1 ; RUN: llc -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI-SAFE -check-prefix=SI -check-prefix=FUNC %s
2 ; RUN: llc -enable-no-nans-fp-math -enable-unsafe-fp-math -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI-NONAN -check-prefix=SI -check-prefix=FUNC %s
3 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
5 ; FIXME: Should replace unsafe-fp-math with no signed zeros.
7 declare i32 @llvm.r600.read.tidig.x() #1
9 ; FUNC-LABEL: @test_fmin_legacy_f32
11 ; SI-SAFE: v_min_legacy_f32_e32
12 ; SI-NONAN: v_min_f32_e32
13 define void @test_fmin_legacy_f32(<4 x float> addrspace(1)* %out, <4 x float> inreg %reg0) #0 {
14 %r0 = extractelement <4 x float> %reg0, i32 0
15 %r1 = extractelement <4 x float> %reg0, i32 1
16 %r2 = fcmp uge float %r0, %r1
17 %r3 = select i1 %r2, float %r1, float %r0
18 %vec = insertelement <4 x float> undef, float %r3, i32 0
19 store <4 x float> %vec, <4 x float> addrspace(1)* %out, align 16
23 ; FUNC-LABEL: @test_fmin_legacy_ule_f32
24 ; SI: buffer_load_dword [[A:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
25 ; SI: buffer_load_dword [[B:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
26 ; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]]
27 ; SI-NONAN: v_min_f32_e32 {{v[0-9]+}}, [[B]], [[A]]
28 define void @test_fmin_legacy_ule_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
29 %tid = call i32 @llvm.r600.read.tidig.x() #1
30 %gep.0 = getelementptr float addrspace(1)* %in, i32 %tid
31 %gep.1 = getelementptr float addrspace(1)* %gep.0, i32 1
33 %a = load float addrspace(1)* %gep.0, align 4
34 %b = load float addrspace(1)* %gep.1, align 4
36 %cmp = fcmp ule float %a, %b
37 %val = select i1 %cmp, float %a, float %b
38 store float %val, float addrspace(1)* %out, align 4
42 ; FUNC-LABEL: @test_fmin_legacy_ole_f32
43 ; SI: buffer_load_dword [[A:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
44 ; SI: buffer_load_dword [[B:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
45 ; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]]
46 ; SI-NONAN: v_min_f32_e32 {{v[0-9]+}}, [[B]], [[A]]
47 define void @test_fmin_legacy_ole_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
48 %tid = call i32 @llvm.r600.read.tidig.x() #1
49 %gep.0 = getelementptr float addrspace(1)* %in, i32 %tid
50 %gep.1 = getelementptr float addrspace(1)* %gep.0, i32 1
52 %a = load float addrspace(1)* %gep.0, align 4
53 %b = load float addrspace(1)* %gep.1, align 4
55 %cmp = fcmp ole float %a, %b
56 %val = select i1 %cmp, float %a, float %b
57 store float %val, float addrspace(1)* %out, align 4
61 ; FUNC-LABEL: @test_fmin_legacy_olt_f32
62 ; SI: buffer_load_dword [[A:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
63 ; SI: buffer_load_dword [[B:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
64 ; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]]
65 ; SI-NONAN: v_min_f32_e32 {{v[0-9]+}}, [[B]], [[A]]
66 define void @test_fmin_legacy_olt_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
67 %tid = call i32 @llvm.r600.read.tidig.x() #1
68 %gep.0 = getelementptr float addrspace(1)* %in, i32 %tid
69 %gep.1 = getelementptr float addrspace(1)* %gep.0, i32 1
71 %a = load float addrspace(1)* %gep.0, align 4
72 %b = load float addrspace(1)* %gep.1, align 4
74 %cmp = fcmp olt float %a, %b
75 %val = select i1 %cmp, float %a, float %b
76 store float %val, float addrspace(1)* %out, align 4
80 ; FUNC-LABEL: @test_fmin_legacy_ult_f32
81 ; SI: buffer_load_dword [[A:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
82 ; SI: buffer_load_dword [[B:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
83 ; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]]
84 ; SI-NONAN: v_min_f32_e32 {{v[0-9]+}}, [[B]], [[A]]
85 define void @test_fmin_legacy_ult_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
86 %tid = call i32 @llvm.r600.read.tidig.x() #1
87 %gep.0 = getelementptr float addrspace(1)* %in, i32 %tid
88 %gep.1 = getelementptr float addrspace(1)* %gep.0, i32 1
90 %a = load float addrspace(1)* %gep.0, align 4
91 %b = load float addrspace(1)* %gep.1, align 4
93 %cmp = fcmp ult float %a, %b
94 %val = select i1 %cmp, float %a, float %b
95 store float %val, float addrspace(1)* %out, align 4
99 ; FUNC-LABEL: @test_fmin_legacy_ole_f32_multi_use
100 ; SI: buffer_load_dword [[A:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
101 ; SI: buffer_load_dword [[B:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
104 ; SI-NEXT: v_cndmask_b32
107 define void @test_fmin_legacy_ole_f32_multi_use(float addrspace(1)* %out0, i1 addrspace(1)* %out1, float addrspace(1)* %in) #0 {
108 %tid = call i32 @llvm.r600.read.tidig.x() #1
109 %gep.0 = getelementptr float addrspace(1)* %in, i32 %tid
110 %gep.1 = getelementptr float addrspace(1)* %gep.0, i32 1
112 %a = load float addrspace(1)* %gep.0, align 4
113 %b = load float addrspace(1)* %gep.1, align 4
115 %cmp = fcmp ole float %a, %b
116 %val0 = select i1 %cmp, float %a, float %b
117 store float %val0, float addrspace(1)* %out0, align 4
118 store i1 %cmp, i1 addrspace(1)* %out1
122 attributes #0 = { nounwind }
123 attributes #1 = { nounwind readnone }