1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
| ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SI,FUNC %s
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,FUNC %s
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9,FUNC %s
; FUNC-LABEL: {{^}}s_uaddo_i64_zext:
; GCN: s_add_u32
; GCN: s_addc_u32
; GCN: v_cmp_lt_u64_e32 vcc
; EG: ADDC_UINT
; EG: ADDC_UINT
define amdgpu_kernel void @s_uaddo_i64_zext(i64 addrspace(1)* %out, i64 %a, i64 %b) #0 {
%uadd = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
%val = extractvalue { i64, i1 } %uadd, 0
%carry = extractvalue { i64, i1 } %uadd, 1
%ext = zext i1 %carry to i64
%add2 = add i64 %val, %ext
store i64 %add2, i64 addrspace(1)* %out, align 8
ret void
}
; FIXME: Could do scalar
; FUNC-LABEL: {{^}}s_uaddo_i32:
; SI: v_add_i32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
; VI: v_add_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
; GFX9: v_add_co_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
; EG: ADDC_UINT
; EG: ADD_INT
define amdgpu_kernel void @s_uaddo_i32(i32 addrspace(1)* %out, i1 addrspace(1)* %carryout, i32 %a, i32 %b) #0 {
%uadd = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
%val = extractvalue { i32, i1 } %uadd, 0
%carry = extractvalue { i32, i1 } %uadd, 1
store i32 %val, i32 addrspace(1)* %out, align 4
store i1 %carry, i1 addrspace(1)* %carryout
ret void
}
; FUNC-LABEL: {{^}}v_uaddo_i32:
; SI: v_add_i32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
; VI: v_add_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
; GFX9: v_add_co_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
; EG: ADDC_UINT
; EG: ADD_INT
define amdgpu_kernel void @v_uaddo_i32(i32 addrspace(1)* %out, i1 addrspace(1)* %carryout, i32 addrspace(1)* %a.ptr, i32 addrspace(1)* %b.ptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%tid.ext = sext i32 %tid to i64
%a.gep = getelementptr inbounds i32, i32 addrspace(1)* %a.ptr
%b.gep = getelementptr inbounds i32, i32 addrspace(1)* %b.ptr
%a = load i32, i32 addrspace(1)* %a.gep, align 4
%b = load i32, i32 addrspace(1)* %b.gep, align 4
%uadd = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
%val = extractvalue { i32, i1 } %uadd, 0
%carry = extractvalue { i32, i1 } %uadd, 1
store i32 %val, i32 addrspace(1)* %out, align 4
store i1 %carry, i1 addrspace(1)* %carryout
ret void
}
; FUNC-LABEL: {{^}}v_uaddo_i32_novcc:
; SI: v_add_i32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
; VI: v_add_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
; GFX9: v_add_co_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
; EG: ADDC_UINT
; EG: ADD_INT
define amdgpu_kernel void @v_uaddo_i32_novcc(i32 addrspace(1)* %out, i1 addrspace(1)* %carryout, i32 addrspace(1)* %a.ptr, i32 addrspace(1)* %b.ptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%tid.ext = sext i32 %tid to i64
%a.gep = getelementptr inbounds i32, i32 addrspace(1)* %a.ptr
%b.gep = getelementptr inbounds i32, i32 addrspace(1)* %b.ptr
%a = load i32, i32 addrspace(1)* %a.gep, align 4
%b = load i32, i32 addrspace(1)* %b.gep, align 4
%uadd = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
%val = extractvalue { i32, i1 } %uadd, 0
%carry = extractvalue { i32, i1 } %uadd, 1
store volatile i32 %val, i32 addrspace(1)* %out, align 4
call void asm sideeffect "", "~{vcc}"() #0
store volatile i1 %carry, i1 addrspace(1)* %carryout
ret void
}
; FUNC-LABEL: {{^}}s_uaddo_i64:
; GCN: s_add_u32
; GCN: s_addc_u32
; EG: ADDC_UINT
; EG: ADD_INT
define amdgpu_kernel void @s_uaddo_i64(i64 addrspace(1)* %out, i1 addrspace(1)* %carryout, i64 %a, i64 %b) #0 {
%uadd = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
%val = extractvalue { i64, i1 } %uadd, 0
%carry = extractvalue { i64, i1 } %uadd, 1
store i64 %val, i64 addrspace(1)* %out, align 8
store i1 %carry, i1 addrspace(1)* %carryout
ret void
}
; FUNC-LABEL: {{^}}v_uaddo_i64:
; SI: v_add_i32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
; SI: v_addc_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_add_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
; VI: v_addc_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_add_co_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
; GFX9: v_addc_co_u32_e32 v{{[0-9]+}}, vcc,
; EG: ADDC_UINT
