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
232
233
234
235
236
237
238
239
240
|
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,SI %s
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI,GFX89 %s
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9,GFX89 %s
declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone speculatable
; GCN-LABEL: {{^}}s_sub_i32:
; GCN: s_load_dwordx4 s[[[#LOAD:]]:{{[0-9]+}}]
; GCN: s_sub_i32 s{{[0-9]+}}, s[[#LOAD + 2]], s[[#LOAD + 3]]
define amdgpu_kernel void @s_sub_i32(ptr addrspace(1) %out, i32 %a, i32 %b) {
%result = sub i32 %a, %b
store i32 %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}s_sub_imm_i32:
; GCN: s_load_dword [[A:s[0-9]+]]
; GCN: s_sub_i32 s{{[0-9]+}}, 0x4d2, [[A]]
define amdgpu_kernel void @s_sub_imm_i32(ptr addrspace(1) %out, i32 %a) {
%result = sub i32 1234, %a
store i32 %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}test_sub_i32:
; SI: v_subrev_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
; GFX9: v_sub_u32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
define amdgpu_kernel void @test_sub_i32(ptr addrspace(1) %out, ptr addrspace(1) %in) {
%b_ptr = getelementptr i32, ptr addrspace(1) %in, i32 1
%a = load i32, ptr addrspace(1) %in
%b = load i32, ptr addrspace(1) %b_ptr
%result = sub i32 %a, %b
store i32 %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}test_sub_imm_i32:
; SI: v_sub_i32_e32 v{{[0-9]+}}, vcc, 0x7b, v{{[0-9]+}}
; GFX9: v_sub_u32_e32 v{{[0-9]+}}, 0x7b, v{{[0-9]+}}
define amdgpu_kernel void @test_sub_imm_i32(ptr addrspace(1) %out, ptr addrspace(1) %in) {
%a = load i32, ptr addrspace(1) %in
%result = sub i32 123, %a
store i32 %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}test_sub_v2i32:
; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
; SI: v_subrev_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
; GFX9: v_sub_u32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; GFX9: v_sub_u32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
define amdgpu_kernel void @test_sub_v2i32(ptr addrspace(1) %out, ptr addrspace(1) %in) {
%b_ptr = getelementptr <2 x i32>, ptr addrspace(1) %in, i32 1
%a = load <2 x i32>, ptr addrspace(1) %in
%b = load <2 x i32>, ptr addrspace(1) %b_ptr
%result = sub <2 x i32> %a, %b
store <2 x i32> %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}test_sub_v4i32:
; SI: v_subrev_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
; GFX9: v_sub_u32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; GFX9: v_sub_u32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; GFX9: v_sub_u32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; GFX9: v_sub_u32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
define amdgpu_kernel void @test_sub_v4i32(ptr addrspace(1) %out, ptr addrspace(1) %in) {
%b_ptr = getelementptr <4 x i32>, ptr addrspace(1) %in, i32 1
%a = load <4 x i32>, ptr addrspace(1) %in
%b = load <4 x i32>, ptr addrspace(1) %b_ptr
%result = sub <4 x i32> %a, %b
store <4 x i32> %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}test_sub_i16:
; SI: v_sub_i32_e32 v{{[0-9]+}}, vcc,
; GFX89: v_sub_u16_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
define amdgpu_kernel void @test_sub_i16(ptr addrspace(1) %out, ptr addrspace(1) %in) {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr i16, ptr addrspace(1) %in, i32 %tid
%b_ptr = getelementptr i16, ptr addrspace(1) %gep, i32 1
%a = load volatile i16, ptr addrspace(1) %gep
%b = load volatile i16, ptr addrspace(1) %b_ptr
%result = sub i16 %a, %b
store i16 %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}test_sub_v2i16:
; VI: v_sub_u16_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; VI: v_sub_u16_sdwa v{{[0-9]+, v[0-9]+, v[0-9]+}}
; GFX9: v_pk_sub_i16
define amdgpu_kernel void @test_sub_v2i16(ptr addrspace(1) %out, ptr addrspace(1) %in) {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <2 x i16>, ptr addrspace(1) %in, i32 %tid
%b_ptr = getelementptr <2 x i16>, ptr addrspace(1) %gep, i16 1
%a = load <2 x i16>, ptr addrspace(1) %gep
%b = load <2 x i16>, ptr addrspace(1) %b_ptr
%result = sub <2 x i16> %a, %b
store <2 x i16> %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}test_sub_v4i16:
; VI: v_sub_u16_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; VI: v_sub_u16_sdwa v{{[0-9]+, v[0-9]+, v[0-9]+}}
; VI: v_sub_u16_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
; VI: v_sub_u16_sdwa v{{[0-9]+, v[0-9]+, v[0-9]+}}
; GFX9: v_pk_sub_i16
; GFX9: v_pk_sub_i16
