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
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
; Load argument depends on waitcnt which should be skipped.
define amdgpu_kernel void @call_memory_arg_load(i32 addrspace(3)* %ptr, i32) #0 {
; GCN-LABEL: call_memory_arg_load:
; GCN: ; %bb.0:
; GCN-NEXT: s_load_dword s6, s[6:7], 0x0
; GCN-NEXT: s_add_u32 flat_scratch_lo, s8, s11
; GCN-NEXT: s_addc_u32 flat_scratch_hi, s9, 0
; GCN-NEXT: s_add_u32 s0, s0, s11
; GCN-NEXT: s_addc_u32 s1, s1, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v0, s6
; GCN-NEXT: ds_read_b32 v0, v0
; GCN-NEXT: s_mov_b64 s[6:7], s[4:5]
; GCN-NEXT: s_mov_b32 s32, 0
; GCN-NEXT: s_getpc_b64 s[8:9]
; GCN-NEXT: s_add_u32 s8, s8, func@rel32@lo+4
; GCN-NEXT: s_addc_u32 s9, s9, func@rel32@hi+12
; GCN-NEXT: s_swappc_b64 s[30:31], s[8:9]
; GCN-NEXT: s_endpgm
%vgpr = load volatile i32, i32 addrspace(3)* %ptr
call void @func(i32 %vgpr)
ret void
}
; Memory waitcnt with no register dependence on the call
define amdgpu_kernel void @call_memory_no_dep(i32 addrspace(1)* %ptr, i32) #0 {
; GCN-LABEL: call_memory_no_dep:
; GCN: ; %bb.0:
; GCN-NEXT: s_load_dwordx2 s[6:7], s[6:7], 0x0
; GCN-NEXT: s_add_u32 flat_scratch_lo, s8, s11
; GCN-NEXT: s_addc_u32 flat_scratch_hi, s9, 0
; GCN-NEXT: s_add_u32 s0, s0, s11
; GCN-NEXT: v_mov_b32_e32 v0, 0
; GCN-NEXT: s_addc_u32 s1, s1, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: global_store_dword v0, v0, s[6:7]
; GCN-NEXT: s_mov_b64 s[6:7], s[4:5]
; GCN-NEXT: v_mov_b32_e32 v0, 0
; GCN-NEXT: s_mov_b32 s32, 0
; GCN-NEXT: s_getpc_b64 s[8:9]
; GCN-NEXT: s_add_u32 s8, s8, func@rel32@lo+4
; GCN-NEXT: s_addc_u32 s9, s9, func@rel32@hi+12
; GCN-NEXT: s_swappc_b64 s[30:31], s[8:9]
; GCN-NEXT: s_endpgm
store i32 0, i32 addrspace(1)* %ptr
call void @func(i32 0)
ret void
}
; Should not wait after the call before memory
define amdgpu_kernel void @call_no_wait_after_call(i32 addrspace(1)* %ptr, i32) #0 {
; GCN-LABEL: call_no_wait_after_call:
; GCN: ; %bb.0:
; GCN-NEXT: s_add_u32 flat_scratch_lo, s8, s11
; GCN-NEXT: s_load_dwordx2 s[34:35], s[6:7], 0x0
; GCN-NEXT: s_addc_u32 flat_scratch_hi, s9, 0
; GCN-NEXT: s_add_u32 s0, s0, s11
; GCN-NEXT: s_addc_u32 s1, s1, 0
; GCN-NEXT: s_mov_b64 s[6:7], s[4:5]
; GCN-NEXT: v_mov_b32_e32 v0, 0
; GCN-NEXT: s_mov_b32 s32, 0
; GCN-NEXT: s_getpc_b64 s[8:9]
; GCN-NEXT: s_add_u32 s8, s8, func@rel32@lo+4
; GCN-NEXT: s_addc_u32 s9, s9, func@rel32@hi+12
; GCN-NEXT: v_mov_b32_e32 v40, 0
; GCN-NEXT: s_swappc_b64 s[30:31], s[8:9]
; GCN-NEXT: global_store_dword v40, v40, s[34:35]
; GCN-NEXT: s_endpgm
call void @func(i32 0)
store i32 0, i32 addrspace(1)* %ptr
ret void
}
define amdgpu_kernel void @call_no_wait_after_call_return_val(i32 addrspace(1)* %ptr, i32) #0 {
