1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512ifma,+avx512vl --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512ifma,+avx512vl --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64
5 declare <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(<2 x i64>, <2 x i64>, <2 x i64>)
7 define <2 x i64>@test_int_x86_avx512_mask_vpmadd52h_uq_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3) {
8 ; X86-LABEL: test_int_x86_avx512_mask_vpmadd52h_uq_128:
10 ; X86-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
11 ; X86-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm3 # encoding: [0x62,0xf2,0xf5,0x08,0xb5,0xda]
12 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
13 ; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
14 ; X86-NEXT: vmovdqa %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xe0]
15 ; X86-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm4 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xb5,0xe2]
16 ; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
17 ; X86-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xb5,0xc2]
18 ; X86-NEXT: vpaddq %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xd4,0xc0]
19 ; X86-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb5,0xd2]
20 ; X86-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
21 ; X86-NEXT: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
22 ; X86-NEXT: retl # encoding: [0xc3]
24 ; X64-LABEL: test_int_x86_avx512_mask_vpmadd52h_uq_128:
26 ; X64-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
27 ; X64-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm3 # encoding: [0x62,0xf2,0xf5,0x08,0xb5,0xda]
28 ; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
29 ; X64-NEXT: vmovdqa %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xe0]
30 ; X64-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm4 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xb5,0xe2]
31 ; X64-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
32 ; X64-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xb5,0xc2]
33 ; X64-NEXT: vpaddq %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xd4,0xc0]
34 ; X64-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb5,0xd2]
35 ; X64-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
36 ; X64-NEXT: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
37 ; X64-NEXT: retq # encoding: [0xc3]
39 %1 = call <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2)
40 %2 = bitcast i8 %x3 to <8 x i1>
41 %extract2 = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
42 %3 = select <2 x i1> %extract2, <2 x i64> %1, <2 x i64> %x0
43 %4 = call <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> zeroinitializer)
44 %5 = bitcast i8 %x3 to <8 x i1>
45 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <2 x i32> <i32 0, i32 1>
46 %6 = select <2 x i1> %extract1, <2 x i64> %4, <2 x i64> %x0
47 %7 = call <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(<2 x i64> zeroinitializer, <2 x i64> %x1, <2 x i64> zeroinitializer)
48 %8 = bitcast i8 %x3 to <8 x i1>
49 %extract = shufflevector <8 x i1> %8, <8 x i1> %8, <2 x i32> <i32 0, i32 1>
50 %9 = select <2 x i1> %extract, <2 x i64> %7, <2 x i64> zeroinitializer
51 %10 = call <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2)
52 %res4 = add <2 x i64> %3, %6
53 %res5 = add <2 x i64> %10, %9
54 %res6 = add <2 x i64> %res5, %res4
58 declare <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(<4 x i64>, <4 x i64>, <4 x i64>)
60 define <4 x i64>@test_int_x86_avx512_mask_vpmadd52h_uq_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %x3) {
61 ; X86-LABEL: test_int_x86_avx512_mask_vpmadd52h_uq_256:
63 ; X86-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
64 ; X86-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm3 # encoding: [0x62,0xf2,0xf5,0x28,0xb5,0xda]
65 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
66 ; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
67 ; X86-NEXT: vmovdqa %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xe0]
68 ; X86-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm4 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xb5,0xe2]
69 ; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
70 ; X86-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xb5,0xc2]
71 ; X86-NEXT: vpaddq %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xd4,0xc0]
72 ; X86-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb5,0xd2]
73 ; X86-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
74 ; X86-NEXT: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
75 ; X86-NEXT: retl # encoding: [0xc3]
77 ; X64-LABEL: test_int_x86_avx512_mask_vpmadd52h_uq_256:
79 ; X64-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
80 ; X64-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm3 # encoding: [0x62,0xf2,0xf5,0x28,0xb5,0xda]
81 ; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
82 ; X64-NEXT: vmovdqa %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xe0]
83 ; X64-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm4 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xb5,0xe2]
84 ; X64-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
