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
|
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GCN,WORKAROUND %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GCN,WORKAROUND %s
; Does not apply to wave64
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+wavefrontsize64 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GCN,NOWORKAROUND %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+wavefrontsize64 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GCN,NOWORKAROUND %s
; Does not apply to gfx1101
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1101 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GCN,NOWORKAROUND %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1101 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GCN,NOWORKAROUND %s
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1102 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GCN,WORKAROUND %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1102 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GCN,WORKAROUND %s
; There aren't any stack objects, but we still enable the
; private_segment_wavefront_offset to get to 16, and the workgroup ID
; is in s14.
; private_segment_buffer + workgroup_id_x = 5, + 11 padding
; GCN-LABEL: {{^}}minimal_kernel_inputs:
; WORKAROUND: v_mov_b32_e32 [[V:v[0-9]+]], s15
; NOWORKAROUND: v_mov_b32_e32 [[V:v[0-9]+]], s0
; GCN-NEXT: global_store_b32 v{{\[[0-9]+:[0-9]+\]}}, [[V]], off
; GCN: .amdhsa_kernel minimal_kernel_inputs
; WORKAROUND: .amdhsa_user_sgpr_count 15
; NOWORKAROUND: .amdhsa_user_sgpr_count 0
; GCN-NEXT: .amdhsa_user_sgpr_dispatch_ptr 0
; GCN-NEXT: .amdhsa_user_sgpr_queue_ptr 0
; GCN-NEXT: .amdhsa_user_sgpr_kernarg_segment_ptr 0
; GCN-NEXT: .amdhsa_user_sgpr_dispatch_id 0
; GCN-NEXT: .amdhsa_user_sgpr_private_segment_size 0
; GCN-NEXT: .amdhsa_wavefront_size32
; GCN-NEXT: .amdhsa_uses_dynamic_stack 0
; GCN-NEXT: .amdhsa_enable_private_segment 0
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_x 1
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_y 0
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_z 0
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_info 0
; GCN-NEXT: .amdhsa_system_vgpr_workitem_id 0
; WORKAROUND: ; COMPUTE_PGM_RSRC2:USER_SGPR: 15
; NOWORKAROUND: ; COMPUTE_PGM_RSRC2:USER_SGPR: 0
define amdgpu_kernel void @minimal_kernel_inputs() {
%id = call i32 @llvm.amdgcn.workgroup.id.x()
store volatile i32 %id, i32 addrspace(1)* undef
ret void
}
; GCN-LABEL: {{^}}minimal_kernel_inputs_with_stack:
; WORKAROUND: v_mov_b32_e32 [[V:v[0-9]+]], s15
; NOWORKAROUND: v_mov_b32_e32 [[V:v[0-9]+]], s0
; GCN: global_store_b32 v{{\[[0-9]+:[0-9]+\]}}, [[V]], off
; GCN: .amdhsa_kernel minimal_kernel_inputs
; WORKAROUND: .amdhsa_user_sgpr_count 15
; NOWORKAROUND: .amdhsa_user_sgpr_count 0
; GCN-NEXT: .amdhsa_user_sgpr_dispatch_ptr 0
; GCN-NEXT: .amdhsa_user_sgpr_queue_ptr 0
; GCN-NEXT: .amdhsa_user_sgpr_kernarg_segment_ptr 0
; GCN-NEXT: .amdhsa_user_sgpr_dispatch_id 0
; GCN-NEXT: .amdhsa_user_sgpr_private_segment_size 0
; GCN-NEXT: .amdhsa_wavefront_size32
; GCN-NEXT: .amdhsa_uses_dynamic_stack 0
; GCN-NEXT: .amdhsa_enable_private_segment 1
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_x 1
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_y 0
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_z 0
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_info 0
; GCN-NEXT: .amdhsa_system_vgpr_workitem_id 0
; WORKAROUND: ; COMPUTE_PGM_RSRC2:USER_SGPR: 15
; NOWORKAROUND: ; COMPUTE_PGM_RSRC2:USER_SGPR: 0
define amdgpu_kernel void @minimal_kernel_inputs_with_stack() {
%alloca = alloca i32, addrspace(5)
%id = call i32 @llvm.amdgcn.workgroup.id.x()
store volatile i32 %id, i32 addrspace(1)* undef
store volatile i32 0, i32 addrspace(5)* %alloca
ret void
}
; GCN-LABEL: {{^}}queue_ptr:
; GCN: global_load_u8 v{{[0-9]+}}, v{{[0-9]+}}, s[0:1]
; WORKAROUND: v_mov_b32_e32 [[V:v[0-9]+]], s15
; NOWORKAROUND: v_mov_b32_e32 [[V:v[0-9]+]], s2
; GCN-NEXT: global_store_b32 v{{\[[0-9]+:[0-9]+\]}}, [[V]], off
; GCN: .amdhsa_kernel queue_ptr
; WORKAROUND: .amdhsa_user_sgpr_count 15
; NOWORKAROUND: .amdhsa_user_sgpr_count 2
; GCN-NEXT: .amdhsa_user_sgpr_dispatch_ptr 0
