1 ; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2 ; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
4 declare float @llvm.fma.f32(float, float, float) nounwind readnone
5 declare <2 x float> @llvm.fma.v2f32(<2 x float>, <2 x float>, <2 x float>) nounwind readnone
6 declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) nounwind readnone
8 ; FUNC-LABEL: {{^}}fma_f32:
9 ; SI: v_fma_f32 {{v[0-9]+, v[0-9]+, v[0-9]+, v[0-9]+}}
11 ; EG: MEM_RAT_{{.*}} STORE_{{.*}} [[RES:T[0-9]\.[XYZW]]], {{T[0-9]\.[XYZW]}},
12 ; EG: FMA {{\*? *}}[[RES]]
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:
27 ; EG: MEM_RAT_{{.*}} STORE_{{.*}} [[RES:T[0-9]]].[[CHLO:[XYZW]]][[CHHI:[XYZW]]], {{T[0-9]\.[XYZW]}},
28 ; EG-DAG: FMA {{\*? *}}[[RES]].[[CHLO]]
29 ; EG-DAG: FMA {{\*? *}}[[RES]].[[CHHI]]
30 define void @fma_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %in1,
31 <2 x float> addrspace(1)* %in2, <2 x float> addrspace(1)* %in3) {
32 %r0 = load <2 x float> addrspace(1)* %in1
33 %r1 = load <2 x float> addrspace(1)* %in2
34 %r2 = load <2 x float> addrspace(1)* %in3
35 %r3 = tail call <2 x float> @llvm.fma.v2f32(<2 x float> %r0, <2 x float> %r1, <2 x float> %r2)
36 store <2 x float> %r3, <2 x float> addrspace(1)* %out
40 ; FUNC-LABEL: {{^}}fma_v4f32:
46 ; EG: MEM_RAT_{{.*}} STORE_{{.*}} [[RES:T[0-9]]].{{[XYZW][XYZW][XYZW][XYZW]}}, {{T[0-9]\.[XYZW]}},
47 ; EG-DAG: FMA {{\*? *}}[[RES]].X
48 ; EG-DAG: FMA {{\*? *}}[[RES]].Y
49 ; EG-DAG: FMA {{\*? *}}[[RES]].Z
50 ; EG-DAG: FMA {{\*? *}}[[RES]].W
51 define void @fma_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in1,
52 <4 x float> addrspace(1)* %in2, <4 x float> addrspace(1)* %in3) {
53 %r0 = load <4 x float> addrspace(1)* %in1
54 %r1 = load <4 x float> addrspace(1)* %in2
55 %r2 = load <4 x float> addrspace(1)* %in3
56 %r3 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %r0, <4 x float> %r1, <4 x float> %r2)
57 store <4 x float> %r3, <4 x float> addrspace(1)* %out