1; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 -enable-unsafe-fp-math | FileCheck %s
2; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 | FileCheck -check-prefix=CHECK-SAFE %s
3target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
4target triple = "powerpc64-unknown-linux-gnu"
5
6declare double @llvm.sqrt.f64(double)
7declare float @llvm.sqrt.f32(float)
8declare <4 x float> @llvm.sqrt.v4f32(<4 x float>)
9
10define double @foo(double %a, double %b) nounwind {
11entry:
12  %x = call double @llvm.sqrt.f64(double %b)
13  %r = fdiv double %a, %x
14  ret double %r
15
16; CHECK: @foo
17; CHECK-DAG: frsqrte
18; CHECK-DAG: fnmsub
19; CHECK: fmul
20; CHECK: fmadd
21; CHECK: fmul
22; CHECK: fmul
23; CHECK: fmadd
24; CHECK: fmul
25; CHECK: fmul
26; CHECK: blr
27
28; CHECK-SAFE: @foo
29; CHECK-SAFE: fsqrt
30; CHECK-SAFE: fdiv
31; CHECK-SAFE: blr
32}
33
34define double @foof(double %a, float %b) nounwind {
35entry:
36  %x = call float @llvm.sqrt.f32(float %b)
37  %y = fpext float %x to double
38  %r = fdiv double %a, %y
39  ret double %r
40
41; CHECK: @foof
42; CHECK-DAG: frsqrtes
43; CHECK-DAG: fnmsubs
44; CHECK: fmuls
45; CHECK: fmadds
46; CHECK: fmuls
47; CHECK: fmul
48; CHECK: blr
49
50; CHECK-SAFE: @foof
51; CHECK-SAFE: fsqrts
52; CHECK-SAFE: fdiv
53; CHECK-SAFE: blr
54}
55
56define float @food(float %a, double %b) nounwind {
57entry:
58  %x = call double @llvm.sqrt.f64(double %b)
59  %y = fptrunc double %x to float
60  %r = fdiv float %a, %y
61  ret float %r
62
63; CHECK: @foo
64; CHECK-DAG: frsqrte
65; CHECK-DAG: fnmsub
66; CHECK: fmul
67; CHECK: fmadd
68; CHECK: fmul
69; CHECK: fmul
70; CHECK: fmadd
71; CHECK: fmul
72; CHECK: frsp
73; CHECK: fmuls
74; CHECK: blr
75
76; CHECK-SAFE: @foo
77; CHECK-SAFE: fsqrt
78; CHECK-SAFE: fdivs
79; CHECK-SAFE: blr
80}
81
82define float @goo(float %a, float %b) nounwind {
83entry:
84  %x = call float @llvm.sqrt.f32(float %b)
85  %r = fdiv float %a, %x
86  ret float %r
87
88; CHECK: @goo
89; CHECK-DAG: frsqrtes
90; CHECK-DAG: fnmsubs
91; CHECK: fmuls
92; CHECK: fmadds
93; CHECK: fmuls
94; CHECK: fmuls
95; CHECK: blr
96
97; CHECK-SAFE: @goo
98; CHECK-SAFE: fsqrts
99; CHECK-SAFE: fdivs
100; CHECK-SAFE: blr
101}
102
103define <4 x float> @hoo(<4 x float> %a, <4 x float> %b) nounwind {
104entry:
105  %x = call <4 x float> @llvm.sqrt.v4f32(<4 x float> %b)
106  %r = fdiv <4 x float> %a, %x
107  ret <4 x float> %r
108
109; CHECK: @hoo
110; CHECK: vrsqrtefp
111
112; CHECK-SAFE: @hoo
113; CHECK-SAFE-NOT: vrsqrtefp
114; CHECK-SAFE: blr
115}
116
117define double @foo2(double %a, double %b) nounwind {
118entry:
119  %r = fdiv double %a, %b
120  ret double %r
121
122; CHECK: @foo2
123; CHECK-DAG: fre
124; CHECK-DAG: fnmsub
125; CHECK: fmadd
126; CHECK: fnmsub
127; CHECK: fmadd
128; CHECK: fmul
129; CHECK: blr
130
131; CHECK-SAFE: @foo2
132; CHECK-SAFE: fdiv
133; CHECK-SAFE: blr
134}
135
136define float @goo2(float %a, float %b) nounwind {
137entry:
138  %r = fdiv float %a, %b
139  ret float %r
140
141; CHECK: @goo2
142; CHECK-DAG: fres
143; CHECK-DAG: fnmsubs
144; CHECK: fmadds
145; CHECK: fmuls
146; CHECK: blr
147
148; CHECK-SAFE: @goo2
149; CHECK-SAFE: fdivs
150; CHECK-SAFE: blr
151}
152
153define <4 x float> @hoo2(<4 x float> %a, <4 x float> %b) nounwind {
154entry:
155  %r = fdiv <4 x float> %a, %b
156  ret <4 x float> %r
157
158; CHECK: @hoo2
159; CHECK: vrefp
160
161; CHECK-SAFE: @hoo2
162; CHECK-SAFE-NOT: vrefp
163; CHECK-SAFE: blr
164}
165
166define double @foo3(double %a) nounwind {
167entry:
168  %r = call double @llvm.sqrt.f64(double %a)
169  ret double %r
170
171; CHECK: @foo3
172; CHECK: fcmpu
173; CHECK-DAG: frsqrte
174; CHECK-DAG: fnmsub
175; CHECK: fmul
176; CHECK: fmadd
177; CHECK: fmul
178; CHECK: fmul
179; CHECK: fmadd
180; CHECK: fmul
181; CHECK: fre
182; CHECK: fnmsub
183; CHECK: fmadd
184; CHECK: fnmsub
185; CHECK: fmadd
186; CHECK: blr
187
188; CHECK-SAFE: @foo3
189; CHECK-SAFE: fsqrt
190; CHECK-SAFE: blr
191}
192
193define float @goo3(float %a) nounwind {
194entry:
195  %r = call float @llvm.sqrt.f32(float %a)
196  ret float %r
197
198; CHECK: @goo3
199; CHECK: fcmpu
200; CHECK-DAG: frsqrtes
201; CHECK-DAG: fnmsubs
202; CHECK: fmuls
203; CHECK: fmadds
204; CHECK: fmuls
205; CHECK: fres
206; CHECK: fnmsubs
207; CHECK: fmadds
208; CHECK: blr
209
210; CHECK-SAFE: @goo3
211; CHECK-SAFE: fsqrts
212; CHECK-SAFE: blr
213}
214
215define <4 x float> @hoo3(<4 x float> %a) nounwind {
216entry:
217  %r = call <4 x float> @llvm.sqrt.v4f32(<4 x float> %a)
218  ret <4 x float> %r
219
220; CHECK: @hoo3
221; CHECK: vrsqrtefp
222; CHECK-DAG: vrefp
223; CHECK-DAG: vcmpeqfp
224
225; CHECK-SAFE: @hoo3
226; CHECK-SAFE-NOT: vrsqrtefp
227; CHECK-SAFE: blr
228}
229
230