1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512vp2intersect --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vp2intersect --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64
5 define void @test_mm512_2intersect_epi32(<8 x i64> %a, <8 x i64> %b, i16* nocapture %m0, i16* nocapture %m1) {
6 ; X86-LABEL: test_mm512_2intersect_epi32:
7 ; X86: # %bb.0: # %entry
8 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
9 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx # encoding: [0x8b,0x4c,0x24,0x04]
10 ; X86-NEXT: vp2intersectd %zmm1, %zmm0, %k0 # encoding: [0x62,0xf2,0x7f,0x48,0x68,0xc1]
11 ; X86-NEXT: kmovw %k0, (%ecx) # encoding: [0xc5,0xf8,0x91,0x01]
12 ; X86-NEXT: kmovw %k1, (%eax) # encoding: [0xc5,0xf8,0x91,0x08]
13 ; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
14 ; X86-NEXT: retl # encoding: [0xc3]
16 ; X64-LABEL: test_mm512_2intersect_epi32:
17 ; X64: # %bb.0: # %entry
18 ; X64-NEXT: vp2intersectd %zmm1, %zmm0, %k0 # encoding: [0x62,0xf2,0x7f,0x48,0x68,0xc1]
19 ; X64-NEXT: kmovw %k0, (%rdi) # encoding: [0xc5,0xf8,0x91,0x07]
20 ; X64-NEXT: kmovw %k1, (%rsi) # encoding: [0xc5,0xf8,0x91,0x0e]
21 ; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
22 ; X64-NEXT: retq # encoding: [0xc3]
24 %0 = bitcast <8 x i64> %a to <16 x i32>
25 %1 = bitcast <8 x i64> %b to <16 x i32>
26 %2 = tail call { <16 x i1>, <16 x i1> } @llvm.x86.avx512.vp2intersect.d.512(<16 x i32> %0, <16 x i32> %1)
27 %3 = extractvalue { <16 x i1>, <16 x i1> } %2, 0
28 %4 = bitcast i16* %m0 to <16 x i1>*
29 store <16 x i1> %3, <16 x i1>* %4, align 16
30 %5 = extractvalue { <16 x i1>, <16 x i1> } %2, 1
31 %6 = bitcast i16* %m1 to <16 x i1>*
32 store <16 x i1> %5, <16 x i1>* %6, align 16
36 define void @test_mm512_2intersect_epi64(<8 x i64> %a, <8 x i64> %b, i8* nocapture %m0, i8* nocapture %m1) {
37 ; X86-LABEL: test_mm512_2intersect_epi64:
38 ; X86: # %bb.0: # %entry
39 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
40 ; X86-NEXT: vp2intersectq %zmm1, %zmm0, %k0 # encoding: [0x62,0xf2,0xff,0x48,0x68,0xc1]
41 ; X86-NEXT: kmovw %k1, %ecx # encoding: [0xc5,0xf8,0x93,0xc9]
42 ; X86-NEXT: kmovw %k0, %edx # encoding: [0xc5,0xf8,0x93,0xd0]
43 ; X86-NEXT: movb %dl, (%eax) # encoding: [0x88,0x10]
44 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
45 ; X86-NEXT: movb %cl, (%eax) # encoding: [0x88,0x08]
46 ; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
47 ; X86-NEXT: retl # encoding: [0xc3]
49 ; X64-LABEL: test_mm512_2intersect_epi64:
50 ; X64: # %bb.0: # %entry
51 ; X64-NEXT: vp2intersectq %zmm1, %zmm0, %k0 # encoding: [0x62,0xf2,0xff,0x48,0x68,0xc1]
52 ; X64-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
53 ; X64-NEXT: kmovw %k0, %ecx # encoding: [0xc5,0xf8,0x93,0xc8]
54 ; X64-NEXT: movb %cl, (%rdi) # encoding: [0x88,0x0f]
55 ; X64-NEXT: movb %al, (%rsi) # encoding: [0x88,0x06]
56 ; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
57 ; X64-NEXT: retq # encoding: [0xc3]
59 %0 = tail call { <8 x i1>, <8 x i1> } @llvm.x86.avx512.vp2intersect.q.512(<8 x i64> %a, <8 x i64> %b)
60 %1 = extractvalue { <8 x i1>, <8 x i1> } %0, 0
61 %2 = bitcast i8* %m0 to <8 x i1>*
62 store <8 x i1> %1, <8 x i1>* %2, align 8
