1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
2; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
3
4; GCN-LABEL: {{^}}v_clamp_add_src_f32:
5; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]]
6; GCN-NOT: [[A]]
7; GCN: v_add_f32_e64 v{{[0-9]+}}, [[A]], 1.0 clamp{{$}}
8define amdgpu_kernel void @v_clamp_add_src_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
9  %tid = call i32 @llvm.amdgcn.workitem.id.x()
10  %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
11  %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
12  %a = load float, float addrspace(1)* %gep0
13  %add = fadd float %a, 1.0
14  %max = call float @llvm.maxnum.f32(float %add, float 0.0)
15  %clamp = call float @llvm.minnum.f32(float %max, float 1.0)
16  store float %clamp, float addrspace(1)* %out.gep
17  ret void
18}
19
20; GCN-LABEL: {{^}}v_clamp_multi_use_src_f32:
21; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]]
22; GCN: v_add_f32_e32 [[ADD:v[0-9]+]], 1.0, [[A]]{{$}}
23; GCN: v_max_f32_e64 v{{[0-9]+}}, [[ADD]], [[ADD]] clamp{{$}}
24define amdgpu_kernel void @v_clamp_multi_use_src_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
25  %tid = call i32 @llvm.amdgcn.workitem.id.x()
26  %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
27  %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
28  %a = load float, float addrspace(1)* %gep0
29  %add = fadd float %a, 1.0
30  %max = call float @llvm.maxnum.f32(float %add, float 0.0)
31  %clamp = call float @llvm.minnum.f32(float %max, float 1.0)
32  store float %clamp, float addrspace(1)* %out.gep
33  store volatile float %add, float addrspace(1)* undef
34  ret void
35}
36
37; GCN-LABEL: {{^}}v_clamp_dbg_use_src_f32:
38; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]]
39; GCN-NOT: [[A]]
40; GCN: v_add_f32_e64 v{{[0-9]+}}, [[A]], 1.0 clamp{{$}}
41define amdgpu_kernel void @v_clamp_dbg_use_src_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
42  %tid = call i32 @llvm.amdgcn.workitem.id.x()
43  %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
44  %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
45  %a = load float, float addrspace(1)* %gep0
46  %add = fadd float %a, 1.0
47  call void @llvm.dbg.value(metadata float %add, i64 0, metadata !4, metadata !9), !dbg !10
48  %max = call float @llvm.maxnum.f32(float %add, float 0.0)
49  %clamp = call float @llvm.minnum.f32(float %max, float 1.0)
50  store float %clamp, float addrspace(1)* %out.gep
51  ret void
52}
53
54; GCN-LABEL: {{^}}v_clamp_add_neg_src_f32:
55; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]]
56; GCN: v_floor_f32_e32 [[FLOOR:v[0-9]+]], [[A]]
57; GCN: v_max_f32_e64 v{{[0-9]+}}, -[[FLOOR]], -[[FLOOR]] clamp{{$}}
58define amdgpu_kernel void @v_clamp_add_neg_src_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
59  %tid = call i32 @llvm.amdgcn.workitem.id.x()
60  %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
61  %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
62  %a = load float, float addrspace(1)* %gep0
63  %floor = call float @llvm.floor.f32(float %a)
64  %neg.floor = fsub float -0.0, %floor
65  %max = call float @llvm.maxnum.f32(float %neg.floor, float 0.0)
66  %clamp = call float @llvm.minnum.f32(float %max, float 1.0)
67  store float %clamp, float addrspace(1)* %out.gep
68  ret void
69}
70
71; GCN-LABEL: {{^}}v_non_clamp_max_f32:
72; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]]
73; GCN: v_add_f32_e32 [[ADD:v[0-9]+]], 1.0, [[A]]{{$}}
74; GCN: v_max_f32_e32 v{{[0-9]+}}, 0, [[ADD]]{{$}}
75define amdgpu_kernel void @v_non_clamp_max_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
76  %tid = call i32 @llvm.amdgcn.workitem.id.x()
77  %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
78  %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
79  %a = load float, float addrspace(1)* %gep0
80  %add = fadd float %a, 1.0
81  %max = call float @llvm.maxnum.f32(float %add, float 0.0)
82  store float %max, float addrspace(1)* %out.gep
83  ret void
84}
85
86; GCN-LABEL: {{^}}v_clamp_add_src_f32_denormals:
87; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]]
88; GCN: v_add_f32_e64 [[ADD:v[0-9]+]], [[A]], 1.0 clamp{{$}}
89define amdgpu_kernel void @v_clamp_add_src_f32_denormals(float addrspace(1)* %out, float addrspace(1)* %aptr) #2 {
90  %tid = call i32 @llvm.amdgcn.workitem.id.x()
91  %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
92  %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
93  %a = load float, float addrspace(1)* %gep0
94  %add = fadd float %a, 1.0
95  %max = call float @llvm.maxnum.f32(float %add, float 0.0)
96  %clamp = call float @llvm.minnum.f32(float %max, float 1.0)
97  store float %clamp, float addrspace(1)* %out.gep
98  ret void
99}
100
101; GCN-LABEL: {{^}}v_clamp_add_src_f16_denorm:
102; GCN: {{buffer|flat}}_load_ushort [[A:v[0-9]+]]
103; VI: v_add_f16_e64 [[ADD:v[0-9]+]], [[A]], 1.0 clamp{{$}}
104
105; SI: v_cvt_f32_f16_e32 [[CVT:v[0-9]+]], [[A]]
106; SI: v_add_f32_e64 [[ADD:v[0-9]+]], [[CVT]], 1.0 clamp{{$}}
107; SI: v_cvt_f16_f32_e32 v{{[0-9]+}}, [[ADD]]
108define amdgpu_kernel void @v_clamp_add_src_f16_denorm(half addrspace(1)* %out, half addrspace(1)* %aptr) #0 {
109  %tid = call i32 @llvm.amdgcn.workitem.id.x()
110  %gep0 = getelementptr half, half addrspace(1)* %aptr, i32 %tid
111  %out.gep = getelementptr half, half addrspace(1)* %out, i32 %tid
112  %a = load half, half addrspace(1)* %gep0
113  %add = fadd half %a, 1.0
114  %max = call half @llvm.maxnum.f16(half %add, half 0.0)
115  %clamp = call half @llvm.minnum.f16(half %max, half 1.0)
116  store half %clamp, half addrspace(1)* %out.gep
117  ret void
118}
119
120; GCN-LABEL: {{^}}v_clamp_add_src_f16_no_denormals:
121; GCN: {{buffer|flat}}_load_ushort [[A:v[0-9]+]]
122; VI-NOT: [[A]]
123; VI: v_add_f16_e64 v{{[0-9]+}}, [[A]], 1.0 clamp{{$}}
124
125; SI: v_cvt_f32_f16_e32 [[CVT:v[0-9]+]], [[A]]
126; SI: v_add_f32_e64 [[ADD:v[0-9]+]], [[CVT]], 1.0 clamp{{$}}
127; SI: v_cvt_f16_f32_e32 v{{[0-9]+}}, [[ADD]]
128define amdgpu_kernel void @v_clamp_add_src_f16_no_denormals(half addrspace(1)* %out, half addrspace(1)* %aptr) #3 {
129  %tid = call i32 @llvm.amdgcn.workitem.id.x()
130  %gep0 = getelementptr half, half addrspace(1)* %aptr, i32 %tid
131  %out.gep = getelementptr half, half addrspace(1)* %out, i32 %tid
132  %a = load half, half addrspace(1)* %gep0
133  %add = fadd half %a, 1.0
134  %max = call half @llvm.maxnum.f16(half %add, half 0.0)
135  %clamp = call half @llvm.minnum.f16(half %max, half 1.0)
136  store half %clamp, half addrspace(1)* %out.gep
137  ret void
138}
139
140; GCN-LABEL: {{^}}v_clamp_add_src_v2f32:
141; GCN: {{buffer|flat}}_load_dwordx2 v{{\[}}[[A:[0-9]+]]:[[B:[0-9]+]]{{\]}}
142; GCN-DAG: v_add_f32_e64 v{{[0-9]+}}, v[[A]], 1.0 clamp{{$}}
