100
|
1 ; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck %s
|
|
2 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck %s
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
3
|
100
|
4 ; CHECK-LABEL: {{^}}v_fadd_f64:
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
5 ; CHECK: v_add_f64 {{v[[0-9]+:[0-9]+]}}, {{v[[0-9]+:[0-9]+]}}, {{v[[0-9]+:[0-9]+]}}
|
121
|
6 define amdgpu_kernel void @v_fadd_f64(double addrspace(1)* %out, double addrspace(1)* %in1,
|
100
|
7 double addrspace(1)* %in2) {
|
121
|
8 %tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
9 %gep1 = getelementptr inbounds double, double addrspace(1)* %in1, i32 %tid
|
|
10 %gep2 = getelementptr inbounds double, double addrspace(1)* %in2, i32 %tid
|
|
11 %r0 = load double, double addrspace(1)* %gep1
|
|
12 %r1 = load double, double addrspace(1)* %gep2
|
100
|
13 %r2 = fadd double %r0, %r1
|
|
14 store double %r2, double addrspace(1)* %out
|
|
15 ret void
|
|
16 }
|
|
17
|
|
18 ; CHECK-LABEL: {{^}}s_fadd_f64:
|
121
|
19 ; CHECK: v_add_f64 {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}
|
|
20 define amdgpu_kernel void @s_fadd_f64(double addrspace(1)* %out, double %r0, double %r1) {
|
100
|
21 %r2 = fadd double %r0, %r1
|
|
22 store double %r2, double addrspace(1)* %out
|
|
23 ret void
|
|
24 }
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
25
|
100
|
26 ; CHECK-LABEL: {{^}}v_fadd_v2f64:
|
|
27 ; CHECK: v_add_f64
|
|
28 ; CHECK: v_add_f64
|
121
|
29 ; CHECK: _store_dwordx4
|
|
30 define amdgpu_kernel void @v_fadd_v2f64(<2 x double> addrspace(1)* %out, <2 x double> addrspace(1)* %in1,
|
100
|
31 <2 x double> addrspace(1)* %in2) {
|
|
32 %r0 = load <2 x double>, <2 x double> addrspace(1)* %in1
|
|
33 %r1 = load <2 x double>, <2 x double> addrspace(1)* %in2
|
|
34 %r2 = fadd <2 x double> %r0, %r1
|
|
35 store <2 x double> %r2, <2 x double> addrspace(1)* %out
|
|
36 ret void
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
37 }
|
100
|
38
|
|
39 ; CHECK-LABEL: {{^}}s_fadd_v2f64:
|
|
40 ; CHECK: v_add_f64 {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}
|
|
41 ; CHECK: v_add_f64 {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}
|
121
|
42 ; CHECK: _store_dwordx4
|
|
43 define amdgpu_kernel void @s_fadd_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %r0, <2 x double> %r1) {
|
100
|
44 %r2 = fadd <2 x double> %r0, %r1
|
|
45 store <2 x double> %r2, <2 x double> addrspace(1)* %out
|
|
46 ret void
|
|
47 }
|
121
|
48
|
|
49 declare i32 @llvm.amdgcn.workitem.id.x() #1
|
|
50
|
|
51 attributes #0 = { nounwind }
|
|
52 attributes #1 = { nounwind readnone }
|