63 %3 = extractvalue { <8 x i1>, <8 x i1> } %0, 1
64 %4 = bitcast i8* %m1 to <8 x i1>*
65 store <8 x i1> %3, <8 x i1>* %4, align 8
69 define void @test_mm512_2intersect_epi32_p(<8 x i64>* nocapture readonly %a, <8 x i64>* nocapture readonly %b, i16* nocapture %m0, i16* nocapture %m1) {
70 ; X86-LABEL: test_mm512_2intersect_epi32_p:
71 ; X86: # %bb.0: # %entry
72 ; X86-NEXT: pushl %esi # encoding: [0x56]
73 ; X86-NEXT: .cfi_def_cfa_offset 8
74 ; X86-NEXT: .cfi_offset %esi, -8
75 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x14]
76 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx # encoding: [0x8b,0x4c,0x24,0x10]
77 ; X86-NEXT: movl {{[0-9]+}}(%esp), %edx # encoding: [0x8b,0x54,0x24,0x0c]
78 ; X86-NEXT: movl {{[0-9]+}}(%esp), %esi # encoding: [0x8b,0x74,0x24,0x08]
79 ; X86-NEXT: vmovaps (%esi), %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0x06]
80 ; X86-NEXT: vp2intersectd (%edx), %zmm0, %k0 # encoding: [0x62,0xf2,0x7f,0x48,0x68,0x02]
81 ; X86-NEXT: kmovw %k0, (%ecx) # encoding: [0xc5,0xf8,0x91,0x01]
82 ; X86-NEXT: kmovw %k1, (%eax) # encoding: [0xc5,0xf8,0x91,0x08]
83 ; X86-NEXT: popl %esi # encoding: [0x5e]
84 ; X86-NEXT: .cfi_def_cfa_offset 4
85 ; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
86 ; X86-NEXT: retl # encoding: [0xc3]
88 ; X64-LABEL: test_mm512_2intersect_epi32_p:
89 ; X64: # %bb.0: # %entry
90 ; X64-NEXT: vmovaps (%rdi), %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0x07]
91 ; X64-NEXT: vp2intersectd (%rsi), %zmm0, %k0 # encoding: [0x62,0xf2,0x7f,0x48,0x68,0x06]
92 ; X64-NEXT: kmovw %k0, (%rdx) # encoding: [0xc5,0xf8,0x91,0x02]
93 ; X64-NEXT: kmovw %k1, (%rcx) # encoding: [0xc5,0xf8,0x91,0x09]
94 ; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
95 ; X64-NEXT: retq # encoding: [0xc3]
97 %0 = bitcast <8 x i64>* %a to <16 x i32>*
98 %1 = load <16 x i32>, <16 x i32>* %0, align 64
99 %2 = bitcast <8 x i64>* %b to <16 x i32>*
100 %3 = load <16 x i32>, <16 x i32>* %2, align 64
101 %4 = tail call { <16 x i1>, <16 x i1> } @llvm.x86.avx512.vp2intersect.d.512(<16 x i32> %1, <16 x i32> %3)
102 %5 = extractvalue { <16 x i1>, <16 x i1> } %4, 0
103 %6 = bitcast i16* %m0 to <16 x i1>*
104 store <16 x i1> %5, <16 x i1>* %6, align 16
105 %7 = extractvalue { <16 x i1>, <16 x i1> } %4, 1
106 %8 = bitcast i16* %m1 to <16 x i1>*
107 store <16 x i1> %7, <16 x i1>* %8, align 16
111 define void @test_mm512_2intersect_epi64_p(<8 x i64>* nocapture readonly %a, <8 x i64>* nocapture readonly %b, i8* nocapture %m0, i8* nocapture %m1) {
112 ; X86-LABEL: test_mm512_2intersect_epi64_p:
113 ; X86: # %bb.0: # %entry
114 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x0c]
115 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx # encoding: [0x8b,0x4c,0x24,0x08]
116 ; X86-NEXT: movl {{[0-9]+}}(%esp), %edx # encoding: [0x8b,0x54,0x24,0x04]
117 ; X86-NEXT: vmovaps (%edx), %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0x02]
118 ; X86-NEXT: vp2intersectq (%ecx), %zmm0, %k0 # encoding: [0x62,0xf2,0xff,0x48,0x68,0x01]
119 ; X86-NEXT: kmovw %k1, %ecx # encoding: [0xc5,0xf8,0x93,0xc9]
120 ; X86-NEXT: kmovw %k0, %edx # encoding: [0xc5,0xf8,0x93,0xd0]
121 ; X86-NEXT: movb %dl, (%eax) # encoding: [0x88,0x10]
