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
| ; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SI,NOTGFX9 %s
; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,CIPLUS,NOTGFX9 %s
; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,CIPLUS,NOTGFX9 %s
; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,CIPLUS,GFX9 %s
; GCN-LABEL: {{^}}ds_append_lds:
; GCN: s_load_dword [[PTR:s[0-9]+]]
; GCN: s_mov_b32 m0, [[PTR]]
; GCN: ds_append [[RESULT:v[0-9]+]]{{$}}
; GCN-NOT: buffer_wbinvl1
; GCN: {{.*}}store{{.*}} [[RESULT]]
define amdgpu_kernel void @ds_append_lds(i32 addrspace(3)* %lds, i32 addrspace(1)* %out) #0 {
%val = call i32 @llvm.amdgcn.ds.append.p3i32(i32 addrspace(3)* %lds, i1 false)
store i32 %val, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}ds_append_lds_max_offset:
; GCN: s_load_dword [[PTR:s[0-9]+]]
; GCN: s_mov_b32 m0, [[PTR]]
; GCN: ds_append [[RESULT:v[0-9]+]] offset:65532{{$}}
; GCN-NOT: buffer_wbinvl1
; GCN: {{.*}}store{{.*}} [[RESULT]]
define amdgpu_kernel void @ds_append_lds_max_offset(i32 addrspace(3)* %lds, i32 addrspace(1)* %out) #0 {
%gep = getelementptr inbounds i32, i32 addrspace(3)* %lds, i32 16383
%val = call i32 @llvm.amdgcn.ds.append.p3i32(i32 addrspace(3)* %gep, i1 false)
store i32 %val, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}ds_append_no_fold_offset_si:
; GCN: s_load_dword [[PTR:s[0-9]+]]
; SI: s_add_i32 [[PTR]], [[PTR]], 16
; SI: s_mov_b32 m0, [[PTR]]
; SI: ds_append [[RESULT:v[0-9]+]]{{$}}
; CIPLUS: s_mov_b32 m0, [[PTR]]
; CIPLUS: ds_append [[RESULT:v[0-9]+]] offset:16{{$}}
; GCN-NOT: buffer_wbinvl1
; GCN: {{.*}}store{{.*}} [[RESULT]]
define amdgpu_kernel void @ds_append_no_fold_offset_si(i32 addrspace(3)* addrspace(4)* %lds.ptr, i32 addrspace(1)* %out) #0 {
%lds = load i32 addrspace(3)*, i32 addrspace(3)* addrspace(4)* %lds.ptr, align 4
%gep = getelementptr inbounds i32, i32 addrspace(3)* %lds, i32 4
%val = call i32 @llvm.amdgcn.ds.append.p3i32(i32 addrspace(3)* %gep, i1 false)
store i32 %val, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}ds_append_lds_over_max_offset:
; GCN: s_load_dword [[PTR:s[0-9]+]]
; SI: s_bitset1_b32 [[PTR]], 16
; CIPLUS: s_add_i32 [[PTR]], [[PTR]], 0x10000
; GCN: s_mov_b32 m0, [[PTR]]
; GCN: ds_append [[RESULT:v[0-9]+]]{{$}}
; GCN-NOT: buffer_wbinvl1
; GCN: {{.*}}store{{.*}} [[RESULT]]
define amdgpu_kernel void @ds_append_lds_over_max_offset(i32 addrspace(3)* %lds, i32 addrspace(1)* %out) #0 {
%gep = getelementptr inbounds i32, i32 addrspace(3)* %lds, i32 16384
%val = call i32 @llvm.amdgcn.ds.append.p3i32(i32 addrspace(3)* %gep, i1 false)
store i32 %val, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}ds_append_lds_vgpr_addr:
; GCN: v_readfirstlane_b32 [[READLANE:s[0-9]+]], v0
; GCN: s_mov_b32 m0, [[READLANE]]
; GCN: ds_append [[RESULT:v[0-9]+]]{{$}}
; GCN-NOT: buffer_wbinvl1
; GCN: {{.*}}store{{.*}} [[RESULT]]
define void @ds_append_lds_vgpr_addr(i32 addrspace(3)* %lds, i32 addrspace(1)* %out) #0 {
%val = call i32 @llvm.amdgcn.ds.append.p3i32(i32 addrspace(3)* %lds, i1 false)
store i32 %val, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}ds_append_gds:
; GCN: s_load_dword [[PTR:s[0-9]+]]
; GCN: s_mov_b32 m0, [[PTR]]
; GCN: ds_append [[RESULT:v[0-9]+]] gds{{$}}
; GCN-NOT: buffer_wbinvl1
; GCN: {{.*}}store{{.*}} [[RESULT]]
define amdgpu_kernel void @ds_append_gds(i32 addrspace(2)* %gds, i32 addrspace(1)* %out) #0 {
%val = call i32 @llvm.amdgcn.ds.append.p2i32(i32 addrspace(2)* %gds, i1 false)
store i32 %val, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}ds_append_gds_max_offset:
; GCN: s_load_dword [[PTR:s[0-9]+]]
; GCN: s_mov_b32 m0, [[PTR]]
; GCN: ds_append [[RESULT:v[0-9]+]] offset:65532 gds{{$}}
; GCN-NOT: buffer_wbinvl1
; GCN: {{.*}}store{{.*}} [[RESULT]]
define amdgpu_kernel void @ds_append_gds_max_offset(i32 addrspace(2)* %gds, i32 addrspace(1)* %out) #0 {
%gep = getelementptr inbounds i32, i32 addrspace(2)* %gds, i32 16383
%val = call i32 @llvm.amdgcn.ds.append.p2i32(i32 addrspace(2)* %gep, i1 false)
store i32 %val, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}ds_append_gds_over_max_offset:
; GCN-NOT: buffer_wbinvl1
define amdgpu_kernel void @ds_append_gds_over_max_offset(i32 addrspace(2)* %gds, i32 addrspace(1)* %out) #0 {
%gep = getelementptr inbounds i32, i32 addrspace(2)* %gds, i32 16384
%val = call i32 @llvm.amdgcn.ds.append.p2i32(i32 addrspace(2)* %gep, i1 false)
store i32 %val, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}ds_append_lds_m0_restore:
; GCN: s_load_dword [[PTR:s[0-9]+]]
; GCN: s_mov_b32 m0, [[PTR]]
; GCN: ds_append [[RESULT:v[0-9]+]]{{$}}
; GCN-NOT: buffer_wbinvl1
; NOTGFX9: s_mov_b32 m0, -1
; GFX9-NOT: m0
; GCN: _store_dword
; GCN: ds_read_b32
define amdgpu_kernel void @ds_append_lds_m0_restore(i32 addrspace(3)* %lds, i32 addrspace(1)* %out) #0 {
%val0 = call i32 @llvm.amdgcn.ds.append.p3i32(i32 addrspace(3)* %lds, i1 false)
store i32 %val0, i32 addrspace(1)* %out
%val1 = load volatile i32, i32 addrspace(3)* %lds
ret void
}
declare i32 @llvm.amdgcn.ds.append.p3i32(i32 addrspace(3)* nocapture, i1) #1
declare i32 @llvm.amdgcn.ds.append.p2i32(i32 addrspace(2)* nocapture, i1) #1
attributes #0 = { nounwind }
attributes #1 = { argmemonly convergent nounwind }
|