85 ; X64-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xb5,0xc2]
86 ; X64-NEXT: vpaddq %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xd4,0xc0]
87 ; X64-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb5,0xd2]
88 ; X64-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
89 ; X64-NEXT: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
90 ; X64-NEXT: retq # encoding: [0xc3]
92 %1 = call <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2)
93 %2 = bitcast i8 %x3 to <8 x i1>
94 %extract2 = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
95 %3 = select <4 x i1> %extract2, <4 x i64> %1, <4 x i64> %x0
96 %4 = call <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> zeroinitializer)
97 %5 = bitcast i8 %x3 to <8 x i1>
98 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
99 %6 = select <4 x i1> %extract1, <4 x i64> %4, <4 x i64> %x0
100 %7 = call <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(<4 x i64> zeroinitializer, <4 x i64> %x1, <4 x i64> zeroinitializer)
101 %8 = bitcast i8 %x3 to <8 x i1>
102 %extract = shufflevector <8 x i1> %8, <8 x i1> %8, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
103 %9 = select <4 x i1> %extract, <4 x i64> %7, <4 x i64> zeroinitializer
104 %10 = call <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2)
105 %res4 = add <4 x i64> %3, %6
106 %res5 = add <4 x i64> %10, %9
107 %res6 = add <4 x i64> %res5, %res4
111 define <2 x i64>@test_int_x86_avx512_maskz_vpmadd52h_uq_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3) {
112 ; X86-LABEL: test_int_x86_avx512_maskz_vpmadd52h_uq_128:
114 ; X86-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
115 ; X86-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm3 # encoding: [0x62,0xf2,0xf5,0x08,0xb5,0xda]
116 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
117 ; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
118 ; X86-NEXT: vmovdqa %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xe0]
119 ; X86-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm4 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb5,0xe2]
120 ; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
121 ; X86-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb5,0xc2]
122 ; X86-NEXT: vpaddq %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xd4,0xc0]
123 ; X86-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb5,0xd2]
124 ; X86-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
125 ; X86-NEXT: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
126 ; X86-NEXT: retl # encoding: [0xc3]
128 ; X64-LABEL: test_int_x86_avx512_maskz_vpmadd52h_uq_128:
130 ; X64-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
131 ; X64-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm3 # encoding: [0x62,0xf2,0xf5,0x08,0xb5,0xda]
132 ; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
133 ; X64-NEXT: vmovdqa %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xe0]
134 ; X64-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm4 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb5,0xe2]
135 ; X64-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
136 ; X64-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb5,0xc2]
137 ; X64-NEXT: vpaddq %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xd4,0xc0]
138 ; X64-NEXT: vpmadd52huq %xmm2, %xmm1, %xmm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb5,0xd2]
139 ; X64-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
140 ; X64-NEXT: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
141 ; X64-NEXT: retq # encoding: [0xc3]
143 %1 = call <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2)
144 %2 = bitcast i8 %x3 to <8 x i1>
145 %extract2 = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
146 %3 = select <2 x i1> %extract2, <2 x i64> %1, <2 x i64> zeroinitializer
147 %4 = call <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> zeroinitializer)
148 %5 = bitcast i8 %x3 to <8 x i1>
149 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <2 x i32> <i32 0, i32 1>
150 %6 = select <2 x i1> %extract1, <2 x i64> %4, <2 x i64> zeroinitializer
151 %7 = call <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(<2 x i64> zeroinitializer, <2 x i64> %x1, <2 x i64> zeroinitializer)
152 %8 = bitcast i8 %x3 to <8 x i1>
153 %extract = shufflevector <8 x i1> %8, <8 x i1> %8, <2 x i32> <i32 0, i32 1>
154 %9 = select <2 x i1> %extract, <2 x i64> %7, <2 x i64> zeroinitializer
155 %10 = call <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2)
156 %res4 = add <2 x i64> %3, %6
157 %res5 = add <2 x i64> %10, %9
158 %res6 = add <2 x i64> %res5, %res4
162 define <4 x i64>@test_int_x86_avx512_maskz_vpmadd52h_uq_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %x3) {
163 ; X86-LABEL: test_int_x86_avx512_maskz_vpmadd52h_uq_256:
165 ; X86-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
166 ; X86-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm3 # encoding: [0x62,0xf2,0xf5,0x28,0xb5,0xda]
167 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
168 ; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
169 ; X86-NEXT: vmovdqa %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xe0]