; GCN-NEXT: .amdhsa_user_sgpr_queue_ptr 1
; GCN-NEXT: .amdhsa_user_sgpr_kernarg_segment_ptr 0
; GCN-NEXT: .amdhsa_user_sgpr_dispatch_id 0
; GCN-NEXT: .amdhsa_user_sgpr_private_segment_size 0
; GCN-NEXT: .amdhsa_wavefront_size32
; GCN-NEXT: .amdhsa_uses_dynamic_stack 0
; GCN-NEXT: .amdhsa_enable_private_segment 0
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_x 1
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_y 0
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_z 0
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_info 0
; GCN-NEXT: .amdhsa_system_vgpr_workitem_id 0
; WORKAROUND: ; COMPUTE_PGM_RSRC2:USER_SGPR: 15
; NOWORKAROUND: ; COMPUTE_PGM_RSRC2:USER_SGPR: 2
define amdgpu_kernel void @queue_ptr() {
%queue.ptr = call noalias i8 addrspace(4)* @llvm.amdgcn.queue.ptr() #0
%load = load volatile i8, i8 addrspace(4)* %queue.ptr
%id = call i32 @llvm.amdgcn.workgroup.id.x()
store volatile i32 %id, i32 addrspace(1)* undef
ret void
}
; GCN-LABEL: {{^}}all_inputs:
; WORKAROUND: v_mov_b32_e32 [[V_X:v[0-9]+]], s13
; WORKAROUND: v_mov_b32_e32 [[V_Y:v[0-9]+]], s14
; WORKAROUND: v_mov_b32_e32 [[V_Z:v[0-9]+]], s15
; NOWORKAROUND: v_mov_b32_e32 [[V_X:v[0-9]+]], s8
; NOWORKAROUND: v_mov_b32_e32 [[V_Y:v[0-9]+]], s9
; NOWORKAROUND: v_mov_b32_e32 [[V_Z:v[0-9]+]], s10
; GCN: global_load_u8 v{{[0-9]+}}, v{{[0-9]+}}, s[0:1]
; GCN: global_load_u8 v{{[0-9]+}}, v{{[0-9]+}}, s[2:3]
; GCN: global_load_u8 v{{[0-9]+}}, v{{[0-9]+}}, s[4:5]
; GCN-DAG: v_mov_b32_e32 v[[DISPATCH_LO:[0-9]+]], s6
; GCN-DAG: v_mov_b32_e32 v[[DISPATCH_HI:[0-9]+]], s7
; GCN: global_store_b32 v{{\[[0-9]+:[0-9]+\]}}, [[V_X]], off
; GCN: global_store_b32 v{{\[[0-9]+:[0-9]+\]}}, [[V_Y]], off
; GCN: global_store_b32 v{{\[[0-9]+:[0-9]+\]}}, [[V_Z]], off
; GCN: global_store_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[DISPATCH_LO]]:[[DISPATCH_HI]]{{\]}}, off
; GCN: .amdhsa_kernel all_inputs
; WORKAROUND: .amdhsa_user_sgpr_count 13
; NOWORKAROUND: .amdhsa_user_sgpr_count 8
; GCN-NEXT: .amdhsa_user_sgpr_dispatch_ptr 1
; GCN-NEXT: .amdhsa_user_sgpr_queue_ptr 1
; GCN-NEXT: .amdhsa_user_sgpr_kernarg_segment_ptr 1
; GCN-NEXT: .amdhsa_user_sgpr_dispatch_id 1
; GCN-NEXT: .amdhsa_user_sgpr_private_segment_size 0
; GCN-NEXT: .amdhsa_wavefront_size32
; GCN-NEXT: .amdhsa_uses_dynamic_stack 0
; GCN-NEXT: .amdhsa_enable_private_segment 1
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_x 1
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_y 1
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_id_z 1
; GCN-NEXT: .amdhsa_system_sgpr_workgroup_info 0
; GCN-NEXT: .amdhsa_system_vgpr_workitem_id 0
; WORKAROUND: ; COMPUTE_PGM_RSRC2:USER_SGPR: 13
; NOWORKAROUND: ; COMPUTE_PGM_RSRC2:USER_SGPR: 8
define amdgpu_kernel void @all_inputs() {
%alloca = alloca i32, addrspace(5)
store volatile i32 0, i32 addrspace(5)* %alloca
%dispatch.ptr = call noalias i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
%load.dispatch = load volatile i8, i8 addrspace(4)* %dispatch.ptr
%queue.ptr = call noalias i8 addrspace(4)* @llvm.amdgcn.queue.ptr()
%load.queue = load volatile i8, i8 addrspace(4)* %queue.ptr
%implicitarg.ptr = call noalias i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr()
%load.implicitarg = load volatile i8, i8 addrspace(4)* %implicitarg.ptr
%id.x = call i32 @llvm.amdgcn.workgroup.id.x()
store volatile i32 %id.x, i32 addrspace(1)* undef
%id.y = call i32 @llvm.amdgcn.workgroup.id.y()
store volatile i32 %id.y, i32 addrspace(1)* undef
%id.z = call i32 @llvm.amdgcn.workgroup.id.z()
store volatile i32 %id.z, i32 addrspace(1)* undef
%dispatch.id = call i64 @llvm.amdgcn.dispatch.id()
store volatile i64 %dispatch.id, i64 addrspace(1)* undef
ret void
}
declare i32 @llvm.amdgcn.workgroup.id.x() #0
declare i32 @llvm.amdgcn.workgroup.id.y() #0
declare i32 @llvm.amdgcn.workgroup.id.z() #0
declare align 4 i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr() #0
declare align 4 i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr() #0
declare align 4 i8 addrspace(4)* @llvm.amdgcn.queue.ptr() #0
declare align 4 i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr() #0
declare i64 @llvm.amdgcn.dispatch.id() #0
attributes #0 = { nounwind readnone speculatable willreturn }
|
