xref: /llvm-project/llvm/test/CodeGen/X86/avx512vnni-intrinsics.ll (revision 2f448bf509432c1a19ec46ab8cbc7353c03c6280) 
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>, <16 x i32> } @test_int_x86_avx512_mask_vpdpbusd_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) {
17; X86-LABEL: test_int_x86_avx512_mask_vpdpbusd_512:
18; X86:       # %bb.0:
19; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
20; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
21; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
22; X86-NEXT:    vpdpbusd (%eax), %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x50,0x00]
23; X86-NEXT:    vpdpbusd %zmm2, %zmm1, %zmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x50,0xda]
24; X86-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
25; X86-NEXT:    retl # encoding: [0xc3]
26;
27; X64-LABEL: test_int_x86_avx512_mask_vpdpbusd_512:
28; X64:       # %bb.0:
29; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
30; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
31; X64-NEXT:    vpdpbusd (%rdi), %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x50,0x07]
32; X64-NEXT:    vpdpbusd %zmm2, %zmm1, %zmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x50,0xda]
33; X64-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
34; X64-NEXT:    retq # encoding: [0xc3]
35  %x2 = load <16 x i32>, ptr %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  %res1 = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> %3, 0
43  %res2 = insertvalue { <16 x i32>, <16 x i32> }  %res1, <16 x i32> %6, 1
44  ret { <16 x i32>, <16 x i32> } %res2
45}
46
47declare <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32>, <16 x i32>, <16 x i32>)
48
49define <16 x i32>@test_int_x86_avx512_vpdpbusds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
50; CHECK-LABEL: test_int_x86_avx512_vpdpbusds_512:
51; CHECK:       # %bb.0:
52; CHECK-NEXT:    vpdpbusds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x51,0xc2]
53; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
54  %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
55  ret <16 x i32> %1
56}
57
58define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpbusds_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) {
59; X86-LABEL: test_int_x86_avx512_mask_vpdpbusds_512:
60; X86:       # %bb.0:
61; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
62; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
63; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
64; X86-NEXT:    vpdpbusds (%eax), %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x51,0x00]
65; X86-NEXT:    vpdpbusds %zmm2, %zmm1, %zmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x51,0xda]
66; X86-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
67; X86-NEXT:    retl # encoding: [0xc3]
68;
69; X64-LABEL: test_int_x86_avx512_mask_vpdpbusds_512:
70; X64:       # %bb.0:
71; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
72; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
73; X64-NEXT:    vpdpbusds (%rdi), %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x51,0x07]
74; X64-NEXT:    vpdpbusds %zmm2, %zmm1, %zmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x51,0xda]
75; X64-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
76; X64-NEXT:    retq # encoding: [0xc3]
77  %x2 = load <16 x i32>, ptr %x2p
78  %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
79  %2 = bitcast i16 %x3 to <16 x i1>
80  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
81  %4 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
82  %5 = bitcast i16 %x3 to <16 x i1>
83  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
84  %res1 = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> %3, 0
85  %res2 = insertvalue { <16 x i32>, <16 x i32> }  %res1, <16 x i32> %6, 1
86  ret { <16 x i32>, <16 x i32> } %res2
87}
88
89declare <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32>, <16 x i32>, <16 x i32>)
90
91define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
92; CHECK-LABEL: test_int_x86_avx512_vpdpwssd_512:
93; CHECK:       # %bb.0:
94; CHECK-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x52,0xc2]
95; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
96  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
97  ret <16 x i32> %1
98}
99
100define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) {
101; X86-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
102; X86:       # %bb.0:
103; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
104; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
105; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
106; X86-NEXT:    vpdpwssd (%eax), %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x52,0x00]
107; X86-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x52,0xda]
108; X86-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
109; X86-NEXT:    retl # encoding: [0xc3]
110;
111; X64-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
112; X64:       # %bb.0:
113; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
114; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
115; X64-NEXT:    vpdpwssd (%rdi), %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x52,0x07]
116; X64-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x52,0xda]
117; X64-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
118; X64-NEXT:    retq # encoding: [0xc3]
119  %x2 = load <16 x i32>, ptr %x2p
120  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
121  %2 = bitcast i16 %x3 to <16 x i1>
122  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
123  %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
124  %5 = bitcast i16 %x3 to <16 x i1>
125  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
126  %res1 = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> %3, 0
127  %res2 = insertvalue { <16 x i32>, <16 x i32> }  %res1, <16 x i32> %6, 1
128  ret { <16 x i32>, <16 x i32> } %res2
129}
130
131declare <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>)
132
133define <16 x i32>@test_int_x86_avx512_ask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
134; CHECK-LABEL: test_int_x86_avx512_ask_vpdpwssds_512:
135; CHECK:       # %bb.0:
136; CHECK-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x53,0xc2]
137; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
138  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
139  ret <16 x i32> %1
140}
141
142define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) {
143; X86-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
144; X86:       # %bb.0:
145; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
146; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
147; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
148; X86-NEXT:    vpdpwssds (%eax), %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x53,0x00]
149; X86-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x53,0xda]
150; X86-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
151; X86-NEXT:    retl # encoding: [0xc3]
152;
153; X64-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
154; X64:       # %bb.0:
155; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
156; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
157; X64-NEXT:    vpdpwssds (%rdi), %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x53,0x07]
158; X64-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x53,0xda]
159; X64-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
160; X64-NEXT:    retq # encoding: [0xc3]
161  %x2 = load <16 x i32>, ptr %x2p
162  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
163  %2 = bitcast i16 %x3 to <16 x i1>
164  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
165  %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
166  %5 = bitcast i16 %x3 to <16 x i1>
167  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
168  %res1 = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> %3, 0
169  %res2 = insertvalue { <16 x i32>, <16 x i32> }  %res1, <16 x i32> %6, 1
170  ret { <16 x i32>, <16 x i32> } %res2
171}
172