1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512vnni --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vnni --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64
4
5declare <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32>, <16 x i32>, <16 x i32>)
6
7define <16 x i32>@test_int_x86_avx512_ask_vpdpbusd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
8; CHECK-LABEL: test_int_x86_avx512_ask_vpdpbusd_512:
9; CHECK:       # %bb.0:
10; CHECK-NEXT:    vpdpbusd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x50,0xc2]
11; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
12  %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
13  ret <16 x i32> %1
14}
15
16define <16 x i32>@test_int_x86_avx512_mask_vpdpbusd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, <16 x i32> %x4, i16 %x3) {
17; X86-LABEL: test_int_x86_avx512_mask_vpdpbusd_512:
18; X86:       # %bb.0:
19; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
20; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
21; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
22; X86-NEXT:    vpdpbusd (%eax), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x50,0x18]
23; X86-NEXT:    vpdpbusd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x50,0xc2]
24; X86-NEXT:    vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0]
25; X86-NEXT:    retl # encoding: [0xc3]
26;
27; X64-LABEL: test_int_x86_avx512_mask_vpdpbusd_512:
28; X64:       # %bb.0:
29; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
30; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
31; X64-NEXT:    vpdpbusd (%rdi), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x50,0x1f]
32; X64-NEXT:    vpdpbusd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x50,0xc2]
33; X64-NEXT:    vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0]
34; X64-NEXT:    retq # encoding: [0xc3]
35  %x2 = load <16 x i32>, <16 x i32>* %x2p
36  %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
37  %2 = bitcast i16 %x3 to <16 x i1>
38  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
39  %4 = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
40  %5 = bitcast i16 %x3 to <16 x i1>
41  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
42  %res3 = add <16 x i32> %3, %6
43  ret <16 x i32> %res3
44}
45
46declare <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32>, <16 x i32>, <16 x i32>)
47
48define <16 x i32>@test_int_x86_avx512_vpdpbusds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
49; CHECK-LABEL: test_int_x86_avx512_vpdpbusds_512:
50; CHECK:       # %bb.0:
51; CHECK-NEXT:    vpdpbusds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x51,0xc2]
52; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
53  %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
54  ret <16 x i32> %1
55}
56
57define <16 x i32>@test_int_x86_avx512_mask_vpdpbusds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, <16 x i32> %x4, i16 %x3) {
58; X86-LABEL: test_int_x86_avx512_mask_vpdpbusds_512:
59; X86:       # %bb.0:
60; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
61; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
62; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
63; X86-NEXT:    vpdpbusds (%eax), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x51,0x18]
64; X86-NEXT:    vpdpbusds %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x51,0xc2]
65; X86-NEXT:    vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0]
66; X86-NEXT:    retl # encoding: [0xc3]
67;
68; X64-LABEL: test_int_x86_avx512_mask_vpdpbusds_512:
69; X64:       # %bb.0:
70; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
71; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
72; X64-NEXT:    vpdpbusds (%rdi), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x51,0x1f]
73; X64-NEXT:    vpdpbusds %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x51,0xc2]
74; X64-NEXT:    vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0]
75; X64-NEXT:    retq # encoding: [0xc3]
76  %x2 = load <16 x i32>, <16 x i32>* %x2p
77  %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
78  %2 = bitcast i16 %x3 to <16 x i1>
79  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
80  %4 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
81  %5 = bitcast i16 %x3 to <16 x i1>
82  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
83  %res3 = add <16 x i32> %3, %6
84  ret <16 x i32> %res3
85}
86
87declare <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32>, <16 x i32>, <16 x i32>)
88
89define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
90; CHECK-LABEL: test_int_x86_avx512_vpdpwssd_512:
91; CHECK:       # %bb.0:
92; CHECK-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x52,0xc2]
93; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
94  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
95  ret <16 x i32> %1
96}
97
98define <16 x i32>@test_int_x86_avx512_mask_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, <16 x i32> %x4, i16 %x3) {
99; X86-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
100; X86:       # %bb.0:
101; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
102; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
103; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
104; X86-NEXT:    vpdpwssd (%eax), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x52,0x18]
105; X86-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x52,0xc2]
106; X86-NEXT:    vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0]
107; X86-NEXT:    retl # encoding: [0xc3]
108;
109; X64-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
110; X64:       # %bb.0:
111; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
112; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
113; X64-NEXT:    vpdpwssd (%rdi), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x52,0x1f]
114; X64-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x52,0xc2]
115; X64-NEXT:    vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0]
116; X64-NEXT:    retq # encoding: [0xc3]
117  %x2 = load <16 x i32>, <16 x i32>* %x2p
118  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
119  %2 = bitcast i16 %x3 to <16 x i1>
120  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
121  %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
122  %5 = bitcast i16 %x3 to <16 x i1>
123  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
124  %res3 = add <16 x i32> %3, %6
125  ret <16 x i32> %res3
126}
127
128declare <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>)
129
130define <16 x i32>@test_int_x86_avx512_ask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
131; CHECK-LABEL: test_int_x86_avx512_ask_vpdpwssds_512:
132; CHECK:       # %bb.0:
133; CHECK-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x53,0xc2]
134; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
135  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
136  ret <16 x i32> %1
137}
138
139define <16 x i32>@test_int_x86_avx512_mask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, <16 x i32> %x4, i16 %x3) {
140; X86-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
141; X86:       # %bb.0:
142; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
143; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
144; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
145; X86-NEXT:    vpdpwssds (%eax), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x53,0x18]
146; X86-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x53,0xc2]
147; X86-NEXT:    vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0]
148; X86-NEXT:    retl # encoding: [0xc3]
149;
150; X64-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
151; X64:       # %bb.0:
152; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
153; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
154; X64-NEXT:    vpdpwssds (%rdi), %zmm1, %zmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x53,0x1f]
155; X64-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x53,0xc2]
156; X64-NEXT:    vpaddd %zmm0, %zmm3, %zmm0 # encoding: [0x62,0xf1,0x65,0x48,0xfe,0xc0]
157; X64-NEXT:    retq # encoding: [0xc3]
158  %x2 = load <16 x i32>, <16 x i32>* %x2p
159  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
160  %2 = bitcast i16 %x3 to <16 x i1>
161  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
162  %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
163  %5 = bitcast i16 %x3 to <16 x i1>
164  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
165  %res3 = add <16 x i32> %3, %6
166  ret <16 x i32> %res3
167}
168