1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s
3
4declare <2 x i64> @llvm.cttz.v2i64(<2 x i64>, i1)
5declare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>, i1)
6declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64>)
7
8define <2 x i64> @footz(<2 x i64> %a) nounwind {
9; CHECK-LABEL: footz:
10; CHECK:       # %bb.0:
11; CHECK-NEXT:    pcmpeqd %xmm1, %xmm1
12; CHECK-NEXT:    paddq %xmm0, %xmm1
13; CHECK-NEXT:    pandn %xmm1, %xmm0
14; CHECK-NEXT:    movdqa %xmm0, %xmm1
15; CHECK-NEXT:    psrlw $1, %xmm1
16; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
17; CHECK-NEXT:    psubb %xmm1, %xmm0
18; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
19; CHECK-NEXT:    movdqa %xmm0, %xmm2
20; CHECK-NEXT:    pand %xmm1, %xmm2
21; CHECK-NEXT:    psrlw $2, %xmm0
22; CHECK-NEXT:    pand %xmm1, %xmm0
23; CHECK-NEXT:    paddb %xmm2, %xmm0
24; CHECK-NEXT:    movdqa %xmm0, %xmm1
25; CHECK-NEXT:    psrlw $4, %xmm1
26; CHECK-NEXT:    paddb %xmm1, %xmm0
27; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
28; CHECK-NEXT:    pxor %xmm1, %xmm1
29; CHECK-NEXT:    psadbw %xmm1, %xmm0
30; CHECK-NEXT:    retq
31  %c = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
32  ret <2 x i64> %c
33
34}
35define <2 x i64> @foolz(<2 x i64> %a) nounwind {
36; CHECK-LABEL: foolz:
37; CHECK:       # %bb.0:
38; CHECK-NEXT:    movdqa %xmm0, %xmm1
39; CHECK-NEXT:    psrlq $1, %xmm1
40; CHECK-NEXT:    por %xmm0, %xmm1
41; CHECK-NEXT:    movdqa %xmm1, %xmm0
42; CHECK-NEXT:    psrlq $2, %xmm0
43; CHECK-NEXT:    por %xmm1, %xmm0
44; CHECK-NEXT:    movdqa %xmm0, %xmm1
45; CHECK-NEXT:    psrlq $4, %xmm1
46; CHECK-NEXT:    por %xmm0, %xmm1
47; CHECK-NEXT:    movdqa %xmm1, %xmm0
48; CHECK-NEXT:    psrlq $8, %xmm0
49; CHECK-NEXT:    por %xmm1, %xmm0
50; CHECK-NEXT:    movdqa %xmm0, %xmm1
51; CHECK-NEXT:    psrlq $16, %xmm1
52; CHECK-NEXT:    por %xmm0, %xmm1
53; CHECK-NEXT:    movdqa %xmm1, %xmm0
54; CHECK-NEXT:    psrlq $32, %xmm0
55; CHECK-NEXT:    por %xmm1, %xmm0
56; CHECK-NEXT:    pcmpeqd %xmm1, %xmm1
57; CHECK-NEXT:    pxor %xmm0, %xmm1
58; CHECK-NEXT:    movdqa %xmm1, %xmm0
59; CHECK-NEXT:    psrlw $1, %xmm0
60; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
61; CHECK-NEXT:    psubb %xmm0, %xmm1
62; CHECK-NEXT:    movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
63; CHECK-NEXT:    movdqa %xmm1, %xmm2
64; CHECK-NEXT:    pand %xmm0, %xmm2
65; CHECK-NEXT:    psrlw $2, %xmm1
66; CHECK-NEXT:    pand %xmm0, %xmm1
67; CHECK-NEXT:    paddb %xmm2, %xmm1
68; CHECK-NEXT:    movdqa %xmm1, %xmm2
69; CHECK-NEXT:    psrlw $4, %xmm2
70; CHECK-NEXT:    paddb %xmm1, %xmm2
71; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
72; CHECK-NEXT:    pxor %xmm0, %xmm0
73; CHECK-NEXT:    psadbw %xmm2, %xmm0
