4_4_9_9_11_11_13_13_12_12_17_17_19_19_21_21_20_20_25_25_27_27_29_29_28_28(<32 x i16> %a, <32 x i16> %b) {
163 ; KNL-LABEL: shuffle_v32i16_1_1_0_0_5_5_4_4_9_9_11_11_13_13_12_12_17_17_19_19_21_21_20_20_25_25_27_27_29_29_28_28:
164 ; KNL: ## %bb.0:
165 ; KNL-NEXT: vpshuflw {{.*#+}} ymm1 = ymm0[1,1,0,0,4,5,6,7,9,9,8,8,12,13,14,15]
166 ; KNL-NEXT: vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,5,5,4,4,8,9,10,11,13,13,12,12]
167 ; KNL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
168 ; KNL-NEXT: vpshuflw {{.*#+}} ymm0 = ymm0[1,1,0,0,4,5,6,7,9,9,8,8,12,13,14,15]
169 ; KNL-NEXT: vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,5,5,4,4,8,9,10,11,13,13,12,12]
170 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
171 ; KNL-NEXT: retq
172 ;
173 ; SKX-LABEL: shuffle_v32i16_1_1_0_0_5_5_4_4_9_9_11_11_13_13_12_12_17_17_19_19_21_21_20_20_25_25_27_27_29_29_28_28:
174 ; SKX: ## %bb.0:
175 ; SKX-NEXT: vpshufb {{.*#+}} zmm0 = zmm0[2,3,2,3,0,1,0,1,10,11,10,11,8,9,8,9,18,19,18,19,16,17,16,17,26,27,26,27,24,25,24,25,34,35,34,35,32,33,32,33,42,43,42,43,40,41,40,41,50,51,50,51,48,49,48,49,58,59,58,59,56,57,56,57]
176 ; SKX-NEXT: retq
177 %c = shufflevector <32 x i16> %a, <32 x i16> zeroinitializer, <32 x i32> <i32 1, i32 1, i32 0, i32 0, i32 5, i32 5, i32 4, i32 4, i32 9, i32 9, i32 8, i32 8, i32 13, i32 13, i32 12, i32 12, i32 17, i32 17, i32 16, i32 16, i32 21, i32 21, i32 20, i32 20, i32 25, i32 25, i32 24, i32 24, i32 29, i32 29, i32 28, i32 28>
178 ret <32 x i16> %c
179 }
181 define <32 x i16> @shuffle_v32i16_01_00_03_02_05_04_07_06_09_08_11_10_13_12_15_14_17_16_19_18_21_20_23_22_25_24_27_26_29_28_31_30(<32 x i16> %a) {
182 ; ALL-LABEL: shuffle_v32i16_01_00_03_02_05_04_07_06_09_08_11_10_13_12_15_14_17_16_19_18_21_20_23_22_25_24_27_26_29_28_31_30:
183 ; ALL: ## %bb.0:
184 ; ALL-NEXT: vprold $16, %zmm0, %zmm0
185 ; ALL-NEXT: retq
186 %shuffle = shufflevector <32 x i16> %a, <32 x i16> undef, <32 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14, i32 17, i32 16, i32 19, i32 18, i32 21, i32 20, i32 23, i32 22, i32 25, i32 24, i32 27, i32 26, i32 29, i32 28, i32 31, i32 30>
187 ret <32 x i16> %shuffle
188 }
190 define <32 x i16> @shuffle_v32i16_03_00_01_02_07_04_05_06_11_08_09_10_15_12_13_14_19_16_17_18_23_20_21_22_27_24_25_26_31_28_29_30(<32 x i16> %a) {
191 ; ALL-LABEL: shuffle_v32i16_03_00_01_02_07_04_05_06_11_08_09_10_15_12_13_14_19_16_17_18_23_20_21_22_27_24_25_26_31_28_29_30:
192 ; ALL: ## %bb.0:
193 ; ALL-NEXT: vprolq $16, %zmm0, %zmm0
194 ; ALL-NEXT: retq
195 %shuffle = shufflevector <32 x i16> %a, <32 x i16> undef, <32 x i32> <i32 3, i32 0, i32 1, i32 2, i32 7, i32 4, i32 5, i32 6, i32 11, i32 8, i32 9, i32 10, i32 15, i32 12, i32 13, i32 14, i32 19, i32 16, i32 17, i32 18, i32 23, i32 20, i32 21, i32 22, i32 27, i32 24, i32 25, i32 26, i32 31, i32 28, i32 29, i32 30>