122 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x10]
123 ; X86-NEXT: movb %cl, (%eax) # encoding: [0x88,0x08]
124 ; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
125 ; X86-NEXT: retl # encoding: [0xc3]
127 ; X64-LABEL: test_mm512_2intersect_epi64_p:
128 ; X64: # %bb.0: # %entry
129 ; X64-NEXT: vmovaps (%rdi), %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0x07]
130 ; X64-NEXT: vp2intersectq (%rsi), %zmm0, %k0 # encoding: [0x62,0xf2,0xff,0x48,0x68,0x06]
131 ; X64-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
132 ; X64-NEXT: kmovw %k0, %esi # encoding: [0xc5,0xf8,0x93,0xf0]
133 ; X64-NEXT: movb %sil, (%rdx) # encoding: [0x40,0x88,0x32]
134 ; X64-NEXT: movb %al, (%rcx) # encoding: [0x88,0x01]
135 ; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
136 ; X64-NEXT: retq # encoding: [0xc3]
139 %0 = load <8 x i64>, <8 x i64>* %a, align 64
140 %1 = load <8 x i64>, <8 x i64>* %b, align 64
141 %2 = tail call { <8 x i1>, <8 x i1> } @llvm.x86.avx512.vp2intersect.q.512(<8 x i64> %0, <8 x i64> %1)
142 %3 = extractvalue { <8 x i1>, <8 x i1> } %2, 0
143 %4 = bitcast i8* %m0 to <8 x i1>*
144 store <8 x i1> %3, <8 x i1>* %4, align 8
145 %5 = extractvalue { <8 x i1>, <8 x i1> } %2, 1
146 %6 = bitcast i8* %m1 to <8 x i1>*
147 store <8 x i1> %5, <8 x i1>* %6, align 8
151 define void @test_mm512_2intersect_epi32_b(i32* nocapture readonly %a, i32* nocapture readonly %b, i16* nocapture %m0, i16* nocapture %m1) {
152 ; X86-LABEL: test_mm512_2intersect_epi32_b:
153 ; X86: # %bb.0: # %entry
154 ; X86-NEXT: pushl %esi # encoding: [0x56]
155 ; X86-NEXT: .cfi_def_cfa_offset 8
156 ; X86-NEXT: .cfi_offset %esi, -8
157 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x14]
158 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx # encoding: [0x8b,0x4c,0x24,0x10]
159 ; X86-NEXT: movl {{[0-9]+}}(%esp), %edx # encoding: [0x8b,0x54,0x24,0x0c]
160 ; X86-NEXT: movl {{[0-9]+}}(%esp), %esi # encoding: [0x8b,0x74,0x24,0x08]
161 ; X86-NEXT: vbroadcastss (%esi), %zmm0 # encoding: [0x62,0xf2,0x7d,0x48,0x18,0x06]
162 ; X86-NEXT: vp2intersectd (%edx){1to16}, %zmm0, %k0 # encoding: [0x62,0xf2,0x7f,0x58,0x68,0x02]
163 ; X86-NEXT: kmovw %k0, (%ecx) # encoding: [0xc5,0xf8,0x91,0x01]
164 ; X86-NEXT: kmovw %k1, (%eax) # encoding: [0xc5,0xf8,0x91,0x08]
165 ; X86-NEXT: popl %esi # encoding: [0x5e]
166 ; X86-NEXT: .cfi_def_cfa_offset 4
167 ; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
168 ; X86-NEXT: retl # encoding: [0xc3]
170 ; X64-LABEL: test_mm512_2intersect_epi32_b:
171 ; X64: # %bb.0: # %entry
172 ; X64-NEXT: vbroadcastss (%rdi), %zmm0 # encoding: [0x62,0xf2,0x7d,0x48,0x18,0x07]
173 ; X64-NEXT: vp2intersectd (%rsi){1to16}, %zmm0, %k0 # encoding: [0x62,0xf2,0x7f,0x58,0x68,0x06]
174 ; X64-NEXT: kmovw %k0, (%rdx) # encoding: [0xc5,0xf8,0x91,0x02]
175 ; X64-NEXT: kmovw %k1, (%rcx) # encoding: [0xc5,0xf8,0x91,0x09]
176 ; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
177 ; X64-NEXT: retq # encoding: [0xc3]
179 %0 = load i32, i32* %a, align 4
180 %vecinit.i = insertelement <16 x i32> undef, i32 %0, i32 0
181 %vecinit15.i = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
182 %1 = load i32, i32* %b, align 4
183 %vecinit.i2 = insertelement <16 x i32> undef, i32 %1, i32 0