74; CHECK-NEXT:    retq
75  %c = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
76  ret <2 x i64> %c
77
78}
79
80define <2 x i64> @foopop(<2 x i64> %a) nounwind {
81; CHECK-LABEL: foopop:
82; CHECK:       # %bb.0:
83; CHECK-NEXT:    movdqa %xmm0, %xmm1
84; CHECK-NEXT:    psrlw $1, %xmm1
85; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
86; CHECK-NEXT:    psubb %xmm1, %xmm0
87; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
88; CHECK-NEXT:    movdqa %xmm0, %xmm2
89; CHECK-NEXT:    pand %xmm1, %xmm2
90; CHECK-NEXT:    psrlw $2, %xmm0
91; CHECK-NEXT:    pand %xmm1, %xmm0
92; CHECK-NEXT:    paddb %xmm2, %xmm0
93; CHECK-NEXT:    movdqa %xmm0, %xmm1
94; CHECK-NEXT:    psrlw $4, %xmm1
95; CHECK-NEXT:    paddb %xmm1, %xmm0
96; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
97; CHECK-NEXT:    pxor %xmm1, %xmm1
98; CHECK-NEXT:    psadbw %xmm1, %xmm0
99; CHECK-NEXT:    retq
100  %c = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
101  ret <2 x i64> %c
102}
103
104declare <2 x i32> @llvm.cttz.v2i32(<2 x i32>, i1)
105declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1)
106declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32>)
107
108define <2 x i32> @promtz(<2 x i32> %a) nounwind {
109; CHECK-LABEL: promtz:
110; CHECK:       # %bb.0:
111; CHECK-NEXT:    pcmpeqd %xmm1, %xmm1
112; CHECK-NEXT:    paddd %xmm0, %xmm1
113; CHECK-NEXT:    pandn %xmm1, %xmm0
114; CHECK-NEXT:    movdqa %xmm0, %xmm1
115; CHECK-NEXT:    psrlw $1, %xmm1
116; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
117; CHECK-NEXT:    psubb %xmm1, %xmm0
118; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
119; CHECK-NEXT:    movdqa %xmm0, %xmm2
120; CHECK-NEXT:    pand %xmm1, %xmm2
121; CHECK-NEXT:    psrlw $2, %xmm0
122; CHECK-NEXT:    pand %xmm1, %xmm0
123; CHECK-NEXT:    paddb %xmm2, %xmm0
124; CHECK-NEXT:    movdqa %xmm0, %xmm1
125; CHECK-NEXT:    psrlw $4, %xmm1
126; CHECK-NEXT:    paddb %xmm0, %xmm1
127; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
128; CHECK-NEXT:    pxor %xmm0, %xmm0
129; CHECK-NEXT:    movdqa %xmm1, %xmm2
130; CHECK-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
131; CHECK-NEXT:    psadbw %xmm0, %xmm2
132; CHECK-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
133; CHECK-NEXT:    psadbw %xmm0, %xmm1
134; CHECK-NEXT:    packuswb %xmm2, %xmm1
135; CHECK-NEXT:    movdqa %xmm1, %xmm0
136; CHECK-NEXT:    retq
137  %c = call <2 x i32> @llvm.cttz.v2i32(<2 x i32> %a, i1 false)
138  ret <2 x i32> %c
139
140}
141define <2 x i32> @promlz(<2 x i32> %a) nounwind {
142; CHECK-LABEL: promlz:
143; CHECK:       # %bb.0:
144; CHECK-NEXT:    movdqa %xmm0, %xmm1
145; CHECK-NEXT:    psrld $1, %xmm1
146; CHECK-NEXT:    por %xmm0, %xmm1
147; CHECK-NEXT:    movdqa %xmm1, %xmm0