196 ret <32 x i16> %shuffle
197 }
199 define <32 x i16> @shuffle_v32i16_0zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz(<32 x i16> %a) {
200 ; KNL-LABEL: shuffle_v32i16_0zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz:
201 ; KNL: ## %bb.0:
202 ; KNL-NEXT: vmovdqa {{.*#+}} xmm1 = [65535,0,0,0]
203 ; KNL-NEXT: vpandq %zmm1, %zmm0, %zmm0
204 ; KNL-NEXT: retq
205 ;
206 ; SKX-LABEL: shuffle_v32i16_0zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz:
207 ; SKX: ## %bb.0:
208 ; SKX-NEXT: vmovaps {{.*#+}} xmm1 = [65535,0,0,0]
209 ; SKX-NEXT: vandps %zmm1, %zmm0, %zmm0
210 ; SKX-NEXT: retq
211 %shuffle = shufflevector <32 x i16> %a, <32 x i16> zeroinitializer, <32 x i32> <i32 0, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>
212 ret <32 x i16> %shuffle
213 }
215 define <32 x i16> @shuffle_v32i16_ashr_00_02_04_06_32_34_36_38_08_10_12_14_40_42_44_46_16_18_20_22_48_50_52_54_24_26_28_30_56_58_60_62(<16 x i32> %a0, <16 x i32> %a1) nounwind {
216 ; KNL-LABEL: shuffle_v32i16_ashr_00_02_04_06_32_34_36_38_08_10_12_14_40_42_44_46_16_18_20_22_48_50_52_54_24_26_28_30_56_58_60_62:
217 ; KNL: ## %bb.0:
218 ; KNL-NEXT: vpsrad $25, %zmm0, %zmm0
219 ; KNL-NEXT: vpsrad $25, %zmm1, %zmm1
220 ; KNL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
221 ; KNL-NEXT: vextracti64x4 $1, %zmm0, %ymm3
222 ; KNL-NEXT: vpackssdw %ymm2, %ymm3, %ymm2
223 ; KNL-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
224 ; KNL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
225 ; KNL-NEXT: retq
226 ;
227 ; SKX-LABEL: shuffle_v32i16_ashr_00_02_04_06_32_34_36_38_08_10_12_14_40_42_44_46_16_18_20_22_48_50_52_54_24_26_28_30_56_58_60_62:
228 ; SKX: ## %bb.0:
229 ; SKX-NEXT: vpsrad $25, %zmm0, %zmm0
230 ; SKX-NEXT: vpsrad $25, %zmm1, %zmm1
231 ; SKX-NEXT: vpackssdw %zmm1, %zmm0, %zmm0
232 ; SKX-NEXT: retq
233 %1 = ashr <16 x i32> %a0, <i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25>
234 %2 = ashr <16 x i32> %a1, <i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25>
235 %3 = bitcast <16 x i32> %1 to <32 x i16>
236 %4 = bitcast <16 x i32> %2 to <32 x i16>
237 %5 = shufflevector <32 x i16> %3, <32 x i16> %4, <32 x i32> <i32 0, i32 2, i32 4, i32 6, i32 32, i32 34, i32 36, i32 38, i32 8, i32 10, i32 12, i32 14, i32 40, i32 42, i32 44, i32 46, i32 16, i32 18, i32 20, i32 22, i32 48, i32 50, i32 52, i32 54, i32 24, i32 26, i32 28, i32 30, i32 56, i32 58, i32 60, i32 62>
238 ret <32 x i16> %5
239 }
241 define <32 x i16> @shuffle_v32i16_lshr_00_02_04_06_32_34_36_38_08_10_12_14_40_42_44_46_16_18_20_22_48_50_52_54_24_26_28_30_56_58_60_62(<16 x i32> %a0, <16 x i32> %a1) nounwind {
242 ; KNL-LABEL: shuffle_v32i16_lshr_00_02_04_06_32_34_36_38_08_10_12_14_40_42_44_46_16_18_20_22_48_50_52_54_24_26_28_30_56_58_60_62:
243 ; KNL: ## %bb.0:
244 ; KNL-NEXT: vpsrld $25, %zmm0, %zmm0
245 ; KNL-NEXT: vpsrld $25, %zmm1, %zmm1
