121
|
1 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN-SAFE -check-prefix=GCN -check-prefix=SI %s
|
|
2 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN-SAFE -check-prefix=GCN -check-prefix=CI %s
|
|
3 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN-SAFE -check-prefix=GCN -check-prefix=FUNC %s
|
|
4 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN-UNSAFE -check-prefix=GCN %s
|
|
5 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN-UNSAFE -check-prefix=GCN %s
|
100
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
6
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
7 declare float @llvm.fabs.f32(float) #0
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
8 declare float @llvm.floor.f32(float) #0
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
9
|
120
|
10 ; GCN-LABEL: {{^}}fract_f32:
|
|
11 ; GCN-SAFE: v_floor_f32_e32 [[FLR:v[0-9]+]], [[INPUT:v[0-9]+]]
|
121
|
12 ; GCN-SAFE: v_sub_f32_e32 [[RESULT:v[0-9]+]], [[INPUT]], [[FLR]]
|
120
|
13
|
|
14 ; GCN-UNSAFE: v_fract_f32_e32 [[RESULT:v[0-9]+]], [[INPUT:v[0-9]+]]
|
|
15
|
100
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
16 ; GCN: buffer_store_dword [[RESULT]]
|
121
|
17 define amdgpu_kernel void @fract_f32(float addrspace(1)* %out, float addrspace(1)* %src) #1 {
|
100
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
18 %x = load float, float addrspace(1)* %src
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
19 %floor.x = call float @llvm.floor.f32(float %x)
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
20 %fract = fsub float %x, %floor.x
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
21 store float %fract, float addrspace(1)* %out
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
22 ret void
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
23 }
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
24
|
120
|
25 ; GCN-LABEL: {{^}}fract_f32_neg:
|
|
26 ; GCN-SAFE: v_floor_f32_e64 [[FLR:v[0-9]+]], -[[INPUT:v[0-9]+]]
|
|
27 ; GCN-SAFE: v_sub_f32_e64 [[RESULT:v[0-9]+]], -[[INPUT]], [[FLR]]
|
|
28
|
|
29 ; GCN-UNSAFE: v_fract_f32_e64 [[RESULT:v[0-9]+]], -[[INPUT:v[0-9]+]]
|
|
30
|
100
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
31 ; GCN: buffer_store_dword [[RESULT]]
|
121
|
32 define amdgpu_kernel void @fract_f32_neg(float addrspace(1)* %out, float addrspace(1)* %src) #1 {
|
100
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
33 %x = load float, float addrspace(1)* %src
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
34 %x.neg = fsub float -0.0, %x
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
35 %floor.x.neg = call float @llvm.floor.f32(float %x.neg)
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
36 %fract = fsub float %x.neg, %floor.x.neg
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
37 store float %fract, float addrspace(1)* %out
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
38 ret void
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
39 }
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
40
|
120
|
41 ; GCN-LABEL: {{^}}fract_f32_neg_abs:
|
|
42 ; GCN-SAFE: v_floor_f32_e64 [[FLR:v[0-9]+]], -|[[INPUT:v[0-9]+]]|
|
|
43 ; GCN-SAFE: v_sub_f32_e64 [[RESULT:v[0-9]+]], -|[[INPUT]]|, [[FLR]]
|
|
44
|
|
45 ; GCN-UNSAFE: v_fract_f32_e64 [[RESULT:v[0-9]+]], -|[[INPUT:v[0-9]+]]|
|
|
46
|
100
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
47 ; GCN: buffer_store_dword [[RESULT]]
|
121
|
48 define amdgpu_kernel void @fract_f32_neg_abs(float addrspace(1)* %out, float addrspace(1)* %src) #1 {
|
100
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
49 %x = load float, float addrspace(1)* %src
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
50 %abs.x = call float @llvm.fabs.f32(float %x)
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
51 %neg.abs.x = fsub float -0.0, %abs.x
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
52 %floor.neg.abs.x = call float @llvm.floor.f32(float %neg.abs.x)
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
53 %fract = fsub float %neg.abs.x, %floor.neg.abs.x
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
54 store float %fract, float addrspace(1)* %out
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
55 ret void
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
56 }
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
57
|
120
|
58 ; GCN-LABEL: {{^}}multi_use_floor_fract_f32:
|
|
59 ; GCN-UNSAFE-DAG: v_floor_f32_e32 [[FLOOR:v[0-9]+]], [[INPUT:v[0-9]+]]
|
|
60 ; GCN-UNSAFE-DAG: v_fract_f32_e32 [[FRACT:v[0-9]+]], [[INPUT:v[0-9]+]]
|
|
61
|
|
62 ; GCN-UNSAFE: buffer_store_dword [[FLOOR]]
|
|
63 ; GCN-UNSAFE: buffer_store_dword [[FRACT]]
|
121
|
64 define amdgpu_kernel void @multi_use_floor_fract_f32(float addrspace(1)* %out, float addrspace(1)* %src) #1 {
|
120
|
65 %x = load float, float addrspace(1)* %src
|
|
66 %floor.x = call float @llvm.floor.f32(float %x)
|
|
67 %fract = fsub float %x, %floor.x
|
|
68 store volatile float %floor.x, float addrspace(1)* %out
|
|
69 store volatile float %fract, float addrspace(1)* %out
|
|
70 ret void
|
|
71 }
|
|
72
|
100
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
73 attributes #0 = { nounwind readnone }
|
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
74 attributes #1 = { nounwind }
|