1; RUN: llc -march=amdgcn -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 -mattr=-fp32-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=SI-SAFE -check-prefix=SI -check-prefix=FUNC %s
3; XUN: llc -march=amdgcn -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
8declare float @llvm.amdgcn.rcp.f32(float) #0
9declare double @llvm.amdgcn.rcp.f64(double) #0
10
11declare double @llvm.sqrt.f64(double) #0
12declare float @llvm.sqrt.f32(float) #0
13
14
15; FUNC-LABEL: {{^}}rcp_f32:
16; SI: v_rcp_f32_e32
17define void @rcp_f32(float addrspace(1)* %out, float %src) #1 {
18  %rcp = call float @llvm.amdgcn.rcp.f32(float %src) #0
19  store float %rcp, float addrspace(1)* %out, align 4
20  ret void
21}
22
23; FUNC-LABEL: {{^}}rcp_pat_f32:
24
25; SI-SAFE: v_rcp_f32_e32
26; XSI-SAFE-SPDENORM-NOT: v_rcp_f32_e32
27define void @rcp_pat_f32(float addrspace(1)* %out, float %src) #1 {
28  %rcp = fdiv float 1.0, %src
29  store float %rcp, float addrspace(1)* %out, align 4
30  ret void
31}
32
33; FUNC-LABEL: {{^}}rsq_rcp_pat_f32:
34; SI-UNSAFE: v_rsq_f32_e32
35; SI-SAFE: v_sqrt_f32_e32
36; SI-SAFE: v_rcp_f32_e32
37define void @rsq_rcp_pat_f32(float addrspace(1)* %out, float %src) #1 {
38  %sqrt = call float @llvm.sqrt.f32(float %src) #0
39  %rcp = call float @llvm.amdgcn.rcp.f32(float %sqrt) #0
40  store float %rcp, float addrspace(1)* %out, align 4
41  ret void
42}
43
44; FUNC-LABEL: {{^}}rcp_f64:
45; SI: v_rcp_f64_e32
46define void @rcp_f64(double addrspace(1)* %out, double %src) #1 {
47  %rcp = call double @llvm.amdgcn.rcp.f64(double %src) #0
48  store double %rcp, double addrspace(1)* %out, align 8
49  ret void
50}
51
52; FUNC-LABEL: {{^}}rcp_pat_f64:
53; SI: v_rcp_f64_e32
54define void @rcp_pat_f64(double addrspace(1)* %out, double %src) #1 {
55  %rcp = fdiv double 1.0, %src
56  store double %rcp, double addrspace(1)* %out, align 8
57  ret void
58}
59
60; FUNC-LABEL: {{^}}rsq_rcp_pat_f64:
61; SI-UNSAFE: v_rsq_f64_e32
62; SI-SAFE-NOT: v_rsq_f64_e32
63; SI-SAFE: v_sqrt_f64
64; SI-SAFE: v_rcp_f64
65define void @rsq_rcp_pat_f64(double addrspace(1)* %out, double %src) #1 {
66  %sqrt = call double @llvm.sqrt.f64(double %src) #0
67  %rcp = call double @llvm.amdgcn.rcp.f64(double %sqrt) #0
68  store double %rcp, double addrspace(1)* %out, align 8
69  ret void
70}
71
72; FUNC-LABEL: {{^}}rcp_undef_f32:
73; SI-NOT: v_rcp_f32
74define void @rcp_undef_f32(float addrspace(1)* %out) #1 {
75  %rcp = call float @llvm.amdgcn.rcp.f32(float undef) #0
76  store float %rcp, float addrspace(1)* %out, align 4
77  ret void
78}
79
80attributes #0 = { nounwind readnone }
81attributes #1 = { nounwind }
82