170 ; X86-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm4 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb5,0xe2]
171 ; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
172 ; X86-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb5,0xc2]
173 ; X86-NEXT: vpaddq %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xd4,0xc0]
174 ; X86-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb5,0xd2]
175 ; X86-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
176 ; X86-NEXT: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
177 ; X86-NEXT: retl # encoding: [0xc3]
179 ; X64-LABEL: test_int_x86_avx512_maskz_vpmadd52h_uq_256:
181 ; X64-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
182 ; X64-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm3 # encoding: [0x62,0xf2,0xf5,0x28,0xb5,0xda]
183 ; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
184 ; X64-NEXT: vmovdqa %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xe0]
185 ; X64-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm4 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb5,0xe2]
186 ; X64-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
187 ; X64-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb5,0xc2]
188 ; X64-NEXT: vpaddq %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xd4,0xc0]
189 ; X64-NEXT: vpmadd52huq %ymm2, %ymm1, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb5,0xd2]
190 ; X64-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
191 ; X64-NEXT: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
192 ; X64-NEXT: retq # encoding: [0xc3]
194 %1 = call <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2)
195 %2 = bitcast i8 %x3 to <8 x i1>
196 %extract2 = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
197 %3 = select <4 x i1> %extract2, <4 x i64> %1, <4 x i64> zeroinitializer
198 %4 = call <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> zeroinitializer)
199 %5 = bitcast i8 %x3 to <8 x i1>
200 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
201 %6 = select <4 x i1> %extract1, <4 x i64> %4, <4 x i64> zeroinitializer
202 %7 = call <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(<4 x i64> zeroinitializer, <4 x i64> %x1, <4 x i64> zeroinitializer)
203 %8 = bitcast i8 %x3 to <8 x i1>
204 %extract = shufflevector <8 x i1> %8, <8 x i1> %8, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
205 %9 = select <4 x i1> %extract, <4 x i64> %7, <4 x i64> zeroinitializer
206 %10 = call <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2)
207 %res4 = add <4 x i64> %3, %6
208 %res5 = add <4 x i64> %10, %9
209 %res6 = add <4 x i64> %res5, %res4
213 declare <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(<2 x i64>, <2 x i64>, <2 x i64>)
215 define <2 x i64>@test_int_x86_avx512_mask_vpmadd52l_uq_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3) {
216 ; X86-LABEL: test_int_x86_avx512_mask_vpmadd52l_uq_128:
218 ; X86-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
219 ; X86-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm3 # encoding: [0x62,0xf2,0xf5,0x08,0xb4,0xda]
220 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
221 ; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
222 ; X86-NEXT: vmovdqa %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xe0]
223 ; X86-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm4 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xb4,0xe2]
224 ; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
225 ; X86-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xb4,0xc2]
226 ; X86-NEXT: vpaddq %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xd4,0xc0]
227 ; X86-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb4,0xd2]
228 ; X86-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
229 ; X86-NEXT: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
230 ; X86-NEXT: retl # encoding: [0xc3]
232 ; X64-LABEL: test_int_x86_avx512_mask_vpmadd52l_uq_128:
234 ; X64-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
235 ; X64-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm3 # encoding: [0x62,0xf2,0xf5,0x08,0xb4,0xda]
236 ; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
237 ; X64-NEXT: vmovdqa %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xe0]
238 ; X64-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm4 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xb4,0xe2]
239 ; X64-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
240 ; X64-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xb4,0xc2]
241 ; X64-NEXT: vpaddq %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xd4,0xc0]
242 ; X64-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb4,0xd2]
243 ; X64-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
244 ; X64-NEXT: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
245 ; X64-NEXT: retq # encoding: [0xc3]
247 %1 = call <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2)
248 %2 = bitcast i8 %x3 to <8 x i1>
249 %extract2 = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
250 %3 = select <2 x i1> %extract2, <2 x i64> %1, <2 x i64> %x0
251 %4 = call <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> zeroinitializer)
252 %5 = bitcast i8 %x3 to <8 x i1>