184 %vecinit15.i3 = shufflevector <16 x i32> %vecinit.i2, <16 x i32> undef, <16 x i32> zeroinitializer
185 %2 = tail call { <16 x i1>, <16 x i1> } @llvm.x86.avx512.vp2intersect.d.512(<16 x i32> %vecinit15.i, <16 x i32> %vecinit15.i3)
186 %3 = extractvalue { <16 x i1>, <16 x i1> } %2, 0
187 %4 = bitcast i16* %m0 to <16 x i1>*
188 store <16 x i1> %3, <16 x i1>* %4, align 16
189 %5 = extractvalue { <16 x i1>, <16 x i1> } %2, 1
190 %6 = bitcast i16* %m1 to <16 x i1>*
191 store <16 x i1> %5, <16 x i1>* %6, align 16
195 define void @test_mm512_2intersect_epi64_b(i64* nocapture readonly %a, i64* nocapture readonly %b, i8* nocapture %m0, i8* nocapture %m1) {
196 ; X86-LABEL: test_mm512_2intersect_epi64_b:
197 ; X86: # %bb.0: # %entry
198 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x0c]
199 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx # encoding: [0x8b,0x4c,0x24,0x08]
200 ; X86-NEXT: movl {{[0-9]+}}(%esp), %edx # encoding: [0x8b,0x54,0x24,0x04]
201 ; X86-NEXT: vbroadcastsd (%edx), %zmm0 # encoding: [0x62,0xf2,0xfd,0x48,0x19,0x02]
202 ; X86-NEXT: vbroadcastsd (%ecx), %zmm1 # encoding: [0x62,0xf2,0xfd,0x48,0x19,0x09]
203 ; X86-NEXT: vp2intersectq %zmm1, %zmm0, %k0 # encoding: [0x62,0xf2,0xff,0x48,0x68,0xc1]
204 ; X86-NEXT: kmovw %k1, %ecx # encoding: [0xc5,0xf8,0x93,0xc9]
205 ; X86-NEXT: kmovw %k0, %edx # encoding: [0xc5,0xf8,0x93,0xd0]
206 ; X86-NEXT: movb %dl, (%eax) # encoding: [0x88,0x10]
207 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x10]
208 ; X86-NEXT: movb %cl, (%eax) # encoding: [0x88,0x08]
209 ; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
210 ; X86-NEXT: retl # encoding: [0xc3]
212 ; X64-LABEL: test_mm512_2intersect_epi64_b:
213 ; X64: # %bb.0: # %entry
214 ; X64-NEXT: vbroadcastsd (%rdi), %zmm0 # encoding: [0x62,0xf2,0xfd,0x48,0x19,0x07]
215 ; X64-NEXT: vp2intersectq (%rsi){1to8}, %zmm0, %k0 # encoding: [0x62,0xf2,0xff,0x58,0x68,0x06]
216 ; X64-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1]
217 ; X64-NEXT: kmovw %k0, %esi # encoding: [0xc5,0xf8,0x93,0xf0]
218 ; X64-NEXT: movb %sil, (%rdx) # encoding: [0x40,0x88,0x32]
219 ; X64-NEXT: movb %al, (%rcx) # encoding: [0x88,0x01]
220 ; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
221 ; X64-NEXT: retq # encoding: [0xc3]
223 %0 = load i64, i64* %a, align 8
224 %vecinit.i = insertelement <8 x i64> undef, i64 %0, i32 0
225 %vecinit7.i = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
226 %1 = load i64, i64* %b, align 8
227 %vecinit.i2 = insertelement <8 x i64> undef, i64 %1, i32 0
228 %vecinit7.i3 = shufflevector <8 x i64> %vecinit.i2, <8 x i64> undef, <8 x i32> zeroinitializer
229 %2 = tail call { <8 x i1>, <8 x i1> } @llvm.x86.avx512.vp2intersect.q.512(<8 x i64> %vecinit7.i, <8 x i64> %vecinit7.i3)
230 %3 = extractvalue { <8 x i1>, <8 x i1> } %2, 0
231 %4 = bitcast i8* %m0 to <8 x i1>*
232 store <8 x i1> %3, <8 x i1>* %4, align 8
233 %5 = extractvalue { <8 x i1>, <8 x i1> } %2, 1
234 %6 = bitcast i8* %m1 to <8 x i1>*
235 store <8 x i1> %5, <8 x i1>* %6, align 8
239 declare { <16 x i1>, <16 x i1> } @llvm.x86.avx512.vp2intersect.d.512(<16 x i32>, <16 x i32>)
240 declare { <8 x i1>, <8 x i1> } @llvm.x86.avx512.vp2intersect.q.512(<8 x i64>, <8 x i64>)