1; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck %s
2
3declare float @llvm.fmuladd.f32(float, float, float)
4declare double @llvm.fmuladd.f64(double, double, double)
5declare i32 @llvm.r600.read.tidig.x() nounwind readnone
6declare float @llvm.fabs.f32(float) nounwind readnone
7
8; CHECK-LABEL: {{^}}fmuladd_f32:
9; CHECK: v_mad_f32 {{v[0-9]+, v[0-9]+, v[0-9]+, v[0-9]+}}
10
11define void @fmuladd_f32(float addrspace(1)* %out, float addrspace(1)* %in1,
12                         float addrspace(1)* %in2, float addrspace(1)* %in3) {
13   %r0 = load float, float addrspace(1)* %in1
14   %r1 = load float, float addrspace(1)* %in2
15   %r2 = load float, float addrspace(1)* %in3
16   %r3 = tail call float @llvm.fmuladd.f32(float %r0, float %r1, float %r2)
17   store float %r3, float addrspace(1)* %out
18   ret void
19}
20
21; CHECK-LABEL: {{^}}fmuladd_f64:
22; CHECK: v_fma_f64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\]}}
23
24define void @fmuladd_f64(double addrspace(1)* %out, double addrspace(1)* %in1,
25                         double addrspace(1)* %in2, double addrspace(1)* %in3) {
26   %r0 = load double, double addrspace(1)* %in1
27   %r1 = load double, double addrspace(1)* %in2
28   %r2 = load double, double addrspace(1)* %in3
29   %r3 = tail call double @llvm.fmuladd.f64(double %r0, double %r1, double %r2)
30   store double %r3, double addrspace(1)* %out
31   ret void
32}
33
34; CHECK-LABEL: {{^}}fmuladd_2.0_a_b_f32
35; CHECK-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
36; CHECK-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
37; CHECK: v_mad_f32 [[RESULT:v[0-9]+]], 2.0, [[R1]], [[R2]]
38; CHECK: buffer_store_dword [[RESULT]]
39define void @fmuladd_2.0_a_b_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
40  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
41  %gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
42  %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
43  %gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
44
45  %r1 = load float, float addrspace(1)* %gep.0
46  %r2 = load float, float addrspace(1)* %gep.1
47
48  %r3 = tail call float @llvm.fmuladd.f32(float 2.0, float %r1, float %r2)
49  store float %r3, float addrspace(1)* %gep.out
50  ret void
51}
52
53; CHECK-LABEL: {{^}}fmuladd_a_2.0_b_f32
54; CHECK-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
55; CHECK-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
56; CHECK: v_mad_f32 [[RESULT:v[0-9]+]], 2.0, [[R1]], [[R2]]
57; CHECK: buffer_store_dword [[RESULT]]
58define void @fmuladd_a_2.0_b_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
59  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
60  %gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
61  %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
62  %gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
63
64  %r1 = load float, float addrspace(1)* %gep.0
65  %r2 = load float, float addrspace(1)* %gep.1
66
67  %r3 = tail call float @llvm.fmuladd.f32(float %r1, float 2.0, float %r2)
68  store float %r3, float addrspace(1)* %gep.out
69  ret void
70}
71
72; CHECK-LABEL: {{^}}fadd_a_a_b_f32:
73; CHECK-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
74; CHECK-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
75; CHECK: v_mad_f32 [[RESULT:v[0-9]+]], 2.0, [[R1]], [[R2]]
76; CHECK: buffer_store_dword [[RESULT]]
77define void @fadd_a_a_b_f32(float addrspace(1)* %out,
78                            float addrspace(1)* %in1,
79                            float addrspace(1)* %in2) {
80  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
81  %gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
82  %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
83  %gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
84
85  %r0 = load float, float addrspace(1)* %gep.0
86  %r1 = load float, float addrspace(1)* %gep.1
87
88  %add.0 = fadd float %r0, %r0
89  %add.1 = fadd float %add.0, %r1
90  store float %add.1, float addrspace(1)* %out
91  ret void
92}
93
94; CHECK-LABEL: {{^}}fadd_b_a_a_f32:
95; CHECK-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
96; CHECK-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
97; CHECK: v_mad_f32 [[RESULT:v[0-9]+]], 2.0, [[R1]], [[R2]]
98; CHECK: buffer_store_dword [[RESULT]]