define amdgpu_kernel void @test_sub_v4i16(ptr addrspace(1) %out, ptr addrspace(1) %in) {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <4 x i16>, ptr addrspace(1) %in, i32 %tid
%b_ptr = getelementptr <4 x i16>, ptr addrspace(1) %gep, i16 1
%a = load <4 x i16>, ptr addrspace(1) %gep
%b = load <4 x i16>, ptr addrspace(1) %b_ptr
%result = sub <4 x i16> %a, %b
store <4 x i16> %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}s_sub_i64:
; GCN: s_sub_u32
; GCN: s_subb_u32
define amdgpu_kernel void @s_sub_i64(ptr addrspace(1) noalias %out, i64 %a, i64 %b) nounwind {
%result = sub i64 %a, %b
store i64 %result, ptr addrspace(1) %out, align 8
ret void
}
; GCN-LABEL: {{^}}v_sub_i64:
; SI: v_sub_i32_e32
; SI: v_subb_u32_e32
; VI: v_sub_u32_e32
; VI: v_subb_u32_e32
; GFX9: v_sub_co_u32_e32
; GFX9: v_subb_co_u32_e32
define amdgpu_kernel void @v_sub_i64(ptr addrspace(1) noalias %out, ptr addrspace(1) noalias %inA, ptr addrspace(1) noalias %inB) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x() readnone
%a_ptr = getelementptr i64, ptr addrspace(1) %inA, i32 %tid
%b_ptr = getelementptr i64, ptr addrspace(1) %inB, i32 %tid
%a = load i64, ptr addrspace(1) %a_ptr
%b = load i64, ptr addrspace(1) %b_ptr
%result = sub i64 %a, %b
store i64 %result, ptr addrspace(1) %out, align 8
ret void
}
; GCN-LABEL: {{^}}v_test_sub_v2i64:
; SI: v_sub_i32_e32 v{{[0-9]+}}, vcc,
; SI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; SI: v_sub_i32_e32 v{{[0-9]+}}, vcc,
; SI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_sub_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_sub_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_sub_co_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_subb_co_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_sub_co_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_subb_co_u32_e32 v{{[0-9]+}}, vcc,
define amdgpu_kernel void @v_test_sub_v2i64(ptr addrspace(1) %out, ptr addrspace(1) noalias %inA, ptr addrspace(1) noalias %inB) {
%tid = call i32 @llvm.amdgcn.workitem.id.x() readnone
%a_ptr = getelementptr <2 x i64>, ptr addrspace(1) %inA, i32 %tid
%b_ptr = getelementptr <2 x i64>, ptr addrspace(1) %inB, i32 %tid
%a = load <2 x i64>, ptr addrspace(1) %a_ptr
%b = load <2 x i64>, ptr addrspace(1) %b_ptr
%result = sub <2 x i64> %a, %b
store <2 x i64> %result, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: {{^}}v_test_sub_v4i64:
; SI: v_sub_i32_e32 v{{[0-9]+}}, vcc,
; SI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; SI: v_sub_i32_e32 v{{[0-9]+}}, vcc,
; SI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; SI: v_sub_i32_e32 v{{[0-9]+}}, vcc,
; SI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; SI: v_sub_i32_e32 v{{[0-9]+}}, vcc,
; SI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_sub_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_sub_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_sub_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_sub_u32_e32 v{{[0-9]+}}, vcc,
; VI: v_subb_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_sub_co_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_subb_co_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_sub_co_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_subb_co_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_sub_co_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_subb_co_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_sub_co_u32_e32 v{{[0-9]+}}, vcc,
; GFX9: v_subb_co_u32_e32 v{{[0-9]+}}, vcc,
define amdgpu_kernel void @v_test_sub_v4i64(ptr addrspace(1) %out, ptr addrspace(1) noalias %inA, ptr addrspace(1) noalias %inB) {
%tid = call i32 @llvm.amdgcn.workitem.id.x() readnone
%a_ptr = getelementptr <4 x i64>, ptr addrspace(1) %inA, i32 %tid
%b_ptr = getelementptr <4 x i64>, ptr addrspace(1) %inB, i32 %tid
%a = load <4 x i64>, ptr addrspace(1) %a_ptr
%b = load <4 x i64>, ptr addrspace(1) %b_ptr
%result = sub <4 x i64> %a, %b
store <4 x i64> %result, ptr addrspace(1) %out
ret void
}
; Make sure the VOP3 form of sub is initially selected. Otherwise pair
; of opies from/to VCC would be necessary
; GCN-LABEL: {{^}}sub_select_vop3:
; SI: v_subrev_i32_e64 v0, s[0:1], s0, v0
; VI: v_subrev_u32_e64 v0, s[0:1], s0, v0
; GFX9: v_subrev_u32_e32 v0, s0, v0
; GCN: ; def vcc
; GCN: ds_write_b32
; GCN: ; use vcc
define amdgpu_ps void @sub_select_vop3(i32 inreg %s, i32 %v) {
%vcc = call i64 asm sideeffect "; def vcc", "={vcc}"()
%sub = sub i32 %v, %s
store i32 %sub, ptr addrspace(3) undef
call void asm sideeffect "; use vcc", "{vcc}"(i64 %vcc)
ret void
}
|
