1 ; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
3 declare float @llvm.fma.f32(float, float, float) nounwind readnone
4 declare <2 x float> @llvm.fma.v2f32(<2 x float>, <2 x float>, <2 x float>) nounwind readnone
5 declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) nounwind readnone
7 declare double @llvm.fma.f64(double, double, double) nounwind readnone
8 declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) nounwind readnone
9 declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>) nounwind readnone
11 ; FUNC-LABEL: @fma_f32
12 ; SI: V_FMA_F32 {{v[0-9]+, v[0-9]+, v[0-9]+, v[0-9]+}}
13 define void @fma_f32(float addrspace(1)* %out, float addrspace(1)* %in1,
14 float addrspace(1)* %in2, float addrspace(1)* %in3) {
15 %r0 = load float addrspace(1)* %in1
16 %r1 = load float addrspace(1)* %in2
17 %r2 = load float addrspace(1)* %in3
18 %r3 = tail call float @llvm.fma.f32(float %r0, float %r1, float %r2)
19 store float %r3, float addrspace(1)* %out
23 ; FUNC-LABEL: @fma_v2f32
26 define void @fma_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %in1,
27 <2 x float> addrspace(1)* %in2, <2 x float> addrspace(1)* %in3) {
28 %r0 = load <2 x float> addrspace(1)* %in1
29 %r1 = load <2 x float> addrspace(1)* %in2
30 %r2 = load <2 x float> addrspace(1)* %in3
31 %r3 = tail call <2 x float> @llvm.fma.v2f32(<2 x float> %r0, <2 x float> %r1, <2 x float> %r2)
32 store <2 x float> %r3, <2 x float> addrspace(1)* %out
36 ; FUNC-LABEL: @fma_v4f32
41 define void @fma_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in1,
42 <4 x float> addrspace(1)* %in2, <4 x float> addrspace(1)* %in3) {
43 %r0 = load <4 x float> addrspace(1)* %in1
44 %r1 = load <4 x float> addrspace(1)* %in2
45 %r2 = load <4 x float> addrspace(1)* %in3
46 %r3 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %r0, <4 x float> %r1, <4 x float> %r2)
47 store <4 x float> %r3, <4 x float> addrspace(1)* %out
51 ; FUNC-LABEL: @fma_f64
52 ; SI: V_FMA_F64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\]}}
53 define void @fma_f64(double addrspace(1)* %out, double addrspace(1)* %in1,
54 double addrspace(1)* %in2, double addrspace(1)* %in3) {
55 %r0 = load double addrspace(1)* %in1
56 %r1 = load double addrspace(1)* %in2
57 %r2 = load double addrspace(1)* %in3
58 %r3 = tail call double @llvm.fma.f64(double %r0, double %r1, double %r2)
59 store double %r3, double addrspace(1)* %out
63 ; FUNC-LABEL: @fma_v2f64
66 define void @fma_v2f64(<2 x double> addrspace(1)* %out, <2 x double> addrspace(1)* %in1,
67 <2 x double> addrspace(1)* %in2, <2 x double> addrspace(1)* %in3) {
68 %r0 = load <2 x double> addrspace(1)* %in1
69 %r1 = load <2 x double> addrspace(1)* %in2
70 %r2 = load <2 x double> addrspace(1)* %in3
71 %r3 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %r0, <2 x double> %r1, <2 x double> %r2)
72 store <2 x double> %r3, <2 x double> addrspace(1)* %out
76 ; FUNC-LABEL: @fma_v4f64
81 define void @fma_v4f64(<4 x double> addrspace(1)* %out, <4 x double> addrspace(1)* %in1,
82 <4 x double> addrspace(1)* %in2, <4 x double> addrspace(1)* %in3) {
83 %r0 = load <4 x double> addrspace(1)* %in1
84 %r1 = load <4 x double> addrspace(1)* %in2
85 %r2 = load <4 x double> addrspace(1)* %in3
86 %r3 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %r0, <4 x double> %r1, <4 x double> %r2)
87 store <4 x double> %r3, <4 x double> addrspace(1)* %out