|
120
|
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
95
|
2 ; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=ALL --check-prefix=X32
|
|
|
3 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=ALL --check-prefix=X64
|
|
|
4
|
|
|
5 ;
|
|
|
6 ; Library Functions
|
|
|
7 ;
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
8
|
|
120
|
9 define float @tst1(float %a, float %b) nounwind {
|
|
|
10 ; X32-LABEL: tst1:
|
|
|
11 ; X32: # BB#0:
|
|
|
12 ; X32-NEXT: subl $8, %esp
|
|
|
13 ; X32-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
|
|
|
14 ; X32-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
|
|
|
15 ; X32-NEXT: movss %xmm1, {{[0-9]+}}(%esp)
|
|
|
16 ; X32-NEXT: movss %xmm0, (%esp)
|
|
|
17 ; X32-NEXT: calll copysignf
|
|
|
18 ; X32-NEXT: addl $8, %esp
|
|
|
19 ; X32-NEXT: retl
|
|
95
|
20 ;
|
|
120
|
21 ; X64-LABEL: tst1:
|
|
|
22 ; X64: # BB#0:
|
|
|
23 ; X64-NEXT: movaps %xmm0, %xmm2
|
|
|
24 ; X64-NEXT: movaps %xmm1, %xmm0
|
|
|
25 ; X64-NEXT: movaps %xmm2, %xmm1
|
|
|
26 ; X64-NEXT: jmp copysignf # TAILCALL
|
|
95
|
27 %tmp = tail call float @copysignf( float %b, float %a )
|
|
|
28 ret float %tmp
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
29 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
30
|
|
120
|
31 define double @tst2(double %a, float %b, float %c) nounwind {
|
|
|
32 ; X32-LABEL: tst2:
|
|
|
33 ; X32: # BB#0:
|
|
|
34 ; X32-NEXT: subl $16, %esp
|
|
|
35 ; X32-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
|
|
|
36 ; X32-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
|
|
|
37 ; X32-NEXT: addss {{[0-9]+}}(%esp), %xmm1
|
|
|
38 ; X32-NEXT: cvtss2sd %xmm1, %xmm1
|
|
|
39 ; X32-NEXT: movsd %xmm0, (%esp)
|
|
|
40 ; X32-NEXT: movsd %xmm1, {{[0-9]+}}(%esp)
|
|
|
41 ; X32-NEXT: calll copysign
|
|
|
42 ; X32-NEXT: addl $16, %esp
|
|
|
43 ; X32-NEXT: retl
|
|
95
|
44 ;
|
|
120
|
45 ; X64-LABEL: tst2:
|
|
|
46 ; X64: # BB#0:
|
|
|
47 ; X64-NEXT: addss %xmm2, %xmm1
|
|
|
48 ; X64-NEXT: cvtss2sd %xmm1, %xmm1
|
|
|
49 ; X64-NEXT: jmp copysign # TAILCALL
|
|
95
|
50 %tmp1 = fadd float %b, %c
|
|
|
51 %tmp2 = fpext float %tmp1 to double
|
|
|
52 %tmp = tail call double @copysign( double %a, double %tmp2 )
|
|
|
53 ret double %tmp
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
54 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
55
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
56 declare float @copysignf(float, float)
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
57 declare double @copysign(double, double)
|
|
95
|
58
|
|
|
59 ;
|
|
|
60 ; LLVM Intrinsic
|
|
|
61 ;
|
|
|
62
|
|
120
|
63 define float @int1(float %a, float %b) nounwind {
|
|
|
64 ; X32-LABEL: int1:
|
|
|
65 ; X32: # BB#0:
|
|
|
66 ; X32-NEXT: pushl %eax
|
|
|
67 ; X32-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
|
|
|
68 ; X32-NEXT: andps {{\.LCPI.*}}, %xmm0
|
|
|
69 ; X32-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
|
|
|
70 ; X32-NEXT: andps {{\.LCPI.*}}, %xmm1
|
|
|
71 ; X32-NEXT: orps %xmm0, %xmm1
