• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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
6
7declare float @llvm.fabs.f32(float) #0
8declare float @llvm.floor.f32(float) #0
9
10; GCN-LABEL: {{^}}fract_f32:
11; GCN-SAFE: v_floor_f32_e32 [[FLR:v[0-9]+]], [[INPUT:v[0-9]+]]
12; GCN-SAFE: v_sub_f32_e32 [[RESULT:v[0-9]+]], [[INPUT]], [[FLR]]
13
14; GCN-UNSAFE: v_fract_f32_e32 [[RESULT:v[0-9]+]], [[INPUT:v[0-9]+]]
15
16; GCN: buffer_store_dword [[RESULT]]
17define amdgpu_kernel void @fract_f32(float addrspace(1)* %out, float addrspace(1)* %src) #1 {
18  %x = load float, float addrspace(1)* %src
19  %floor.x = call float @llvm.floor.f32(float %x)
20  %fract = fsub float %x, %floor.x
21  store float %fract, float addrspace(1)* %out
22  ret void
23}
24
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
31; GCN: buffer_store_dword [[RESULT]]
32define amdgpu_kernel void @fract_f32_neg(float addrspace(1)* %out, float addrspace(1)* %src) #1 {
33  %x = load float, float addrspace(1)* %src
34  %x.neg = fsub float -0.0, %x
35  %floor.x.neg = call float @llvm.floor.f32(float %x.neg)
36  %fract = fsub float %x.neg, %floor.x.neg
37  store float %fract, float addrspace(1)* %out
38  ret void
39}
40
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
47; GCN: buffer_store_dword [[RESULT]]
48define amdgpu_kernel void @fract_f32_neg_abs(float addrspace(1)* %out, float addrspace(1)* %src) #1 {
49  %x = load float, float addrspace(1)* %src
50  %abs.x = call float @llvm.fabs.f32(float %x)
51  %neg.abs.x = fsub float -0.0, %abs.x
52  %floor.neg.abs.x = call float @llvm.floor.f32(float %neg.abs.x)
53  %fract = fsub float %neg.abs.x, %floor.neg.abs.x
54  store float %fract, float addrspace(1)* %out
55  ret void
56}
57
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]]
64define amdgpu_kernel void @multi_use_floor_fract_f32(float addrspace(1)* %out, float addrspace(1)* %src) #1 {
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
73attributes #0 = { nounwind readnone }
74attributes #1 = { nounwind }
75