83
|
1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
2
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
3 declare <2 x i64> @llvm.cttz.v2i64(<2 x i64>, i1)
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
4 declare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>, i1)
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
5 declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64>)
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
6
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
7 define <2 x i64> @footz(<2 x i64> %a) nounwind {
|
95
|
8 ; CHECK-LABEL: footz:
|
|
9 ; CHECK: # BB#0:
|
|
10 ; CHECK-NEXT: movd %xmm0, %rax
|
|
11 ; CHECK-NEXT: bsfq %rax, %rax
|
|
12 ; CHECK-NEXT: movd %rax, %xmm1
|
|
13 ; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
|
|
14 ; CHECK-NEXT: movd %xmm0, %rax
|
|
15 ; CHECK-NEXT: bsfq %rax, %rax
|
|
16 ; CHECK-NEXT: movd %rax, %xmm0
|
|
17 ; CHECK-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
|
|
18 ; CHECK-NEXT: movdqa %xmm1, %xmm0
|
|
19 ; CHECK-NEXT: retq
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
20 %c = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
21 ret <2 x i64> %c
|
83
|
22
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
23 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
24 define <2 x i64> @foolz(<2 x i64> %a) nounwind {
|
95
|
25 ; CHECK-LABEL: foolz:
|
|
26 ; CHECK: # BB#0:
|
|
27 ; CHECK-NEXT: movd %xmm0, %rax
|
|
28 ; CHECK-NEXT: bsrq %rax, %rax
|
|
29 ; CHECK-NEXT: xorq $63, %rax
|
|
30 ; CHECK-NEXT: movd %rax, %xmm1
|
|
31 ; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
|
|
32 ; CHECK-NEXT: movd %xmm0, %rax
|
|
33 ; CHECK-NEXT: bsrq %rax, %rax
|
|
34 ; CHECK-NEXT: xorq $63, %rax
|
|
35 ; CHECK-NEXT: movd %rax, %xmm0
|
|
36 ; CHECK-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
|
|
37 ; CHECK-NEXT: movdqa %xmm1, %xmm0
|
|
38 ; CHECK-NEXT: retq
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
39 %c = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
40 ret <2 x i64> %c
|
83
|
41
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
42 }
|
83
|
43
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
44 define <2 x i64> @foopop(<2 x i64> %a) nounwind {
|
95
|
45 ; CHECK-LABEL: foopop:
|
|
46 ; CHECK: # BB#0:
|
|
47 ; CHECK-NEXT: movdqa %xmm0, %xmm1
|
|
48 ; CHECK-NEXT: psrlq $1, %xmm1
|
|
49 ; CHECK-NEXT: pand {{.*}}(%rip), %xmm1
|
|
50 ; CHECK-NEXT: psubq %xmm1, %xmm0
|
|
51 ; CHECK-NEXT: movdqa {{.*#+}} xmm1 = [3689348814741910323,3689348814741910323]
|
|
52 ; CHECK-NEXT: movdqa %xmm0, %xmm2
|
|
53 ; CHECK-NEXT: pand %xmm1, %xmm2
|
|
54 ; CHECK-NEXT: psrlq $2, %xmm0
|
|
55 ; CHECK-NEXT: pand %xmm1, %xmm0
|
|
56 ; CHECK-NEXT: paddq %xmm2, %xmm0
|
|
57 ; CHECK-NEXT: movdqa %xmm0, %xmm1
|
|
58 ; CHECK-NEXT: psrlq $4, %xmm1
|
|
59 ; CHECK-NEXT: paddq %xmm0, %xmm1
|
|
60 ; CHECK-NEXT: pand {{.*}}(%rip), %xmm1
|
|
61 ; CHECK-NEXT: pxor %xmm0, %xmm0
|
|
62 ; CHECK-NEXT: psadbw %xmm0, %xmm1
|
|
63 ; CHECK-NEXT: movdqa %xmm1, %xmm0
|
|
64 ; CHECK-NEXT: retq
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