|
|
|
72 ; X32-NEXT: movss %xmm1, (%esp)
|
|
|
73 ; X32-NEXT: flds (%esp)
|
|
|
74 ; X32-NEXT: popl %eax
|
|
|
75 ; X32-NEXT: retl
|
|
95
|
76 ;
|
|
120
|
77 ; X64-LABEL: int1:
|
|
|
78 ; X64: # BB#0:
|
|
|
79 ; X64-NEXT: andps {{.*}}(%rip), %xmm0
|
|
|
80 ; X64-NEXT: andps {{.*}}(%rip), %xmm1
|
|
|
81 ; X64-NEXT: orps %xmm1, %xmm0
|
|
|
82 ; X64-NEXT: retq
|
|
95
|
83 %tmp = tail call float @llvm.copysign.f32( float %b, float %a )
|
|
|
84 ret float %tmp
|
|
|
85 }
|
|
|
86
|
|
120
|
87 define double @int2(double %a, float %b, float %c) nounwind {
|
|
|
88 ; X32-LABEL: int2:
|
|
|
89 ; X32: # BB#0:
|
|
|
90 ; X32-NEXT: pushl %ebp
|
|
|
91 ; X32-NEXT: movl %esp, %ebp
|
|
|
92 ; X32-NEXT: andl $-8, %esp
|
|
|
93 ; X32-NEXT: subl $8, %esp
|
|
|
94 ; X32-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
|
|
|
95 ; X32-NEXT: addss 20(%ebp), %xmm0
|
|
|
96 ; X32-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
|
|
121
|
97 ; X32-NEXT: andps {{\.LCPI.*}}, %xmm1
|
|
120
|
98 ; X32-NEXT: cvtss2sd %xmm0, %xmm0
|
|
121
|
99 ; X32-NEXT: andps {{\.LCPI.*}}, %xmm0
|
|
|
100 ; X32-NEXT: orps %xmm1, %xmm0
|
|
|
101 ; X32-NEXT: movlps %xmm0, (%esp)
|
|
120
|
102 ; X32-NEXT: fldl (%esp)
|
|
|
103 ; X32-NEXT: movl %ebp, %esp
|
|
|
104 ; X32-NEXT: popl %ebp
|
|
|
105 ; X32-NEXT: retl
|
|
95
|
106 ;
|
|
120
|
107 ; X64-LABEL: int2:
|
|
|
108 ; X64: # BB#0:
|
|
|
109 ; X64-NEXT: addss %xmm2, %xmm1
|
|
|
110 ; X64-NEXT: cvtss2sd %xmm1, %xmm1
|
|
|
111 ; X64-NEXT: andps {{.*}}(%rip), %xmm1
|
|
|
112 ; X64-NEXT: andps {{.*}}(%rip), %xmm0
|
|
|
113 ; X64-NEXT: orps %xmm1, %xmm0
|
|
|
114 ; X64-NEXT: retq
|
|
95
|
115 %tmp1 = fadd float %b, %c
|
|
|
116 %tmp2 = fpext float %tmp1 to double
|
|
|
117 %tmp = tail call double @llvm.copysign.f64( double %a, double %tmp2 )
|
|
|
118 ret double %tmp
|
|
|
119 }
|
|
|
120
|
|
120
|
121 define float @cst1() nounwind {
|
|
|
122 ; X32-LABEL: cst1:
|
|
|
123 ; X32: # BB#0:
|
|
|
124 ; X32-NEXT: fld1
|
|
|
125 ; X32-NEXT: fchs
|
|
|
126 ; X32-NEXT: retl
|
|
95
|
127 ;
|
|
120
|
128 ; X64-LABEL: cst1:
|
|
|
129 ; X64: # BB#0:
|
|
|
130 ; X64-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
|
|
|
131 ; X64-NEXT: retq
|
|
95
|
132 %tmp = tail call float @llvm.copysign.f32( float 1.0, float -2.0 )
|
|
|
133 ret float %tmp
|
|
|
134 }
|
|
|
135
|
|
120
|
136 define double @cst2() nounwind {
|
|
|
137 ; X32-LABEL: cst2:
|
|
|
138 ; X32: # BB#0:
|
|
|
139 ; X32-NEXT: fldz
|
|
|
140 ; X32-NEXT: fchs
|
|
|
141 ; X32-NEXT: retl
|
|
95
|
142 ;
|
|
120
|
143 ; X64-LABEL: cst2:
|
|
|
144 ; X64: # BB#0:
|
|
|
145 ; X64-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
|
|
|
146 ; X64-NEXT: retq
|
|
95
|
147 %tmp1 = fadd float -1.0, -1.0
|
|
|
148 %tmp2 = fpext float %tmp1 to double
|
|
|
149 %tmp = tail call double @llvm.copysign.f64( double 0.0, double %tmp2 )
|
|
|
150 ret double %tmp
|
|
|
151 }
|
|
|
152
|
|
|
153 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
|
|
|
154 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
|