246 ; KNL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
247 ; KNL-NEXT: vextracti64x4 $1, %zmm0, %ymm3
248 ; KNL-NEXT: vpackusdw %ymm2, %ymm3, %ymm2
249 ; KNL-NEXT: vpackusdw %ymm1, %ymm0, %ymm0
250 ; KNL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
251 ; KNL-NEXT: retq
252 ;
253 ; SKX-LABEL: shuffle_v32i16_lshr_00_02_04_06_32_34_36_38_08_10_12_14_40_42_44_46_16_18_20_22_48_50_52_54_24_26_28_30_56_58_60_62:
254 ; SKX: ## %bb.0:
255 ; SKX-NEXT: vpsrld $25, %zmm0, %zmm0
256 ; SKX-NEXT: vpsrld $25, %zmm1, %zmm1
257 ; SKX-NEXT: vpackusdw %zmm1, %zmm0, %zmm0
258 ; SKX-NEXT: retq
259 %1 = lshr <16 x i32> %a0, <i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25>
260 %2 = lshr <16 x i32> %a1, <i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25>
261 %3 = bitcast <16 x i32> %1 to <32 x i16>
262 %4 = bitcast <16 x i32> %2 to <32 x i16>
263 %5 = shufflevector <32 x i16> %3, <32 x i16> %4, <32 x i32> <i32 0, i32 2, i32 4, i32 6, i32 32, i32 34, i32 36, i32 38, i32 8, i32 10, i32 12, i32 14, i32 40, i32 42, i32 44, i32 46, i32 16, i32 18, i32 20, i32 22, i32 48, i32 50, i32 52, i32 54, i32 24, i32 26, i32 28, i32 30, i32 56, i32 58, i32 60, i32 62>
264 ret <32 x i16> %5
265 }
267 define <32 x i16> @insert_dup_mem_v32i16_i32(ptr %ptr) {
268 ; KNL-LABEL: insert_dup_mem_v32i16_i32:
269 ; KNL: ## %bb.0:
270 ; KNL-NEXT: vpbroadcastw (%rdi), %ymm0
271 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
272 ; KNL-NEXT: retq
273 ;
274 ; SKX-LABEL: insert_dup_mem_v32i16_i32:
275 ; SKX: ## %bb.0:
276 ; SKX-NEXT: vpbroadcastw (%rdi), %zmm0
277 ; SKX-NEXT: retq
278 %tmp = load i32, ptr %ptr, align 4
279 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0
280 %tmp2 = bitcast <4 x i32> %tmp1 to <8 x i16>
281 %tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> undef, <32 x i32> zeroinitializer
282 ret <32 x i16> %tmp3
283 }
285 define <32 x i16> @insert_dup_mem_v32i16_sext_i16(ptr %ptr) {
286 ; KNL-LABEL: insert_dup_mem_v32i16_sext_i16:
287 ; KNL: ## %bb.0:
288 ; KNL-NEXT: vpbroadcastw (%rdi), %ymm0
289 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
290 ; KNL-NEXT: retq
291 ;
292 ; SKX-LABEL: insert_dup_mem_v32i16_sext_i16:
293 ; SKX: ## %bb.0:
294 ; SKX-NEXT: vpbroadcastw (%rdi), %zmm0
295 ; SKX-NEXT: retq
296 %tmp = load i16, ptr %ptr, align 2
297 %tmp1 = sext i16 %tmp to i32
298 %tmp2 = insertelement <4 x i32> zeroinitializer, i32 %tmp1, i32 0
299 %tmp3 = bitcast <4 x i32> %tmp2 to <8 x i16>
300 %tmp4 = shufflevector <8 x i16> %tmp3, <8 x i16> undef, <32 x i32> zeroinitializer
301 ret <32 x i16> %tmp4
302 }
304 define <32 x i16> @insert_dup_elt1_mem_v32i16_i32(ptr %ptr) #0 {
305 ; KNL-LABEL: insert_dup_elt1_mem_v32i16_i32:
306 ; KNL: ## %bb.0:
307 ; KNL-NEXT: vpbroadcastw 2(%rdi), %ymm0
308 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
309 ; KNL-NEXT: retq
310 ;