253 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <2 x i32> <i32 0, i32 1>
254 %6 = select <2 x i1> %extract1, <2 x i64> %4, <2 x i64> %x0
255 %7 = call <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(<2 x i64> zeroinitializer, <2 x i64> %x1, <2 x i64> zeroinitializer)
256 %8 = bitcast i8 %x3 to <8 x i1>
257 %extract = shufflevector <8 x i1> %8, <8 x i1> %8, <2 x i32> <i32 0, i32 1>
258 %9 = select <2 x i1> %extract, <2 x i64> %7, <2 x i64> zeroinitializer
259 %10 = call <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2)
260 %res4 = add <2 x i64> %3, %6
261 %res5 = add <2 x i64> %10, %9
262 %res6 = add <2 x i64> %res5, %res4
266 declare <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(<4 x i64>, <4 x i64>, <4 x i64>)
268 define <4 x i64>@test_int_x86_avx512_mask_vpmadd52l_uq_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %x3) {
269 ; X86-LABEL: test_int_x86_avx512_mask_vpmadd52l_uq_256:
271 ; X86-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
272 ; X86-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm3 # encoding: [0x62,0xf2,0xf5,0x28,0xb4,0xda]
273 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
274 ; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
275 ; X86-NEXT: vmovdqa %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xe0]
276 ; X86-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm4 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xb4,0xe2]
277 ; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
278 ; X86-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xb4,0xc2]
279 ; X86-NEXT: vpaddq %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xd4,0xc0]
280 ; X86-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb4,0xd2]
281 ; X86-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
282 ; X86-NEXT: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
283 ; X86-NEXT: retl # encoding: [0xc3]
285 ; X64-LABEL: test_int_x86_avx512_mask_vpmadd52l_uq_256:
287 ; X64-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
288 ; X64-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm3 # encoding: [0x62,0xf2,0xf5,0x28,0xb4,0xda]
289 ; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
290 ; X64-NEXT: vmovdqa %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xe0]
291 ; X64-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm4 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xb4,0xe2]
292 ; X64-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
293 ; X64-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xb4,0xc2]
294 ; X64-NEXT: vpaddq %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xd4,0xc0]
295 ; X64-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb4,0xd2]
296 ; X64-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
297 ; X64-NEXT: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
298 ; X64-NEXT: retq # encoding: [0xc3]
300 %1 = call <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2)
301 %2 = bitcast i8 %x3 to <8 x i1>
302 %extract2 = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
303 %3 = select <4 x i1> %extract2, <4 x i64> %1, <4 x i64> %x0
304 %4 = call <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> zeroinitializer)
305 %5 = bitcast i8 %x3 to <8 x i1>
306 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
307 %6 = select <4 x i1> %extract1, <4 x i64> %4, <4 x i64> %x0
308 %7 = call <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(<4 x i64> zeroinitializer, <4 x i64> %x1, <4 x i64> zeroinitializer)
309 %8 = bitcast i8 %x3 to <8 x i1>
310 %extract = shufflevector <8 x i1> %8, <8 x i1> %8, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
311 %9 = select <4 x i1> %extract, <4 x i64> %7, <4 x i64> zeroinitializer
312 %10 = call <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2)
313 %res4 = add <4 x i64> %3, %6
314 %res5 = add <4 x i64> %10, %9
315 %res6 = add <4 x i64> %res5, %res4
319 define <2 x i64>@test_int_x86_avx512_maskz_vpmadd52l_uq_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3) {
320 ; X86-LABEL: test_int_x86_avx512_maskz_vpmadd52l_uq_128:
322 ; X86-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
323 ; X86-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm3 # encoding: [0x62,0xf2,0xf5,0x08,0xb4,0xda]
324 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
325 ; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
326 ; X86-NEXT: vmovdqa %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xe0]
327 ; X86-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm4 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb4,0xe2]
328 ; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
329 ; X86-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb4,0xc2]
330 ; X86-NEXT: vpaddq %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xd4,0xc0]
331 ; X86-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb4,0xd2]
332 ; X86-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
333 ; X86-NEXT: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
334 ; X86-NEXT: retl # encoding: [0xc3]
336 ; X64-LABEL: test_int_x86_avx512_maskz_vpmadd52l_uq_128:
338 ; X64-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