65 %c = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
66 ret <2 x i64> %c
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
67 }
|
83
|
68
|
|
69 declare <2 x i32> @llvm.cttz.v2i32(<2 x i32>, i1)
|
|
70 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1)
|
|
71 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32>)
|
|
72
|
|
73 define <2 x i32> @promtz(<2 x i32> %a) nounwind {
|
95
|
74 ; CHECK-LABEL: promtz:
|
|
75 ; CHECK: # BB#0:
|
|
76 ; CHECK-NEXT: por {{.*}}(%rip), %xmm0
|
|
77 ; CHECK-NEXT: movd %xmm0, %rax
|
|
78 ; CHECK-NEXT: bsfq %rax, %rax
|
|
79 ; CHECK-NEXT: movl $64, %ecx
|
|
80 ; CHECK-NEXT: cmoveq %rcx, %rax
|
|
81 ; CHECK-NEXT: movd %rax, %xmm1
|
|
82 ; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
|
|
83 ; CHECK-NEXT: movd %xmm0, %rax
|
|
84 ; CHECK-NEXT: bsfq %rax, %rax
|
|
85 ; CHECK-NEXT: cmoveq %rcx, %rax
|
|
86 ; CHECK-NEXT: movd %rax, %xmm0
|
|
87 ; CHECK-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
|
|
88 ; CHECK-NEXT: movdqa %xmm1, %xmm0
|
|
89 ; CHECK-NEXT: retq
|
83
|
90 %c = call <2 x i32> @llvm.cttz.v2i32(<2 x i32> %a, i1 false)
|
|
91 ret <2 x i32> %c
|
|
92
|
|
93 }
|
|
94 define <2 x i32> @promlz(<2 x i32> %a) nounwind {
|
95
|
95 ; CHECK-LABEL: promlz:
|
|
96 ; CHECK: # BB#0:
|
|
97 ; CHECK-NEXT: pand {{.*}}(%rip), %xmm0
|
|
98 ; CHECK-NEXT: movd %xmm0, %rax
|
|
99 ; CHECK-NEXT: bsrq %rax, %rax
|
|
100 ; CHECK-NEXT: movl $127, %ecx
|
|
101 ; CHECK-NEXT: cmoveq %rcx, %rax
|
|
102 ; CHECK-NEXT: xorq $63, %rax
|
|
103 ; CHECK-NEXT: movd %rax, %xmm1
|
|
104 ; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
|
|
105 ; CHECK-NEXT: movd %xmm0, %rax
|
|
106 ; CHECK-NEXT: bsrq %rax, %rax
|
|
107 ; CHECK-NEXT: cmoveq %rcx, %rax
|
|
108 ; CHECK-NEXT: xorq $63, %rax
|
|
109 ; CHECK-NEXT: movd %rax, %xmm0
|
|
110 ; CHECK-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
|
|
111 ; CHECK-NEXT: psubq {{.*}}(%rip), %xmm1
|
|
112 ; CHECK-NEXT: movdqa %xmm1, %xmm0
|
|
113 ; CHECK-NEXT: retq
|
83
|
114 %c = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %a, i1 false)
|
|
115 ret <2 x i32> %c
|
|
116
|
|
117 }
|
|
118
|
|
119 define <2 x i32> @prompop(<2 x i32> %a) nounwind {
|
95
|
120 ; CHECK-LABEL: prompop:
|
|
121 ; CHECK: # BB#0:
|
|
122 ; CHECK-NEXT: pand {{.*}}(%rip), %xmm0
|
|
123 ; CHECK-NEXT: movdqa %xmm0, %xmm1
|
|
124 ; CHECK-NEXT: psrlq $1, %xmm1
|
|
125 ; CHECK-NEXT: pand {{.*}}(%rip), %xmm1
|
|
126 ; CHECK-NEXT: psubq %xmm1, %xmm0
|
|
127 ; CHECK-NEXT: movdqa {{.*#+}} xmm1 = [3689348814741910323,3689348814741910323]
|
|
128 ; CHECK-NEXT: movdqa %xmm0, %xmm2
|
|
129 ; CHECK-NEXT: pand %xmm1, %xmm2
|
|
130 ; CHECK-NEXT: psrlq $2, %xmm0
|
|
131 ; CHECK-NEXT: pand %xmm1, %xmm0
|
|
132 ; CHECK-NEXT: paddq %xmm2, %xmm0
|
|
133 ; CHECK-NEXT: movdqa %xmm0, %xmm1
|
|
134 ; CHECK-NEXT: psrlq $4, %xmm1
|
|
135 ; CHECK-NEXT: paddq %xmm0, %xmm1
|
|
136 ; CHECK-NEXT: pand {{.*}}(%rip), %xmm1
|
|
137 ; CHECK-NEXT: pxor %xmm0, %xmm0
|
|
138 ; CHECK-NEXT: psadbw %xmm0, %xmm1
|
|
139 ; CHECK-NEXT: movdqa %xmm1, %xmm0
|
|
140 ; CHECK-NEXT: retq
|
83
|
141 %c = call <2 x i32> @llvm.ctpop.v2i32(<2 x i32> %a)
|
|
142 ret <2 x i32> %c
|
|
143 }
|