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
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1250 -O3 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1250 -O3 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-GISEL %s
declare i32 @llvm.amdgcn.cluster.load.b32.i32.p1(ptr addrspace(1), i32 %cpol, i32 %mask)
declare <2 x i32> @llvm.amdgcn.cluster.load.b64.v2i32.p1(ptr addrspace(1), i32 %cpol, i32 %mask)
declare <4 x i32> @llvm.amdgcn.cluster.load.b128.v4i32.p1(ptr addrspace(1), i32 %cpol, i32 %mask)
define amdgpu_ps void @cluster_load_b32_vaddr(ptr addrspace(1) %addr, ptr addrspace(1) %use, i32 %mask) {
; GFX1250-LABEL: cluster_load_b32_vaddr:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-NEXT: s_mov_b32 m0, s0
; GFX1250-NEXT: cluster_load_b32 v0, v[0:1], off offset:32 th:TH_LOAD_NT
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b32 v[2:3], v0, off
; GFX1250-NEXT: s_endpgm
entry:
%gep = getelementptr i64, ptr addrspace(1) %addr, i32 4
%val = call i32 @llvm.amdgcn.cluster.load.b32.i32.p1(ptr addrspace(1) %gep, i32 1, i32 %mask)
store i32 %val, ptr addrspace(1) %use
ret void
}
define amdgpu_ps void @cluster_load_b32_vaddr_imm_mask(ptr addrspace(1) %addr, ptr addrspace(1) %use) {
; GFX1250-LABEL: cluster_load_b32_vaddr_imm_mask:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: s_mov_b32 m0, 7
; GFX1250-NEXT: cluster_load_b32 v0, v[0:1], off offset:32 th:TH_LOAD_HT scope:SCOPE_SE
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b32 v[2:3], v0, off
; GFX1250-NEXT: s_endpgm
entry:
%gep = getelementptr i64, ptr addrspace(1) %addr, i32 4
%val = call i32 @llvm.amdgcn.cluster.load.b32.i32.p1(ptr addrspace(1) %gep, i32 10, i32 7)
store i32 %val, ptr addrspace(1) %use
ret void
}
define amdgpu_ps void @cluster_load_b32_saddr(ptr addrspace(1) inreg %addr, ptr addrspace(1) %use, i32 inreg %mask) {
; GFX1250-LABEL: cluster_load_b32_saddr:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: v_mov_b32_e32 v2, 0
; GFX1250-NEXT: s_mov_b32 m0, s2
; GFX1250-NEXT: cluster_load_b32 v2, v2, s[0:1] offset:32 th:TH_LOAD_NT_HT scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b32 v[0:1], v2, off
; GFX1250-NEXT: s_endpgm
entry:
%gep = getelementptr i64, ptr addrspace(1) %addr, i32 4
%val = call i32 @llvm.amdgcn.cluster.load.b32.i32.p1(ptr addrspace(1) %gep, i32 22, i32 %mask)
store i32 %val, ptr addrspace(1) %use
ret void
}
define amdgpu_ps void @cluster_load_monitor_b32_saddr_scale_offset(ptr addrspace(1) inreg %addr, ptr addrspace(1) %use, i32 inreg %mask, i32 %idx) {
; GFX1250-LABEL: cluster_load_monitor_b32_saddr_scale_offset:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: s_mov_b32 m0, s2
; GFX1250-NEXT: cluster_load_b32 v2, v2, s[0:1] scale_offset th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b32 v[0:1], v2, off
; GFX1250-NEXT: s_endpgm
entry:
%idxprom = sext i32 %idx to i64
%gep = getelementptr i32, ptr addrspace(1) %addr, i64 %idxprom
%val = call i32 @llvm.amdgcn.cluster.load.b32.i32.p1(ptr addrspace(1) %gep, i32 27, i32 inreg %mask)
store i32 %val, ptr addrspace(1) %use
ret void
}
define amdgpu_ps void @cluster_load_b64_vaddr(ptr addrspace(1) %addr, ptr addrspace(1) %use, i32 %mask) {
; GFX1250-LABEL: cluster_load_b64_vaddr:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-NEXT: s_mov_b32 m0, s0
; GFX1250-NEXT: cluster_load_b64 v[0:1], v[0:1], off offset:32 th:TH_LOAD_NT
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b64 v[2:3], v[0:1], off
; GFX1250-NEXT: s_endpgm
entry:
%gep = getelementptr i64, ptr addrspace(1) %addr, i32 4
%val = call <2 x i32> @llvm.amdgcn.cluster.load.b64.v2i32.p1(ptr addrspace(1) %gep, i32 1, i32 %mask)
store <2 x i32> %val, ptr addrspace(1) %use
ret void
}
define amdgpu_ps void @cluster_load_b64_vaddr_imm_mask(ptr addrspace(1) %addr, ptr addrspace(1) %use) {
; GFX1250-LABEL: cluster_load_b64_vaddr_imm_mask:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: s_mov_b32 m0, 0x10007
