1; RUN: llc -march=amdgcn -mcpu=SI -mattr=-fp32-denormals -enable-unsafe-fp-math -verify-machineinstrs < %s | FileCheck -check-prefix=SI-UNSAFE -check-prefix=SI -check-prefix=FUNC %s
2; RUN: llc -march=amdgcn -mcpu=SI -mattr=-fp32-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=SI-SAFE -check-prefix=SI -check-prefix=FUNC %s
3; XUN: llc -march=amdgcn -mcpu=SI -mattr=+fp32-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=SI-SAFE-SPDENORM -check-prefix=SI -check-prefix=FUNC %s
4; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-fp32-denormals -enable-unsafe-fp-math -verify-machineinstrs < %s | FileCheck -check-prefix=SI-UNSAFE -check-prefix=SI -check-prefix=FUNC %s
5; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-fp32-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=SI-SAFE -check-prefix=SI -check-prefix=FUNC %s
6; XUN: llc -march=amdgcn -mcpu=tonga -mattr=+fp32-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=SI-SAFE-SPDENORM -check-prefix=SI -check-prefix=FUNC %s
7
8; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG-SAFE -check-prefix=FUNC %s
9; RUN: llc -march=r600 -mcpu=cayman -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
10
11declare float @llvm.AMDGPU.rcp.f32(float) nounwind readnone
12declare double @llvm.AMDGPU.rcp.f64(double) nounwind readnone
13
14declare float @llvm.sqrt.f32(float) nounwind readnone
15
16; FUNC-LABEL: {{^}}rcp_f32:
17; SI: v_rcp_f32_e32
18; EG: RECIP_IEEE
19define void @rcp_f32(float addrspace(1)* %out, float %src) nounwind {
20  %rcp = call float @llvm.AMDGPU.rcp.f32(float %src) nounwind readnone
21  store float %rcp, float addrspace(1)* %out, align 4
22  ret void
23}
24
25; FIXME: Evergreen only ever does unsafe fp math.
26; FUNC-LABEL: {{^}}rcp_pat_f32:
27
28; SI-SAFE: v_rcp_f32_e32
29; XSI-SAFE-SPDENORM-NOT: v_rcp_f32_e32
30
31; EG: RECIP_IEEE
32
33define void @rcp_pat_f32(float addrspace(1)* %out, float %src) nounwind {
34  %rcp = fdiv float 1.0, %src
35  store float %rcp, float addrspace(1)* %out, align 4
36  ret void
37}
38
39; FUNC-LABEL: {{^}}rsq_rcp_pat_f32:
40; SI-UNSAFE: v_rsq_f32_e32
41; SI-SAFE: v_sqrt_f32_e32
42; SI-SAFE: v_rcp_f32_e32
43
44; EG: RECIPSQRT_IEEE
45define void @rsq_rcp_pat_f32(float addrspace(1)* %out, float %src) nounwind {
46  %sqrt = call float @llvm.sqrt.f32(float %src) nounwind readnone
47  %rcp = call float @llvm.AMDGPU.rcp.f32(float %sqrt) nounwind readnone
48  store float %rcp, float addrspace(1)* %out, align 4
49  ret void
50}
51