99define void @fadd_b_a_a_f32(float addrspace(1)* %out,
100                            float addrspace(1)* %in1,
101                            float addrspace(1)* %in2) {
102  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
103  %gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
104  %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
105  %gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
106
107  %r0 = load float, float addrspace(1)* %gep.0
108  %r1 = load float, float addrspace(1)* %gep.1
109
110  %add.0 = fadd float %r0, %r0
111  %add.1 = fadd float %r1, %add.0
112  store float %add.1, float addrspace(1)* %out
113  ret void
114}
115
116; CHECK-LABEL: {{^}}fmuladd_neg_2.0_a_b_f32
117; CHECK-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
118; CHECK-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
119; CHECK: v_mad_f32 [[RESULT:v[0-9]+]], -2.0, [[R1]], [[R2]]
120; CHECK: buffer_store_dword [[RESULT]]
121define void @fmuladd_neg_2.0_a_b_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
122  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
123  %gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
124  %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
125  %gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
126
127  %r1 = load float, float addrspace(1)* %gep.0
128  %r2 = load float, float addrspace(1)* %gep.1
129
130  %r3 = tail call float @llvm.fmuladd.f32(float -2.0, float %r1, float %r2)
131  store float %r3, float addrspace(1)* %gep.out
132  ret void
133}
134
135
136; CHECK-LABEL: {{^}}fmuladd_neg_2.0_neg_a_b_f32
137; CHECK-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
138; CHECK-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
139; CHECK: v_mad_f32 [[RESULT:v[0-9]+]], 2.0, [[R1]], [[R2]]
140; CHECK: buffer_store_dword [[RESULT]]
141define void @fmuladd_neg_2.0_neg_a_b_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
142  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
143  %gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
144  %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
145  %gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
146
147  %r1 = load float, float addrspace(1)* %gep.0
148  %r2 = load float, float addrspace(1)* %gep.1
149
150  %r1.fneg = fsub float -0.000000e+00, %r1
151
152  %r3 = tail call float @llvm.fmuladd.f32(float -2.0, float %r1.fneg, float %r2)
153  store float %r3, float addrspace(1)* %gep.out
154  ret void
155}
156
157
158; CHECK-LABEL: {{^}}fmuladd_2.0_neg_a_b_f32
159; CHECK-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
160; CHECK-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
161; CHECK: v_mad_f32 [[RESULT:v[0-9]+]], -2.0, [[R1]], [[R2]]
162; CHECK: buffer_store_dword [[RESULT]]
163define void @fmuladd_2.0_neg_a_b_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
164  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
165  %gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
166  %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
167  %gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
168
169  %r1 = load float, float addrspace(1)* %gep.0
170  %r2 = load float, float addrspace(1)* %gep.1
171
172  %r1.fneg = fsub float -0.000000e+00, %r1
173
174  %r3 = tail call float @llvm.fmuladd.f32(float 2.0, float %r1.fneg, float %r2)
175  store float %r3, float addrspace(1)* %gep.out
176  ret void
177}
178
179
180; CHECK-LABEL: {{^}}fmuladd_2.0_a_neg_b_f32
181; CHECK-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
182; CHECK-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
183; CHECK: v_mad_f32 [[RESULT:v[0-9]+]], 2.0, [[R1]], -[[R2]]
184; CHECK: buffer_store_dword [[RESULT]]
185define void @fmuladd_2.0_a_neg_b_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
186  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
187  %gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
188  %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
189  %gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
190
191  %r1 = load float, float addrspace(1)* %gep.0
192  %r2 = load float, float addrspace(1)* %gep.1
193
194  %r2.fneg = fsub float -0.000000e+00, %r2
195
196  %r3 = tail call float @llvm.fmuladd.f32(float 2.0, float %r1, float %r2.fneg)
197  store float %r3, float addrspace(1)* %gep.out
198  ret void
199}
200