; EG: ADD_INT
define amdgpu_kernel void @v_uaddo_i64(i64 addrspace(1)* %out, i1 addrspace(1)* %carryout, i64 addrspace(1)* %a.ptr, i64 addrspace(1)* %b.ptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%tid.ext = sext i32 %tid to i64
%a.gep = getelementptr inbounds i64, i64 addrspace(1)* %a.ptr
%b.gep = getelementptr inbounds i64, i64 addrspace(1)* %b.ptr
%a = load i64, i64 addrspace(1)* %a.gep
%b = load i64, i64 addrspace(1)* %b.gep
%uadd = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
%val = extractvalue { i64, i1 } %uadd, 0
%carry = extractvalue { i64, i1 } %uadd, 1
store i64 %val, i64 addrspace(1)* %out
store i1 %carry, i1 addrspace(1)* %carryout
ret void
}
; FUNC-LABEL: {{^}}v_uaddo_i16:
; VI: v_add_u16_e32
; VI: v_cmp_lt_u16_e32
; GFX9: v_add_u16_e32
; GFX9: v_cmp_lt_u16_e32
define amdgpu_kernel void @v_uaddo_i16(i16 addrspace(1)* %out, i1 addrspace(1)* %carryout, i16 addrspace(1)* %a.ptr, i16 addrspace(1)* %b.ptr) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%tid.ext = sext i32 %tid to i64
%a.gep = getelementptr inbounds i16, i16 addrspace(1)* %a.ptr
%b.gep = getelementptr inbounds i16, i16 addrspace(1)* %b.ptr
%a = load i16, i16 addrspace(1)* %a.gep
%b = load i16, i16 addrspace(1)* %b.gep
%uadd = call { i16, i1 } @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
%val = extractvalue { i16, i1 } %uadd, 0
%carry = extractvalue { i16, i1 } %uadd, 1
store i16 %val, i16 addrspace(1)* %out
store i1 %carry, i1 addrspace(1)* %carryout
ret void
}
; FUNC-LABEL: {{^}}v_uaddo_v2i32:
; SICIVI: v_cmp_lt_i32
; SICIVI: v_cmp_lt_i32
; SICIVI: v_add_{{[iu]}}32
; SICIVI: v_cmp_lt_i32
; SICIVI: v_cmp_lt_i32
; SICIVI: v_add_{{[iu]}}32
define amdgpu_kernel void @v_uaddo_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %carryout, <2 x i32> addrspace(1)* %aptr, <2 x i32> addrspace(1)* %bptr) nounwind {
%a = load <2 x i32>, <2 x i32> addrspace(1)* %aptr, align 4
%b = load <2 x i32>, <2 x i32> addrspace(1)* %bptr, align 4
%sadd = call { <2 x i32>, <2 x i1> } @llvm.uadd.with.overflow.v2i32(<2 x i32> %a, <2 x i32> %b) nounwind
%val = extractvalue { <2 x i32>, <2 x i1> } %sadd, 0
%carry = extractvalue { <2 x i32>, <2 x i1> } %sadd, 1
store <2 x i32> %val, <2 x i32> addrspace(1)* %out, align 4
%carry.ext = zext <2 x i1> %carry to <2 x i32>
store <2 x i32> %carry.ext, <2 x i32> addrspace(1)* %carryout
ret void
}
; FUNC-LABEL: {{^}}s_uaddo_clamp_bit:
; GCN: v_add_{{i|u|co_u}}32_e32
; GCN: s_endpgm
define amdgpu_kernel void @s_uaddo_clamp_bit(i32 addrspace(1)* %out, i1 addrspace(1)* %carryout, i32 %a, i32 %b) #0 {
entry:
%uadd = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
%val = extractvalue { i32, i1 } %uadd, 0
%carry = extractvalue { i32, i1 } %uadd, 1
%c2 = icmp eq i1 %carry, false
%cc = icmp eq i32 %a, %b
br i1 %cc, label %exit, label %if
if:
br label %exit
exit:
%cout = phi i1 [false, %entry], [%c2, %if]
store i32 %val, i32 addrspace(1)* %out, align 4
store i1 %cout, i1 addrspace(1)* %carryout
ret void
}
; FUNC-LABEL: {{^}}v_uaddo_clamp_bit:
; GCN: v_add_{{i|u|co_u}}32_e64
; GCN: s_endpgm
define amdgpu_kernel void @v_uaddo_clamp_bit(i32 addrspace(1)* %out, i1 addrspace(1)* %carryout, i32 addrspace(1)* %a.ptr, i32 addrspace(1)* %b.ptr) #0 {
entry:
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%tid.ext = sext i32 %tid to i64
%a.gep = getelementptr inbounds i32, i32 addrspace(1)* %a.ptr
%b.gep = getelementptr inbounds i32, i32 addrspace(1)* %b.ptr
%a = load i32, i32 addrspace(1)* %a.gep
%b = load i32, i32 addrspace(1)* %b.gep
%uadd = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
%val = extractvalue { i32, i1 } %uadd, 0
%carry = extractvalue { i32, i1 } %uadd, 1
%c2 = icmp eq i1 %carry, false
%cc = icmp eq i32 %a, %b
br i1 %cc, label %exit, label %if
if:
br label %exit
exit:
%cout = phi i1 [false, %entry], [%c2, %if]
store i32 %val, i32 addrspace(1)* %out, align 4
store i1 %cout, i1 addrspace(1)* %carryout
ret void
}
declare i32 @llvm.amdgcn.workitem.id.x() #1
declare { i16, i1 } @llvm.uadd.with.overflow.i16(i16, i16) #1
declare { i32, i1 } @llvm.uadd.with.overflow.i32(i32, i32) #1
declare { i64, i1 } @llvm.uadd.with.overflow.i64(i64, i64) #1
declare { <2 x i32>, <2 x i1> } @llvm.uadd.with.overflow.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }
|