; GCN-LABEL: call_no_wait_after_call_return_val:
; GCN: ; %bb.0:
; GCN-NEXT: s_add_u32 flat_scratch_lo, s8, s11
; GCN-NEXT: s_load_dwordx2 s[34:35], s[6:7], 0x0
; GCN-NEXT: s_addc_u32 flat_scratch_hi, s9, 0
; GCN-NEXT: s_add_u32 s0, s0, s11
; GCN-NEXT: s_addc_u32 s1, s1, 0
; GCN-NEXT: s_mov_b64 s[6:7], s[4:5]
; GCN-NEXT: v_mov_b32_e32 v0, 0
; GCN-NEXT: s_mov_b32 s32, 0
; GCN-NEXT: s_getpc_b64 s[8:9]
; GCN-NEXT: s_add_u32 s8, s8, func.return@rel32@lo+4
; GCN-NEXT: s_addc_u32 s9, s9, func.return@rel32@hi+12
; GCN-NEXT: v_mov_b32_e32 v40, 0
; GCN-NEXT: s_swappc_b64 s[30:31], s[8:9]
; GCN-NEXT: global_store_dword v40, v0, s[34:35]
; GCN-NEXT: s_endpgm
%rv = call i32 @func.return(i32 0)
store i32 %rv, i32 addrspace(1)* %ptr
ret void
}
; Need to wait for the address dependency
define amdgpu_kernel void @call_got_load(i32 addrspace(1)* %ptr, i32) #0 {
; GCN-LABEL: call_got_load:
; GCN: ; %bb.0:
; GCN-NEXT: s_add_u32 flat_scratch_lo, s8, s11
; GCN-NEXT: s_addc_u32 flat_scratch_hi, s9, 0
; GCN-NEXT: s_add_u32 s0, s0, s11
; GCN-NEXT: s_addc_u32 s1, s1, 0
; GCN-NEXT: s_getpc_b64 s[6:7]
; GCN-NEXT: s_add_u32 s6, s6, got.func@gotpcrel32@lo+4
; GCN-NEXT: s_addc_u32 s7, s7, got.func@gotpcrel32@hi+12
; GCN-NEXT: s_load_dwordx2 s[8:9], s[6:7], 0x0
; GCN-NEXT: s_mov_b64 s[6:7], s[4:5]
; GCN-NEXT: v_mov_b32_e32 v0, 0
; GCN-NEXT: s_mov_b32 s32, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_swappc_b64 s[30:31], s[8:9]
; GCN-NEXT: s_endpgm
call void @got.func(i32 0)
ret void
}
; Need to wait for the address dependency
define void @tailcall_got_load(i32 addrspace(1)* %ptr, i32) #0 {
; GCN-LABEL: tailcall_got_load:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: s_getpc_b64 s[4:5]
; GCN-NEXT: s_add_u32 s4, s4, got.func@gotpcrel32@lo+4
; GCN-NEXT: s_addc_u32 s5, s5, got.func@gotpcrel32@hi+12
; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
; GCN-NEXT: v_mov_b32_e32 v0, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_setpc_b64 s[4:5]
tail call void @got.func(i32 0)
ret void
}
; No need to wait for the load.
define void @tail_call_memory_arg_load(i32 addrspace(3)* %ptr, i32) #0 {
; GCN-LABEL: tail_call_memory_arg_load:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: ds_read_b32 v0, v0
; GCN-NEXT: s_getpc_b64 s[4:5]
; GCN-NEXT: s_add_u32 s4, s4, func@rel32@lo+4
; GCN-NEXT: s_addc_u32 s5, s5, func@rel32@hi+12
; GCN-NEXT: s_setpc_b64 s[4:5]
%vgpr = load volatile i32, i32 addrspace(3)* %ptr
tail call void @func(i32 %vgpr)
ret void
}
declare hidden void @func(i32) #0
declare hidden i32 @func.return(i32) #0
declare void @got.func(i32) #0
attributes #0 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
|
