1 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2 ; RUN: llc -amdgpu-scalarize-global-loads=false -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 declare i32 @llvm.r600.read.tidig.x() nounwind readnone
10 ; FUNC-LABEL: {{^}}fma_f32:
11 ; SI: v_fma_f32 {{v[0-9]+, v[0-9]+, v[0-9]+, v[0-9]+}}
13 ; EG: MEM_RAT_{{.*}} STORE_{{.*}} [[RES:T[0-9]\.[XYZW]]], {{T[0-9]\.[XYZW]}},
14 ; EG: FMA {{\*? *}}[[RES]]
15 define amdgpu_kernel void @fma_f32(float addrspace(1)* %out, float addrspace(1)* %in1,
16 float addrspace(1)* %in2, float addrspace(1)* %in3) {
17 %r0 = load float, float addrspace(1)* %in1
18 %r1 = load float, float addrspace(1)* %in2
19 %r2 = load float, float addrspace(1)* %in3
20 %r3 = tail call float @llvm.fma.f32(float %r0, float %r1, float %r2)
21 store float %r3, float addrspace(1)* %out
25 ; FUNC-LABEL: {{^}}fma_v2f32:
29 ; EG: MEM_RAT_{{.*}} STORE_{{.*}} [[RES:T[0-9]]].[[CHLO:[XYZW]]][[CHHI:[XYZW]]], {{T[0-9]\.[XYZW]}},
30 ; EG-DAG: FMA {{\*? *}}[[RES]].[[CHLO]]
31 ; EG-DAG: FMA {{\*? *}}[[RES]].[[CHHI]]
32 define amdgpu_kernel void @fma_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %in1,
33 <2 x float> addrspace(1)* %in2, <2 x float> addrspace(1)* %in3) {
34 %r0 = load <2 x float>, <2 x float> addrspace(1)* %in1
35 %r1 = load <2 x float>, <2 x float> addrspace(1)* %in2
36 %r2 = load <2 x float>, <2 x float> addrspace(1)* %in3
37 %r3 = tail call <2 x float> @llvm.fma.v2f32(<2 x float> %r0, <2 x float> %r1, <2 x float> %r2)
38 store <2 x float> %r3, <2 x float> addrspace(1)* %out
42 ; FUNC-LABEL: {{^}}fma_v4f32:
48 ; EG: MEM_RAT_{{.*}} STORE_{{.*}} [[RES:T[0-9]]].{{[XYZW][XYZW][XYZW][XYZW]}}, {{T[0-9]\.[XYZW]}},
49 ; EG-DAG: FMA {{\*? *}}[[RES]].X
50 ; EG-DAG: FMA {{\*? *}}[[RES]].Y
51 ; EG-DAG: FMA {{\*? *}}[[RES]].Z
52 ; EG-DAG: FMA {{\*? *}}[[RES]].W
53 define amdgpu_kernel void @fma_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in1,
54 <4 x float> addrspace(1)* %in2, <4 x float> addrspace(1)* %in3) {
55 %r0 = load <4 x float>, <4 x float> addrspace(1)* %in1
56 %r1 = load <4 x float>, <4 x float> addrspace(1)* %in2
57 %r2 = load <4 x float>, <4 x float> addrspace(1)* %in3
58 %r3 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %r0, <4 x float> %r1, <4 x float> %r2)
59 store <4 x float> %r3, <4 x float> addrspace(1)* %out
63 ; FUNC-LABEL: @fma_commute_mul_inline_imm_f32
64 ; SI: v_fma_f32 {{v[0-9]+}}, {{v[0-9]+}}, 2.0, {{v[0-9]+}}
65 define amdgpu_kernel void @fma_commute_mul_inline_imm_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in.a, float addrspace(1)* noalias %in.b) nounwind {
66 %tid = tail call i32 @llvm.r600.read.tidig.x() nounwind readnone
67 %in.a.gep = getelementptr float, float addrspace(1)* %in.a, i32 %tid
68 %in.b.gep = getelementptr float, float addrspace(1)* %in.b, i32 %tid
69 %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
71 %a = load float, float addrspace(1)* %in.a.gep, align 4
72 %b = load float, float addrspace(1)* %in.b.gep, align 4
74 %fma = call float @llvm.fma.f32(float %a, float 2.0, float %b)
75 store float %fma, float addrspace(1)* %out.gep, align 4
79 ; FUNC-LABEL: @fma_commute_mul_s_f32
80 define amdgpu_kernel void @fma_commute_mul_s_f32(float addrspace(1)* noalias %out, float addrspace(1)* noalias %in.a, float addrspace(1)* noalias %in.b, float %b) nounwind {
81 %tid = tail call i32 @llvm.r600.read.tidig.x() nounwind readnone
82 %in.a.gep = getelementptr float, float addrspace(1)* %in.a, i32 %tid
83 %in.b.gep = getelementptr float, float addrspace(1)* %in.b, i32 %tid
84 %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
86 %a = load float, float addrspace(1)* %in.a.gep, align 4
87 %c = load float, float addrspace(1)* %in.b.gep, align 4
89 %fma = call float @llvm.fma.f32(float %a, float %b, float %c)
90 store float %fma, float addrspace(1)* %out.gep, align 4