311 ; SKX-LABEL: insert_dup_elt1_mem_v32i16_i32:
312 ; SKX: ## %bb.0:
313 ; SKX-NEXT: vpbroadcastw 2(%rdi), %zmm0
314 ; SKX-NEXT: retq
315 %tmp = load i32, ptr %ptr, align 4
316 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0
317 %tmp2 = bitcast <4 x i32> %tmp1 to <8 x i16>
318 %tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> undef, <32 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
319 ret <32 x i16> %tmp3
320 }
322 define <32 x i16> @insert_dup_elt3_mem_v32i16_i32(ptr %ptr) #0 {
323 ; KNL-LABEL: insert_dup_elt3_mem_v32i16_i32:
324 ; KNL: ## %bb.0:
325 ; KNL-NEXT: vpbroadcastw 2(%rdi), %ymm0
326 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
327 ; KNL-NEXT: retq
328 ;
329 ; SKX-LABEL: insert_dup_elt3_mem_v32i16_i32:
330 ; SKX: ## %bb.0:
331 ; SKX-NEXT: vpbroadcastw 2(%rdi), %zmm0
332 ; SKX-NEXT: retq
333 %tmp = load i32, ptr %ptr, align 4
334 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 1
335 %tmp2 = bitcast <4 x i32> %tmp1 to <8 x i16>
336 %tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> undef, <32 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
337 ret <32 x i16> %tmp3
338 }
340 define <32 x i16> @insert_dup_mem_v16i16_i64(ptr %ptr) {
341 ; KNL-LABEL: insert_dup_mem_v16i16_i64:
342 ; KNL: ## %bb.0:
343 ; KNL-NEXT: vpbroadcastw (%rdi), %ymm0
344 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
345 ; KNL-NEXT: retq
346 ;
347 ; SKX-LABEL: insert_dup_mem_v16i16_i64:
348 ; SKX: ## %bb.0:
349 ; SKX-NEXT: vpbroadcastw (%rdi), %zmm0
350 ; SKX-NEXT: retq
351 %tmp = load i64, ptr %ptr, align 4
352 %tmp1 = insertelement <2 x i64> zeroinitializer, i64 %tmp, i32 0
353 %tmp2 = bitcast <2 x i64> %tmp1 to <8 x i16>
354 %tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> undef, <32 x i32> zeroinitializer
355 ret <32 x i16> %tmp3
356 }
358 define <32 x i16> @insert_dup_elt1_mem_v16i16_i64(ptr %ptr) {
359 ; KNL-LABEL: insert_dup_elt1_mem_v16i16_i64:
360 ; KNL: ## %bb.0:
361 ; KNL-NEXT: vpbroadcastw 2(%rdi), %ymm0
362 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
363 ; KNL-NEXT: retq
364 ;
365 ; SKX-LABEL: insert_dup_elt1_mem_v16i16_i64:
366 ; SKX: ## %bb.0:
367 ; SKX-NEXT: vpbroadcastw 2(%rdi), %zmm0
368 ; SKX-NEXT: retq
369 %tmp = load i64, ptr %ptr, align 4
370 %tmp1 = insertelement <2 x i64> zeroinitializer, i64 %tmp, i32 0
371 %tmp2 = bitcast <2 x i64> %tmp1 to <8 x i16>
372 %tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> undef, <32 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
373 ret <32 x i16> %tmp3
374 }
376 define <32 x i16> @insert_dup_elt3_mem_v16i16_i64(ptr %ptr) {
377 ; KNL-LABEL: insert_dup_elt3_mem_v16i16_i64:
378 ; KNL: ## %bb.0:
379 ; KNL-NEXT: vpbroadcastw 6(%rdi), %ymm0
380 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
381 ; KNL-NEXT: retq
382 ;
383 ; SKX-LABEL: insert_dup_elt3_mem_v16i16_i64:
384 ; SKX: ## %bb.0:
385 ; SKX-NEXT: vpbroadcastw 6(%rdi), %zmm0
386 ; SKX-NEXT: retq
387 %tmp = load i64, ptr %ptr, align 4
388 %tmp1 = insertelement <2 x i64> zeroinitializer, i64 %tmp, i32 0
389 %tmp2 = bitcast <2 x i64> %tmp1 to <8 x i16>
390 %tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> undef, <32 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
391 ret <32 x i16> %tmp3
392 }
394 define <32 x i16> @insert_dup_elt7_mem_v16i16_i64(ptr %ptr) {
395 ; KNL-LABEL: insert_dup_elt7_mem_v16i16_i64:
396 ; KNL: ## %bb.0:
397 ; KNL-NEXT: vpbroadcastw 6(%rdi), %ymm0
398 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
399 ; KNL-NEXT: retq
400 ;
401 ; SKX-LABEL: insert_dup_elt7_mem_v16i16_i64:
402 ; SKX: ## %bb.0:
403 ; SKX-NEXT: vpbroadcastw 6(%rdi), %zmm0
404 ; SKX-NEXT: retq
405 %tmp = load i64, ptr %ptr, align 4
406 %tmp1 = insertelement <2 x i64> zeroinitializer, i64 %tmp, i32 1
407 %tmp2 = bitcast <2 x i64> %tmp1 to <8 x i16>
408 %tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> undef, <32 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
409 ret <32 x i16> %tmp3
410 }
412 define <32 x i16> @insert_dup_mem_v16i16_sext_i16_i64(ptr %ptr) {
413 ; KNL-LABEL: insert_dup_mem_v16i16_sext_i16_i64:
414 ; KNL: ## %bb.0:
415 ; KNL-NEXT: vpbroadcastw (%rdi), %ymm0
416 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
417 ; KNL-NEXT: retq
418 ;
419 ; SKX-LABEL: insert_dup_mem_v16i16_sext_i16_i64:
420 ; SKX: ## %bb.0:
421 ; SKX-NEXT: vpbroadcastw (%rdi), %zmm0
422 ; SKX-NEXT: retq
423 %tmp = load i16, ptr %ptr, align 2
424 %tmp1 = sext i16 %tmp to i64
425 %tmp2 = insertelement <2 x i64> zeroinitializer, i64 %tmp1, i32 0
426 %tmp3 = bitcast <2 x i64> %tmp2 to <8 x i16>
427 %tmp4 = shufflevector <8 x i16> %tmp3, <8 x i16> undef, <32 x i32> zeroinitializer
428 ret <32 x i16> %tmp4
429 }
431 define <32 x i16> @shuffle_v32i16_32_zz_zz_zz_33_zz_zz_zz_34_zz_zz_zz_35_zz_zz_zz_36_zz_zz_zz_37_zz_zz_zz_38_zz_zz_zz_39_zz_zz_zz(<32 x i16> %a) {
432 ; ALL-LABEL: shuffle_v32i16_32_zz_zz_zz_33_zz_zz_zz_34_zz_zz_zz_35_zz_zz_zz_36_zz_zz_zz_37_zz_zz_zz_38_zz_zz_zz_39_zz_zz_zz:
433 ; ALL: ## %bb.0:
434 ; ALL-NEXT: vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
435 ; ALL-NEXT: retq
436 %shuffle = shufflevector <32 x i16> zeroinitializer, <32 x i16> %a, <32 x i32> <i32 32, i32 0, i32 0, i32 0, i32 33, i32 0, i32 0, i32 0, i32 34, i32 0, i32 0, i32 0, i32 35, i32 0, i32 0, i32 0, i32 36, i32 0, i32 0, i32 0, i32 37, i32 0, i32 0, i32 0, i32 38, i32 0, i32 0, i32 0, i32 39, i32 0, i32 0, i32 0>
437 ret <32 x i16> %shuffle
438 }
440 define <32 x i16> @shuffle_v32i16_32_zz_33_zz_34_zz_35_zz_36_zz_37_zz_38_zz_39_zz_40_zz_41_zz_42_zz_43_zz_44_zz_45_zz_46_zz_47_zz(<32 x i16> %a) {
441 ; ALL-LABEL: shuffle_v32i16_32_zz_33_zz_34_zz_35_zz_36_zz_37_zz_38_zz_39_zz_40_zz_41_zz_42_zz_43_zz_44_zz_45_zz_46_zz_47_zz:
442 ; ALL: ## %bb.0:
443 ; ALL-NEXT: vpmovzxwd {{.*#+}} 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
444 ; ALL-NEXT: retq
445 %shuffle = shufflevector <32 x i16> zeroinitializer, <32 x i16> %a, <32 x i32> <i32 32, i32 0, i32 33, i32 0, i32 34, i32 0, i32 35, i32 0, i32 36, i32 0, i32 37, i32 0, i32 38, i32 0, i32 39, i32 0, i32 40, i32 0, i32 41, i32 0, i32 42, i32 0, i32 43, i32 0, i32 44, i32 0, i32 45, i32 0, i32 46, i32 0, i32 47, i32 0>
446 ret <32 x i16> %shuffle
447 }
449 define <8 x i16> @pr32967(<32 x i16> %v) {
450 ; KNL-LABEL: pr32967:
451 ; KNL: ## %bb.0:
452 ; KNL-NEXT: vpsrlq $16, %zmm0, %zmm0
453 ; KNL-NEXT: vpmovqw %zmm0, %xmm0
454 ; KNL-NEXT: retq
455 ;
456 ; SKX-LABEL: pr32967:
457 ; SKX: ## %bb.0:
458 ; SKX-NEXT: vpsrlq $16, %zmm0, %zmm0
459 ; SKX-NEXT: vpmovqw %zmm0, %xmm0
460 ; SKX-NEXT: vzeroupper
461 ; SKX-NEXT: retq
462 %shuffle = shufflevector <32 x i16> %v, <32 x i16> undef, <8 x i32> <i32 1,i32 5,i32 9,i32 13,i32 17,i32 21,i32 25,i32 29>
463 ret <8 x i16> %shuffle
464 }
466 define <32 x i16> @shuffle_v32i16_07_zz_05_zz_03_zz_01_zz_15_zz_13_zz_11_zz_09_zz_23_zz_21_zz_19_zz_17_zz_31_zz_29_zz_27_zz_25_zz(<32 x i16> %a) {
467 ; KNL-LABEL: shuffle_v32i16_07_zz_05_zz_03_zz_01_zz_15_zz_13_zz_11_zz_09_zz_23_zz_21_zz_19_zz_17_zz_31_zz_29_zz_27_zz_25_zz:
468 ; KNL: ## %bb.0:
469 ; KNL-NEXT: vpshufb {{.*#+}} ymm1 = ymm0[14,15],zero,zero,ymm0[10,11],zero,zero,ymm0[6,7],zero,zero,ymm0[2,3],zero,zero,ymm0[30,31],zero,zero,ymm0[26,27],zero,zero,ymm0[22,23],zero,zero,ymm0[18,19],zero,zero
470 ; KNL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
471 ; KNL-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[14,15],zero,zero,ymm0[10,11],zero,zero,ymm0[6,7],zero,zero,ymm0[2,3],zero,zero,ymm0[30,31],zero,zero,ymm0[26,27],zero,zero,ymm0[22,23],zero,zero,ymm0[20,21],zero,zero
472 ; KNL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
473 ; KNL-NEXT: retq
474 ;
475 ; SKX-LABEL: shuffle_v32i16_07_zz_05_zz_03_zz_01_zz_15_zz_13_zz_11_zz_09_zz_23_zz_21_zz_19_zz_17_zz_31_zz_29_zz_27_zz_25_zz:
476 ; SKX: ## %bb.0:
477 ; SKX-NEXT: vpshufb {{.*#+}} zmm0 = zmm0[14,15],zero,zero,zmm0[10,11],zero,zero,zmm0[6,7],zero,zero,zmm0[2,3],zero,zero,zmm0[30,31],zero,zero,zmm0[26,27],zero,zero,zmm0[22,23],zero,zero,zmm0[18,19],zero,zero,zmm0[46,47],zero,zero,zmm0[42,43],zero,zero,zmm0[38,39],zero,zero,zmm0[34,35],zero,zero,zmm0[62,63],zero,zero,zmm0[58,59],zero,zero,zmm0[54,55],zero,zero,zmm0[52,53],zero,zero
478 ; SKX-NEXT: retq
479 %shuffle = shufflevector <32 x i16> zeroinitializer, <32 x i16> %a, <32 x i32> <i32 39, i32 0, i32 37, i32 0, i32 35, i32 0, i32 33, i32 0, i32 47, i32 0, i32 45, i32 0, i32 43, i32 0, i32 41, i32 0, i32 55, i32 0, i32 53, i32 0, i32 51, i32 0, i32 49, i32 0, i32 63, i32 0, i32 61, i32 0, i32 59, i32 0, i32 58, i32 0>
480 ret <32 x i16> %shuffle
481 }