1; RUN: llc < %s -mtriple=i386-apple-darwin10  -mattr=+fma,-fma4  | FileCheck %s --check-prefix=CHECK-FMA-INST
2; RUN: llc < %s -mtriple=i386-apple-darwin10  -mattr=-fma,-fma4  | FileCheck %s --check-prefix=CHECK-FMA-CALL
3; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -mattr=+fma,-fma4 | FileCheck %s --check-prefix=CHECK-FMA-INST
4; RUN: llc < %s -mtriple=x86_64-apple-darwin10  -mattr=-fma,-fma4 | FileCheck %s --check-prefix=CHECK-FMA-CALL
5; RUN: llc < %s -march=x86 -mcpu=bdver2 -mattr=-fma4  | FileCheck %s --check-prefix=CHECK-FMA-INST
6; RUN: llc < %s -march=x86 -mcpu=bdver2 -mattr=-fma,-fma4 | FileCheck %s --check-prefix=CHECK-FMA-CALL
7
8; CHECK: test_f32
9; CHECK-FMA-INST: vfmadd213ss
10; CHECK-FMA-CALL: fmaf
11
12define float @test_f32(float %a, float %b, float %c) nounwind readnone ssp {
13entry:
14  %call = tail call float @llvm.fma.f32(float %a, float %b, float %c) nounwind readnone
15  ret float %call
16}
17
18; CHECK: test_f64
19; CHECK-FMA-INST: vfmadd213sd
20; CHECK-FMA-CALL: fma
21
22define double @test_f64(double %a, double %b, double %c) nounwind readnone ssp {
23entry:
24  %call = tail call double @llvm.fma.f64(double %a, double %b, double %c) nounwind readnone
25  ret double %call
26}
27
28; CHECK: test_f80
29; CHECK: fmal
30
31define x86_fp80 @test_f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c) nounwind readnone ssp {
32entry:
33  %call = tail call x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c) nounwind readnone
34  ret x86_fp80 %call
35}
36
37; CHECK: test_f32_cst
38; CHECK-NOT: fma
39define float @test_f32_cst() nounwind readnone ssp {
40entry:
41  %call = tail call float @llvm.fma.f32(float 3.0, float 3.0, float 3.0) nounwind readnone
42  ret float %call
43}
44
45; Test FMA3 variant selection
46; CHECK-FMA-INST: fma3_select231ssX:
47; CHECK-FMA-INST: vfmadd231ss %xmm
48define float @fma3_select231ssX(float %x, float %y) #0 {
49entry:
50  br label %while.body
51while.body:                                       ; preds = %while.body, %while.body
52  %acc.01 = phi float [ 0.000000e+00, %entry ], [ %acc, %while.body ]
53  %acc = tail call float @llvm.fma.f32(float %x, float %y, float %acc.01) nounwind readnone
54  %b = fcmp ueq float %acc, 0.0
55  br i1 %b, label %while.body, label %while.end
56while.end:                                        ; preds = %while.body, %entry
57  ret float %acc
58}
59
60; Test FMA3 variant selection
61; CHECK-FMA-INST: fma3_select231pdY:
62; CHECK-FMA-INST: vfmadd231pd %ymm
63define <4 x double> @fma3_select231pdY(<4 x double> %x, <4 x double> %y) #0 {
64entry:
65  br label %while.body
66while.body:                                       ; preds = %entry, %while.body
67  %acc.04 = phi <4 x double> [ zeroinitializer, %entry ], [ %add, %while.body ]
68  %add = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %x, <4 x double> %y, <4 x double> %acc.04)
69  %vecext = extractelement <4 x double> %add, i32 0
70  %cmp = fcmp oeq double %vecext, 0.000000e+00
71  br i1 %cmp, label %while.body, label %while.end
72
73while.end:                                        ; preds = %while.body
74  ret <4 x double> %add
75}
76
77declare float @llvm.fma.f32(float, float, float) nounwind readnone
78declare double @llvm.fma.f64(double, double, double) nounwind readnone
79declare x86_fp80 @llvm.fma.f80(x86_fp80, x86_fp80, x86_fp80) nounwind readnone
80declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>) nounwind readnone
81