• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -strict-whitespace -check-prefix=SI -check-prefix=FUNC %s
2; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -strict-whitespace -check-prefix=SI -check-prefix=FUNC %s
3; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
4
5declare float @llvm.fabs.f32(float) nounwind readnone
6declare float @llvm.AMDGPU.clamp.f32(float, float, float) nounwind readnone
7
8; FUNC-LABEL: {{^}}clamp_0_1_f32:
9; SI: s_load_dword [[ARG:s[0-9]+]],
10; SI: v_add_f32_e64 [[RESULT:v[0-9]+]], 0, [[ARG]] clamp{{$}}
11; SI: buffer_store_dword [[RESULT]]
12; SI: s_endpgm
13
14; EG: MOV_SAT
15define void @clamp_0_1_f32(float addrspace(1)* %out, float %src) nounwind {
16  %clamp = call float @llvm.AMDGPU.clamp.f32(float %src, float 0.0, float 1.0) nounwind readnone
17  store float %clamp, float addrspace(1)* %out, align 4
18  ret void
19}
20
21; FUNC-LABEL: {{^}}clamp_fabs_0_1_f32:
22; SI: s_load_dword [[ARG:s[0-9]+]],
23; SI: v_add_f32_e64 [[RESULT:v[0-9]+]], 0, |[[ARG]]| clamp{{$}}
24; SI: buffer_store_dword [[RESULT]]
25; SI: s_endpgm
26define void @clamp_fabs_0_1_f32(float addrspace(1)* %out, float %src) nounwind {
27  %src.fabs = call float @llvm.fabs.f32(float %src) nounwind readnone
28  %clamp = call float @llvm.AMDGPU.clamp.f32(float %src.fabs, float 0.0, float 1.0) nounwind readnone
29  store float %clamp, float addrspace(1)* %out, align 4
30  ret void
31}
32
33; FUNC-LABEL: {{^}}clamp_fneg_0_1_f32:
34; SI: s_load_dword [[ARG:s[0-9]+]],
35; SI: v_add_f32_e64 [[RESULT:v[0-9]+]], 0, -[[ARG]] clamp{{$}}
36; SI: buffer_store_dword [[RESULT]]
37; SI: s_endpgm
38define void @clamp_fneg_0_1_f32(float addrspace(1)* %out, float %src) nounwind {
39  %src.fneg = fsub float -0.0, %src
40  %clamp = call float @llvm.AMDGPU.clamp.f32(float %src.fneg, float 0.0, float 1.0) nounwind readnone
41  store float %clamp, float addrspace(1)* %out, align 4
42  ret void
43}
44
45; FUNC-LABEL: {{^}}clamp_fneg_fabs_0_1_f32:
46; SI: s_load_dword [[ARG:s[0-9]+]],
47; SI: v_add_f32_e64 [[RESULT:v[0-9]+]], 0, -|[[ARG]]| clamp{{$}}
48; SI: buffer_store_dword [[RESULT]]
49; SI: s_endpgm
50define void @clamp_fneg_fabs_0_1_f32(float addrspace(1)* %out, float %src) nounwind {
51  %src.fabs = call float @llvm.fabs.f32(float %src) nounwind readnone
52  %src.fneg.fabs = fsub float -0.0, %src.fabs
53  %clamp = call float @llvm.AMDGPU.clamp.f32(float %src.fneg.fabs, float 0.0, float 1.0) nounwind readnone
54  store float %clamp, float addrspace(1)* %out, align 4
55  ret void
56}
57