143; GCN-DAG: v_add_f32_e64 v{{[0-9]+}}, v[[B]], 1.0 clamp{{$}}
144define amdgpu_kernel void @v_clamp_add_src_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %aptr) #0 {
145  %tid = call i32 @llvm.amdgcn.workitem.id.x()
146  %gep0 = getelementptr <2 x float>, <2 x float> addrspace(1)* %aptr, i32 %tid
147  %out.gep = getelementptr <2 x float>, <2 x float> addrspace(1)* %out, i32 %tid
148  %a = load <2 x float>, <2 x float> addrspace(1)* %gep0
149  %add = fadd <2 x float> %a, <float 1.0, float 1.0>
150  %max = call <2 x float> @llvm.maxnum.v2f32(<2 x float> %add, <2 x float> zeroinitializer)
151  %clamp = call <2 x float> @llvm.minnum.v2f32(<2 x float> %max, <2 x float> <float 1.0, float 1.0>)
152  store <2 x float> %clamp, <2 x float> addrspace(1)* %out.gep
153  ret void
154}
155
156; GCN-LABEL: {{^}}v_clamp_add_src_f64:
157; GCN: {{buffer|flat}}_load_dwordx2 [[A:v\[[0-9]+:[0-9]+\]]]
158; GCN: v_add_f64 v{{\[[0-9]+:[0-9]+\]}}, [[A]], 1.0 clamp{{$}}
159define amdgpu_kernel void @v_clamp_add_src_f64(double addrspace(1)* %out, double addrspace(1)* %aptr) #0 {
160  %tid = call i32 @llvm.amdgcn.workitem.id.x()
161  %gep0 = getelementptr double, double addrspace(1)* %aptr, i32 %tid
162  %out.gep = getelementptr double, double addrspace(1)* %out, i32 %tid
163  %a = load double, double addrspace(1)* %gep0
164  %add = fadd double %a, 1.0
165  %max = call double @llvm.maxnum.f64(double %add, double 0.0)
166  %clamp = call double @llvm.minnum.f64(double %max, double 1.0)
167  store double %clamp, double addrspace(1)* %out.gep
168  ret void
169}
170
171declare i32 @llvm.amdgcn.workitem.id.x() #1
172declare float @llvm.fabs.f32(float) #1
173declare float @llvm.floor.f32(float) #1
174declare float @llvm.minnum.f32(float, float) #1
175declare float @llvm.maxnum.f32(float, float) #1
176declare float @llvm.amdgcn.fmed3.f32(float, float, float) #1
177declare double @llvm.fabs.f64(double) #1
178declare double @llvm.minnum.f64(double, double) #1
179declare double @llvm.maxnum.f64(double, double) #1
180declare half @llvm.fabs.f16(half) #1
181declare half @llvm.minnum.f16(half, half) #1
182declare half @llvm.maxnum.f16(half, half) #1
183declare <2 x float> @llvm.minnum.v2f32(<2 x float>, <2 x float>) #1
184declare <2 x float> @llvm.maxnum.v2f32(<2 x float>, <2 x float>) #1
185declare void @llvm.dbg.value(metadata, i64, metadata, metadata) #1
186
187attributes #0 = { nounwind }
188attributes #1 = { nounwind readnone }
189attributes #2 = { nounwind "target-features"="+fp32-denormals" }
190attributes #3 = { nounwind "target-features"="-fp64-fp16-denormals" }
191
192!llvm.dbg.cu = !{!0}
193!llvm.module.flags = !{!2, !3}
194
195!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, isOptimized: true, runtimeVersion: 0, emissionKind: NoDebug)
196!1 = !DIFile(filename: "/tmp/foo.cl", directory: "/dev/null")
197!2 = !{i32 2, !"Dwarf Version", i32 4}
198!3 = !{i32 2, !"Debug Info Version", i32 3}
199!4 = !DILocalVariable(name: "add", arg: 1, scope: !5, file: !1, line: 1)
200!5 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !6, isLocal: false, isDefinition: true, scopeLine: 2, flags: DIFlagPrototyped, isOptimized: true, unit: !0)
201!6 = !DISubroutineType(types: !7)
202!7 = !{null, !8}
203!8 = !DIBasicType(name: "float", size: 32, align: 32)
204!9 = !DIExpression()
205!10 = !DILocation(line: 1, column: 42, scope: !5)
206