339 ; X64-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm3 # encoding: [0x62,0xf2,0xf5,0x08,0xb4,0xda]
340 ; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
341 ; X64-NEXT: vmovdqa %xmm0, %xmm4 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xe0]
342 ; X64-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm4 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb4,0xe2]
343 ; X64-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
344 ; X64-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb4,0xc2]
345 ; X64-NEXT: vpaddq %xmm0, %xmm4, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xd9,0xd4,0xc0]
346 ; X64-NEXT: vpmadd52luq %xmm2, %xmm1, %xmm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xb4,0xd2]
347 ; X64-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
348 ; X64-NEXT: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
349 ; X64-NEXT: retq # encoding: [0xc3]
351 %1 = call <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2)
352 %2 = bitcast i8 %x3 to <8 x i1>
353 %extract2 = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
354 %3 = select <2 x i1> %extract2, <2 x i64> %1, <2 x i64> zeroinitializer
355 %4 = call <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> zeroinitializer)
356 %5 = bitcast i8 %x3 to <8 x i1>
357 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <2 x i32> <i32 0, i32 1>
358 %6 = select <2 x i1> %extract1, <2 x i64> %4, <2 x i64> zeroinitializer
359 %7 = call <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(<2 x i64> zeroinitializer, <2 x i64> %x1, <2 x i64> zeroinitializer)
360 %8 = bitcast i8 %x3 to <8 x i1>
361 %extract = shufflevector <8 x i1> %8, <8 x i1> %8, <2 x i32> <i32 0, i32 1>
362 %9 = select <2 x i1> %extract, <2 x i64> %7, <2 x i64> zeroinitializer
363 %10 = call <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2)
364 %res4 = add <2 x i64> %3, %6
365 %res5 = add <2 x i64> %10, %9
366 %res6 = add <2 x i64> %res5, %res4
370 define <4 x i64>@test_int_x86_avx512_maskz_vpmadd52l_uq_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %x3) {
371 ; X86-LABEL: test_int_x86_avx512_maskz_vpmadd52l_uq_256:
373 ; X86-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
374 ; X86-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm3 # encoding: [0x62,0xf2,0xf5,0x28,0xb4,0xda]
375 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
376 ; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
377 ; X86-NEXT: vmovdqa %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xe0]
378 ; X86-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm4 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb4,0xe2]
379 ; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
380 ; X86-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb4,0xc2]
381 ; X86-NEXT: vpaddq %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xd4,0xc0]
382 ; X86-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb4,0xd2]
383 ; X86-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
384 ; X86-NEXT: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
385 ; X86-NEXT: retl # encoding: [0xc3]
387 ; X64-LABEL: test_int_x86_avx512_maskz_vpmadd52l_uq_256:
389 ; X64-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
390 ; X64-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm3 # encoding: [0x62,0xf2,0xf5,0x28,0xb4,0xda]
391 ; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
392 ; X64-NEXT: vmovdqa %ymm0, %ymm4 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xe0]
393 ; X64-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm4 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb4,0xe2]
394 ; X64-NEXT: vpxor %xmm2, %xmm2, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2]
395 ; X64-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb4,0xc2]
396 ; X64-NEXT: vpaddq %ymm0, %ymm4, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xdd,0xd4,0xc0]
397 ; X64-NEXT: vpmadd52luq %ymm2, %ymm1, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xb4,0xd2]
398 ; X64-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
399 ; X64-NEXT: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
400 ; X64-NEXT: retq # encoding: [0xc3]
402 %1 = call <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2)
403 %2 = bitcast i8 %x3 to <8 x i1>
404 %extract2 = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
405 %3 = select <4 x i1> %extract2, <4 x i64> %1, <4 x i64> zeroinitializer
406 %4 = call <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> zeroinitializer)
407 %5 = bitcast i8 %x3 to <8 x i1>
408 %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
409 %6 = select <4 x i1> %extract1, <4 x i64> %4, <4 x i64> zeroinitializer
410 %7 = call <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(<4 x i64> zeroinitializer, <4 x i64> %x1, <4 x i64> zeroinitializer)
411 %8 = bitcast i8 %x3 to <8 x i1>
412 %extract = shufflevector <8 x i1> %8, <8 x i1> %8, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
413 %9 = select <4 x i1> %extract, <4 x i64> %7, <4 x i64> zeroinitializer
414 %10 = call <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2)
415 %res4 = add <4 x i64> %3, %6
416 %res5 = add <4 x i64> %10, %9
417 %res6 = add <4 x i64> %res5, %res4