148; CHECK-NEXT:    psrld $2, %xmm0
149; CHECK-NEXT:    por %xmm1, %xmm0
150; CHECK-NEXT:    movdqa %xmm0, %xmm1
151; CHECK-NEXT:    psrld $4, %xmm1
152; CHECK-NEXT:    por %xmm0, %xmm1
153; CHECK-NEXT:    movdqa %xmm1, %xmm0
154; CHECK-NEXT:    psrld $8, %xmm0
155; CHECK-NEXT:    por %xmm1, %xmm0
156; CHECK-NEXT:    movdqa %xmm0, %xmm1
157; CHECK-NEXT:    psrld $16, %xmm1
158; CHECK-NEXT:    por %xmm0, %xmm1
159; CHECK-NEXT:    pcmpeqd %xmm2, %xmm2
160; CHECK-NEXT:    pxor %xmm1, %xmm2
161; CHECK-NEXT:    movdqa %xmm2, %xmm0
162; CHECK-NEXT:    psrlw $1, %xmm0
163; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
164; CHECK-NEXT:    psubb %xmm0, %xmm2
165; CHECK-NEXT:    movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
166; CHECK-NEXT:    movdqa %xmm2, %xmm1
167; CHECK-NEXT:    pand %xmm0, %xmm1
168; CHECK-NEXT:    psrlw $2, %xmm2
169; CHECK-NEXT:    pand %xmm0, %xmm2
170; CHECK-NEXT:    paddb %xmm1, %xmm2
171; CHECK-NEXT:    movdqa %xmm2, %xmm0
172; CHECK-NEXT:    psrlw $4, %xmm0
173; CHECK-NEXT:    paddb %xmm2, %xmm0
174; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
175; CHECK-NEXT:    pxor %xmm1, %xmm1
176; CHECK-NEXT:    movdqa %xmm0, %xmm2
177; CHECK-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
178; CHECK-NEXT:    psadbw %xmm1, %xmm2
179; CHECK-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
180; CHECK-NEXT:    psadbw %xmm1, %xmm0
181; CHECK-NEXT:    packuswb %xmm2, %xmm0
182; CHECK-NEXT:    retq
183  %c = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %a, i1 false)
184  ret <2 x i32> %c
185
186}
187
188define <2 x i32> @prompop(<2 x i32> %a) nounwind {
189; CHECK-LABEL: prompop:
190; CHECK:       # %bb.0:
191; CHECK-NEXT:    movdqa %xmm0, %xmm1
192; CHECK-NEXT:    psrlw $1, %xmm1
193; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
194; CHECK-NEXT:    psubb %xmm1, %xmm0
195; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
196; CHECK-NEXT:    movdqa %xmm0, %xmm2
197; CHECK-NEXT:    pand %xmm1, %xmm2
198; CHECK-NEXT:    psrlw $2, %xmm0
199; CHECK-NEXT:    pand %xmm1, %xmm0
200; CHECK-NEXT:    paddb %xmm2, %xmm0
201; CHECK-NEXT:    movdqa %xmm0, %xmm1
202; CHECK-NEXT:    psrlw $4, %xmm1
203; CHECK-NEXT:    paddb %xmm0, %xmm1
204; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
205; CHECK-NEXT:    pxor %xmm0, %xmm0
206; CHECK-NEXT:    movdqa %xmm1, %xmm2
207; CHECK-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
208; CHECK-NEXT:    psadbw %xmm0, %xmm2
209; CHECK-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
210; CHECK-NEXT:    psadbw %xmm0, %xmm1
211; CHECK-NEXT:    packuswb %xmm2, %xmm1
212; CHECK-NEXT:    movdqa %xmm1, %xmm0
213; CHECK-NEXT:    retq
214  %c = call <2 x i32> @llvm.ctpop.v2i32(<2 x i32> %a)
215  ret <2 x i32> %c
216}
217