; GFX1250-NEXT: cluster_load_b64 v[0:1], v[0:1], off offset:32 th:TH_LOAD_HT scope:SCOPE_SE
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b64 v[2:3], v[0:1], off
; GFX1250-NEXT: s_endpgm
entry:
%gep = getelementptr i64, ptr addrspace(1) %addr, i32 4
%val = call <2 x i32> @llvm.amdgcn.cluster.load.b64.v2i32.p1(ptr addrspace(1) %gep, i32 10, i32 65543)
store <2 x i32> %val, ptr addrspace(1) %use
ret void
}
define amdgpu_ps void @cluster_load_b64_saddr(ptr addrspace(1) inreg %addr, ptr addrspace(1) %use, i32 inreg %mask) {
; GFX1250-LABEL: cluster_load_b64_saddr:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: v_mov_b32_e32 v2, 0
; GFX1250-NEXT: s_mov_b32 m0, s2
; GFX1250-NEXT: cluster_load_b64 v[2:3], v2, s[0:1] offset:32 th:TH_LOAD_NT_HT scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b64 v[0:1], v[2:3], off
; GFX1250-NEXT: s_endpgm
entry:
%gep = getelementptr i64, ptr addrspace(1) %addr, i32 4
%val = call <2 x i32> @llvm.amdgcn.cluster.load.b64.v2i32.p1(ptr addrspace(1) %gep, i32 22, i32 %mask)
store <2 x i32> %val, ptr addrspace(1) %use
ret void
}
define amdgpu_ps void @cluster_load_monitor_b64_saddr_scale_offset(ptr addrspace(1) inreg %addr, ptr addrspace(1) %use, i32 inreg %mask, i32 %idx) {
; GFX1250-LABEL: cluster_load_monitor_b64_saddr_scale_offset:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: s_mov_b32 m0, s2
; GFX1250-NEXT: cluster_load_b64 v[2:3], v2, s[0:1] scale_offset th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b64 v[0:1], v[2:3], off
; GFX1250-NEXT: s_endpgm
entry:
%idxprom = sext i32 %idx to i64
%gep = getelementptr i64, ptr addrspace(1) %addr, i64 %idxprom
%val = call <2 x i32> @llvm.amdgcn.cluster.load.b64.v2i32.p1(ptr addrspace(1) %gep, i32 27, i32 inreg %mask)
store <2 x i32> %val, ptr addrspace(1) %use
ret void
}
define amdgpu_ps void @cluster_load_b128_vaddr(ptr addrspace(1) %addr, ptr addrspace(1) %use, i32 %mask) {
; GFX1250-LABEL: cluster_load_b128_vaddr:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-NEXT: s_mov_b32 m0, s0
; GFX1250-NEXT: cluster_load_b128 v[4:7], v[0:1], off offset:32 th:TH_LOAD_NT
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b128 v[2:3], v[4:7], off
; GFX1250-NEXT: s_endpgm
entry:
%gep = getelementptr i64, ptr addrspace(1) %addr, i32 4
%val = call <4 x i32> @llvm.amdgcn.cluster.load.b128.v4i32.p1(ptr addrspace(1) %gep, i32 1, i32 %mask)
store <4 x i32> %val, ptr addrspace(1) %use
ret void
}
define amdgpu_ps void @cluster_load_b128_vaddr_imm_mask(ptr addrspace(1) %addr, ptr addrspace(1) %use) {
; GFX1250-LABEL: cluster_load_b128_vaddr_imm_mask:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: s_mov_b32 m0, 15
; GFX1250-NEXT: cluster_load_b128 v[4:7], v[0:1], off offset:32 th:TH_LOAD_HT scope:SCOPE_SE
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b128 v[2:3], v[4:7], off
; GFX1250-NEXT: s_endpgm
entry:
%gep = getelementptr i64, ptr addrspace(1) %addr, i32 4
%val = call <4 x i32> @llvm.amdgcn.cluster.load.b128.v4i32.p1(ptr addrspace(1) %gep, i32 10, i32 15)
store <4 x i32> %val, ptr addrspace(1) %use
ret void
}
define amdgpu_ps void @cluster_load_b128_saddr(ptr addrspace(1) inreg %addr, ptr addrspace(1) %use, i32 inreg %mask) {
; GFX1250-LABEL: cluster_load_b128_saddr:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
; GFX1250-NEXT: v_mov_b32_e32 v2, 0
; GFX1250-NEXT: s_mov_b32 m0, s2
; GFX1250-NEXT: cluster_load_b128 v[2:5], v2, s[0:1] offset:32 th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b128 v[0:1], v[2:5], off
; GFX1250-NEXT: s_endpgm
entry:
%gep = getelementptr i64, ptr addrspace(1) %addr, i32 4
%val = call <4 x i32> @llvm.amdgcn.cluster.load.b128.v4i32.p1(ptr addrspace(1) %gep, i32 27, i32 inreg %mask)
store <4 x i32> %val, ptr addrspace(1) %use
ret void
}
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GFX1250-GISEL: {{.*}}
; GFX1250-SDAG: {{.*}}
|
